COMMUNITY MULTISCALE AIR QUALITY (CMAQ) MODELING OF THE
ATMOSPHERIC QUALITY AND POLLUTANT DEPOSITION OVER THE
TERRACE-KITIMAT VALLEY OF NORTHWESTERN BRITISH
COLUMBIA, CANADA
by
Chibuike Onwukwe
B. Tech., Federal University of Technology Owerri, 2005
M. Sc., École Centrale de Nantes, 2015

DISSERTATION SUBMITTED IN PARTIAL FULFILLMENT OF
THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY
IN
NATURAL RESOURCES AND ENVIRONMENTAL STUDIES

UNIVERSITY OF NORTHERN BRITISH COLUMBIA
September 2020
© Chibuike Onwukwe, 2020

Abstract
Tracking the effects of air pollution from industries is important for
developing

management

strategies

under

changing

emissions.

However,

computational tools for air pollution assessment often do not elucidate modeling
uncertainty, making it difficult for environmental policy-makers to know how much
confidence to put in model results, which also hampers aspects that may need
improving. This study examined how the WRF-SMOKE-CMAQ modeling system
with various planetary boundary-layer (PBL) schemes and atmospheric datasets
mimics the local meteorology, air quality and acidic deposition at 1 km horizontal
resolution over the industrializing Terrace-Kitimat Valley of northwestern British
Columbia. Quantitative and qualitative correspondence of model outputs with
observational data varied with station location, the nature of pollutant emissions, and
quantity of chemical species. Valid model outputs were used to delineate present
compliance with objectives on ambient fine particulate matter, and baseline
exceedance of critical loads of sulfur and nitrogen deposition for the forest ecosystem.
Spatial impacts of anticipated industrial emissions on the environment were also
assessed. An additional 15 tonnes day-1 permissible SO2 emission from an aluminum
smelter in Kitimat was projected to result in 50–88 % increase in aerial exceedance of
the limit for protection of lichen, and 37–67 % increase in spatial exceedance of
acidic deposition to soils. Cumulatively, 16–18 km2 of plant habitat, and 10–11 km2
of soil in an area contiguous with the smelter site will likely be damaged by its SO2
emission under the latest regulation. Should two Liquefied Natural Gas projects
commence operations, cumulative NOx concentrations are expected to remain below
harmful levels, while pre-existing areal exceedance of nitrogen deposition will barely
increase (0–1 km2). An additional 4 km2 area will be exposed to SO2 concentrations

ii

that are directly harmful to vegetation, while 13–14 km2 total area with an average of
29.7–35.0 kg ha-1 yr-1 excess sulfur deposition was estimated. These projections
assumed all future emissions of NOx, SO2 and other air pollutants will be from
elevated point sources.

iii

Preface
This dissertation contains the original research and analyses conducted by the author,
Chibuike Onwukwe, under the guidance of Peter Jackson. It contains many figures, some
of which are maps. All maps were created by the author using various software such as
NCL (http://www.ncl.ucar.edu/), google maps javascript (https://developers.google.com/
maps/documentation/javascript/tutorial?), QGIS (http://www.qgis.org/en/site/), and
map viewer of the Integrated Land and Resource Registry, British Columbia Ministry of
Forests, Lands, Natural Resource Operations and Rural Development.
Chapter 2 of this dissertation is published in Volume 59, Issue 8 of J. Appl. Meteorol. Climatol., Chapter 3 is published in Volume 233 of Atmos. Environ., Chapter 4 is in Press at
the J. Air Waste Manage. Assoc., while Chapter 5 is published in Atmos.Pol. Res.

iv

Table of Contents
Page
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

ii

Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

iv

List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

x

List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii
Acronyms, Notations & Units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xvii
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xx
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1

1.1 Geo-ecological imperatives for atmospheric pollution research: an overview

1

1.2 Motivation for present study . . . . . . . . . . . . . . . . . . . . . . . . . .

2

1.3 Study area background . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

4

1.4 Research questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

7

1.5 Research Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2 Meteorological downscaling with WRF model version 4.0 and comparative
evaluation of planetary boundary layer schemes over a complex coastal
airshed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.2.1 Description of the WRF model and PBL schemes . . . . . . . . . . 16
2.2.2 Domain configuration, model initialization and settings . . . . . . 19
2.2.3 Observational data . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.2.4 Statistical evaluation indices . . . . . . . . . . . . . . . . . . . . . . 22
2.3 Results and analyses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

v

2.3.1 Synoptic and mesoscale representations . . . . . . . . . . . . . . . 23
2.3.2 Performance evaluations for surface variables . . . . . . . . . . . . 27
2.3.3 Spatio-differential examinations of surface meteorological fields . 34
2.4 Discussion and Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3 Evaluation of CMAQ modeling sensitivity to planetary boundary layer parameterizations for gaseous and particulate pollutants over a fjord valley . 50
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.2 Methods and data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.2.1 PBL schemes and experiments set-up . . . . . . . . . . . . . . . . . 53
3.2.2 Air quality monitoring data . . . . . . . . . . . . . . . . . . . . . . 56
3.2.3 Statistical evaluation indices . . . . . . . . . . . . . . . . . . . . . . 57
3.3 Results and analyses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.3.1 Ambient air quality . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.3.2 Model performance evaluations . . . . . . . . . . . . . . . . . . . . 63
3.3.3 Annual evaluations and spatial concentration differences . . . . . 73
3.4 Discussion and conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
4 Gridded bias correction of modeled PM 2.5 for exposure assessment and
estimation of background concentrations over a coastal valley in northwestern British Columbia, Canada . . . . . . . . . . . . . . . . . . . . . . . . . 83
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
4.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
4.2.1 Model simulations and observations data . . . . . . . . . . . . . . 86
4.2.2 Bias correction formulations . . . . . . . . . . . . . . . . . . . . . . 88
4.2.3 Evaluation measures . . . . . . . . . . . . . . . . . . . . . . . . . . 88
4.3 Results and analyses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
4.3.1 Accuracy of bias corrected outputs . . . . . . . . . . . . . . . . . . 90
4.3.2 Evaluations for fitness with compliance metrics . . . . . . . . . . . 92
4.3.3 Estimation of background concentrations . . . . . . . . . . . . . . 96
4.4 Discussion and conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
5 Acid wet-deposition modeling sensitivity to WRF-CMAQ planetary boundary layer schemes and exceedance of critical loads over a coastal mountain
valley area of northwestern British Columbia, Canada . . . . . . . . . . . . . 105

vi

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
5.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
5.2.1 Deposition simulations . . . . . . . . . . . . . . . . . . . . . . . . . 110
5.2.2 Measurement data and performance measures . . . . . . . . . . . 111
5.3 Results and analyses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
5.3.1 Performance evaluation for wet deposition of acidifying species . 114
5.3.2 Comparison of total nitrogen and sulfur deposition among PBL
schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
5.3.3 2017 critical-load exceedances for forest ecosystems . . . . . . . . 124
5.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
5.5 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
6 Intercomparison of atmospheric datasets and PBL schemes for precipitation downscaling over a coastal mountain valley of northern British Columbia,
Canada . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
6.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
6.2.1 Description of datasets . . . . . . . . . . . . . . . . . . . . . . . . . 138
6.2.2 Modeling configurations and experiments . . . . . . . . . . . . . . 139
6.2.3 Observational data and evaluation protocol . . . . . . . . . . . . . 141
6.3 Results and analyses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
6.3.1 Quantitative biases for datasets and PBL schemes . . . . . . . . . . 143
6.3.2 Spatio-temporal verifications . . . . . . . . . . . . . . . . . . . . . . 147
6.3.3 Predictive evaluations versus summary distributions . . . . . . . 151
6.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
6.5 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
7 Modeling terrestrial ecosystems exposure to incremental smelter sulfur
dioxide emissions and deposition in a complex coastal valley airshed . . . . 162
7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
7.2 Modeling framework and procedure . . . . . . . . . . . . . . . . . . . . . 166
7.3 Results and analyses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
7.3.1 Baseline SO 2 levels and changes . . . . . . . . . . . . . . . . . . . . 168
7.3.2 Critical level exceedances and mapping . . . . . . . . . . . . . . . 170
7.3.3 Critical load of acidity exceedances and mapping . . . . . . . . . . 172
vii

7.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
7.5 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
8 Quantifying incremental and cumulative terrestrial ecosystems impacts of
NOx and SO2 emissions from LNG operations in the Terrace-Kitimat valley of northwestern British Columbia . . . . . . . . . . . . . . . . . . . . . . . 181
8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
8.2 LNG emissions and numerical modeling set-up . . . . . . . . . . . . . . . 185
8.3 Results and analyses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
8.3.1 Relative changes in ambient NO x and SO2 and exceedance of critical levels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
8.3.2 Relative changes in nitrogen and sulfur deposition and exceedance
of critical loads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
8.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
8.5 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
9 Summary and Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
9.0.1 Recapitulation of research findings . . . . . . . . . . . . . . . . . . 203
9.0.2 Significance of study findings . . . . . . . . . . . . . . . . . . . . . 208
9.0.3 Directions for future research . . . . . . . . . . . . . . . . . . . . . 210
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
Appendix
A Selection of meteorological year for model simulations . . . . . . . . . . . . 243
B WRF physics settings and model heights . . . . . . . . . . . . . . . . . . . . . 246
C Formulae for statistical measures . . . . . . . . . . . . . . . . . . . . . . . . . 248
D Verification of atmospheric sounding . . . . . . . . . . . . . . . . . . . . . . . 249
E Summer and winter days comparisons of air temperature and wind speed
to seasonal averages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
F Performance statistics for simulation of surface meteorological variables . . 252
G Centreline profile profiles of key WRF output variables . . . . . . . . . . . . 257

viii

H Formulae for indices of precipitation prediction . . . . . . . . . . . . . . . . 259
I

The implication of sulfur exceedance on vegetated land . . . . . . . . . . . . 260

ix

List of Tables
2.1

Properties of WRF domains with ERA5 . . . . . . . . . . . . . . . . . . . . . 21

2.2

Stations for verification of WRF-ERA5 downscaling . . . . . . . . . . . . . . 22

2.3

Performance benchmarks for surface meteorological variables . . . . . . . . 23

3.1

WRF-SMOKE-CMAQ domains attributes . . . . . . . . . . . . . . . . . . . . 56

3.2

Air quality stations having valid data in 2017 . . . . . . . . . . . . . . . . . . 57

3.3

Benchmarks for air quality modeling . . . . . . . . . . . . . . . . . . . . . . . 58

3.4

Averages of observed pollutant concentrations in peak seasons . . . . . . . . 59

3.5

Modeling performance: SO 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64

3.6

Modeling performance: PM2.5 . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

3.7

Modeling performance: NO2

3.8

Modeled annual peak concentrations of air pollutants . . . . . . . . . . . . . 76

4.1

PM2.5 monitoring data completeness . . . . . . . . . . . . . . . . . . . . . . . 87

4.2

Statistical evaluation of bias-corrections compared to original model output

4.3

Accuracy of bias corrected model outputs for compliance indicators . . . . . 93

4.4

Tiers of air quality management based on ambient PM2.5 . . . . . . . . . . . 94

. . . . . . . . . . . . . . . . . . . . . . . . . . . 71

x

91

5.1

Model performance statistics for wet deposition of ammonium, nitrate and
sulfate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118

5.2

Percentage contribution of wet deposition in 2017 to S and N annual deposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124

5.3

Exceedances of critical loads of acidity (CLA) and nutrient nitrogen . . . . . 127

6.1

Nesting with NAM_ANL versus ERA5/ NARR . . . . . . . . . . . . . . . . 140

6.2

Observational data locations for evaluation of precipitation simulations . . . 142

6.3

Fitness measures for precipitation simulations . . . . . . . . . . . . . . . . . 143

6.4

Percent bias for 2017 precipitation simulation at station locations . . . . . . . 145

6.5

Contributions of snow water equivalent to total precipitation in observation and simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148

6.6

Predictive scores of daily precipitation for all events ( ≥ 0.2 mm) and events

≥ mean amounts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
7.1

Meteorological and emissions domain sizes and attributes for SMOKE-CMAQ
modeling with ERA5 and NAM_ANL data . . . . . . . . . . . . . . . . . . . 168

7.2

Estimates of aerial SO2 exceedance of critical levels of vegetation and lichen
exposures due to smelter emissions changes . . . . . . . . . . . . . . . . . . . 173

7.3

Estimates of sulfur deposition exceedance from aluminum smelter’s SO2
emissions in Kitimat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175

8.1

Estimated exceedances of critical level (3.6 ppb) of SO2 lichen exposure
with, and without LNG emissions

8.2

. . . . . . . . . . . . . . . . . . . . . . . . 192

Estimated exceedances of critical load of soil nutrient nitrogen with, and
without LNG emissions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194

xi

8.3

Estimated exceedances of critical load of sulfur with, and without LNG
emissions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197

A.1 Ranking of differences for meteorological year selection . . . . . . . . . . . . 244
B.1 Settings for WRF physics/dynamics . . . . . . . . . . . . . . . . . . . . . . . 246
B.2 WRF model layer heights . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
F.1

Performance statistics of WRF simulation of air temperature . . . . . . . . . 253

F.2

Performance statistics of WRF simulation of specific humidity . . . . . . . . 254

F.3

Performance statistics of WRF simulation of wind direction . . . . . . . . . . 255

F.4

Performance statistics of WRF simulation of wind speed

. . . . . . . . . . . 256

H.1 Contingency table for precipitation events forecast . . . . . . . . . . . . . . . 259

xii

List of Figures
1.1

Description of study area . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

6

1.2

Surficial geology and landcover of the TKV area . . . . . . . . . . . . . . . .

7

1.3

Land use and development in the TKV . . . . . . . . . . . . . . . . . . . . . .

8

2.1

WRF nesting with ERA5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

2.2

Simulated summertime meteorological fields . . . . . . . . . . . . . . . . . . 26

2.3

Simulated wintertime meteorological fields . . . . . . . . . . . . . . . . . . . 28

2.4

Diurnal air temperature bias with ERA5 . . . . . . . . . . . . . . . . . . . . . 29

2.5

Diurnal wind speed bias with ERA5 . . . . . . . . . . . . . . . . . . . . . . . 33

2.6

Spatial differences amongst PBL schemes for meteorological variables: nighttime . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

2.7

Spatial differences amongst PBL schemes for meteorological variables: daytime . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

2.8

Vertical water vapor profiles for simulations with various PBL schemes . . . 44

2.9

Vertical profiles of potential temperature for simulations with various PBL
schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

2.10 Effect of altenative surface-layer schemes on the MYNN3 PBL scheme . . . 49
3.1

WRF-SMOKE-CMAQ domains for ERA5 simulations . . . . . . . . . . . . . 55

xiii

3.2

Observed meteorology relevant to pollutant dispersion . . . . . . . . . . . . 61

3.3

Wind dependency of air pollutants . . . . . . . . . . . . . . . . . . . . . . . . 62

3.4

SO2 modeling evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66

3.5

PM2.5 modeling evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

3.6

NO2 modeling evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

3.7

Pollutant modeling evaluation of concentration quantiles . . . . . . . . . . . 75

3.8

Spatial distribution of modeled annual concentrations of air pollutants . . . 77

4.1

PM2.5 bias correction formulae and quantile plots of modeled and observed
PM2.5 concentrations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89

4.2

PM2.5 evaluation for categorical correspondence . . . . . . . . . . . . . . . . 95

4.3

Spatial plot of bias-corrected annual PM2.5 concentrations . . . . . . . . . . . 97

4.4

Background PM2.5 estimation from bias-corrected centreline concentrations

4.5

Classification of exposure to outdoor PM2.5 in 2017 . . . . . . . . . . . . . . . 102

5.1

Nesting set-up for WRF-CMAQ deposition modeling . . . . . . . . . . . . . 112

5.2

Weekly time series of observed wet deposition . . . . . . . . . . . . . . . . . 115

5.3

Spatial distribution of annual total nitrogen and sulfur deposition . . . . . . 120

5.4

Pairwise differences in annual total nitogen and sulfur deposition between

98

PBL schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
5.5

Domain-wide contribution of wet deposition as a ratio of total deposition . 123

5.6

Exceedances of critical loads of acidity and nutrient nitrogen . . . . . . . . . 126

6.1

WRF nesting with NAM_ANL versus ERA5/ NARR

6.2

Distribution of monthly precipitation in 2017 . . . . . . . . . . . . . . . . . . 145

xiv

. . . . . . . . . . . . . 141

6.3

Spatial plots of dataset-normalized annual precipitation estimates . . . . . . 146

6.4

Spatial plots of PBL schemes-normalized annual precipitation estimates . . 147

6.5

Monthly series of observations (OBS) at gauge stations, and simulations
with the MYJ PBL scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149

6.6

Pairwise stations-normalized 2017 precipitation . . . . . . . . . . . . . . . . 150

6.7

Pairwise station-normalized 2017 precipitation . . . . . . . . . . . . . . . . . 152

6.8

Categorical evaluation of simulated precipitation amounts by various datasets154

6.9

Cumulative mean distributions of simulated and observed daily precipitation155

7.1

SO2 concentrations distribution and relative increase . . . . . . . . . . . . . . 170

7.2

SO2 spatial exceedance for lichen in Kitimat . . . . . . . . . . . . . . . . . . . 172

7.3

Modeled sulfur deposition exceedance from aluminum smelter’s SO2 emissions in Kitimat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174

7.4

Annual mean concentrations from maximum SO2 emissions of 42 tonnes
day−1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177

8.1

Locations of two LNG projects in the Kitimat area . . . . . . . . . . . . . . . 187

8.2

Annual emission estimates of major air pollutants from proposed LNG
projects in Kitimat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188

8.3

NO x and SO2 concentration changes relative to levels without contributions from the LNG industry . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190

8.4

Exceedance of critical level (3.6 ppb) of lichen exposure to SO2 with, and
without contributions from LNG industry emissions . . . . . . . . . . . . . . 191

8.5

Nitrogen and sulfur deposition changes relative to the loads without contributions from LNG industry emissions . . . . . . . . . . . . . . . . . . . . . 193

xv

8.6

Modeled spatial exceedance of critical load of acidity with, and without
contributions from LNG industry . . . . . . . . . . . . . . . . . . . . . . . . . 196

8.7

Along-valley NO x and SO2 concentrations from cumulative emissions . . . 198

A.1 Observed wind speed, precipitation and air temperature in 2017 versus the
averages for 2006–2018 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
D.1 Verification of regional atmospheric sounding . . . . . . . . . . . . . . . . . . 249
E.1 Winterday and summerday comparisons to observed seasonal averages of
air temperature and wind . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
G.1 Valley centreline profiles of WRF diagnosed meteorological variables . . . . 258

xvi

Acronyms, Notations & Units
Acronyms
ACM2

Asymetric Convective Model version 2

BC

British Columbia

BCMOECS

BC Ministry of Environment and Climate Change Strategy

CCME

Canadian Council of Ministers of the Environment

CMAQ

Community Multiscale Air Quality

ECCC

Environment and Climate Change Canada

ECMWF

European Centre for Medium-range Weather Forecasts

ERA5

ECMWF fifth major global reanalyses

FAC2

Modeled values within a factor of 2 of observations

LNG

Liquefied natural gas

MM5

Fifth-generation Penn State/NCAR mesoscale model

MYJ

Mellor–Yamada–Janjić

MYNN3

Mellor–Yamada–Nakanishi–Niino level 3

NAM-ANL

North American Mesoscale Analyses

NARR

North American Regional Reanalysis

xvii

NOx

Nitrogen oxides

NO2

Nitrogen dioxide

PBL

Planetary boundary layer

PM2.5

Fine particulate matter

PST

Pacific Standard Time

SH

Shin–Hong

SMOKE

Sparse Matrix Operator Kernel Emissions

SO2

Sulfur dioxide

SST

Sea surface temperature

TKE

Turbulent kinetic energy

TKV

Terrace–Kitimat Valley

US

United States

USEPA

US Environmental Protection Agency

UTC

Universal Time Coordinated

UW

University of Washington

VOC

Volatile organic compounds

WHO

World Health Organization
xviii

WRF

Weather Research and Forecasting

YSU

Yonsei University

Notations
CLA

Critical load of acidity

CLNnut

Critical load of nitrogen deposition

NH4 +

Ammonium

NO3 −

Nitrate

SO4 2−

Sulfate

Units
g kg-1

grams per kilogram

kg ha −1 yr−1 kilograms per hectare per year
m s-1

meters per second

µg m-3

micrograms per cubic meter

ppb

parts per billion

xix

Acknowledgements
I use this opportunity to express my deep appreciation to Peter Jackson. Exceptionally
beneficient, Peter’s handling of students’ concerns is admirable. As supervisor, his guidance, candor, affableness, and absolute support saw to the completion of this study, and
for this, I am eternally indebted. My immense gratitude extends to other members of the
Supervisory Committee: Bruce Ainslie through whom I acquired the technical skills to
approach this subject, Stephen Déry for providing critical feedbacks, and Mark Groulx
for his insightful perspectives all the way. Collectively, I thank you for your time.
I sincerely acknowledge financial awards for research and conference travels received
from the UNBC Office of Graduate Studies. Reminiscing now, those travels served opportunities for meeting and exchanging ideas with atmospheric scientists from near and
far, much of which proved valuable to my work. They also offered precious, albeit fleeting moments of relief, sustaining my health for which I am grateful.
Much thanks goes to contacts in Kitimat / Terrace through whom local, experiential information on air quality was tapped, especially Steven and Lis Stannus and Pam Vollrath
of the Kitimat–Terrace Air Quality Coalition, and Shawn Zettler of Rio Tinto Alcan. The
assistance from you all is cherished.
Lastly, I would like to thank colleagues with whom I shared office at some point: Rachel
Hay and Brayden Nilson, neighbors: Aseem Sharma, Hadleigh Thompson and Nazrul
Islam; and Prince Geoerge resident, Janelle Wright. Your company made studies bearable
and I wish each one of you fulfilling careers.

xx

Chapter One
Introduction
1.1

Geo-ecological imperatives for atmospheric pollution
research: an overview

Shared access to the atmosphere for disposal of gaseous wastes has resulted in the potential for unintended adverse impacts on ecosystems and societies. Apart from public
health concerns (Kurt et al. 2016), air pollutants have been linked to harm on vegetation
and soils. Sulfur dioxide (SO2 ) can induce bleaching and injury on foliage (Legge and
Krupa 2002). High intake of nitrogen oxides (NOx ) by plants can reduce photosynthesis
and yield (Hu et al. 2015). Exposure of trees to mixtures of several air pollutants can impair their resistance to stresses such as frost and snow storms (Vorobeichik et al. 2014).
Further, excessive deposition of sulfur- and nitrogen-containing air pollutants to soils can
lead to acidification, and nutrient enrichment (Augustaitis 2011), subsequently causing
changes in composition of natural vegetation (Bobbink et al. 2010, Gilliam 2019). The
severity of ecological effects of air pollution varies with plant species and habitats. Long

1

periods of cold temperatures and wet climates with frequent fog, low clouds and low
light conditions can enhance the toxicity of gaseous air pollutants for flora (Augustaitis
2011, WHO 2000). Natural recovery of pollution-impacted ecosystems can be slow, and
original plant communities may remain absent long after remedial measures are implemented (Vorobeichik et al. 2014).
To prevent deleterious consequences on the environment, various jurisdictions have focused on inventorying anthropogenic air pollutants, and targeting large contributors for
reduction. Commitments to large-scale reductions in emissions that contribute to crossborder acidifying pollution such as the 1991 Canada–U.S. Air Quality Agreement (ECCC
2018a), have complemented national and provincial actions to mitigate ecosystem degradation. At the core of measures for long-term management of acid rain in Canada (CCME
2014b), is monitoring of precursor atmospheric pollutants, and prescribing and enforcing
ambient air quality standards. Airshed planning and management strategies are often
based on data from pollution monitoring stations.

1.2

Motivation for present study

Air pollutant levels could be very variable in space, thereby limiting the usefulness of
fixed-site monitors. Even in the present era of low-cost sensors, it is inconceivable to have
fixed monitors at every location. Reliable and accurate ground-based monitoring can be
challenging and costly to set-up and operate for long periods especially in mountainous
regions. Atmospheric-chemistry models such as CMAQ, that are capable of generating

2

gridded concentrations of a pollutant can, if shown to be sufficiently realistic, alleviate
the challenge of field monitoring. These computational tools model the fate of pollutant
releases from all known anthropogenic and biogenic sources, including their advection,
chemical and physical transformation, and loss. They are suitable for multi-pollutant
scenarios but are especially useful for accounting for pollution by species that are not directly released (secondary pollution). By simulating the sophisticated oxidant chemistry
in the deposition of acidifying species, atmospheric chemistry models are increasingly
being used to qualify habitat exposures to contaminants, thereby providing an integrated
approach for projecting the effect of air emissions at various spatio-temporal scales. In
addition, they are cost-saving since they can be used to assess alternative emissions scenarios which otherwise would be difficult to perform in the real world. Retrospective
analysis can also be accomplished with them to provide understanding of atmospheric
phenomena driving high pollutant concentrations. For these reasons, atmospheric chemistry models are widely accepted science tools for airshed planning and management.
Canadian applications of atmospheric-chemistry models for acid deposition assessment
(e.g Moran et al. 2008) ,however, have mainly been over parts of the eastern provinces
where extensive harm to natural habitats attributable to both the lack of natural alkalinity of bedrock material, and to historically significant SO2 and NOx emissions from
within and nearby, occur. The few modeling studies in Western Canada have either
been over large areas of mostly flat terrain (e.g. Makar et al. 2018) or for wide, deindustrialized basins (e.g. Nasr et al. 2010). Coastal parts of British Columbia have low
acid-buffering soils (ECCC 1991). Consequently, SO2 and NO x increase in places with

3

complicated topography could leave these terrestrial systems vulnerable to impacts of
acid deposition. Modeling the effect of acid precursor release from industrial activity in
fjord topography—areas with long, narrow coastal inlets bounded by steep mountains—
presents unique needs that include discretizing the circulation in deep, narrow valleys.
Accordingly, the Terrace-Kitimat Valley (TKV), an industrializing corridor of northwestern BC, serves an ideal testbed for atmospheric research.

1.3

Study area background

Aligned north-south within the Pacific Coast mountains and flanked by ridges up to 1800
m high, the TKV lies at the head of a 90-km fjord (Fig. 1.1). Roughly 70 km in length, the
valley is broadest around the small city of Terrace, but as narrow as 8 km at some sections further south, near the port town of Kitimat. The valley’s weather is dictated by air
masses originating over both the Pacific Ocean and continental North America. In summer (June–August), when the Pacific High lingers near its northernmost position (Klock
and Mullock 2000), clear, fine weather prevails. Surface wind is mostly southerly from the
Douglas Channel, and land-sea breezes and slope-valley winds are experienced. In winter (December-–February), land-falling mid-latitude cyclones bring heavy snowfall and
the intermittent clash of their onshore winds and northerly outflowing continental arctic
air result in stormy weather (Lange 2003). The mean annual temperature in Kitimat (Terrace) is 6.9 °C (7.4 °C), varying from a minimum of -4 °C average in January to a maximum
of + 22 °C average in July; annual precipitation amount is 2210 mm in Kitimat, but 1170
mm in Terrace (ECCC 2019), in the partial rain shadow of the Coast Mountains. Within
4

and over surrounding mountains are wilderness areas, comprising coniferous forests supported on podzolic soils (Fig. 1.2) (Soil Classification Working Group 1998). Vegetation
on the valley floor, portions of which are drained by the Skeena River and Kitimat River,
is dominated by western hemlock, western red cedar, Sitka spruce, Douglas-fir, amabilis
fir; and mountain hemlock occurring at sub-alpine height (Pojar et al. 1991). Understorey
abundance of bryophytes (Meidinger 2018) is prevalent.
About 28,000 people live in the TKV, of which ∼ 12,000 are in and around the small city of
Terrace, 8,300 in the port town of Kitimat, 600 in the First Nations community of Kitamaat
Village, with the remainder scattered elsewhere (Statistics Canada 2018). Anthropogenic
emissions in the valley are from these residential clusters, and light road and rail traffic,
as well as industrial sources (Fig. 1.3). An existing aluminum smelter in Kitimat emits
large amounts of SO2 . Re-emerging in 2016 after expansion/modernization works that
boosted production (Rio Tinto 2016), daily permissible SO2 however, increased from 27
tonnes to 42 tonnes. Exports processing facilities such as liquefied natural gas plants
(District of Kitimat 2020), have also been proposed in the area, some of which have been
endorsed by the host Haisla and Tsimshian nations. Proposed developments however,
have raised concerns about the air quality and deterioration of the natural environment
(Environmental Appeal Board 2015). Projects when operational, will emit atmospheric
pollutants (SO2 , NO x , particulate matter, etc.) in Kitimat. Pollutant modeling is needed,
not only to determine present impacts but also, to project incremental and cumulative
effects of anticipated emissions.

5

Figure 1.1 The Terrace–Kitimat Valley and surroundings, indicating the monitoring network for surface meteorology, air quality and atmospheric deposition.
On the right is an enlargement of the Kitimat area, where an existing aluminum
smelter (at the location of the industry symbol), about 2 km south of the Haul
Road station, emits large amounts of SO2 . The inset on the left identifies the TKV
in the Coast mountain ranges of northwestern British Columbia
6

Figure 1.2 Surficial geology of the Terrace–Kitimat Valley and surroundings (left),
and land cover distribution (right)

1.4

Research questions

This study investigates the impacts of air pollutant emissions, transport and deposition
in the Terrace-Kitimat Valley and the surrounding area using the WRF-SMOKE-CMAQ
modeling system (more details in subsequent chapters) at fine spatial resolution. Specific
research objectives are highlighted in the following questions:

7

8
Figure 1.3 Land use and development in the TKV

1. What explains the behaviors of alternative PBL schemes in the WRF model for simulating the TKV’s surface meteorology?
In numerical weather models, the prediction of phenomena that are too small to be resolved at the grid scale, such as turbulence are via various simplified formulae called
parameterization schemes. Those for the planetary boundary layer (PBL) could impact
the simulation of meteorological variables that are input to chemical transport models.
PBL schemes are often developed under ideal conditions: flat terrain, and dry, warm environments (e.g. Hong et al. 2006, Nakanishi and Niino 2009). The applicability of PBL
schemes in the very complex TKV area is unknown, therefore, should be evaluated.
2. How does the choice of PBL scheme affect the simulation of PM2.5 , SO2 and NO2 concentrations and what reasons may account for differences in model outputs?
An assessment of the modeling skill of various PBL schemes for pollutant measurements
in the TKV is conducted. This is necessary since modules in coupled numerical modeling
systems are periodically updated and it is in the interest of the science community that
revisions are rigorously evaluated.
3. Could quantile-based bias correction of CMAQ output improve usefulness for assessing regulatory compliance to air quality objectives?
In cases where simulations have large biases, identifying whether post-processing of original outputs render them valid for assessing violations of standards on local air quality is
important.
4. How well do simulations with various PBL parameterizations capture wet deposition
9

of acidifying ions and what are the baseline exceedances of sulfur and nitrogen deposition in the TKV area?
Simulations of wet deposition which is a component of total deposition of acidifying
species, is evaluated by comparing outputs from CMAQ to those of observational data
at two measurement locations in the TKV. The credibility of outputs for various model
runs are examined prior to estimating deposition impacts of existing emission sources in
the valley.
5. What choices of atmospheric forcing data and PBL scheme in the WRF model can reproduce precipitation for the purpose of projecting atmospheric deposition over the TKV
and what is the uncertainty in the precipitation field with the best fit simulations?
Determining the validity of hydrological outputs of meteorological datasets, as substitutes of actual precipitation measurements, is needed for realistic projection of deposition
impacts of future industrial emissions.
6. What incremental and aggregate impacts could arise should the existing smelter at
Kitimat emit 42 tonnes day−1 that is the highest permissible rate?
Although permitted up to 42 tonnes day−1 SO2 release, smelter emissions as at 2017 was
still around the previous limit of 27 tonnes day −1 (ECCC 2020). Hence determining what
pollutant concentrations, and incremental effects that may arise should the smelter emit
at the maximum permissible amount, is crucial to understanding the sensitivity of the
valley’s atmosphere to pollutant emissions.
7. For proposed developments in the TKV,
10

(a) To what additional extent could vegetation be exposed to harmful NOx and SO2
concentrations from LNG projects?
(b) How much direct vegetation exposure to harmful NOx and SO2 concentrations could
result from aggregate industrial emissions?
(c) By how much could soil nitrogen enrichment and acidification change as a result of
the LNG industry?
(d) What aerial exceedances of critical loads of nitrogen and sulfur deposition will arise
from aggregate industrial emissions?

1.5

Research Outline

The rest of the dissertation consist of chapters, each of which address the research questions under topical headlines and in the same order as listed above, are in the form of
manuscripts for peer-reviewed journals.
Chapter 2 examines the influence of turbulence closures built-in to the WRF model in generating representative meteorological fields for the TKV area. It evaluates their emulation
of surface meteorological parameters that affect air quality (e.g. air temperature, specific
humidity and wind) and analyzes differences in outputs.
Chapter 3 evaluates the capability of the WRF-SMOKE-CMAQ modeling system that is
configured with various PBL schemes, to mimic observed air quality measurements in the
valley (fine particulate matter, sulfur dioxide and nitrogen dioxide). It ranks the fitness
of model runs both quantitatively and qualitatively, also discusses aspects for improve11

ment.
Chapter 4 addresses the correction of bias in CMAQ output, and assesses whether improvement is appropriate for benchmarking exposure to ambient pollutant concentrations.
Chapter 5 investigates wet acidic deposition sensitivity to PBL scheme choices. It also
quantifies the uncertainty in exceedances of critical loads of nitrogen and sulfur deposition caused by base emission in TKV, prior to precursor emissions changes from planned
industrial facilities.
Chapter 6 assesses the credibility of different choices of atmospheric forcing data and PBL
schemes for precipitation modeling, such that could affect the accuracy of acid deposition
projections for the TKV area.
Chapter 7 projects the implication of smelter SO2 emissions increase for vegetation exposure to ambient concentrations, and the risk of indirect impacts via soil acidification.
Chapter 8 estimates the aerial impact of future export processing industries in the valley, specifically from planned liquefied natural gas facilities and allied shipping activities.
Effects on atmospheric quality are modeled and evaluated from a cumulative concentrations/deposition assessment perspective.
Chapter 9 provides an overall summary of the study’s findings and concludes the report.

12

Chapter Two
Meteorological downscaling with WRF
model version 4.0 and comparative
evaluation of planetary boundary layer
schemes over a complex coastal airshed
(Also published at https://doi.org/10.1175/JAMC-D-19-0212.1 in J. Appl. Meteorol. Climatol.)

Abstract
Evaluation of downscaled meteorological information is crucial to identifying model behaviors that may propagate to end-applications such as the simulation of local air quality.
This study conducted and assessed year-long simulations of hourly meteorology over the
Terrace–Kitimat Valley of northwestern British Columbia at 1-km horizontal gridding for
six PBL schemes in the WRF model, version 4.0. In terms of key surface meteorologi13

cal variables that affect air quality, simulations over land demonstrated better skill for
specific humidity and wind direction, than for air temperature and wind speed. Spatial
differences in modeled atmospheric properties and vertical profiles, especially for moisture content, were used to diagnose the relative capacity of each PBL scheme to represent
pollution dispersion and dilution. Stable conditions at night increased suppression of
boundary-layer mixing by the nonlocal YSU scheme compared with suppression by the
local eddy-diffusion component of the ACM2 scheme, resulting in decreased wind speed
and ambient temperature, but moister air with the YSU scheme. The weakening of mixing by the MYNN3 scheme with inland distance suggested that higher-order, nonlocal
transport is sensitive to increasing topographic steepness towards the northern part of the
valley. Disparities in mixing strengths amongst PBL schemes were greater in the summer
when conditions were generally less stable with moist, warm air blowing inland than in
winter when the valley channels cold, stable air from the interior. Increased convection
in daytime led to greater entrainment of air from aloft and a thicker PBL with the YSU
scheme than the ACM2 scheme in summer while increasing counter-gradient transport
in the MYNN3 scheme that reduces dilution.

2.1

Introduction

Correct representation of the wind, temperature, moisture and mixing heights are needed
to predict the fate of air pollutants. It is generally recognized that PBL processes impact
the prediction of these variables and many investigations of the sensitivity of coupled
numerical models to PBL schemes has been undertaken in the last decade. Cheng et al.
14

(2012) simulated ozone episodes over Taiwan and found that discrepancies between results were due to vertical mixing strength and entrainment properties of the tested PBL
schemes. Banks and Baldasano (2016) evaluated how outputs from some PBL schemes in
the Weather Research and Forecasting (WRF) model differ from one deployed for operational air quality forecasts over Catalonia, Spain. In India, Gunwani and Mohan (2017)
researched the sensitivity of PBL schemes to different climatic zones based on model
simulation of some meteorological parameters. Others (e.g. Reboredo et al. 2015, Xie
et al. 2012) have quantified meteorological prediction differences to ascertain which PBL
scheme is most appropriate to specific geographical domains.
However, much of the related literature (e.g. Mohan and Gupta 2018, Misenis and Zhang
2010) are for short episodes and do not portray cross-temporal facets of model performance. In some regions, where severe pollution episodes may be uncommon, and routine contaminant exposures from all other periods in a year are more of a concern, airshed
managers could be more interested in simulation of air quality over annual, seasonal or
even diurnal periods. To the author’s knowledge, no studies have been undertaken to
identify the effect of different PBL parameterization schemes in the downscaling of meteorological variables for modeling the air quality in northwestern Canada. Since outputs
from meteorological models are commonly fed into air quality models, ultimately weighing on airshed planning and policies, it is imperative to assess the reliability of downscaled meteorology prior to their use for current assessments and future projections.
This chapter conducts high-resolution simulations of meteorological variables with theWRF
mode (version 4.0), and the fifth major global reanalysis product of the European Centre
15

for Medium-range Weather Forecasts (ERA5), with a focus on evaluating performance of
different PBL formulations. Its objectives are to quantify modeling uncertainties and to
qualify mixing and dispersion capabilities of alternative physical schemes. The remainder of this manuscript is organized as follows. In section 2.2, the experimental set-up,
including the PBL schemes being tested, nested domains, and WRF model configuration are described. Section 2.3 evaluates the uncertainty in modeled surface meteorology
via comparisons with observations. Differences among simulations resulting from PBL
scheme options are also analyzed. Section 2.4 discusses the implications of simulation
similarities and differences for air quality modeling and concludes the chapter.

2.2

Methods

2.2.1 Description of the WRF model and PBL schemes
WRF is a state-of-the-art atmospheric modeling system designed for both meteorological
research and numerical weather prediction. The dynamical solver, known as the Advanced Research WRF (ARW) core, is based on the fully compressible, non-hydrostatic
Euler equations with terrain-following Eta coordinates. WRF can run on a variety of
computing platforms and suits a broad range of applications across scales ranging from
tens of meters to thousands of kilometers (Skamarock et al. 2019).
Turbulence closure schemes, also referred to as PBL schemes, are built-in to meteorological models to represent turbulent fluxes that are not explicitly resolved on the model grid.
PBL processes are parameterized either through local closure or nonlocal closure models.

16

In the case of local closure, only those vertical levels that are adjacent to a given point directly affect variables at that point. Nonlocal closure schemes use multiple vertical levels
to determine variables at a given point (Cohen et al. 2015). For meteorological downscaling with the WRF model, six such schemes are experimented. These schemes are the:
Mellor–Yamada–Janjić (MYJ), Asymmetric Convective Model, version 2 (ACM2), Mellor–Yamada–Nakanishi–Niino level 3 (MYNN3), Shin–Hong (SH), University of Washington (UW), and Yonsei University (YSU).
The YSU scheme is a first-order nonlocal scheme with explicit entrainment at the PBL
top and a parabolic K-profile in an unstable mixed layer. It calculates PBL height from
the surface and uses a threshold Richardson number of 0.25 for stable cases, and zero
for unstable flow (Hong et al. 2006). The SH scheme is a nonlocal ’scale-aware’ formulation for vertical transport in a convective PBL. Vertical mixing in the stable PBL and free
atmosphere follows YSU, in addition to diagnosed turbulent kinetic energy (TKE) and
mixing length, but the explicit treatment of heat flux entrainment is replaced by grid sizedependent terms for nonlocal, and local transport components (Shin and Hong 2015).
The ACM2 scheme is a hybrid scheme that features nonlocal transport from the lowest
level to all other model layers alongside a local eddy-diffusion component, and an asymmetrical layer-by-layer downward transport. It calculates PBL height when the critical
bulk Richardson number above the level of neutral buoyancy exceeds a value of 0.25. For
stable or neutral flows, the scheme shuts off nonlocal transport and uses local closure
(Pleim 2007). The MYJ scheme is a one-and-half-order local closure that includes a prognostic equation for TKE. The PBL height is the height at which the TKE reaches 0.2 m2 s-2

17

(Janjić 1994). The MYNN3 scheme is a second-order closure that expresses stability and
mixing length based on the results of large eddy simulations rather than observations. It
prescribes PBL height when TKE reaches 1.0 × 10-6 m2 s-2 (Nakanishi and Niino 2009).
The UW scheme is a one-and-half-order local closure in which TKE is a diagnosed, rather
than a prognostic variable. It uses moist-conserved variables with an explicit entrainment
closure for convective layers. A threshold of 0.25 for the critical bulk Richardson number
is used to determine PBL height in all stability cases (Bretherton and Park 2009).
The above schemes are a mix of legacy and newer physics modules that have accompanied major updates to the WRF code. For instance, whereas the YSU, MYJ, and ACM2
schemes are quite common in the literature (e.g. Nielsen-Gammon et al. 2010, Zhang
et al. 2013), the UW and SH schemes are fairly recent additions. Consequently, selected
PBL schemes represent the state-of-the-art in options for parameterizing atmospheric processes over predominantly pristine (non-urbanized) areas.
Each PBL formulation can be mapped to just one or more surface-layer schemes. The
surface-layer schemes compute friction velocity and other exchange coefficients for estimating sensible and latent heat fluxes, and momentum flux from land surface models,
and surface stress in the PBL schemes. In this study, the MYJ, MYNN3, and UW PBL
schemes are coupled to the Eta surface-layer scheme (Janjić 2002), while the ACM, YSU,
and SH PBL schemes are mapped to a modified MM5 surface-layer scheme (Jiménez et
al. 2012). Both surface-layer schemes are based on similarity theory but the Eta similarity
includes a parameterization of viscous sub-layer (Janjić 2002).

18

2.2.2 Domain configuration, model initialization and settings
Computational domains assumed a telescopic nesting arrangement (Fig. 2.1). A grid ratio of 1:5 applied to successive parent domains on a Lambert Conformal projection was
used to attain a horizontal resolution of 1 km for the TKV and surrounding areas (Table
2.1). Because there are several lakes in the valley, terrain data were interpolated from
the Moderate Resolution Imaging Spectroradiometer (MODIS) International GeosphereBiosphere Programme (IGBP) 21-category land use and land cover fields with representation for lake bodies. This was to effectively distinguish between inland water bodies
and seas, for appropriate processing and extrapolation of surface air temperatures over
lakes. Atmospheric forcing data (for initial and boundary conditions) for the entire set-up
was ERA5 (ECMWF 2019). Simulations up to a model top of 50 hPa were then performed
for the year 2017. The 2017 meteorological year was selected because it best represents
the climatological average (of wind speed and precipitation) for the valley from 2006 to
the present. Appendix A has the details for this choice. For all simulations, physics settings (Table B.1 in Appendix B) were the Noah land surface model (Tewari et al. 2004),
the Thompson scheme for microphysics (Thompson et al. 2008), the rapid radiative transfer model (Iacono et al. 2008) for longwave and shortwave radiation, and the horizontal
Smagorinsky first-order closure (Talbot et al. 2012) for mixing terms. The Grell-Freitas
ensemble cumulus physics scheme (Grell and Freitas 2014) used for the largest and intermediate domains was not required for the innermost (study) domain (Arakawa et al.
2011) due to its sufficiently high resolution. A one-way nesting (that is, the transfer of information from parent nests without feedback from child nests) was implemented. Since

19

meteorological data would be fed to an air quality model that required at least 10 days
spin up period, WRF outputs were first generated for the period 00:00 UTC 20 December 2016– and 00:00 UTC 31 December 2016. Thereafter, WRF simulations with the ERA5
dataset were initialized monthly, with model spin-up of one day, and a further one day
at the end of each month. The monthly overlap days, including for the period 20–31
December 2016 were discarded when merging hourly output files for the meteorological
analyses.

Figure 2.1 The Terrace– Kitimat Valley (TKV) and surroundings. Left: WRF nests
for downscaling ERA5 data to study area. The immediate bounding nest for the
TKV is d03 in red. Right: Enlargement of the d03 domain. Black outline specifies
the valley channel wherein meteorological observation stations for model evaluation are indicated with black-filled markers. The red line is assumed centerline
through the valley and the coastal channel

20

Table 2.1 Properties of the WRF model domains. Domains are roughly centered
on the valley. Simulations for all three domains were into 39 vertical layers of
varying thickness, with 12 layers in the lowest 2000 m. Geographic data of horizontal resolution similar to grid sizes of respective domains were deployed.

Domain

Horizontal x grid

y grid

Grid centre

Grid centre

Geographic (static)

longitude

data resolution

resolution

points points latitude

d01

25 km

100

100

54.200 ◦ N

128.600 ◦ W

20-arc-minute

d02

5 km

121

121

53.850 ◦ N

128.795 ◦ W

2-arc-minute

d03

1 km

101

121

54.223 ◦ N

128.640 ◦ W

30-arc-second

2.2.3 Observational data
Three surface meteorological variables were selected for the evaluations. These are 2meter air temperature (T2), specific humidity at 2 m (Q2) and 10m-wind (speed and direction). These variables were selected because they influence ambient air quality (Banks and
Baldasano 2016, Dennis et al. 2010) and are monitored at various locations in the valley
(Fig. 2.1, Table 2.2). Observation stations are operated and maintained by several organizations including Environment and Climate Change Canada, BC Ministry of Transportation and Infrastructure, and BC Ministry of Environment and Climate Change Strategy’s
Air Quality Network. Station data can be downloaded from https://data.pacificclimate.org/
portal/pcds/map/. Because each network is designed to meet the specific need of its operating agency, there is variety in site characteristics and sensor equipment. For each
parameter at any station, only records with at least 80 % yearlong completeness in hourly
values are used for evaluation.

21

Table 2.2 Observation data locations for validation of surface meteorology variables for the TKV

Station

Latitude Longitude Elevation

Station Description

Surface variable

0.0

Maritime buoy

Temperature, Wind

128.639

94.0

Residential area

Temperature, Wind

128.538

220.0

Lakeside monitor

(◦ N)

(◦ W)

(m)

Nanakwa

53.830

128.830

Whitesail

54.067

Onion Lake

54.302

Temperature, Wind,
Specific humidity

Terrace

54.522

128.608

81.0

On school property

Temperature, Wind,
Specific humidity

2.2.4 Statistical evaluation indices
For evaluations, values for model grid cells corresponding to ground locations of observations were retrieved. The fitness measures used in this work are the mean bias (MB),
the root-mean-square error (RMSE), Pearson’s correlation coefficient (r) and modified index of agreement (IOA) according to Willmott et al. (2012). These indices are defined in
Appendix C and were selected to reflect a mix of accuracy measurements for paired data.
They also enable comparison to performance benchmarks (Table 2.3) for surface variables
suggested by Emery et al. (2001). The benchmarks were developed over a limited duration in summertime, in an environment that is dissimilar to the TKV, therefore, may
not be ideal for high-resolution, cross-seasonal simulations over complex terrain. MB as
well as RMSE are in units of the parameter of interest and are negatively-oriented indices,

22

which means values closer to zero are better. For r and IOA however, scores range from
-1 to 1, for which values approaching 1 indicate good performance. The identification of
the most suitable PBL scheme is from a count of instances in which it is top-ranked, and
also meets the performance benchmark. This is checked per statistical measure for all the
stations.
Table 2.3 Performance benchmarks (except for r) proposed by Emery et al. (2001)
for surface meteorological variables in mesoscale models. Statistical measures
per season, are on diurnal-hour means (0, 1, .., 23 ) for air temperature and wind
speed, and on all hourly values for wind direction and specific humidity. Seasons
are spring for the months of March to May, summer for June to August, autumn
for September to November, and winter for December to February. RMSE is used
in place of mean error for air temperature and specific humidity
Benchmark
Measure

2.3

Air temperature Specific humidity Wind direction

Wind speed

±0.5 m s−1

MB

±0.5 K

±1 g kg−1

RMSE

≤2K

≤ 2 g kg−1

≤ 2 m s −1

IOA

≥ 0.8

≥ 0.6

≥ 0.6

r

≥ 0.8

≥ 0.8

≥ 0.6

±10 ◦

Results and analyses

2.3.1 Synoptic and mesoscale representations
PBL schemes drive the vertical distribution of fluxes in the atmosphere and soundings
of temperature, humidity and wind alongside modeled profiles from each PBL scheme
23

are examined qualitatively. All simulations reasonably retrace the above aerological variables for different times in winter and summer seasons at the Annette Island radiosonde
station, Alaska, which is the upper air sounding location nearest the TKV (Fig. D.1 in
Appendix D). This station is outside the 1-km grid, but within the 5-km grid. Temperature and dew point profiles modeled by the various schemes virtually match observations. Simulated dew point profiles for 0400 PST on a January day replicated a drying
lower troposphere; coincident profiles of the YSU and SH schemes being drier than observations at 700 hPa height above the surface. Representation of the veering wind is
also reasonable, given that specific days and times could have unusual atmospheric conditions. It needs emphasizing however, that the atmospheric environment around the
Annette Island station, at least for the lowest 2000 m, could differ markedly from that of
the TKV, since the station is west of the Coast Mountain ranges, completely exposed to
Pacific Ocean air masses. Despite physiographical differences between the radiosonde
location and stations in the TKV, the fact of general correspondence for upper air data
attests to a reasonable representation of synoptic information.
Figures 2.2 and 2.3 are plots of meteorological conditions at the surface for chosen summer (July 22) and winter (January 21) days, respectively, whose comparisons to observed
seasonal averages of air temperature and wind are in Fig. E.1 of Appendix E. Specific
times of the spatial plots are hours of diurnal temperature maximum and minimum over
land, which are relevant for assessing mesoscale circulations such as land-sea breezes
and mountain-slope winds. One verification of WRF simulation realism is the altitudinal
decrease of daytime air temperature in summer, from around 18 °C on the valley floor to

24

about 4 °C over the mountains, which is consistent with the expectation of the valley to be
warmer than the ridges (Fig. 2.2). Wind vectors are also more organized and aligned with
the main and tributary valleys than outside them, indicating substantial wind adjustment
to the underlying terrain and the efficacy of simulation at high horizontal resolution. Over
the main valley itself, modeled winds are seasonally nuanced. In summer (Fig. 2.2), the
wind is southerly, which is representative of the up-valley flow that occurs for ∼ 70 % of
non-calms (wind speed ≥ 0.5 m s-1 ) observed at the Terrace station. This pattern is also
dominant overnight (0200 PST) in all the PBL schemes, perhaps due to fewer nighttime
hours in summer for reversal of wind direction, or from stronger synoptic forcing on this
day, but terrain recognition by the WRF model appears satisfactory. Southerly wind vectors turn anticlockwise on approaching the Skeena River, resulting in a flow down the
river channel, towards the coast. Some convergence of vectors towards watercourses that
drain the valley, suggesting downslope winds in tributary valleys is also present in the
simulations at 0200 PST. The afternoon hour (1400 PST) mainly indicates wind directions
out of stream channels, or upslope. Compared to the nighttime period, the southerly flow
in the valley is as would be expected, more intense due to enhancement by sea breezes.
Wind vectors within the valley are straighter and better organized, with slightly more
build-up on valley sidewalls for the MYNN3 and MYJ schemes.
The winter season (Fig. 2.3) when the valley is snow-covered, differs markedly from
summer. At night (0200 PST), the warmest parts are around the shoreline, and over major
inland water surfaces. This is as expected since water bodies cool more slowly than land.
Over the ridges that flank the valley, the MYNN3, MYJ, and UW PBL schemes depict

25

Figure 2.2 Ambient air temperature and wind vectors at the surface for a typical
day in summer, at night (top) and during daylight (bottom) simulated by the
various PBL schemes.

26

warmer temperatures than the ACM2 and YSU PBL schemes, probably due to differences
in surface-layer schemes. By afternoon, the land areas have warmed such that both the
valley and shore areas are mostly above 0 °C. Indeed, at 1400 PST, the northern half of the
valley is warmer than the southern portion, with temperatures at mountain peaks similar
(-9 to -6 °C) amongst PBL schemes. Wind vectors for the winter season are consistent
with the dominance of northerly outflow at this time of a year; however, these are less
organized than during summer. While the flow within the valley remains fairly distinct
from winds over surrounding ridges, wind vectors in the valley are less organized than
in summer suggesting stronger synoptic forcing in the winter season on this day. Once
exiting the valley into the Douglas Channel, wind vectors are more aligned with the fjord,
consistent with the expectation of less drag over water surfaces.

2.3.2 Performance evaluations for surface variables
Ambient temperature
Near-surface temperature (T2) is an important meteorological parameter as it affects the
buoyancy of air pollutants and reaction rates of chemical species. Figure 2.4 shows diurnal sequences of hourly mean biases for summer and winter seasons. In summer, a warm
bias occurs after sunrise, peaking at about 0800 PST, becoming a cool bias after noontime,
and reaches a negative peak at sunset. The adaptation to convective transport by the
nonlocal SH and YSU PBL schemes whose bias profiles are virtually the same, is evident
in larger daytime bias amplitudes than those of the other schemes. At nighttime, YSU,
SH and ACM2 have marginally colder biases than the others, leading to relatively cooler
27

Figure 2.3 Ambient air temperature and wind vectors at the surface for a typical day in winter, at night (top) and during daylight (bottom) simulated by the
various PBL schemes.

28

temperatures over the course of a single day with these schemes.

Figure 2.4 Hourly mean bias of T2 (modeled minus observation) over land for
each PBL scheme in summer and winter seasons. Profiles above 0 K signifies
warm bias. Dark grey plot backgrounds are nighttime periods.

Table F.1 in Appendix F tabulates model performance across four seasons in a year. Going
from the spring to winter period, T2 simulations change from negative biases to predominantly positive biases. Biases are less during the summer and winter seasons, compared
to the spring and autumn seasons. Over land, diurnal MB ranged -2.5 to 1.3 K in spring, 29

0.6 to -0.2 K in summer, 0.5 to 4.1 K in autumn and -0.3 to 0.9 K in winter. Model errors for
T2 are also smaller in summer and winter months; these periods having RMSE values that
are ≤ 2 K at all three land stations. Temporal correlations are good, and the r benchmark
is achieved in 100 % of evaluations from spring through autumn (> 0.9 in summer) and
67 % for winter. Few evaluations meet the IOA criteria; although the majority of values
are within 0.60–0.79 across seasons. Results from all statistical measures indicate that T2
simulations over land are poorest in autumn. Biases and errors for the Nanakwa station in
spring (MB range: -0.5 to 0.1 K, RMSE range: 1.2 to 1.4 K) and summer, (MB range: -1.0 to
-0.4 K, RMSE range: 1.5 to 1.7 K) suggest reasonable accuracy of interpolated sea-surface
temperatures in warmer months. In terms of individual PBL schemes, MYJ marginally
outperforms UW with regard to accuracy and precision indices (MB, RMSE, and IOA),
while MYNN3 performs best for pattern replication (r). The MYJ scheme also has several
satisfactory correlation scores, thus more ranked by all four statistical measures.

Specific humidity
Another important meteorological variable from an air quality perspective, especially in
controlling aerosol formation and transformation processes, is specific humidity (water
vapor content). Performance statistics for specific humidity at 2-m height (Q2) for each
PBL scheme per season, and at two stations is presented in Table F.2 of Appendix F. Across
PBL schemes and relative to observations, the dominant pattern is a dry bias. Biases are
larger at the Onion lake station which receives more precipitation than the Terrace station,
and in summer when more evapotranspiration occurs, than in winter. These further attest

30

to the dynamical consistency of model outputs. At the Onion lake station, for instance,
the MB range is -0.50 to -0.37 g kg-1 in winter, as against -0.68 to -0.19 g kg-1 in summer.
Overall, model performance for specific humidity is remarkable. Ninety-four percent of
evaluations for MB, 100 % for RMSE, 56 % for r and 65 % of IOA fall within respective
cut-off values of satisfactory performance. With regard to individual PBL schemes, the
MYNN3 scheme has the best fit for the Q2 simulation, based on the number of top-ranked
and satisfactory evaluations for it. Indeed, because all schemes underestimate Q2, the
MYNN3 scheme best ameliorates negative moisture biases.

Wind
Surface wind impacts the mixing and transport of air contaminants from source areas to
receptors. Biases in modeled wind direction (WDIR10) at the different stations are presented in Table F.3 of Appendix F. The localized nature of wind over complex terrain is
attested by biases at Onion Lake which is anticlockwise, unlike other stations. Noteworthy is that at the Nanakwa buoy, almost all evaluations meet the ± 10 ° criteria for wind
direction bias. On land, model skill is reduced, arising from increased surface roughness,
and wind flow being affected by multiple terrain elements. Only 8 % of evaluations for
land stations are within the bias range for acceptable skill. However, 75 % of all evaluations for Terrace are ≤ 20 ° and all for Terrace and Whitesail are ≤ 30 °. The largest biases
are at the Onion Lake station, in the central portion of the valley (up to ∼ -60 °). Using the
same rule for assessing suitability, the MYNN3 and UW PBL schemes rank best for wind
direction.

31

Wind speed uncertainty due to PBL scheme choices is also of interest and Fig. 2.5 shows
diurnal sequences of hourly mean biases for 10 m-wind speed (WSPD10) in summer
and winter. Differences between schemes rise to as much as 1.5 m s-1 in summer, but
are smaller (∼ 1 m s-1 ) in winter. The valley atmosphere is more prone to decoupling
the surface layer from layers aloft in winter due to increased stability, thereby lessening
the sensitivity of surface winds to PBL parameterizations. Strong synoptic forcing from
land-falling Pacific mid-latitude cyclones would account for why modeled WSPD10 has
greater biases in winter than in summer. It is reasoned that the general overestimation
of WSPD10 is due to WRF not sufficiently weakening winter storms as they transition
from the smooth ocean to the rugged interior. Indeed, wind speed biases increase from
the coast towards the northern part of the valley. Aside from complications of cross-flow
of the Skeena River (see Fig. 2.1), and urban obstacles that attenuate real surface winds
in Terrace, the general area to the north is dominated by mountainous topography. Wind
speed overestimation has been reported in several experiments using WRF at similar horizontal grid resolutions over complex terrain (e.g. Avolio et al. 2017, Hariprasad et al.
2014, Zhang et al. 2013).
The performance statistics for WSPD10 for all four seasons for each of the PBL schemes
are in Table F.4 of Appendix F. Overestimations occur not only in winter and summer
months but also through spring and autumn. Across land stations, diurnal MB ranged
0.4 to 2.7 m s-1 in spring, -0.2 to -2.5 m s-1 in summer, 1.3 to 3.0 m s-1 in autumn, and 1.3
to 3.7 m -1 in winter. Just 10 out of 72 evaluations (14 %) are within the MB benchmark
for wind speed. Model errors for WSPD10 are also higher in autumn and winter months,

32

Figure 2.5 Hourly mean bias of WSPD10 (modeled minus observations) according to PBL schemes. A value of 0 m s-1 signifies no bias. Dark grey plot backgrounds are nighttime periods.
although the proportion of evaluations within the RMSE criteria is much improved (68
%). Favorable RMSE and r scores are in contrast with negative 1OA values. Whereas 64
% of evaluations for r are ≥ 0.6, only 11 % — all for Onion Lake— meet the IOA benchmark. Thus, compared to air temperature and humidity, WSPD10 fits poorly with the
benchmarks, especially at Terrace.

33

2.3.3 Spatio-differential examinations of surface meteorological fields
Because diurnal surface heating and cooling impact atmospheric structure, spatial differences in model outputs are examined for the relative effect of individual WRF PBL
schemes on meteorological properties at nighttime and daylight periods. The seasonal
attributes of PBL scheme adjustment to local weather and evolving capacities for mixing
of surface fluxes are also qualified.

Nighttime
Some turbulence closure approaches are conditionally operable. The ACM2 scheme deactivates nonlocal transport for stable conditions, which would prevail at night. Of interest
therefore, is how meteorological quantities for periods when only the local closure component of this scheme operates, compares to those from the fully nonlocal YSU scheme,
given that both PBL schemes are mapped to the same surface layer model. Figure 2.6a
shows that over the valley portion of the domain, the skin temperature (SKINT), T2, and
WSPD10 from the ACM2 scheme are greater than those of the YSU scheme. At higher elevations outside the valley channel, the topography is less confined and the YSU scheme
is more capable of moving cold air from the surface to layers aloft, and downward momentum transport. The relative weakening of nonlocal transport over the valley therefore
derives from the confinement of the land surface by steep mountains, and is also due to
nighttime stable conditions that restrain vertical transport through the entire PBL. In the
corresponding wintertime, when the atmosphere is more stable, the temperature difference over the valley between the YSU scheme and the ACM2 scheme is increased (∼ 0.5

34

°C) but less widespread, since the severest restriction to nonlocal flux exchange would be
at the deepest of cold air pools.
Since the expression of Q2 as a mixing ratio is analogous to that of an air pollutant that
may be discharged at the ground level, its comparison across schemes could be useful
proxies for contaminant accumulation and dispersion. Greater Q2 with the YSU scheme
over most parts of the study domain may be linked to the relative capacity of nonlocal
closure to facilitate vertical transfers between model layers. But the correspondence of
ACM2 /YSU paired differences for PBL heights (PBLH) and Q2 is also noteworthy. Relative to outputs from the YSU scheme, the PBLH simulated by the ACM2 scheme is thicker
over water bodies, which are the same areas where its Q2 values are moister. In the YSU
scheme, vertical diffusion is constrained to the diagnosed PBL depth. Thus, it is less diffusive of Q2 over the land areas. Over water, Q2 from the YSU scheme is subject to a
steeper vertical gradient that enables faster transport of water vapor from the bottom of
the PBL to overlying model layers.
The constraining of nighttime mixing by steep topography is less with local turbulence
closure. However various local mixing approaches exist whose expressions may also
differ with alternative surface-layer parameterizations. Figure 2.6b compares the ACM2
scheme nighttime outputs, to those of the local MYJ scheme that is mapped to the Eta
surface-layer model. Unlike the MYJ scheme, the local closure implementation in the
ACM2 scheme ignores TKE transport. Hence in the winter season when temperature inversions are frequent, along-valley PBLH simulated by the ACM2 scheme is shallower
than that of the MYJ scheme. However, that up to1.0 °C difference for T2 over the TKV is
35

modeled, suggests surface-layer scheme influence. Over the valley, MYJ outputs warmer
T2 than ACM2, but cooler skin temperatures (SKINT), hinting at greater heat flux retention with the Eta surface-layer scheme than with the MM5 surface-layer scheme. Unlike
the MM5 surface-layer scheme, the Eta surface-layer scheme considers thermal roughness
length (Janjić 2002), apparently contributing to warmer ambient temperatures with the
MYJ PBL scheme. Differences in friction velocities (not shown) were similar to patterns
for wind speeds, for which the MYJ scheme values over the valley and coastal channels
are less. Friction velocity is a fraction of wind speed (Camuffo 2014). The friction velocity
limit in previous versions of the MM5 surface-layer scheme ( 0.1 m s-1 ) was considered as
preventing undesirable effects such as excessive ground cooling (Jiménez et al. 2012). The
revised MM5 surface-layer scheme as used in the present study, permits friction velocities
as low as 0.01 m s-1 (Jiménez et al. 2012), while in the Eta surface-layer scheme, a correction is applied so that only non-zero values are obtained (Janjić, 2002). Consequently,
lower friction velocity over the valley and coastal channels with the MYJ scheme possibly
derives from its coupling with the Eta surface-layer scheme, which in turn, could offset
part of the warming effect of thermal roughness length. Cooler SKINT of the MYJ output
relative to that of the ACM2 simulation over the valley channel may have resulted from
such differences in friction velocities.
For other model runs with the Eta surface-layer scheme (Figs. 2.6c-d), differences among
PBL schemes for simulated variables are less than in Fig. 2.6b, where both the surfacelayer scheme and PBL schemes are varied. Nonetheless, the differential subtleties are
worth examining, given that PBL scheme development, testing and revising, has primar-

36

ily centered on weather prediction, rather than on air pollution modeling. Warmer skin
temperature for MYNN3 compared to MYJ translate to a higher T2, but do not result in
greater water vapor retention over the entire grid. MYNN3’s summertime Q2 over the
southern portion of the domain is drier than that of the MYJ scheme. But this portion
roughly coincides with the parts where the height of MYNN3’s PBL surpasses that of the
MYJ scheme (also Fig. G.1 in Appendix G), suggesting greater mixing with the MYNN3
scheme. This seems to impact nighttime WSPD10 as well, with greater (lesser) values for
the MYNN3 scheme coastward (landward). In addition to implementing local eddy diffusion, the MYNN3 scheme, unlike the MYJ scheme, has a nonlocal mixing component.
The suppression of nonlocal vertical exchange in the MYNN3 scheme relative to the MYJ
scheme, is therefore less than for the YSU scheme in relation to the ACM2 scheme in
Fig. 2.6a. There is a seasonal signal with this behavior. In summer, modeled onshore
winds generate instabilities over considerable distances upon which nonlocal mixing is
enhanced. In wintertime, when valley air is colder and drier, and simulated circulations
are northerly, MYNN3’s nonlocal mixing is restrained: the mixing is more diffusive than
in the MYJ scheme only over the Douglas channel that is exposed to warmer, maritime
air.
The UW scheme, designed for simulation of marine stratocumulus-capped boundary layers, neglects TKE storage, and assumes no TKE transport occurs during stable conditions
(Bretherton and Park 2009), hence it simulates the lowest PBL heights and cloud bases
that are nearest to the ground (Figure G.1 in Appendix G). Some modeling experiments
in mountainous regions (e.g. Balzarini et al. 2014) have also reported lower PBL heights

37

Figure 2.6 Spatial differences of averaged skin temperature (SKINT), ambient
temperature (T2), surface wind speed (WSPD10), ambient moisture content (Q2)
and PBL heights at 0200 PST in summer and winter. MYNN3 – MYJ means values
for MYJ subtracted from MYNN3, etc
with the UW scheme. Its output of smaller Q2 than that of the MYJ scheme (not shown)
suggests a relatively rapid moisture loss mechanism. It is worth stating that the UW

38

PBL scheme’s verification and integration into the WRF model noted the diffusion of humidity and chemical species from the lowest model grid level, rather than their mixing
across layers (Bretherton and Park 2009). Because air pollutants are also mass fluxes, less
nighttime ambient moisture with the UW scheme relative to the MY-schemes suggests
pollutant concentrations could be lower with the UW scheme.

Daytime
PBL schemes adjust differently to enhanced convection in the daytime, also impacting
meteorological fields over complex terrain. Comparisons between the ACM2 and YSU
schemes in summer (Fig. 2.7a) indicate that for the most part, the ACM2 scheme produces a higher PBL over land areas, in reverse of the nighttime pattern. Unlike single
estimation of the PBL top in the YSU scheme, the PBLH in the ACM2 scheme is calculated from separate diagnosis of convective and entrainment layers (Pleim 2007), partly
explaining PBLH differences. Convective regimes activate the nonlocal transport component of ACM2, thereby providing a hybrid vertical flux transfer mechanism, as opposed
to the YSU scheme that is nonlocal. Over valley areas adjacent to the marine channel, incursions of maritime air may diminish the development of convective regimes more than
at positions farther inland, hence small portions where the ACM2 scheme’s Q2 output
surpass those of the YSU scheme. WSPD10 differences along the Douglas Channel show
the ACM2 output is windier, also contributing to moister inflow. But for the most part,
ACM2 generates less Q2 in the daytime of summer (see also Fig. 2.8), which suggests
stronger updrafts by the ACM2 than by the YSU scheme in this period. Therefore, the

39

ACM2 scheme could produce less pollutant concentration at the surface than the YSU
scheme, especially in the drier, northern part of the valley. In the cold winter season
(Figs. 2.7a, 2.8), the daytime Q2 difference is negligible (< 0.1 g kg −1 ) and the change of
water vapor with height is comparable between the two PBL schemes. This is because
the ACM2-predicted PBL heights are mainly lower than those of the YSU scheme, and its
nonlocal transport component is inactive in stable/neutral conditions that dominate in
winter.
Figure 2.7b compares functionally-similar PBL schemes for daytime mixing but mapped
to different surface-layer schemes. The MYNN3 scheme, with relatively positive incoming radiation over the valley in summer, simulates lower SKINT than the ACM2 scheme.
However, over the mountains, T2 differences, even in wintertime, are more positive for
the MYNN3 scheme than over the corresponding areas in the SKINT plot. This again
could be due to greater sensible heat flux from the Eta surface-layer scheme. Apart
from differences in surface-layer formulation, dissimilarity in the implementation of hybrid mixing methods would also account for differences in simulated quantities. The
K-theory based ACM2 scheme uses first-order statistics to approximate higher moment
terms, while the MYNN3 scheme is a TKE-based second-order closure. The PBLH diagnosed by the ACM2 scheme is predominantly thinner (thicker) in summer (winter)
than the MYNN3. Adjustment of vertical gradients of momentum and moisture fluxes
with PBLH changes in the ACM2 would also affect WSPD10 and Q2, respectively. The
MYNN3 scheme mostly outputs greater Q2 and WSPD10 than the ACM2 scheme but the
extent to which PBL parameterization alone accounts for this outcome is confounded by

40

the use of alternative surface-layer schemes.
Simulation differences among the TKE-based PBL schemes (Figs. 2.7c-d) show the influence of local and nonlocal closures with the Eta surface-layer scheme. PBLH is greater
with MYJ than with MYNN3 except near the coast. As previously stated, the daytime intrusion of marine air counteracts growing instability arising from solar heating. For this
reason, higher PBLH with the MYNN3 scheme relative to that of the MYJ scheme does
not extend as far inland as it does at night (Figure G.1 in Appendix G). But instead of
a moisture profile that is more vertically homogeneous than those of the MYJ and UW
schemes, the MYNN3 scheme indicates less mixing (Fig. 2.8) due to how it implements
nonlocal transport. The MYNN3 scheme comprises a partial-condensation model for convective activity, in which the buoyancy effects of subgrid-scale clouds are considered using counter-gradient terms (Nakanishi and Niino 2009). The counter-gradient flux facility
in the MYNN3 scheme thus explains why it has greater (lesser) Q2 (WSPD10) values than
the outputs from the MYJ and UW schemes.
Recent revisions to mixing-length formulations in the MYNN3 scheme for which the
current code further decreases the downward mixing of momentum (Olson et al. 2019),
would also contribute to its generation of WSPD10 and Q2 values that are less biased from
station observations than those of the UW and MYJ schemes. Because of increased linking of the MYNN3 scheme to cloud processes and to other physics modules, such as radiation schemes (beginning from WRF version 3.8), feedbacks from simulated boundary
layer clouds are probably more influential on surface meteorological variables than with
other PBL schemes. For instance, the UW scheme outputs greater cloud coverage than
41

Figure 2.7 Spatial differences of averaged skin temperature (SKINT), ambient
temperature (T2), radiation reaching the surface (INRAD), surface wind speed
(WSPD10), ambient moisture content (Q2) and PBL heights at 1400 PST in summer and winter. MYNN3 – MYJ means values for MYJ subtracted from MYNN3,
etc.
the MYNN3 scheme (Figure G.1 in Appendix G), which ought to result in less radiation
reaching the surface with the UW scheme. Instead, incoming solar radiation is smaller
with the MYNN3 scheme, consequently its colder T2. The consideration of moist convection by the MYNN3 scheme is also evident in it having the biggest gradient for daytime

42

potential temperature (Fig. 2.9). It is reasoned that the facilitation of cloud condensation
in the MYNN3 scheme, enabling latent release and warming of the air aloft, alongside less
solar radiation to the surface and cooler near-surface temperature, causes a stronger temperature gradient. The least vertically-homogenized moisture gradient within the lowest
500 m for the MYNN3 scheme, may partly be due to this stability-enforcing feedback.
In the wintertime when absolute water vapor content is less, and convective activity is
reduced, MYNN3’s Q2 surplus compared to those of MYJ and UW in the valley, is diminished or in deficit. The vertical moisture gradient at this time is slightly steeper for
the MYJ scheme than for the MYNN3 scheme. The above analysis point to the seasonal
modulation of MYNN3’s nonlocal mixing.

2.4

Discussion and Conclusion

Performance scores of the various PBL schemes for surface meteorological parameters
are generally within, or close to values that are found in literature, either as NWP experiments (García-Díez et al. 2013, Hu et al. 2010, Zhang et al. 2013) or in the validation of
meteorological inputs to air quality models in mountainous regions (Cheng et al. 2012,
Hedley and Singleton 1997, Mues et al. 2018). It is not unusual for performance to vary
between models, model versions, grid resolutions, episode durations, and performance is
also subject to operating environments and needs. As an example, the evaluation benchmarks in Emery et al. (2001) were based on MM5 model application at 4-km grids to two
ozone episodes, each lasting a week, in Texas, US. Further, there usually is a minimum
amount of scatter or representativeness error that is inherent in meteorological downscal43

Figure 2.8 Averaged vertical profiles of simulations for mean water vapour mixing ratio in the lowest 1000 m at 0200 and 1400 PST in summer and winter
ing and which cannot be removed. Rife et al. (2004) estimated representativeness error
in a model grid-box size of 1.33 km2 to be about 1 m s-1 for near-surface wind speed in a
well-mixed boundary layer over complex terrain. This type of error would increase with
a smaller model grid size such as in the present study. Therefore, error and bias estimations previously presented reflect the status of the WRF model (version 4.0) and ERA5 to
simulate surface variables in the TKV without observation nudging.
Due to atmospheric circulations constrained by steep topography, the performances of

44

Figure 2.9 Averaged vertical profiles of simulations for mean potential temperatures at 0200 and 1400 PST in summer and winter. Computed gradients are for
the lowest 1500 m
the PBL schemes for wind direction were reasonably comparable. Therefore, the choice
of one scheme over another may not be critical to representing the relative contributions
of wind sectors to pollutant concentrations at specific locations. Performances for surface
air temperature and wind speeds, on the other hand, were more varied. Overpredicted
wind speeds in particular could cause an underestimation of ambient pollutant concen-

45

trations in areas near major emissions sources. Indeed, the finding that PBL schemes
were able to simulate surface wind with sufficient directional accuracy, yet poorly represent the observed speed, summarizes the challenge of validating wind fields at high
spatial resolution over heterogeneous terrain for input to coupled numerical models. As
was noted previously, dynamical downscaling aligned gridded coarse meteorology to the
valley and fjord channels, hence the general correspondence of modeled wind direction
with observations, while ambient wind speed biases reflected uncertainties that can arise
from smoothening of rugged relief in the simulations.
Model terrain height errors were a few to several dozen meters, and terrain data quality and resolutions may have affected individual simulations differently. It was noticed
however that performances for the YSU and SH schemes were similar in nearly all aspects, hence differences in concentrations of air pollutants that may arise from selecting
either scheme would probably be negligible. Both schemes are founded on nonlocal Kprofile assumptions, but the SH PBL scheme additionally addresses turbulent transport
in the ’gray zone’ (Wyngaard 2004), which is the range of grid spacing below which parameterizations have been developed, but above which the relevant processes are entirely
resolved. Since simulations at 1-km horizontal grid spacing as used in the present study
may be coarser than gray-zone resolution, the grid-size dependency of sub-grid scale,
nonlocal heat and momentum transport that is unique to the SH scheme, was probably
not optimized. In a real-case verification of the SH scheme by Shin and Hong (2015),
the simulation of convective rolls was structurally more distinct from those of the YSU
scheme at less than 1-km grid spacing. Because of the 0.281 ° spatial resolution of the

46

ERA5 forcing dataset, implementing a smaller grid size for differential modeling over the
TKV was not done since that would have resulted in more computing costs, and sharp
elevation gradients causing numerical instabilities in the simulations. Further, convective
periods in the TKV are restricted to daytime in the summer, and involve moist turbulence
which is not represented by the SH scheme.
Distinctions in model performance generally evolved with season changes. In summer
when Pacific frontal passages are rare and local thermally-driven circulations are enhanced, the diurnal signatures of individual PBL schemes were fairly unique. Particularly
for air pollutants that peak in the warm season such as ozone, the selection of one PBL
scheme over another may be significant. In winter, synoptic forcing is stronger, the daylight period is shorter and diurnal variability is weaker than in summer. More extended
decoupling of the atmosphere in wintertime diminished the influence of model configurations on near-surface variables, and all schemes performed poorly in this period. But
because atmospheric phenomena such as temperature inversions are not always surfacebased, and in the absence of radiosonde measurements, ensemble projections of the air
quality in the cold season rather than the output from just one PBL scheme may be better.
The UW PBL scheme that consistently produced the shallowest boundary-layer heights
may need further investigation to know how realistic such representations are for air
quality parameters in the region.
In essence, PBL schemes are highly integrated with other physics and dynamics options.
Some of the PBL schemes that were tested in this work can be mapped to other surfacelayer models in WRF. Specifically, the ACM2 PBL scheme can also be mapped to the
47

Pleim-Xiu surface-layer scheme, while the MM5 surface-layer scheme can be used for
both the MYNN3 and UW PBL schemes. As an example, compared to the Eta surfacelayer scheme, MYNN3 mapped to the MM5 surface-layer scheme produced larger negative biases for temperature, specific humidity and wind speed (Fig. 2.10). Where a turbulence closure scheme can link to more than one surface-layer model option, the practice
was to map TKE-based PBL schemes to the Eta surface-layer scheme, and first-order closure schemes to the MM5 surface-layer scheme. Figure 2.10 partially strengthens this
practice.
This study refrained from comparing modeled values to domain-wide averaged observation data. One shortcoming in several past studies was limited objective verification
of target meteorological parameters through time and non-clarity as to what qualifies as
good performance of mesoscale models in regions of strong surface heterogeneity. The
present approach has been to assess performance at discrete locations and to supplement
conventional metrics such as MB, RMSE and r scores with a modified IOA (bounded
between -1 and 1) that evaluates the strength of model predicted variability relative to
observed variability. As an example, summertime RMSE and r values for wind speed at
the Onion Lake and Whitesail stations were comparable and within performance benchmarks. However, while IOA values for Onion Lake were good, IOA for Whitesail was
poor and attributable to errors (overestimation) that are much larger than the variability
in observations. This is in spite of the fact that the Whitesail station is located nearer the
shoreline and should ideally match better with strong breezes that were simulated. It
is important to consider such error disparities in downscaled meteorology over complex

48

coastal areas when simulating long-term chemical concentrations.

Figure 2.10 Summer and winter hourly mean bias plots of T2, Q2 and WSPD10
for Eta similarity (solid lines), and revised MM5 similarity (dashed lines) surfacelayer schemes respectively, mapped to MYNN3 PBL scheme.
49

Chapter Three
Evaluation of CMAQ modeling
sensitivity to planetary boundary layer
parameterizations for gaseous and
particulate pollutants over a fjord valley
(Also published at https://doi.org/10.1016/j.atmosenv.2020.117607 in Atmos. Environ.)

Abstract
Three-dimensional chemical transport models are useful for spatial and temporal analyses of outdoor air quality. However, the suitability of boundary-layer parameterizations
for air pollution modeling over deep, coastal valleys has seldom been tested. An evaluation of the Community Multiscale Air Quality (CMAQ) model performance for five planetary boundary-layer schemes (PBL) with the Weather Research and Forecasting (WRF)
meteorological driver was conducted at 1-km horizontal resolution for fine particulate
50

matter (PM2.5 ), sulfur dioxide (SO2 ) and nitrogen dioxide (NO2 ) over the Terrace-Kitimat
Valley. The top-ranked schemes were Mellor-Yamada-Nakanishi-Niino Level 3 (MYNN3)
for PM2.5 and Mellor-Yamada-Janjic for NO2 . Both schemes ranked high for absolute SO2
levels, but the MYJ and Asymmetric Convective Model, version 2 (ACM2) schemes qualitatively emulated peak summertime diurnal concentrations in the near field of elevated
point sources. Greater nighttime SO2 concentrations with MYNN3 and Yonsei University
PBL schemes, in less agreement with station monitoring 8 km downwind of emissions
from tall stacks, suggested sustained pollutant mixing and downward transport within
the nocturnal boundary layer. Consequently, for these two schemes with representations
of nonlocal mass flux transfers between model layers, inland penetrations of pollutant
plumes were farther than those of ACM2, MYJ, and University of Washington schemes.
For NO2 and PM2.5 that mainly discharged passively from fugitive, ground-level sources,
hence are less accurately quantified than SO2 emissions, the fully local MYJ, and semilocal MYNN3 PBL schemes more reasonably reproduced peak season concentrations than
other schemes. It is concluded that for air pollution modeling in rugged, remote areas, the
mode of pollutant emissions is important for the choice of a PBL scheme. PM2.5 was consistently underestimated by the various PBL schemes, and aspects for improving CMAQ
simulations for a complex environment are discussed.

3.1

Introduction

Advanced computations coupled to numerical weather prediction (NWP) models involve
several physics modules for convection, cloud formation, boundary-layer turbulence, ra51

diation and land-surface processes, that control circulations at various spatial scales, and
are important for the evolution of near-surface meteorological parameters. Of these, the
representation of subgrid-scale turbulence, which determines the stability of the lowest
portion of the atmosphere, is essential to environmental applications, such as simulating air quality. Especially for areas with sparse pollution observational networks, the
choice of a planetary boundary layer (PBL) scheme, whose optimality may depend on
weather regimes and phenomena, can affect the prediction accuracy of a chemical transport model.
Numerical simulations with horizontal grid of several kilometers are often insufficient
for addressing more localized air quality concerns, hence many have researched the influence of PBL parameterizations with finer (1–2 km) horizontal meshes. Li et al. (2019)
evaluated two different PBL schemes in the Weather Research and Forecasting model
coupled with Chemistry (WRF-Chem) and a Large-Eddy Simulation (LES) PBL model
over the Baltimore-Washington region of the US. They found that the LES model afforded
more realistic reactive chemistry for secondary pollutants formation. Pérez et al. (2006)
analyzed air chemistry predictions from three PBL schemes over Barcelona, Spain and related daily maximum pollutant concentrations to mixing heights, temperature and wind
speeds generated by each scheme. Parra (2018) assessed six PBL schemes for Cuenca,
Ecuador and found that two schemes were satisfactory for short-term air quality observations. However, the majority focus of past studies has been for short durations and in
areas with high background pollution, whose conclusions may not be valid for pristine
regions.

52

The focus of this chapter is on the application of the Community Multiscale Air Quality
(CMAQ) model (Appel et al. 2018) over the TKV. This model will be run at a spatial resolution not often used (1 km), and for pollutants that are routinely monitored within the
airshed namely fine particulate matter (PM2.5 ), sulfur dioxide (SO2 ) and nitrogen dioxide (NO2 ). The objectives of this study are to (i) evaluate the overall performance of the
CMAQ model across various monitoring stations in the TKV, (ii) identify the most suitable PBL scheme(s) for simulating air pollutants, (iii) examine the extent to which major
air pollutants are sensitive to PBL parameterizations, (iv) suggest where improvements
to chemistry-transport modeling may be needed. The remainder of this chapter is presented in three sections. The modeling approach, measurement data, and performance
evaluation indices are described in the next section. Section 3.3, conducts a comparative
analysis of results. Section 3.4 discusses the spatio-temporal context of biases and uncertainties relating to changes in PBL schemes and concludes the chapter.

3.2

Methods and data

3.2.1 PBL schemes and experiments set-up
The downscaled meteorology from five PBL parameterizations schemes namely: MYNN3,
MYJ, UW, YSU, and ACM2 in Chapter 2 were deployed. The Meteorology-Chemistry
Interface Processor (Otte and Pleim 2010), version 4.3 was used to create two emissions
grids from the WRF outputs (Fig. 3.1, Table 3.1). The Sparse Matrix Operator Kernel Emissions (SMOKE) modeling system, version 2.6 (Carolina Environmental Program 2009)

53

was used to prepare anthropogenic emissions for input to CMAQ. For each PBL scheme,
emissions were modeled for the 5 km and 1 km grids. The British Columbia portion of
Canada-wide emissions inventory, updated to 2017 for point emissions in the TKV was
used. Stack emissions were processed as elevated point sources. Unique emissions in
point source format, such as from aircraft at airports, were treated as ground sources.
Spatial allocation of area (e.g. agricultural, road and construction dust) and mobile (onroad vehicles and off-road equipment) emissions with SMOKE was implemented with
124 gridding factors (surrogates). Among the gridding surrogates were population and
housing, road classes and length, waterways, and mineral extraction, all resulting in allotment of provincial emissions to grid cells that intersect respective domains. Temporal
splitting of area emissions with SMOKE into hourly time steps required by CMAQ was
done based on local knowledge of activity patterns at diurnal, weekly and monthly intervals. Other important files necessitated by the long (1-year) simulation duration and
geographical settings of our study, such as a seasonal switch file (for biogenic emissions),
a land-sea mask (for sea salt), and a land-use grid (for agricultural dust emissions), were
generated with either SMOKE or spatial allocator tools. The CMAQ chemical-transport
model, version 5.2 (Appel et al. 2018) was run from 00 UTC 20 December, 2016, to 00 UTC
2 January, 2018. To implement a nesting mode with CMAQ runs (see Fig. 3.1), simulations
were performed in two sets. The first set was for generating boundary concentrations
for each PBL scheme from their respective coarse (5-km) emission grids. Thus, boundary
conditions over the local (TKV) domain varied for each hour, as opposed to directly using
clean, invariant North American profiles that are packaged with CMAQ. For the second

54

Figure 3.1 Telescopic grid nesting set-up for WRF-CMAQ modeling over the TKV.
Left: 2-nests level of WRF domains set-up. The red box is the TKV’s meteorological domain. Right: 1-nest level of SMOKE emissions/CMAQ domains set-up.
DOM_BC (in orange), smaller than the white bounds of d02 on the right, specifies the areal extent of emissions from which boundary concentrations to TKV’s
air quality modeling domain (in black) were derived. The red box remains TKV’s
meteorological domain.
set of simulations which was for the 1-km grid, the first 12 days were model spin-ups and
discarded, retaining only hours in 2017. All model runs used the Aero6 aerosol module
(Appel et al. 2018) and carbon bond 05 for chemical speciation (Yarwood et al. 2005). All
simulations were off-line (that is without feedback from CMAQ to meteorological fields).
However, natural emissions from vegetation (biogenic), sea salt and wind-blown dust
were calculated and treated in-line by CMAQ. Although the same number of layers as
the input meteorology (Table B.2 in Appendix B) was used in CMAQ modeling, outputs
were processed for only the layer nearest the surface since no upper-air pollution moni-

55

Table 3.1 Meteorological and emissions domain sizes and attributes. Emissions
domains, with one level of nesting only, are the same sizes as CMAQ domains
Domains (dimensions)

Meteorology

SMOKE Emissions/ CMAQ Concentrations

Parent (W-E × N-S)

100 ×100

-

Parent grid center

54.200 ◦ N, 128.600 ◦ W

-

Parent grid resolution

25 km

-

1st nested (W-E × N-S)

121 × 121

40 × 60

1st nested grid center

53.850 ◦ N, 128.795 ◦ W

54.200 ◦ N, 128.600 ◦ W

1st nested grid resolution

5 km

5 km

2nd nested (W-E × N-S)

101 × 121

36 × 106

2nd nested grid center

54.223 ◦ N, 128.640 ◦ W

54.200 ◦ N, 128.600 ◦ W

2nd nested grid resolution

1 km

1 km

toring exists in the study area.

3.2.2 Air quality monitoring data
The most challenging air quality problems involve complex multi-pollutant interactions
and coupling between atmospheric chemistry and dynamics. The performances of the
PBL schemes for PM2.5 , SO2 , and NO2 simulations are of interest. These pollutants are
directly released from various sources, but some portion of PM2.5 can be formed from
reactions of other pollutants. The pollution monitoring network in the valley consists of
five unevenly-spaced stations (Fig. 1.1, Table 3.2) categorized as urban, semi-urban, industrial and rural. Most stations do not monitor all three pollutants, for example, NO2
56

Table 3.2 Air quality monitoring stations in the Terrace-Kitimat Valley. All are
ground-based and within the valley.

Station

Latitude Longitude Elevation

Station Description

Air pollutant

0.0

Rural

SO2

128.702

11.0

Industrial (Kitimat)

PM2.5

54.054

128.671

18.0

Semi-urban (Kitimat)

PM2.5

Whitesail

54.067

128.639

94.0

Semi-urban (Kitimat)

SO2

Terrace

54.522

128.608

81.0

Urban

PM2.5 , SO2 , NO2

◦N

◦W

(m)

Kitamaat Village

53.830

128.830

Haul Road

54.029

Riverlodge

is only measured at Terrace. This station is operated by the BC Ministry of Environment
and Climate Change Strategy (BCMOECS). The others are operated and maintained by
an industrial permittee, reporting to the BCMOECS air quality network database (https:
//envistaweb.env.gov.bc.ca/DynamicTable2.aspx?G_ID=327). Sensors in use are those
recognized as meeting the United States Environmental Protection Agency’s Class III
FEM monitors (USEPA 2016) standards. For each pollutant, only stations with at least
80 % completeness in valid hourly values are used for model evaluation. Low-level concentrations were not excluded from measurement data.

3.2.3 Statistical evaluation indices
Statistical measures are used to quantitatively assess the predictive skill of a model. Modeled values for the grid cells coincident with monitoring stations were retrieved. Several
authors (e.g. Chang and Hanna 2004, Emery et al. 2017) recommend the use of multiple
57

fitness measures for air quality model evaluations as no single measure is universally applicable to all conditions. The fitness measures used in this work reflect a mix of accuracy
measurements for paired data and are listed and defined in Table 3.3.
Performance evaluations are conducted by pairwise comparison of the diurnal profiles
of measured, and modeled concentrations, for peak seasons. Where they exist, model
performance criteria (Boylan and Russell 2006, Emery et al. 2017) are also used to rank
statistical scores. An important caveat is that existing benchmarks are mostly from studies conducted with coarser grid resolutions, over flatter, more homogeneous terrains than
the TKV area (see Emery et al. 2017), hence, may not be fitting for the present study.
Table 3.3 Fitness measures based on paired hourly means (0, 1, .., 23). Formulas
are defined in Appendix C. FAC2 and NMB criteria are from Chang and Hanna
(2004). Performance criteria for PM2.5 are based on Boylan and Russell (2006) and
Emery et al. (2017).
Performance
Measure

Abbreviation
Ideal

Recommended

FAC2

All

≥ 50%

Mean Bias

MB

0 µg m−3 / ppb

Normalized Mean Bias

NMB

0

±0.3

Mean Fractional Bias

MFB

0

±0.6 (for PM2.5 )

NSD

1

Number of predictions within
a factor of 2 of observations

Standard deviation of modeled values
divided by those of observations

58

3.3

Results and analyses

3.3.1 Ambient air quality
Ambient concentrations in pollutant peak seasons are used for analyzing CMAQ simulations. For each station and pollutant, peak season was defined as the 3-month seasonperiod in which the mean hourly pollutant concentration is highest. At all three stations
(Table 3.4) the peak season for ambient SO2 is in summer (June to August). The peak for
NO2 in Terrace is in winter (December to February). For PM2.5 , the peak season for Haul
Road and Terrace stations are in autumn (September to November) while that for Riverlodge is in winter.
Table 3.4 Seasonal mean concentrations of air pollutants at model validation stations. Peak seasons are those with the highest seasonal mean of a pollutant (in
italics).
Pollutant

PM2.5
(µg m−3 )

SO2 (ppb)

NO2 (ppb)

Station

Spring

Summer Autumn

Winter

Annual

Haul Road

2.9

4.0

4.4

3.3

3.6

Riverlodge

3.6

3.8

4.6

5.4

4.4

Terrace

4.0

4.6

10.4

6.9

6.4

Kitamaat

0.31

0.31

0.29

0.27

0.29

Whitesail

0.47

0.61

0.28

0.27

0.41

Terrace

0.51

0.72

0.27

0.32

0.45

Terrace

2.0

1.1

2.5

4.7

2.7

Meteorological and local factors account for differing PM2.5 peak seasons at monitor loca59

tions. Winds are weaker in autumn than during summer and winter seasons. Especially
at the Terrace station, where the frequency of nighttime calms is 30 % (Fig. 3.2), autumn
PM2.5 average is more than twice the summertime value. Northerly wind sectors are
slightly more frequent than southerly winds in autumn, hence the northerly PM2.5 sectors are the more prominent at Terrace, but at the Haul Road station in Kitimat, southerly
sectors dominate PM2.5 levels. Actual PM2.5 concentrations (Fig. 3.3) are also similar
among pollutant sectors at Terrace, unlike at Haul Road where the northerly sectors are
much cleaner than the southerly sectors. Urban releases (e.g. from road traffic, domestic
emissions, commercial outlets) are to all directions around the Terrace monitor, consequently similar qualitative influence. The Haul Road station is located in the industrial
area of Kitimat, just north of a large aluminum smelter, but its precinct is composed of
natural vegetation (uninhabited). Thus, the anthropogenic signal is more conspicuous for
the Haul Road station, even though ambient PM2.5 at that location is much less than at
the Terrace station.
Advection of precursor smelter emissions by predominantly southerly winds in the warm
season would account for summertime SO2 peak at Whitesail and Terrace stations. The
primary source of SO2 in the Kitimat area and the TKV as a whole is the smelter site
which emitted 27 tonnes of SO2 per day in 2017 (ECCC 2020). The strong south-westerly
component at the Kitamaat Village station suggests intermittent, southward deflection of
SO2 plumes, enough to produce mean concentrations that are comparable to other seasons when it is less upwind of the smelter source. Kitamaat Village however, is situated
adjacent to maritime waters and stronger breezes there than inland, disperse pollutants

60

better.

Figure 3.2 Wind roses as wind frequency proportions at Terrace and Whitesail
stations for summer, autumn and winter periods. Radial increments are 15 %,
with the largest radial at 45 %. Mean wind speeds, and percentage of calms
(wind speeds < 0.5 m s−1 ) are denoted by m and c respectively. On the right,
are corresponding diurnal temperature profiles.

61

Figure 3.3 Pollution roses as the relative contribution of wind sectors to mean
peak season concentrations of PM2.5 , SO2 (summer), and NO2 (winter) at monitoring stations. Riverlodge’s PM 2.5 peak season is winter while it is in autumn
for the other stations. Radials (%), are annotated on the wedges. Also shown are
spatial distributions of hourly emissions

62

3.3.2 Model performance evaluations
The adequacy of CMAQ to simulate observed air quality variables can be deduced in a
relative sense, by comparing between the various PBL schemes, across monitor locations,
for specified averaging periods and with model performance benchmarks (where applicable), using several statistical measures. Evaluations are conducted for each pollutant
and are also analyzed graphically.

SO2
SO2 simulation biases are according to location; the majority of PBL schemes overestimate
Whitesail concentrations, and underestimate concentrations at Kitamaat Village and Terrace stations (Table 3.5). Across stations, NMB ranged -0.9 to +1.1, and MFB ranged -1.7
to +0.7. Out of 15 evaluations, FAC2 was deemed satisfactory in five, where all the PBL
schemes are represented, except the UW scheme. NSD values closest to 1 at Whitesail
suggest the best emulation of observed SO2 variability by CMAQ. Both Whitesail and
Terrace are downwind of the major SO2 elevated source in the TKV during summer, but
Whitesail is nearer. The diurnal peak around 1000 PST (Fig. 3.4a), is more delayed in
modeled profiles for Terrace. Differences in emission mode, whereby fugitive, ground
emissions are common in Terrace, likely accounts for discrepancies in the timing of peak
concentrations. As the ground warms during the day and PBL deepens up to heights of
buoyant SO2 plumes from the smelter source, more pollutants are brought into contact
with surface air, leading to a quicker effect on simulated ambient levels for Whitesail.
Indeed, PBL height (PBLH) maxima at Whitesail are more coincident with daytime peak

63

concentrations than at the Terrace station.
Table 3.5 Model performance statistics for SO2 for various PBL parameterization
schemes for the peak summer season. Values in italics indicate best performing
locations per measure or schemes per location. Bolded values are within suggested benchmarks, only applicable to FAC2 and NMB.

Measure

Station

MYJ

MYNN3

UW

ACM2

YSU

Satisfactory
(# out of 15)

0.5 ≤ FAC2

Kitamaat

11

19

0

4

3

≤ 2.0 (# hours

Whitesail

16

8

5

15

9

out of 24)

Terrace

9

22

0

9

18

Kitamaat

-0.15

-0.11

-0.25

-0.20

-0.20

Whitesail

0.21

0.65

-0.35

0.42

0.66

Terrace

-0.41

-0.19

-0.64

-0.39

-0.33

Kitamaat

-0.49

-0.34

-0.80

-0.66

-0.65

Whitesail

0.34

1.06

-0.58

0.69

1.07

Terrace

-0.58

-0.27

-0.89

-0.54

-0.46

Kitamaat

-0.65

-0.41

-1.34

-0.97

-0.95

Whitesail

0.29

0.69

-0.82

0.51

0.70

Terrace

-0.81

-0.31

-1.62

-0.74

-0.60

Kitamaat

0.57

0.40

0.21

0.40

0.42

Whitesail

1.28

0.87

0.96

1.44

1.06

Terrace

0.53

0.63

0.27

0.60

0.32

MB (ppb)

NMB

MFB

NSD

5

1

Daytime MYNN3, YSU and ACM2 outputs close to the stack emissions (i.e. at the White64

sail station) are mostly comparable. Probably due to generating lower wind speeds that
are more in agreement with observations, concentrations maxima for these PBL schemes
are greater than for the MYJ and UW schemes. YSU and ACM2 wind speed bias profiles
are similar, which is likely because they are coupled to the same surface layer scheme
(MM5-similarity). The other (TKE) PBL schemes are coupled to the Eta surface layer
scheme, but it is evident that the general overestimation of onshore wind speeds is reduced in the MYNN3 scheme. Apart from having nonlocal convective mixing as with
the YSU and ACM2 schemes, the MYNN3 scheme implements counter-gradient momentum diffusion (Olson et al. 2019) that mitigates output of excessive wind speeds. At the
Terrace station, where strong winds are output from WRF, daytime MYNN3 wind speed
bias approximates those of YSU and ACM2 schemes, thereby partly contributing to the
MYNN3 scheme having the least SO2 concentration bias.
Considerable nighttime overestimates with the MYNN3 and YSU schemes at Whitesail alludes to active, nonlocal transport in these schemes that are not inhibited by increased atmospheric stability. While the turbulence closure for YSU is purely nonlocal, the MYNN3
scheme includes a nonlocal turbulence production component for buoyancy effects of
cloud-top cooling (Nakanishi and Niino 2009, Olson et al. 2019). The greater cloud cover
amounts with MYNN3 (not shown), would enhance these effects. The strength of the
vertical diffusion in a stable boundary layer is regulated by the buoyancy length scale
and enhanced mixing with MYNN3 causes more downward transport of SO2 from aloft.
Mixing-length formulations for both MYJ and UW schemes are strictly local, hence their
limited capacity for pollutant transport from upper layers to the surface, despite slightly

65

Figure 3.4 (a.) Comparisons of modeled peak season (summer) SO2 profiles at
Terrace and Whitesail stations with observations (OBS), and surface meteorological variables namely wind speed bias, air temperature bias, and PBL height
(PBLH). Thin, black horizontal lines are zero bias axes. (b.) Vertical profiles of
averaged potential temperatures at selected times, as atmospheric stability proxies for the various PBL schemes. The more the vertical homogeneity of potential
temperatures, the less stable the atmosphere. Up to 2.8 ◦ C potential temperature
change within the lowest 500 m is simulated at 0200 PST, Terrace station.
more stable atmospheres than the YSU scheme (Fig. 3.4b). The ACM2 scheme does not
produce as much nighttime concentration as the YSU scheme. For neutral, and stable
conditions that dominate at night, the ACM2 scheme deactivates nonlocal transport of

66

scalar fluxes (Pleim 2007). Vertical exchanges via a purely local component thus result in
pollutant concentrations comparable to the MYJ scheme.

PM2.5
PM2.5 is underestimated (Table 3.6) in all simulations. Across stations, NMB ranged -0.8
to -0.5, and MFB ranged -1.4 to -0.5. Notably, modeled concentrations are not within a factor of 2 of observations (FAC2). In fact, none of the suggested performance benchmarks
is achieved. For individual locations, the best performance is at Haul Road, apparently
due to relatively low concentrations that reduce margins for bias.
Ambient PM2.5 is least between 1000–1500 PST (Fig. 3.5). This should coincide with
deeper PBL in this period but only the MYNN3 and MYJ schemes reasonably represent
the inverse relationship between the daytime dip in station measurement and modeled
PBL rise. Concentrations profiles for the other schemes are flatter, even when their PBLs
rise less. PBLH variation influence on modeled concentrations, it seems, depends on the
absolute height of the PBL itself, which also varies with season. At the Riverlodge station
in Kitimat, where peak PM2.5 is during winter, PBLs are mainly thinner and less variable
than the autumn Terrace profiles, also reflecting in weaker, diurnal variability of modeled concentrations. But it is also known that the pollution intensity at Riverlodge, where
residential wood heating probably accounts for the wintertime PM2.5 peak, is less. The
Kitimat area is more ventilated, with more evenly distributed diurnal PM2.5 levels. Less
modeled concentrations at Riverlodge in comparison to Terrace, alludes to reasonable dynamical consistency of CMAQ simulations.

67

Table 3.6 Model performance statistics for PM2.5 for various PBL parameterization schemes in winter (at Riverlodge) and autumn (Haul Road/Terrace) peak
seasons. Values in italics indicate best performing locations per measure or
schemes per location.

Measure

Station

MYJ

MYNN3

UW

ACM2

YSU

Satisfactory
(# out of 15)

0.5 ≤ FAC2

Haul Road

0

7

0

0

0

≤ 2.0 (# hours

Riverlodge

0

0

0

0

0

out of 24)

Terrace

0

0

0

0

0

Haul Road

-2.71

-2.35

-2.98

-2.73

-2.67

Riverlodge

-3.52

-3.35

-3.94

-3.70

-3.42

Terrace

-7.42

-7.15

-8.53

-8.06

-7.91

Haul Road

-0.62

-0.53

-0.68

-0.62

-0.61

Riverlodge

-0.65

-0.62

-0.73

-0.69

-0.63

Terrace

-0.72

-0.69

-0.82

-0.78

-0.76

Haul Road

-0.89

-0.73

-1.03

-0.90

-0.87

Riverlodge

-0.97

-0.90

-1.15

-1.05

-0.93

Terrace

-1.12

-1.05

-1.40

-1.27

-1.23

Haul Road

0.17

0.54

0.10

0.20

0.19

Riverlodge

0.20

0.31

0.06

0.09

0.24

Terrace

0.22

0.31

0.07

0.12

0.12

MB (µg m−3 )

NMB

MFB

NSD

0

0

0

Slight changes in the order of concentrations from the various PBL schemes, may owe to
such spatial differences in-situ emissions. At Riverlodge, YSU simulates slightly higher

68

Figure 3.5 (a.) Comparisons of modeled peak season PM2.5 profiles at Terrace (in
autumn) and Riverlodge (in winter) with observations (OBS), and surface meteorological variables namely wind speed bias, air temperature bias, and PBL
height (PBLH). Thin, black horizontal line is zero bias axis. (b.) Vertical profiles of
modeled potential temperatures at selected times, as atmospheric stability proxies with the various PBL schemes. The more the vertical homogeneity of potential
temperatures, the less stable the atmosphere. Up to 3.0 ◦ C potential temperature
change within the lowest 500 m is simulated at 0200 PST, Terrace station.
concentrations than MYJ, but less at the Terrace station. Distinct from SO2 , PM2.5 emissions in the TKV are mostly from ground-level sources. Because YSU-predicted PBLH

69

at peak PM2.5 hours are roughly the same (about 350–400 m) at the two stations, smaller
PM2.5 emissions around Riverlodge would entail weaker, vertical concentration gradient
to the top of the PBL, therefore less nonlocal transport of pollutant away from the surface.
Cooler temperatures (by ∼ 1 ° C) with the YSU scheme, possibly reducing the volatility of
particulate constituents, may also have contributed to higher modeled PM2.5 than that of
MYJ at Riverlodge.
If nonlocal transport of PM 2.5 to higher layers is more effective at Terrace for YSU, then
MYNN3 should produce less concentration at this station than the purely local MYJ
scheme. Instead, the MYNN3 scheme not only outputs the highest PM2.5 but also better mimics diurnal levels than MYJ. With MYNN3, the evening peak is higher than the
morning peak, in better agreement with the observation. A likely reason for higher output with MYNN3 than MYJ is the facility of counter-gradient fluxes that applies to any
scalar quantity, in the MYNN3 PBL formulation (Olson et al. 2019). Further, the activation of CMAQ’s wind-blown dust module for all simulations may have contributed more
fine particulates in model layers above the surface for counter-gradient transport with the
MYNN3 scheme.

NO2
Similar to PM2.5 , ground sources (motor vehicles, rail transportation, and other mobile
emissions), account for NO2 at the Terrace station which is the sole, continuous monitoring location for ambient nitrogen oxides in the TKV. Model underestimation of wintertime
levels (Table 3.7) is also evident. Across PBL schemes, NMB range -0.9 to -0.4 while MFB is

70

between -1.7 and -0.5. NSD values, as indicators of variability of modeled concentrations,
are impressive for the MYJ and MYNN3 schemes, but MYJ has the best FAC2 rating.
Table 3.7 Model performance statistics for peak winter NO2 with various PBL
parameterization schemes at the Terrace station. Values in italics indicate best
performing per measure. Bolded values are within suggested benchmarks, only
applicable to FAC2 and NMB.

Measure

MYJ

MYNN3

UW

ACM2

YSU

Satisfactory
(# out of 5)

0.5 ≤ FAC2 ≤ 2.0
16

11

0

0

1

MB (ppb)

-2.10

-2.64

-4.20

-3.54

-3.09

NMB

-0.45

-0.57

-0.90

-0.76

-0.66

MFB

-0.58

-0.79

-1.64

-1.23

-0.99

NSD

1.17

0.84

0.17

0.47

0.62

1

(# hours out of 24)

0

Diurnal pattern replication by the various PBL schemes (Fig. 3.6a) indicates reasonable
success in matching emissions with anthropogenic origins. Simulations reproduce the
morning and evening peak concentration periods, typical of urban vehicular traffic source
of NO2 . Emissions underestimation may have caused the low biases in modeled concentrations particularly for overnight hours, but high hourly wind speed biases seem to
contribute as well. Observed wind speeds at this station, situated in an area ringed by terraces, and where the cross-wise flow of the Skeena River channel relative to the TKV (see
Fig. 1.1) complicates the atmospheric environment, averages 1.5 m s-1 in winter. Model
overestimation of wind speeds by about 3.6 m s-1 on average, signifying greater than pre71

vailing outdoor ventilation, thus diminishes the skill of current PBL schemes to quantify
pollutant concentrations around this deep-lying observatory.

Figure 3.6 (a.) Comparisons of modeled peak season NO2 profiles profiles at Terrace with observations (OBS), and surface meteorological variables namely wind
speed bias, air temperature bias, and PBL height (PBLH). Thin, black horizontal
line is zero bias axis. (b.) Vertical profiles of modeled potential temperatures at
selected times, as atmospheric stability proxies with the various PBL schemes.
The more the vertical homogeneity of potential temperatures, the less stable the
atmosphere. Up to 3.8 ◦ C potential temperature change within the lowest 500 m
is simulated at 0200 PST, Terrace station.

Better agreement of MYJ output with observation than the other PBL schemes differs
from peak SO2 and PM2.5 concentrations for which MYNN3 values were greater. Atmo72

spheric residence times and abundance, chemical reactions, etc. vary between pollutants
and can change between seasons. Daylight period is shortest in the winter season, and
reduced solar heating of the ground affects the thermal structure of the boundary layer.
The greater vertical variation in potential temperatures (Fig 3.6b) compared to other seasons (Figs. 3.4b, 3.5b), points to increased atmospheric stability for all PBL schemes, but
the MYJ scheme is the most stable. This suggests that for modeled concentrations in the
cold season, atmospheric stability is more important than differences in surface meteorological variables such as wind speed and air temperature amongst PBL schemes.
An interesting aspect of NO2 levels from the various scheme relates to modeled PBLH.
Figure 3.6a informs that the order of PBLH values (MYJ > MYNN3 > YSU > ACM2 >
UW) is similar to that for NO2 concentrations. As stated in Chapter 2 (subsection 2.2.1),
each turbulence closure scheme diagnoses PBLH based on certain criteria, hence their
nominal values are essentially not comparable. Perhaps, their utilization within CMAQ
could be for estimating vertical pollutant gradients between concentrations in the surface
layer, and virtually zero concentrations in the free atmosphere above the PBL. The higher
the PBLH, the weaker this gradient, resulting in lesser/slower pollutant transfer from the
surface to higher layers.

3.3.3 Annual evaluations and spatial concentration differences
Many air quality standards are based on top percentile concentrations of monitored pollutants. Therefore, the ability of model simulations to reproduce such levels at an annual
scale is important to what confidence can be placed on their use for supplementing in-situ

73

monitoring. Figure 3.7 shows cumulative mean distribution plots of all hourly measurements and model outputs with the various PBL schemes. For SO2 (Fig. 3.7a) and NO2
(Fig. 3.7c), the mean of top quantiles of MYNN3, MYJ, YSU, and ACM2 PBL schemes are
mainly the same orders of magnitude as observation data. These quite differ from PM2.5
where except for MYNN3, even the annual means of station measurements surpasses the
concentrations for the top quantiles. This implies that PM2.5 model biases in other seasons
hardly compensate autumn/winter underestimations.
Modeled annual mean concentrations of air pollutants over the TKV domain is shown in
Fig. 3.8. Spatial gradients are identifiable with pollutant source types, whereby the elliptical pattern for SO2 elevated point sources, differs from the more concentric contours
for PM2.5 and NO2 . The discrete nature of SO2 emissions is evident in the maximum of
average concentrations per PBL scheme (Table 3.8) that are 1-2 orders of magnitude above
the annual station averages in Table 3.4. The coefficient of variation (CoV) per pollutant,
however, does indicate that there is greater variability amongst PBL schemes with simulating these values for NO2 and SO2 than there is for PM2.5 . Smallest CoV with PM2.5 ,
thus, least concentrations sensitivity, suggests that the choice of a PBL scheme for an annual projection is more crucial for gaseous pollutants than for PM2.5 .
The farthest travel of Kitimat stack emissions is by the MYNN3 scheme. With this PBL
scheme, 0.5-1 ppb SO2 contours (Fig. 3.8) extend to Terrace, compared to about threequarters the distance with the fully nonlocal YSU PBL scheme. Innate approaches to
nonlocal turbulent mixing in these schemes will account for these differences. The YSU
scheme relies on entrainment at the top of the PBL (Hong et al. 2006). As pollutant concen74

Figure 3.7 Annual cumulative mean distributions of modeled, and actual pollutant concentrations at validation stations. The concentration at each octile/sextile
is the mean of all hourly concentration values for it. Quantiles are according to
the number of unique values in station data. Dashed vertical lines are the mean
concentrations of station measurements in 2017.

75

Table 3.8 Highest simulated concentration of pollutants within the TKV modeling domain for various PBL schemes. CoV per pollutant, is the ratio of the mean
to the standard deviation of values for all the schemes.
Pollutant

MYNN3

MYJ

YSU ACM2

UW CoV

PM2.5 (µg m−3 )

7.3

7.6

5.9

5.8

6.8

0.11

SO2 (ppb)

44.0

22.7

23.5

19.6

4.4

0.55

NO2 (ppb)

4.2

3.4

2.5

2.2

2.2

0.27

trations approach zero in upper atmospheric layers, this process would less contibute to
pollutant fluxes at lower layers, compared to the counter-gradient process in the MYNN3
scheme. Inland penetration of SO2 for the hybrid ACM2 scheme is less than the MYNN3
and YSU schemes. Indeed, that ACM2 yields an SO2 pattern that is comparable to the
fully local, MYJ scheme, suggests dominance of stability conditions more permitting of
layer-by-layer (local) transport.
Pollutant levels are least with the UW, consequently the least spatial variability of modeled concentrations. Especially for SO2 , aerial gradient barely exists; the concentration
hotspot is contiguous with the Kitimat industrial source. Because of relatively low boundarylayer heights modeled by the UW scheme, coupled with hot, forceful SO2 ejection from
the smelter source, much of it probably is represented as directly transferred to model
layers above the PBL, that is, to the free atmosphere. The UW scheme was designed for
marine stratocumulus-capped boundary layers (Bretherton and Park 2009) that often are
decoupled, consequently its consistent simulation of shallowest PBLs. Passive groundlevel PM2.5 and NO2 discharges within the PBL would also be subject to a strong vertical

76

Figure 3.8 Spatial distribution of mean annual concentrations of PM2.5 , SO2 and
NO2 with various PBL schemes.

77

concentration gradient that enables rapid losses from the surface to layers aloft. Rather
than mix, the UW scheme diffuse chemical species away from the surface layer (Bretherton and Park 2009), hence small outputs for all three air pollutants.

3.4

Discussion and conclusion

The respective NMB and MFB ranges for peak season air pollutant modeling with different PBL schemes were -0.9 to -0.4 and -1.7 to -0.5 for NO2 ; -0.9 to -1.1 and -1.7 to
+0.7 for SO2 ; -0.8 to -0.5 and -0.7 to -1.4 for PM2.5 . The lesser bias values per pollutant
are close to, or within the range, that is reported in literature for CMAQ modeling over
long durations (Hu et al. 2016, Kota et al. 2018, Syrakov et al. 2016). It needs mentioning, however, that much of recent CMAQ model demonstrations, including the abovementioned studies, are at ’super-regional’ scales (>1000 km) and similarities or differences in performance are not generalizable. Further, recommended benchmarks for regional atmospheric chemistry model performance (Boylan and Russell 2006, Emery et al.
2017) derive mainly from studies over US areas with higher background pollutant levels
and more local sources. Apart from differences in grid resolutions and geography, there
are differences in meteorological forcing data, choice of chemical mechanisms, number of
validation stations, averaging periods and even model versions. Simulations in this study
were performed at high resolution (1 km grid spacing) and used updated meteorologychemistry modules, hence reflect the current status of the WRF-CMAQ modeling system
to emulate key air pollutants over a sparsely populated fjord valley.

78

While compensation of model biases across seasons may have led to better capture of top
percentile concentrations for SO2 than for PM2.5 , peak season performances of individual
pollutants highlighted temporal limitations of present-day PBL parameterizations. PBL
schemes are usually developed and verified under warm, dry atmospheric conditions
(e.g. Hong et al. 2006, Nakanishi and Niino 2009). The summer peak for SO2 coincides
with a period of enhanced mesoscale events such as land-sea breezes, slope-valley winds,
thus are better linked to in-situ contaminant distribution. The diurnal peak occurring at
mid-day also suggested the important role of fumigation, caused by convective downward mixing of SO2 emitted from elevated stacks, on surface concentrations. This occurrence was reasonably mimicked by the various PBL schemes but with some differences
that pointed to slightly better representation of pollutant mixing by some schemes than
others. In colder seasons (autumn and winter) large scale circulations decoupled from the
local terrain dominates, and simulations within a FAC2 for PM 2.5 and NO2 were fewer
than for summertime SO2 evaluations.
The MY-schemes outputs were frequently closest to in-situ pollutant measurements. For
PM2.5 , MYNN3 was closest. Model output with the MYJ scheme was the best fit for
the single-station evaluation for NO2 . PM2.5 and NO2 are predominantly emitted from
ground-level sources. Both PBL schemes also performed well for SO2 , although on the
evidence of good accuracy of maximum and minimum diurnal concentrations in the near
field of stack emissions in Kitimat, the ACM2 and MYJ schemes better emulated temporal
variations. Specifically, the MYJ scheme did not produce as much daytime fumigation as
ACM2, nor excessive nighttime mixing and downward transport as MYNN3 and YSU.

79

However, being that MYJ implements only local mixing between model layers, less pollutant quantity at higher elevations, in comparison to YSU and MYNN3 schemes, was
available for transport within the valley atmosphere. Therefore, for elevated pollutant
point sources that may arise in the Kitimat area, nonlocal PBL parameterization schemes
could be useful for projecting the most spread of emissions.
The physiography of the study area may explain why outputs from the MY-schemes more
closely matched with concentrations of ground-level pollutants. Due to proximity to, and
regular inland advection of the marine boundary layer, Kitimat in particular, is subject to
a shallow PBL. Further, there are many water bodies in the study domain, and much of
the land surface is snow-covered for at least a quarter of a year. Moist surface and nearsurface conditions, probably restrain the formation of strongly convective regimes, and
neutral and stable atmospheric conditions that are likely common in the valley are better
represented by local transport formulations in the MY-schemes. These schemes indicated
stronger vertical potential gradients than the other schemes, hence greater inversion tendencies that may have contributed to them producing the highest pollutant concentrations. The lower ranking of ACM2 performance was quite unexpected since it implements both local and non-local turbulence closure, depending on stability conditions and
is the default scheme in the current version of CMAQ. Some limited-duration dispersion
studies at coastal locations (Rakesh et al. 2013, Srinivas et al. 2015) have reported the suitability of MY-schemes. The MYNN3 and MYJ model runs however, required 1.4 and 1.1
times more storage space, respectively than YSU/ACM2. Thus, it seemed their deployment came with additional computing costs.

80

Modeled daytime concentrations that were within FAC2 at the major pollutant source
areas—Terrace (NO2 ) and Kitimat (SO2 )—indicated CMAQ’s ability to provide reliable
ambient estimates in the absence of fixed monitors. Better quantitative accuracy of the
MY-schemes than others notwithstanding, the achievement of performance benchmarks
for PM2.5 modeling was poor overall. Insufficient characterization of particulate emissions could have contributed to larger biases than for NO2 /SO2 . Specifically, it is likely
that particulate emissions from residential wood stoves—a major source of ambient PM2.5
in the cold season in the interior of British Columbia—is underestimated in the emissions
inventory. PM2.5 underestimations in chemical transport models are widely reported (Li
et al. 2016, Hu et al. 2016, Tessum et al. 2015) but the Canadian context is against a backdrop of province-level compilation of area emissions. Lumping of emissions over large
areas does not allow much flexibility with the SMOKE tool for simulations over domain
sizes as small as the TKV. Air pollutant emissions disaggregated into provincial sub-units
(e.g. regional districts), would enhance the spatial resolution of emissions input to the
CMAQ model. Such emphasis may also benefit emissions inventorying via inverse modeling.
While emissions are a source of modeling uncertainty, other factors could have undermined agreement between station records and modeled concentrations. Observations
themselves are never perfect, and monitoring errors may arise with the aging of sensors,
or for measurements at low ambient pollutant concentrations such as in the TKV. Observation error in pollution monitoring is acknowledged in several studies (Thunis et al.
2012, Pernigotti et al. 2013) and measurement uncertainty ranges are prescribed for differ-

81

ent pollutants (Gerboles and Reuter 2010). Representativeness error (Chang and Hanna
2004, Swall and Foley 2009) is also implicit when comparing point measures to grid cell
volume-averaged estimates and cannot be eliminated. Further, random perturbations in
the valley atmosphere contribute to modeling error. Smoothed topography in the meteorological simulations is also a plausible cause for modeled concentrations deviations from
observations. This especially would be the case for PM2.5 modeling over Terrace which
is in an area surrounded by mountains, and where the across-valley flow of the Skeena
River complicates the wind environment. It is noteworthy that of the three air pollutants
that were modeled, the top concentrations across simulations were most comparable for
PM2.5 , hence the least sensitive to PBL schemes. It is important that updates to PBL parameterization schemes take cognizance of the potential for inherent biases associated
with specific air pollutants, particularly over deep valleys to improve CMAQ modeling.

82

Chapter Four
Gridded bias correction of modeled
PM2.5 for exposure assessment and
estimation of background concentrations
over a coastal valley in northwestern
British Columbia, Canada
(In Press with the same same title at J. Air Waste Manage. Assoc.)

Abstract
Chemical transport models (CTM) can have large biases and errors when simulating pollutant concentrations. To improve the characterization of fine particulate matter (PM 2.5 )
over complex terrain for exposure assessments, three mathematical formulae that utilized
the relationship between modeled and observed quantile concentrations at a monitor sta83

tion were developed. These were then applied to one year of CMAQ model output of
PM2.5 over the Terrace-Kitimat Valley (TKV). The final products enhanced the representation of ambient levels at existing monitoring stations when evaluated with conventional
statistical measures. Better agreement of corrected outputs with observed compliance
metrics was also found. On average, the amended outputs had absolute errors of 11 % and
10 % for the annual mean PM2.5 and 98th percentiles of daily concentrations, respectively,
compared to 45 % and 61 %, respectively, in the original output. These improvements
provided greater confidence to use the reduced-bias outputs to estimate concentrations
at locations without monitors. The fact that pristine conditions dominate the modeling
domain was exploited to derive annual background PM2.5 concentrations of 2.0–2.3 µg
m-3 over the valley. To the author’s knowledge, this is the first study to calculate background PM2.5 concentrations over northern BC coastlands through bias-correction of outputs from an air quality model.

4.1

Introduction

Deteroriation of air quality due to particulate matter with an aerodynamic diameter ≤ 2.5
microns (PM2.5 ) is widely acknowledged. Exposure to PM 2.5 which may be released from
natural and anthropogenic activities such as wildfires and residential wood heating, has
been associated with adverse health effects, notably heart and lung impairment (Atkinson et al. 2015, Crouse et al. 2012). WHO (2016) estimates that globally in 2012, outdoor
(ambient) PM2.5 was responsible for ∼ 3 million deaths in urban and rural areas. Because
of morbidity and mortality impacts, regulations by various jurisdictions to limit ambient
84

levels are stringent, with ambient monitoring focused mainly on cities. However, investigations (Crouse et al. 2012) have also associated early deaths to long-term exposure to
PM2.5 concentrations lower than previously thought to be harmful. Consequently, there
is growing interest in qualifying pollutant exposure in the backcountry where there often
is little monitoring, and underlying problems could remain undetected for long periods.
The British Columbia Ministry of Environment and Climate Strategy has the mandate of
regulating air emissions throughout the province, including developing standards and
objectives on air quality that is protective of public health. They also partner with local
communities in recommending planning goals for airsheds, particularly in defining desirable, voluntary limit values for ambient PM2.5 . However, goal-setting and overall air
quality management in the TKV present some challenges. First, fixed-site pollutant monitoring stations are few, and much of the valley is not captured. Second, despite their regulatory intent, dispersion modeling realism in terms of identifying baseline airshed status
is unknown. Past air quality modeling in the area (e.g ESSA Technologies et al. 2015)
focused on predictions of contaminant concentrations arising from emissions changes.
However, without establishing baseline pollutant concentration, it will be difficult to account for cumulative emission impacts and tracking progress in airshed management
programs. Third, dispersion modeling may underrate or exaggerate the concentration of
particulate pollution over an area. Unless bias correction (Porter et al. 2015, Neal et al.
2014) is performed, use of raw model outputs can misclassify ambient PM2.5 exposures.
Finally, there is need to discriminate particulate levels that can be reduced, from those
that cannot be mitigated. The latter, more appropriately referred to as background con-

85

centrations (McKendry 2006, Veira et al. 2013) is a vital element for consideration when
setting air quality goals and mobilizing resources to achieve them.
In this chapter, objective bias-correction formulas for CMAQ-modeled PM2.5 over the
TKV, for the purpose of deriving baseline ambient exposure and background concentrations are developed and implemented. The subject of generating realistic concentrations of this pollutant at unmonitored locations, and in essence, estimating compliance
to air quality standards are addressed. The remainder of this chapter is organized as
follows. In section 4.2, data types, retrieval, and utilization for the development of corrective schemes, including the analytical methods are described. In section 4.3, the bias
correction results are compared with the raw model outputs and with observations, and
assessed for their closeness to indices of local regulatory compliance. Background concentrations PM2.5 are also derived. Section 4.4 discusses the relevance of final gridded
concentrations for present and future airshed management and concludes the chapter.

4.2

Methods

4.2.1 Model simulations and observations data
Demonstrations use PM2.5 outputs from the WRF-SMOKE-CMAQ modeling by the MYNN3
PBL scheme in chapter 3, which had the closest match with observations. The observations are valid hourly records from three PM2.5 monitoring stations (Table 4.1) for the
same year of simulations and are available at https://envistaweb.env.gov.bc.ca/DynamicTable2.aspx?
G_ID=327. Observational data completeness across stations were mostly > 90 % at sea-

86

sonal intervals. For the purpose of this study, seasons are defined as March-May (spring),
June-August (summer), September-November (autumn), and December-February (winter).
Table 4.1 PM2.5 monitoring locations, and hourly observation data completeness
for year 2017, to the nearest percentage (%)
Station

Latitude ◦ N

Longitude ◦ W Spring

Terrace

54.522

128.608

100

96

98

62

89

Riverlodge

54.054

128.671

100

99

99

100

100

Haul road

54.029

128.702

97

98

98

99

98

Summer Autumn

Winter Total

Prior evaluations indicated significant quantitative differences of modeled PM2.5 levels
from observations. Specifically, quantile distributions showed that the model underestimates the mean of the top quantiles by factors > 2. The errors in modeling that result in
the need for bias correction are from two main sources. The first comes from inexactness
of the emission inventory (amount, location, characteristics and configuration of emission sources); this directly impacts the modeled ambient levels. The other is associated
with how well the model itself emulates atmospheric processes, which includes the representation of meteorological variables, and transport, dispersion, and transformation of
pollutants. Consequently, an adjustment of concentration distribution in the raw output
is necessary to ameliorate the above shortcomings. The original model representation of
observations (see Fig. 4.1) is the basis for bias correction that is explained next.

87

4.2.2 Bias correction formulations
Three separate mathematical equations relating observations to simulations are derived
and analyzed for their effect on raw output: (1) simple linear regression; (2) power transformation (Li 2005) of the dependent variable in (1) based on a maximum likelihood
parameter; and (3) second-degree polynomial. Figure 4.1 describes the three correction
models. The equations are derived from data for the Terrace station. This choice for training the mathematical correction models is because the Terrace station reports the highest
ambient PM2.5 , is affected by diverse pollutant sources (road dust, vehicular traffic, residential sources, etc.); and it is located in an area of significant terrain complexity (deeplying urban observatory). Half the dataset at this station is used to create the equations
using values from every second day, and the remaining data are used for evaluations.
Each of the equations is applied to the original hourly output for the entire grid to generate domain-wide corrections. In the rest of this paper, ’ORIGINAL’ refers to the original
model output, while ’SIMPLE’, ’POWER’ and ’DEG_2’ refers to correction models (1), (2)
and (3) respectively.

4.2.3 Evaluation measures
Common statistical measures namely normalized mean bias (NMB), normalized mean
square error (NMSE), and proportion of model results within a factor of two of observations (FAC2), are used to evaluate whether, and to what extent the final products reduce
original biases. Several authors (e.g. Chang and Hanna 2004, Solazzo and Galmarini
2016) recommend these statistics for model performance evaluations and their formulae
88

Figure 4.1 (a.) Regression plots (quantile mean concentrations of model output
against those of observational data) at Terrace, and derivation of bias correction
formulae. Small open circles are the data points for both. (b.) Quantile plots for
modeled and observed (OBS) data at pollutant monitoring stations. The concentration at each octile is the mean of all hourly concentration values within that
octile. SIMPLE, POWER and DEG_2 are post-correction profiles.
are in Appendix B. NMB and NMSE are bounded between zero and infinity, and values
closer to zero are desired. Benchmarks for good model performance such as FAC2 ≥
50 %, are also calculated. The assessment of modeled-observation data pairs is done for
daily averaged data per season, following recommendations by Emery et al. (2017) for
modeling of PM2.5 . Bias-corrected products are also assessed for fitness with prevailing
air quality standards. Statistics for the latter are summarized as mean errors.

89

4.3

Results and analyses

4.3.1 Accuracy of bias corrected outputs
Figure 4.1b showed the quantile mean concentrations of bias correction vis-a-vis the original simulation and observations across stations. The main effect of the regression schemes
is to shift the distribution profiles closer to observation. Biases for top concentrations are
reduced. However, mixed outcomes are evident; the mean of the top quantile remains
underestimated at Haul Road but overestimated at the other two stations. The commonality of overestimation for the Riverlodge station and the Terrace station that is used for
deriving the corrections, could be because both are near residential areas where PM2.5
emissions exhibit similar diurnal patterns. The Haul Road station in contrast, is more influenced by steady PM2.5 sources such as sea salt and industrial releases.
The effect of bias correction at the seasonal scale is analyzed with paired data (Table 4.2).
Whereas agreement with observation generally improves with corrections, that for the
spring season is mostly more biased than the original output. This outcome nonetheless,
is minor since springtime is not the peak season for ambient PM2.5 in the valley. Peak
PM2.5 levels occur during autumn and winter seasons, and the corrections are quite effective for these periods. Specifically, post-correction NMB and NMSE for the Terrace and
Riverlodge stations are smaller than with the raw output, although the negative biases
for the autumn season remain. Importantly, all FAC2 and almost all NMB values for autumn and winter periods are within 50 % and ± 30 %, respectively, that are suggested
by Emery et al. (2017) as criteria for good model performance. Notwithstanding, PM2.5

90

measurements vary from one location to another and the relevance of bias-correction in
terms of matching compliance metrics is explored.
Table 4.2 Statistical evaluation of bias-corrections in comparison to original
model output, per season. Comparisons are based on daily means. Values in
italics indicate where bias-corrected outputs are worse than the raw output.
Spring

Haul road

Riverlodge

Terrace

Haul road

Riverlodge

Terrace

Haul road

Riverlodge

Terrace

Winter

Terrace

Autumn

Riverlodge

Summer

Haul road

Measure

original

-0.2

-0.2

-0.1

-0.5

-0.3

-0.2

-0.5

-0.5

-0.7

-0.4

-0.6

-0.6

SIMPLE

0.3

0.4

1.1

-0.3

0.2

0.8

-0.3

-0.1

-0.3

0.0

-0.4

-0.3

POWER

0.4

0.4

1.2

-0.2

0.2

0.8

-0.2

-0.1

-0.3

0.1

-0.3

-0.2

DEG_2

0.3

0.4

1.2

-0.2

0.2

0.8

-0.3

-0.1

-0.3

0.1

-0.1

-0.2

original

0.4

0.3

0.4

1.2

0.6

0.6

1.9

1.1

2.1

0.5

1.3

1.9

SIMPLE

0.8

0.7

1.4

0.6

0.4

0.9

1.4

0.6

0.6

0.7

0.8

1.2

POWER

0.8

0.8

1.8

0.5

0.5

1.0

1.2

0.6

0.6

0.6

0.6

1.2

DEG_2

0.8

0.8

2.0

0.5

0.5

1.0

1.3

0.7

0.6

0.6

0.7

1.3

original

76

78

70

60

74

80

54

58

24

72

32

39

SIMPLE

61

67

47

60

71

56

54

57

57

53

54

57

POWER

71

72

43

77

73

59

68

70

59

70

57

64

DEG_2

70

73

48

73

73

59

64

68

59

68

57

64

NMB

NMSE

FAC2

91

4.3.2 Evaluations for fitness with compliance metrics
British Columbia has non-statutory objectives for ambient PM2.5 to protect human health
and to guide decisions on the permitting of new or modified industrial emission sources
(BCMOECS 2020). They prescribe a daily metric (DM) which is the 98th percentile of 24hour mean concentrations over one year (standard = 25 µg m-3 ); and an annual metric
(AM) which is the annual mean of all hourly values over one year (standard = 8 µg m-3 ).
These metrics are calculated for the original model concentrations and bias-corrected outputs and compared to corresponding calculations for observations, to assess fitness with
prevailing regulatory objectives.
Calculations (Table 4.3) indicate that bias correction improves the estimation of compliance metrics. Across the three stations, prediction accuracy of DM with bias-corrected
outputs ranged between -10 % and +20 % (mean error = 10 %) as against -67 % and -62
% (mean error = 61 %) with raw output. Although air quality model evaluation literature
does not suggest what qualifies for a good performance with percentile-based regulatory standards, a bias spread of ±10 % may be considered a desirable performance goal.
Overall, the bias corrections with the different formulas are comparable. Notwithstanding, the higher-order regression models (POWER and DEG_2) have better comparisons
for the Terrace observation data than SIMPLE, suggesting greater usefulness for fitting
percentile-based metrics than a simple, linear scheme. For AM, bias-corrected values are
within ±11 % of observations (mean error = 7 %) compared to between -42 and -50 %
(mean error = 45 %) with the raw output. Because of the sign of model biases, the final
gridded outputs would more likely overestimate DM than AM at unmonitored positions.
92

Table 4.3 Accuracy of corrected values for 24-hour (98th percentile of daily
means) and annual metrics as normalized biases (in %). The mean error is the
absolute ( | |) average of all post-correction biases

Riverlodge

Terrace

Mean error

Haul road

Riverlodge

Terrace

Mean error

Annual metric

Haul road

24-hour metric

original

-62

-55

-67

-61

-42

-44

-50

-45

SIMPLE

-10

12

12

11

-11

-5

7

8

POWER

-6

19

1

9

-3

0

11

5

DEG_2

-7

18

3

9

-6

-9

9

8

Mean error

8

16

5

10

7

5

9

7

Also of interest is whether the corrections could be relied upon for indicating compliance
with ambient PM2.5 standards. A decision-support framework outlined by the Canadian
Council of Ministers of the Environment (CCME 2012) to categorize ambient PM2.5 exposures, and adapted to BC air quality objectives (ESSA Technologies et al. 2015), whereby a
series of management actions are taken as air quality begins to deteriorate is presented for
the TKV (Table 4.4). In applying this framework, exposure levels are color-coded according to concentation ranges of annual and daily metrics. Figure 4.2 describes the agreement
among the original output, corrected values, and observational data. Different from what
is obtained with the original values, all corrected values fall within the same categories as

93

observations. The strongest improvement appears to be for the Terrace station. Without
reducing the biases, the raw outputs will indicate overly optimistic DM and AM compliances, which in turn, could undermine pollutant exposure. That corrections capture the
same management category as observations, indicate reasonable ability to provide useful
guidance at unmonitored locations.
Table 4.4 Tiered air quality management threshold values and actions for ambient
PM2.5 adapted to British Columbia from CCME (2012)
Limits (µg m−3 )
Annual

98th percentile of

mean

24-hour daily mean

8

6

4

Exposure Assessment

Management objective

High

Achieve standard

Moderate

Prevent exceedance

Mild

Improve air quality

Low

Maintain good air quality

25

17.5

10

Figure 4.3 shows post-correction spatial plots of AM for all three schemes. Pollutant concentrations are quite similar among correction schemes, and are below 8 µg m-3 except for
a small area (4–5 grid cells) just west of Terrace. Here, modeled concentrations attain maximum values of 18.3, 12.7 and 29.2 µg m-3 for SIMPLE, POWER and DEG_2 corrections,

94

Figure 4.2 Correspondence of original and corrected outputs with observed
CCME 2012 PM2.5 management categories

95

respectively. Highest post-correction PM2.5 levels for the Terrace area are consistent with
expectations for urban settings, but comparable levels are also around the Haul Road station, near the coastal outlet. This area is an industrial zone where particulate emissions
from an existing aluminum smelter and allied services are probably a major contributor.
Sideward spread of stack emissions is restricted by valley walls to the west of the industrial hub, hence a localized area that is closer to the limit values than the rest of the Kitimat
area. Away from the Terrace and Kitimat areas, modeled AM is less than 3 µg m-3 .

4.3.3 Estimation of background concentrations
Based on the close fit between bias-corrected output with observation for the annual air
quality management metric, background PM2.5 concentrations in the valley can be estimated. Background concentrations are pollutant levels in the absence of local anthropogenic emissions, and represent the sum of concentrations arising from natural processes and those transported into an airshed from afar (McKendry 2006). Ideally, longterm pollutant monitoring at a pristine site would provide background concentrations,
but in its absence in the valley, and based on local knowledge, an unmonitored, uninhabited position in the modeling grid, distant from main emission areas is used. This position
(Fig. 4.4) that is currently devoid of anthropogenic PM2.5 sources and thus representative
of background conditions, is about 30 km from Terrace and Kitimat.
Figure 4.4 indicates that the mean annual background concentration of PM2.5 for the valley is 2.0–2.3 µg m-3 . The AM plots in Fig. 4.3 also hint at this range, but it is worth
mentioning that pristine environments are not confined to within the valley alone. The

96

Figure 4.3 Spatial plot of bias-corrected mean annual PM2.5 concentrations for (a)
SIMPLE (b) POWER and (c) DEG_2
mountain slopes and high-altitude areas are also wildernesses. On the western mountain ridges that are exposed to Pacific air masses, the minimum concentrations are 1.0, 1.8
and 1.4 µg m-3 for SIMPLE, POWER and DEG_2 schemes, respectively. Given the role of
97

transcontinental transport in redistributing atmospheric aerosol and relative isolation of
the TKV area from significant anthropogenic North American sources, these minimums
can be regarded as background concentrations affecting this part of the North American
west coast. These concentrations would result from emissions over the ocean, including sea salt (although this source would have a larger impact area like the valley bottom
that are within the MBL), contributions from Eurasian sources undergoing transpacific
transport, and wind-blown dust within the domain. Episodic particulate transport due
to large-scale, natural, seasonal emergencies might influence background concentrations.
For example, the summer of 2017 was one of the worst forest fire seasons in the BC Interior in recent history, although coastal areas were mostly spared the heaviest smoke
due to predominantly westerly atmospheric circulation. Emissions from wildfires are not
included in the model emission inventory.

Figure 4.4 Background PM2.5 estimation from bias-corrected annual mean centreline concentrations (see Fig. 2.1)

98

4.4

Discussion and conclusion

The application of linear regression models to raw CMAQ model output improved the
representation of ambient PM2.5 over the TKV. The final products from the mathematical formulations were generally comparable, with slightly better agreement of modeled
and stations data for correction that is based on power transform of the original model
output. Improvements are remarkable when it is considered that corrections were with
data from an urban monitor. The POWER formula, for example, reduced autumn NMSE
from 1.9 to 1.2 at Haul road, and from 1.1 to 0.6 at Riverlodge, representing 37 % and
45 % decreases, respectively, from the original modeled values. Djalalova et al. (2015)
reported a 50–75 % reduction in absolute errors with a Kalman filter-analog combination
over the contiguous USA. Nonetheless, it is difficult to compare the present study with
error reduction demonstrations elsewhere because apart from dissimilarities in geographical settings, there are differences in statistical measures employed, including averaging
periods, data collation and observation network densities. It is worth emphasizing that
the majority focus of PM2.5 modeling bias correction literature has been on operational
forecasting for much larger geographical areas, and with higher ambient concentrations
than in the TKV. In China were ambient particulate measurements are much higher, Lyu
et al. (2017) found for a suite of 12-km forecast correction techniques, reductions in normalized mean errors in the ranges of 7.4–19 % for 1-day lead time. Bias corrections in this
study applied retrospectively to simulations on 1-km horizontal grids, thus demonstrates
the efficacy of quantile-based statistical techniques to improve high-resolution chemical
transport model outputs, over complex terrain with sparse monitoring.
99

Degraded springtime FAC2 and NMSE, alongside generally improved statistics for autumn and winter seasons following bias corrections, suggested error compensations that
were beneficial to modeling accuracy for periods of peak ambient PM2.5 . Further, there
was substantial improvement in the representation of compliance metrics for ambient levels. Since the time order of hourly concentrations was not critical to determining whether
the TKV complies with existing PM2.5 standards, the approach of using regression models
that were based on quantiles-mean is valuable to airshed management. Specifically, corrections reduced large biases for the percentile-based daily metric to an average error of
10 %. A comparable effect was also obtained for the annual metric. The significance of improved agreement of simulations with observations is more confidence in base PM2.5 contour maps that are generated. For epidemiological studies on community health effects
of particulate matter, such spatial representations are indispensable, since fixed monitors
have limited aerial coverage. Because instances of post-correction negative biases across
stations was less for the daily metric (3) than the annual metric (5), projections with refined outputs may be more intense for acute (daily) exposure than for chronic (annual)
exposure.
This view especially follows from transforming ambient PM2.5 levels into an exposure
risk ranking (Fig. 4.5) according to thresholds that are enunciated by CCME (2012) and
applicable to British Columbia. Also suggestive is that a small area (∼5 km2 ) just west of
Terrace may be in violation of the provincial PM2.5 standard. At this stage, it is not certain the number of persons that may be impacted if borne out in reality, since this zone is
outside the urban core. Indeed, one shortcoming is that the assessment does not account

100

for effective exposure, which depends on the duration of outdoor activity by individuals. Nonetheless, the evaluation hints at the possibility of an on-going problem. Whereas
collecting ambient data and public education on local air quality are appropriate for all
areas, additional surveillance and analysis of trends can assist with confirming the actual source of the higher levels of PM2.5 . This may warrant ground-truthing over several
months, to preclude costly regulatory measures that might not be required in the present.
PM2.5 constituents are not summed in CMAQ boundary conditions file that dictate modeled levels. This necessitated the derivation of pollutant background from reduced-bias
concentrations within the model domain. Estimated mean annual background PM2.5 concentration over the TKV was 2.0–2.3 µg m-3 which is comparable to the value of 2 µg m-3
suggested by McKendry (2006) for the entire BC province, and 2.5 µg m-3 reported by Vingarzan (2007) for all of Canada. This is also within the range of 1–4 µg m-3 estimated for
the western United States (USEPA 1996) which like the TKV, is affected by trans-Pacific
aerosol transport. It needs mentioning that these works pertain to large geographical areas encompassing diverse physical environments and eco-climatic zones. The lower limit
of background concentration from the present study is 0.2 µg m-3 higher than the lower
end of 1.8–2.5 µg m-3 range calculated for locations in the BC Interior (Veira et al. 2013).
Although the coastal setting of the TKV is exposed to both maritime and continental air
masses, the valley possibly does not have higher background levels than areas further
inland. Perhaps, differences in estimates could be more due to the estimation methods
adopted by the various authors. For example, Veira et al. (2013) was for locations in small
cities and was based on analysis of wind sectors, for which trajectories were assumed

101

Figure 4.5 Color-coded classification of modeled PM2.5 ambient exposure risk
from POWER scheme using thresholds in Table 5. (a) annual metric (b) daily
metric.

102

to have passed over negligible upwind sources of anthropogenic emissions. PM2.5 background concentration in the present study is within the range of 1.7–3.8 µg m-3 by a study
that was based on monitoring data of six remote rural locations in Alberta (Cheng et al.
2000). In fact, a recent dispersion modeling with the CALPUFF model in the TKV (ESSA
Technologies et al. 2015) used background PM2.5 levels ranging 2.2–3.6 µg m-3 for postprocessing of output concentrations at individual receptor locations. These comparisons
allude to the usefulness of bias-corrected outputs of atmospheric-chemistry model simulations over pristine regions to estimate background PM2.5 concentrations.
The desired goal of air quality objectives in BC, especially within residential areas is that
air emissions do not exceed 8 µg m-3 PM2.5 average concentration per annum. The mean
annual background PM2.5 concentrations of 2.0–2.3 µg m-3 in this study, translates to 25–
30 % of the provincial limit. It is reassuring that 2017 ambient concentrations for a vast
portion of the valley are close to this level, implying that a ’‘business as usual’ economic
activity in the area should sustain good air quality. Because aerial exceedance of the
provincial annual metric is also small, isolated issues can be dealt with as they arise.
However, in the event that proposed industries become operational, or work camps are
built in confined neighborhoods, the adoption of local target limits for ambient PM2.5
will be beneficial. Specifically, the Terrace area, where topographical and meteorological
conditions seem less favorable to dispersion of pollutants, and where some of the additional industrial workforce in the TKV could reside, should be prioritized for particulate
emissions abatement. Overall, it is important that airshed planning goals for the TKV,
particularly for PM2.5 , recognize that although anthropogenic emissions are low for the

103

most part at present, the natural background constitutes a portion of the ambient level
that is not amenable to control.

104

Chapter Five
Acid wet-deposition modeling
sensitivity to WRF-CMAQ planetary
boundary layer schemes and exceedance
of critical loads over a coastal mountain
valley area of northwestern British
Columbia, Canada
(Also published at https://doi.org/10.1016/j.apr.2020.09.014 in Atmos. Pol. Res.)

Abstract
Computational tools used to implement the critical-load approach of atmospheric depo-

105

sition impact often do not elucidate modeling uncertainty, making it difficult for environmental policy-makers to know how much confidence to put in its results, also hampering
aspects that may need improving. This study evaluated acid deposition modeling for various parameterizations of the planetary boundary-layer (PBL) over the Terrace Kitimat
Valley (TKV). Of five schemes, simulations with the MYNN3 and MYJ PBL schemes best
captured weekly wet deposition fluxes of acidifying ions (SO4 2− , NO3 − , NH4 + ) within
a factor of 2 of observations at an industrial fence line station. Alongside the YSU PBL
scheme, these two schemes slightly overestimated the chemical species at a station that is
distant from major anthropogenic precursor sources in the valley, hence useful for worstscenario projections of atmospheric deposition on the environment. Forest soils in the
vicinity of a large aluminum smelter in Kitimat was estimated to exceed the critical load
of acidity by 30.1–53.5 kg S ha−1 yr−1 . Exceedance of critical load of nutrient nitrogen
restricted to the Terrace area (≤ 7 km2 ) ranged between none and 0.71 kg ha−1 yr−1 . This
work provides guidance for using PBL schemes in the Weather Research and Forecasting
model that is coupled to a deposition model when calculating critical-load exceedance
over temperate, rugged, coastal geographies.

5.1

Introduction

Aside from degrading ambient air quality, ecosystem impairment is one major concern
of the release of pollutant gases such as sulfur dioxide (SO2 ) and nitrogen dioxide (NO2 ).
These gases, as well as ammonia (NH3 ), are precursors of acidic compounds which include sulfate (SO4 2− ) and nitrate (NO3 − ) that can be deposited in wet form or fall as
106

dry deposition, leading to the acidification of soils and surface waters in regions with
shallow, base-poor soils and low mineral weathering rates (Driscoll et al. 2001, Sullivan
2000). Consequences of acidification include the accumulation of hydrogen ion (H+ ) and
increase in aluminum (Al) concentration in soil water and surface waters, which can adversely impact forest ecosystems and lead to the loss of aquatic biota (Driscoll et al. 2001,
Sullivan 2000). In addition to its acidifying role, increased levels of nitrogen (N) deposition can lead to eutrophication. Although considered to be a limiting nutrient for many
ecosystems, large amounts of N may cause loss of sensitive floral species, alter forest
productivity, and increase leaching of NO3 − to surface waters (Dise et al. 2011, Du et al.
2019). Adverse effects from acidification or eutrophication are usually not immediate due
to the inherent buffering capacity of soils and time-lags in ecosystem response (Cosby
et al. 1985).
The critical-load concept is a useful approach that has been developed to assess whether
pollutant deposition at any location is harmful. Critical load is a quantitative measure of
acid- or eutrophication-buffering capacity of an ecosystem (Pardo 2010). It serves as an
objective metric that can be used to determine both the spatial extent of a region being
subjected to damaging levels of sulfur and nitrogen deposition, and the magnitude of the
acidification (Moran et al. 2008, Pardo 2010). Acidification is likely to occur if annual total
S and N deposition to an ecosystem exceeds the critical loads. Critical loads of acidity
(CLA) and nutrient nitrogen (CLNnut ) is primarily controlled by soil bedrock and geology, and smaller values mean lower acid/eutrophication buffering capacity. CLA fields
for forest ecosystems have been derived at national and regional scales and are being har-

107

nessed for priority protective actions. The combat of acidification problems in eastern
Canada, for instance, has chiefly relied on such information in creating frameworks (e.g.
CCME 2014a, ECCC 2018a) for the reduction of precursor SO2 and nitrogen oxides (NO x )
emissions. Meanwhile, plans for heavy industry in parts of western Canada have brought
into spotlight the potential for ecosystem acidification and more nitrogen deposition.
The use of computational models in quantifying acid deposition and critical loads, and
calculating exceedances is broadly recognized as a useful management tool, in that it permits estimating total acid deposition at unmonitored locations. Such means was recently
deployed in the TKV with the Lagrangian CALPUFF modeling system (ESSA Technologies et al. 2014, Williston et al. 2016), wherein the risk of direct and indirect impacts of SO2
and NOx on terrestrial and aquatic ecosystems for a range of future emissions scenarios
was assessed based on CLA and CLNnut generated from steady-state models and empirical observations. Meanwhile, advanced 3-dimensional Eulerian platforms for simulating complex interactions between meteorology and atmospheric chemistry and pollutant
transport and deposition, have been deployed elsewhere, however with little information on optimal physico-chemical options in such models. Specifically, simulations of
acid deposition models could be sensitive to physics parameterizations such as planetary
boundary-layer schemes in coupled meteorological models. In wet, humid climates and
frequently foggy areas such as the TKV, the removal of pollutants from the atmosphere
by water droplets could be non-negligible, and a model’s handling of moist turbulence is
relevant to the process, hence the importance of examining the influence of PBL schemes
on wet deposition. Several researchers (e.g. Appel et al. 2010, Queen and Zhang 2008,

108

Williams et al. 2017) have explored the validity of CMAQ model-predicted acidifying
quantities, subject to constraints of aerial precision and temporal coverage. For instance,
Queen and Zhang (2008) performed CMAQ simulations for wet acid deposition at 4-,
12- and 36-km horizontal grid resolutions over North Carolina, USA but for two different months. Indeed, there is a lack of research not only on sensitivity of acid deposition
modeling to PBL parameterizations at a time scale of a year or more, but also in characterizing uncertainty of simulation ensembles at high horizontal grid resolution over
complex orography. With increasing recourse to computational systems in environmental decision-making at the local scale, it is imperative that modeling options are evaluated
at finer resolutions for long durations to ascertain their predictive capacities.
In this chapter, wet deposition of acidifying species simulated by CMAQ version 5.2 for
various PBL schemes in input meteorological fields are evaluated against observations for
one year (2017) in the TKV. Indicators of how well wet acidic depositions at the stations
are reproduced, are calculated for each scheme. Spatial differences in total acid depositions between schemes are examined. Portions of the valley that are likely impacted by
deposition of acidifying emissions are also identified, and above critical loads are estimated. Hence, this work also aims at providing baseline acidification status upon which
changes by future emissions can be tracked. The remainder of the chapter is organized
as follows. In the next section, CMAQ deposition modeling set-up, including the PBL
schemes that are tested, the model grids, chemical speciation and output post-processing
are described. In section 5.3, results of simulations, alongside comparative evaluation
of performance for the PBL schemes are presented. Critical load exceedance fields for S

109

and N deposition fluxes are also derived. Section 5.4 provides a discussion of the results,
including spatial variations in contributions of wet deposition and dry deposition across
modeling options. The final section (5.5) summarises and concludes the chapter.

5.2

Methods

5.2.1 Deposition simulations
The numerical experiments to generate deposition fluxes with various PBL schemes namely
the MYNN3, MYJ, UW, ACM2 and YSU PBL parameterizations is the same that was used
for modeling major air contaminants in Chapters 3 and 4. Concisely, the simulation platform consists of three main components namely:
(a) the Weather Research and Forecasting (WRF) model, version 4.0: a meteorological
model,
(b) the Sparse Matrix Operator Kernel Emissions (SMOKE), version 2.6: an emissions
processing tool,
(c) the Community Multiscale Air Quality (CMAQ) model, version 5.2: an atmospheric
chemistry model that also calculates acid deposition.
Hourly CMAQ simulations for atmospheric deposition were retrieved and post-processed
with built-in utility packages and I/O API tools (Coats Jr. 2017) to obtain: accumulated
wet deposition fields for evaluation against 2017 measurements in the valley (Fig. 5.1),
and annual total-sulfur (total-S) deposition and annual total-nitrogen (total-N) deposi110

tion fields for CLA / CLNnut exceedance using critical loads reported in Williston et al.
(2016). For comparison to weekly observed wet deposition data, precursor gas and particle phases of related output species were combined according to formulas provided for
the CB05 chemical deposition mechanism. As an example, sulfate wet deposition included fine and coarse mode sulfate particles, and sulfur dioxide gas. To calculate annual
acid deposition, wet and dry CMAQ deposition flux fields for S and N were summed over
all hours in the year from precursor compounds. Although CMAQ outputs wet deposition separately from dry deposition, both computations involved similar set of gas-phase
and particle species. Computation for total-N applied to a dozen chemical compounds
including ammonium particles and ammonia gas, dinitrogen pentoxide, nitrous acid, nitrochloride and peroxyacetynitrates. Computation for total-S applied to fewer species.

5.2.2 Measurement data and performance measures
Deposition records in the valley for 2017 were obtained from two National Trends Network (NTN) sites (http://nadp.slh.wisc.edu/data/sites/list/?net=NTN). The Haul Road
station is in the industrial area of Kitimat, while the Lakelse Lake station is in the vicinty
of a lake with the same name, 20 km south of Terrace. Both stations perform weekly sampling for acidifying ions (SO4 2− , NO3 − , NH4 + ), and also gauge precipitation volumes.
Each measures deposition from precipitation through a continuously operating wet deposition collector. The collector opens when its sensor detects precipitation, allowing it to
fall into a bucket, and closes when there is none. The collections are weekly after which

111

Figure 5.1 Left: nesting set-up for WRF-CMAQ deposition modeling over the
TKV. Orange boundary specifies northwestern British Columbia region from
which boundary concentrations to TKV’s domain (black) were derived, while
the red boundary is TKV’s meteorological domain. The boundary condition domain is 60 × 40, at 5-km grid spacing. Legend is for elevation above sea-level
in meters. Right: Enlarged TKV area identifying deposition monitoring stations
(with markers). The Haul Road station in Kitimat is within 2 km of an aluminum
smelter that emits large amounts of SO2 . The Lakelse Lake station is about 45 km
away from the smelter and 20 km from Terrace

112

another clean bucket is provided. Each site is also equipped with a weighing bucket rain
gauge to provide a continuous record of rainfall. Out of 52 weeks, measurements for
chemical species were available for 43 and 41 weeks at Haul road and Lakelse Lake respectively.
Precipitation chemistry and co-located precipitation amount data were used to calculate
annual calendar year bulk SO4 2− , NO3 − , and NH4 + deposition rates in 2017:
Deposition (kg ha−1 yr−1 ) = ∑(sub ppt * Cs /100), where
sub ppt is weekly rain gauge reading (mm),
Cs is concentration of SO4 2− , NO3 − or NH4 + (mg litre−1 ) in a sample.
For evaluations, species concentrations at model grid cells corresponding to ground locations of observations were retrieved. In line with recommendations for use of multiple
statistical indicators (Dennis et al. 2010, Solazzo and Galmarini 2016), the number of predictions within a factor of 2 of observations (FAC2), mean bias (MB), normalized mean
bias (NMB), mean fractional bias (MFB), and Pearson’s correlation coefficient r were used
to evaluate fitness between observed and modeled depositions. The formulas for MB,
NMB and MFB are outlined in Appendix C.

113

5.3

Results and analyses

5.3.1 Performance evaluation for wet deposition of acidifying species
Deposition measurements in comparison to simulations are indicated in Fig. 5.2. Modeled profiles have similar orders of magnitude as observations, and consistent with quantities of species (SO4 2− > > NO3 − > > NH4 + ). The deposition of SO4 2− is much underestimated at the Haul road station which may be due to several reasons. As previously
indicated, this station is on the fence line of the smelter site SO2 source in Kitimat. The
SO4 2− discrepancy possibly stems from discretization errors of I-km grid spacing near a
major point emission, also considering the inverse relationship between ambient precursor concentrations and deposition. The total flux of pollutant discharge is a combination
of ambient, deposited and transformed fluxes, and the more quantities are air-borne, the
lesser would be the amounts that can be allocated to deposition by the model. Evaluations
for air pollutants in the Kitimat area (Chapter 3), reported overestimation of SO2 , possibly
resulting in less allotment to the deposition component. Substantial underestimation of
the precipitation in the snowy months of November through February that contributes
nearly half of total annual precipitation, occurs across simulations. Dry precipitation bias
thus contributes to smaller wet SO4 2− deposition than is observed at this station.
But even in instances where weekly precipitation is reasonably simulated by a PBL scheme,
such as in March with the MYJ scheme, 0.3–1.8 kg ha−1 SO4 2− deposition underestimations occur, implying that estimates of other meteorological inputs to CMAQ are also
crucial to the accuracy of wet deposition modeling. Modeled winds in particular, were

114

Figure 5.2 Time series of weekly observed wet deposition of NH4 + , NO3 − , SO4 2− ,
versus simulations from various PBL schemes at Haul Road and Lakelse Lake
stations. Gray shading corresponds to the periods of heavy precipitation in the
valley from November to March. Broken observation profiles are for weeks with
missing measurement.
stronger than field observations, with wind speed overestimates of 2.0–3.5 m s−1 in the
cold, wet season. Because northerly winds are more frequent during winter, overesti-

115

mated wind speed would cause wet deposition of industrial SO2 emissions to be further
coastward, away from the station, hence simulation biases that are larger than those of
the warm, summer period. Winds are southerly in the summer months, but lesser precipitation that results in smaller wet deposition, in addition to more accurate precipitation
and wind speed values, ensures better agreement of modeled SO4 2− deposition with the
Haul Road station measurement than in winter.
At the Lakelse station, 45 km from the smelter site and in pristine surroundings, modeled wet SO4 2− deposition is mostly overestimated even though precipitation is underestimated. Modeled SO4 2− deposition is positively biased with MYJ, MYNN3 and YSU
schemes, and all PBL schemes have > 50 % of paired data within a factor of 2 of observations (Table 5.1). For NO3 − and NH4 + whose quantities are also smaller at the Lakelse
station than at the Haul Road station, model errors, including for normalized biases are
frequently less at the Lakelse station. For instance, the lowest NMB for NO3 − and NH4 +
at Lakelse lake (Haul Road) are 0.01 (0.03) and 0.01 (-0.21) respectively. These comparisons suggest that the nearer the distance of a location to the precursor source, the more
important the accuracy of meteorological inputs to CMAQ wet deposition modeling.
The frequency of best statistical values should be useful in ranking of the performance
of individual PBL schemes, especially for the pristine Lakelse location—a higher management priority from an ecological change perspective. For the three chemical species
however, the best statistical scores are mixed and no one PBL scheme is consistent for
all the measures. Further, a PBL scheme may demonstrate relatively good fitness for a
specific statistic just because it is poor in another. The UW scheme for example, has the
116

best r scores for SO4 2− and NH4 + at this station due to negligible temporal variability
of its small outputs. Given that acidic wet depositions distant from precursor emissions
may vary from year to year, the inherent capacity of a PBL scheme to capture aggregate
deposition, rather than time correlation is more pertinent to environmental protection.
Overall, the MY-schemes (MYJ and MYNN3) output the greatestquantities of acidifying species in wet form, followed by the YSU PBL scheme. This outcome is related to
how much precipitation is generated by the various simulations. Despite frequently producing the highest monthly precipitation and the biggest total precipitation, followed
by the MYNN3 PBL scheme, it is mainly in September that the MYJ PBL scheme generates more chemical fluxes than the other schemes. This again raises the possibility of
properties aside from precipitation amounts uniquely influencing atmospheric wet deposition in the various PBL schemes. Highest SO4 2− deposition with the MYNN3 PBL
scheme is partly due to simulating the greatest amount of ambient water vapor (refer to
Chapter 2) that augments wet deposition via pollutant washout by dew and fog. Unlike
the MYJ scheme, the MYNN3 scheme includes a partial-condensation model for subgridscale clouds (Nakanishi and Niino 2009), consequently simulating cloudier atmosphere
over the study area than with the MYJ scheme. The UW PBL is also designed for moist
turbulent processes (Bretherton and Park 2009), and produces dense, low clouds; however, deposition is less favoured, mainly because of less precipitation amount, and greater
chemical species loss from the lowest model layer with this scheme.

117

Table 5.1 Model performance statistics for NH4 + , NO3 − , and SO4 2− for various
PBL parameterization schemes at Haul road and Lakelse Lake stations. Values in
italics indicate the best performance at each location, per statistic.

MYJ

UW

ACM2

YSU

MYNN3

MYJ

UW

ACM2

YSU

Lakelse Lake

MYNN3
SO4 2−

NO3 −

NH4 +

Precipitation

Haul road

FAC2

50.0

27.1

16.7

16.7

16.7

54.0

48.0

40.0

46.0

46.0

MB (mm)

-8.6

-30.2

-33.7

-34.1

-33.8

-2.1

-11.9

-15.1

-13.8

-12.8

NMB

-0.19

-0.66

-0.73

-0.74

-0.74

-0.07

-0.42

-0.53

-0.49

-0.45

MFB

-0.15

-0.85

-1.00

-1.05

-1.05

-0.08

-0.53

-0.72

-0.64

-0.58

r

0.54

0.65

0.69

0.68

0.64

0.57

0.61

0.57

0.59

0.57

FAC2

57.5

50.0

25.0

35.0

35.0

43.9

46.3

43.9

53.7

51.2

MB (kg ha−1 )

0.00

0.00

-0.01

-0.01

-0.01

0.00

0.00

0.00

0.00

0.00

NMB

-0.21

-0.21

-0.57

-0.51

-0.43

0.13

0.10

-0.38

-0.18

0.01

MFB

-0.28

-0.28

-0.86

-0.72

-0.58

0.05

-0.02

-0.57

-0.29

-0.07

r

0.29

0.28

0.15

0.18

0.26

0.33

0.38

0.50

0.43

0.34

FAC2

45.0

42.5

37.5

27.5

37.5

36.6

46.3

39.0

46.3

39.0

MB (kg ha−1 )

0.01

0.00

-0.01

-0.01

-0.01

0.02

0.01

0.00

0.00

0.01

NMB

0.16

0.03

-0.26

-0.28

-0.17

0.59

0.42

0.01

0.08

0.33

MFB

0.09

-0.01

-0.38

-0.35

-0.21

0.39

0.24

-0.09

-0.01

0.19

r

0.12

0.15

0.10

0.08

0.14

0.23

0.32

0.40

0.43

0.31

FAC2

7.5

7.5

0.0

0.0

2.5

65.9

70.7

56.1

70.7

65.9

MB (kg ha−1 )

-0.59

-0.59

-0.66

-0.64

-0.63

0.01

0.03

-0.04

-0.01

0.02

NMB

-0.81

-0.82

-0.92

-0.89

-0.87

0.15

0.29

-0.48

-0.12

0.19

MFB

-1.35

-1.38

-1.72

-1.62

-1.55

0.11

0.16

-0.72

-0.21

0.12

r

0.11

0.33

0.29

0.21

0.24

0.66

0.66

0.85

0.78

0.73

118

5.3.2 Comparison of total nitrogen and sulfur deposition among PBL
schemes
Aggregate (wet plus dry) sulfur and nitrogen deposition were computed and Fig. 5.3
shows the spatial gradients for model runs with the various PBL schemes. Except for
the UW PBL parameterization, outputs generally reflect anthropogenic sources of N and
S deposition, namely urban and industrial activity, respectively. Modeled annual deposition is mainly < 5 kg ha−1 for S and < 2 kg ha−1 for N. Amounts tend to be greater
on the ridges on the east side of the valley than on the west side, indicating reasonable
accounting of the effect of regional circulations on deposition in the simulations. Eastmoving winds originating from the Pacific Ocean transport the valley’s atmospheric contaminants, some of which are eventually deposited on west-facing slopes. More affected
by the west-east circulations are areas of steeper topography near Terrace (c.f Fig. 5.1),
and around the coastal channel where the drift of marine clouds could be impeded by
the fjord walls. But quite different from S deposition, elevated N deposition on the ridges
around Terrace are comparable to that at its location in the valley. This suggests present
N deposition due to anthropogenic activity is not much greater than occurs naturally.
Figure 5.4 compares the output among PBL schemes, and makes clear that total N deposition for the year in the valley is in the order MYJ > MYNN3 > YSU > ACM2 >
UW, while that for S deposition is YSU > MYNN3 > MYJ > ACM2 > UW. Differences
in the locations of precursor emissions sources, hence seasonality of downwind impacts
would account for varying ranking of N and S deposition intensities. At half-way the dis-

119

Figure 5.3 Spatial plots of annual total-N (top), and annual total-S (bottom) with
the various PBL schemes.
tance between industrial SO2 stack releases in Kitimat and Terrace, the YSU and MYNN3
schemes output 2–4 kg ha−1 more S deposition than those of the MYJ and ACM2 schemes.

120

The TKV interior is subject to S deposition when onshore winds prevail, which is mainly
in the dry, summer season. Thus implicit is that annual S deposition within the valley
is more affected by calculations for the dry deposition component, which in turn derives
from ambient concentrations of precursor species. Comparatively higher summertime
SO2 concentrations were simulated by the YSU and MYNN3 PBL schemes which could
be due to nonlocal mixing, theoretically ideal for convective conditions (Cohen et al. 2015)
being accounted for by these PBL schemes. But it may as well be due to lesser precipitation in the warm season that removes less SO2 from ambient air, thereby making it
more available for transport by the YSU and MYNN3 PBL schemes. In colder seasons, at
which time the MYJ scheme has greater advantage over others in emulating the wet conditions that occur, northerly winds prevail and advected air pollutants are more susceptible to wet removal. Consequently, the MYJ scheme yields the greatest N and S deposition
around the coastal parts of the domain.
Unlike wet deposition, dry deposition is continuous, hence dominates total fluxes in the
vicinity of major emissions (Fig. 5.5). For S deposition largely traceable to smelter site
emissions in the valley, the ratio of wet deposition to annual total deposition (rwet ) increases going from the south towards the northern part of the valley. The dry flux component of total deposition is proportional to ambient precursor concentrations and their
attenuation with distance from the main source, increases the relative contribution of wet
deposition with distance. Hence, except the UW PBL scheme that outputs little S and N,
there is broad similarity across schemes for rwet values. In the central portions rwet for S
deposition is 0.3–0.5, ∼ 0.6 over the Terrace area, and ≥ 0.8 over the surrounding moun-

121

Figure 5.4 Pairwise differences in annual total-N (top), and annual total-S (bottom). Each domain-wide plot comes from subtracting the output of one PBL
scheme from another. For example, MYNN3 – MYJ means the output from MYJ
subtracted from that of MYNN3, etc.
tains, consistent with heavier precipitation at higher elevations.
Moister conditions over large inland water bodies than surrounding land areas such as

122

Figure 5.5 Domain-wide contribution of wet deposition as a ratio of the total-N
(top) and total-S (bottom) for individual PBL schemes.

123

at the Lakelse Lake station (Table 5.2) also results in rwet that is comparable to those of
high elevation areas. For N deposition, precursor concentrations originating from diffused sources (such as roads, residences, natural processes) as opposed to steady, point
sources for S, result in higher, but less orographically nuanced rwet patterns. The range of
rwet for N deposition is 0.3–0.5 over Terrace, 0.6–0.8 in the central areas of the valley, and
generally ≥ 0.8 on the mountains.
Table 5.2 % contribution of wet deposition to S and N annual deposition for individual PBL schemes at selected locations in the TKV.

S

N

Location

MYJ MYNN3

UW ACM2 YSU

Terrace

53.5

52.5

94.8

50.3

53.5

Lakelse Lake

81.3

73.7

86.0

70.7

73.5

Haul road

8.6

5.2

32.3

3.9

3.5

Terrace

30.7

27.0

71.3

29.8

34.6

Lakelse Lake

90.5

88.6

84.0

88.3

87.2

Haul road

74.9

62.1

87.9

61.3

61.9

5.3.3 2017 critical-load exceedances for forest ecosystems
Having totaled S and N deposition fluxes by the various PBL schemes, exceedances of
thresholds for acidification and eutrophication are assessed. The critical load information
that is reported for the TKV area (Williston et al. 2016) is exploited. The average CLA
as estimated from steady-state mass balance models in that study is 181 meq m−2 yr−1

124

which is equivalent to 29 kg S ha−1 yr−1 . The mass balance models used as critical limits,
base cation to aluminum ratios (Bc:Al) of 1 and 6 for the predominant coniferous forests
growing on mineral soils (65 % of the area), and deciduous forest tree species, respectively (ESSA Technologies et al. 2014, Williston et al. 2016). The same study deemed a
critical load of 4 kg ha−1 yr−1 nutrient N (CLNnut ) as protective of semi-natural terrestrial habitats in the area, including lichen communities, based on a synthesis of published
literature. Both the CLA and CLNnut are subtracted from modeled annual S and N deposition fields, respectively, thereby providing the spatial extent of critical load exceedances.
Area-weighted exceedance magnitudes are also calculated.
Figure 5.6 shows year 2017 gridded exceedance fields for the MYNN3 PBL scheme. Exceedance areas for CLA are clearly separate from CLNnut , and in both cases, confined to
grids over, or in close proximity to major anthropogenic precursor emissions. CLN nut
exceedance in the Terrace area is most likely from motor vehicle and rail switch yard
nitrogen oxides emissions. CLA exceedance in the Kitimat area is exclusively from aluminum smelter SO2 releases. These representations indicate that acidifying/eutrophying
depositions are for the most part, well below critical loads for the forest ecosystem. Similarity in spatial exceedance patterns (not shown) also exists for other PBL schemes except
that the UW scheme produced no exceedance, as summarized in Table 5.3. Aerial CLA
exceedances are greater than those for CLNnut ; the greater CLA exceedance mainly due
to continuous, large point emissions, rather than diffuse urban emissions. Apart from
the MYNN3 scheme, the only other PBL scheme that demonstrates CLNnut exeedance, is
the MYJ scheme. For acidity estimates, the MYNN3 scheme produces the most spatial

125

exceedance while the YSU scheme yields the highest area-weighted intensity.

Figure 5.6 Domain-wide exceedances of critical loads of acidity (CLA)(left) and
nutrient N CLNnut (right) with the the MYNN3 PBL scheme.
126

Table 5.3 Exceedances of critical loads of acidity (CLA) and nutrient nitrogen
CLNnut with various PBL schemes in the TKV
MYJ

MYNN3

CLA exceedance area (km2 )

8

10

0

7

7

Area-weighted CLA exceedance (kg S ha−1 yr−1 )

37.5

43.4

0

30.1

53.5

CLnut exceedance area (km2 )

5

7

0

0

0

Area-weighted CLNnut exceedance (kg N ha−1 yr−1 )

0.71

0.69

0

0

0

5.4

UW ACM2

YSU

Discussion

Weekly biases for modeled wet deposition of NH4 + and NO3 − at deposition monitoring
stations were within 0 ± 0.01 kg ha−1 . The near-zero biases were indicative of minimal
background atmospheric pollution in the region. Simulations in relatively clean environments may exhibit near-zero biases for N-containing species with respect to other modeling configurations. Regional CMAQ modeling by Qiao et al. (2015) reported NH4 + mean
biases of 0 kg ha−1 at 12-km, and -0.01 kg ha−1 at 36-km resolutions within a nature reserve in China. It should be mentioned that episode length in their study was not a year,
rather a three-month period (June to August). Guo et al. (2018) found very small biases
in wet deposition (-8.3 × 10−4 kg ha −1 for NH4 + , -3.79 × 10-4 kg ha−1 for NO3 − ) at its
sole validation station in Louisiana, USA over a period of one month. From an accuracy
view point, NH4 + and NO3 − ) quantitative uncertainties in the present study suggest that
the effect of choosing a PBL scheme instead of another, is not trivial for CMAQ-coupled

127

modeling of wet-deposited N species over complex terrain.
For SO4 2− that was the most prevalent of the three acidifying agents, normalized weekly
biases ranged from a maximum underestimate of -0.92 at the fenceline of major industrial
source of precursor SO2 emissions to a maximum overestimate of +0.29 at the pristine location further away. The majority of comparisons at both locations underestimated field
measurements, for which underestimated precipitation contributed. But since some of
the model runs overstated SO4 2− at the background (Lakelse) location, precipitation underestimation was possibly not the sole cause. Big penalties can arise for concentration
fields with tight gradients that are slightly displaced from observations in time or space,
and compensating errors at high (1 km) spatial resolution may have influenced deposition modeling performancee. Overestimation of SO2 in the near source area of stack
emissions may have led to less quantity available for wet deposition. Uncertainty regarding the production and transformations of aerosol species could also have contributed
to negative biases. Nearness to a maritime environment is a major influence on levels
of atmospheric oxidants essential for conversion of SO2 to SO4 2− and CMAQ’s handling
of chemical processes is the current science. Low bias for sulfate modeling is however,
not unusual in industrial zones. Cho et al. (2017) noted 43 % underestimation in CMAQ
simulated wet SO4 2− deposition at the Patricia McInnes site in the Alberta Oil Sands Region (AOSR) of western Canada at 4-km resolution. The majority emulation of sulfate
deposition at Lakelse lake station in the present study, while mimicking the low levels of
other acidifying agents, suggested that longrange deposition to pristine areas from elevated point sources of air pollutants can be reasonably predicted using 1 km horizontal

128

gridding.
In terms of performance, the PBL schemes whose outputs closely matched acid wet deposition species, were the MYNN3, MYJ and YSU PBL schemes. These schemes tended
toward moderate overestimation of chemical species in the far field of emission sources,
hence could be valuable for worst-scenario projections of atmospheric deposition on pristine environments. It is recalled from investigations for major air pollutants (Chapter refchpt3 ) that the MYNN3 and MYJ PBL schemes yielded concentrations that were closest
to peak ambient levels. Consequently, these two schemes have greater contaminant baselines for wet deposition than the other PBL schemes. Both the MYNN3 and MYJ schemes
(MY-shemes) are turbulent kinetic energy (TKE)-based schemes, and the formulation of
MYNN3 is similar to MYJ but with a higher level of closure (2 for MYNN3 compared
to 1.5 for MYJ). The MYNN3 scheme’s influence on deposition in the TKV particularly
derived from its consideration of moist convection, enabling additional pollutant settling
via mists and cloudy boundary layers. The strength of the MYJ scheme, on the other
hand, was due to its yield of the least precipitation deficit on an annual basis. Variants
of the MY-schemes have been advanced for operational weather forecasting (Olson et al.
2019) and pollutant transport and dispersion mapping (Miao et al. 2007, Srinivas et al.
2015) in coastal topographies.
The YSU scheme’s overestimations, especially for sulfate, owed to the position of the
major pollutant source (aluminum smelter) that more opportuned long-range deposition
by onshore winds in the warm season, when precipitation bias is also less.Unlike in the
ACM2 scheme where nonlocal transport of meteorological quantities in the PBL is active
129

only during unstable conditions, nonlocal transport is the default in the YSU scheme.
The wholly, nonlocal eddy diffusion approach of the YSU scheme thus was more supportive of prolonged mixing across model layers than the ACM2. By modeling greater
vertical spread of emitted pollutants, chemical species are longer-lived and travel farther,
hence more positive model biases at the pristine station with the YSU scheme than with
the ACM2 scheme that is the default parameterization in CMAQ version 5.2. The UW
scheme consistently produced the largest biases, apparently due to too little generation
of precursor pollutants in the first place.
Away from the Terrace and Kitimat urban areas, annual deposition fluxes were mostly
comparable among the MYNN3, MYJ, YSU and ACM2 PBL schemes, also consistent with
deposition fluxes in remote areas. The highest annual N deposition of 4.71 kg ha−1 (from
the MYJ scheme) is in the lower end of the range for rural areas in Canada (4.3–11 kg N
ha−1 , Zhang et al. 2009). That study was for sites in central and eastern Canada that are
impacted by industrial NO x emissions from the US. In the present study, only the higher
emission areas of Terrace and Kitimat evidenced predominant dry N and S deposition,
linked to road and rail transport (NO x ) , and smelter (SO2 ) emissions, respectively. Wet
deposition accounted for acidic deposition for much of the study domain. Excluding the
UW scheme, percentage contribution of modeled dry deposition to total N at the Lakelse
Lake station that is a pristine location, ranged 9.5–13 % as against 10–48 % for the rural
Canadian sites study. However, should anthropogenic NOx , SO2 and NH3 emissions increase within the TKV, the contribution of dry deposition to total N and S depositions
may also increase. Eularian grid modeling in areas of western Canada with large indus-

130

trial NOx and SO2 emissions, such as in the AOSR (Cho et al. 2017, Makar et al. 2018),
have reported considerable aerial contribution of dry deposition to total deposition. Cho
et al. (2017) estimates roughly 50 % of total S deposition to be due to dry deposition in
the AOSR, and marginally greater than wet deposition in future projections. The wet deposition ratios and patterns in the present study indirectly inform the proportion of dry
deposition which is a component that is difficult to measure in-situ and in real-time.
Excluding results from the UW scheme, which likely is unrealistic on account of account
of continuous smelting activity (> 60 years), area-averaged CLA exceedance was between 30.1–53.5 kg S ha−1 yr−1 over 7–10 km2 in the vicinity of the major SO 2 source
in Kitimat. The CLNnut exceedance ranged 0–0.71 kg N ha−1 yr−1 over 0–7 km2 around
Terrace. Spatial exceedances , including the largest (by the MYNN3 PBL scheme) were
limited to the valley bottom. In contrast, CMAQ-derived critical load exceedances over
the Georgia Basin of southern BC (Nasr et al., 2010), comprising areas of strong surface
inhomogeneity as the TKV, and receiving modest precipitation (1200–1500 mm annually),
were indicated to be greatest at high elevations and along steep valley slopes. Levels of
socio-economic activity including much smaller industrial and urban S- and N-bearing
emissions in the TKV than in the Lower Fraser Valley would suffice as to why upland
areas are not impacted by the present emissions. Predictive modeling will be relevant
to knowing whether anticipated emissions increase in the TKV and attendant deposition
will enroach upland forests.

131

5.5

Concluding remarks

CMAQ acidic wet deposition modeling sensitivity to parameterized planetary boundarylayer processes over the TKV found outputs to be comparable and credible for various
selections except with the UW scheme that produced very low quantities. The validity
of the majority simulations for wet fluxes, especially NH4 + and NO3 − was evident in
their generation to same order of magnitude as observations, as well as being spatially
correspondent with anthropogenic emissions and precipitation amounts. Overall, model
fitness was better at the Lakelse monitoring station than at the Haul road station; slight
overestimates at the former with the MYNN3, MYJ and YSU PBL suggesting reasonable
fitness of these schemes for future projections. Domain-wide critical load exceedance
calculations were then performed on modeled total depositions, one for annual sulfuronly deposition and another for annual nitrogen-only deposition. Consistent with local
knowledge, small areas in the vicinity of major anthropogenic emissions of SO2 and NO x
were estimated to be in exceedance of critical loads of acidity and nutrient N deposition.

132

Chapter Six
Intercomparison of atmospheric datasets
and PBL schemes for precipitation
downscaling over a coastal mountain
valley of northern British Columbia,
Canada
Abstract
Modeling environmental futures of coastal areas demands processing and linking atmospheric data in a way that adequately represents the role of hydrology on ecological
health. This study examined the capacity of dynamical downscaling to generate reliable precipitation fields over the Terrace–Kitimat Valley, an industrializing corridor in
the Coast Mountains of northern British Columbia, Canada. Precipitation modeling un-

133

certainty was explored via year-long, 1–km resolution WRF model simulations for three
atmospheric datasets and two planetary boundary layer (PBL) schemes. Underestimation by more than 40 % on average across stations with total precipitation ranging 1170–
2380 mm was found for simulations using either the North American Regional Reanalyses for atmospheric forcing or the Mellor-Yamada-Nakanishi-Niino level 3 (MYNN3)
parameterization as PBL scheme. Persistently low bias from model configurations using
these settings suggested that merely selecting an alternative atmospheric forcing dataset
would hardly ameliorate systematic error occasioned by a poor choice of PBL parameterization. Hence, choice of PBL scheme and meteorological dataset are important when
a numerical weather model is used for spatial estimates of precipitation. Model outputs
best corresponded with annual gauge measurements when simulations with the MYJ PBL
scheme were forced with ERA5, outperforming the North American Mesoscale Analyses
(NAM_ANL). The latter however demonstrated better spatial and temporal fitness than
ERA5. Using both datasets therefore may be valuable for projections related to environmental change. With either NAM_ANL or ERA5 for atmospheric forcing and MYJ as a
PBL scheme, the uncertainty in annual simulated precipitation amount ranged between
40 % overestimation and 20 % underestimation of observational data.

6.1

Introduction

The amount and distribution of precipitation are useful eco-climatic identifiers. Much of
the world’s forests are in regions of moderate to high precipitation. A key component
of the water cycle, precipitation is also a principal medium for the removal of pollutants
134

from the atmosphere to the surface. Despite the importance, in-situ precipitation measurement is hampered by expensive gauge installation and operating costs, and poor
siting or relative inaccessibility of favored locations (e.g in remote areas). Additionally,
spatiotemporal discontinuity of precipitation, often warrant reporting at intervals that
may be overly aggregated to serve other purposes. As an example, without further processing, precipitation data archived as daily totals would be incompatible with numerical
schemes requiring meteorological data at sub-daily/hourly time steps.
Meteorological datasets with more spatial coverage and temporal resolution of precipitation values, provide a means to alleviate the challenges of poorly gauged catchments
and aggregated reporting. While some datasets are interpolations of gauge measurements over grids, others employ sophisticated assimilation techniques that combine various observation sources (weather stations, buoys, aircrafts, radars, and satellite products)
with numerical weather forecasts. Many of them can be accessed at no financial cost
(https://rda.ucar.edu/). As meteorological datasets become more accessible, inquiries
about their validity have also increased. A review of global precipitation datasets by
Sun et al. (2018) found discrepancies of up to 300 mm yr−1 among products, with regional variability in the differences. Another global assessment with 22 gridded datasets
by Beck et al. (2017) found gauge-corrected datasets incorporating daily data better calibrated a hydrological model than using temporally coarser gauge data. They also ranked
the fitness of uncorrected datasets. Belo-Pereira et al. (2011) compared four global gridded
datasets to observations within the Iberian Peninsula, and noted that some products better represented the annual cycle and drought than others. Significant differences among

135

precipitation datasets have also been reported for long term comparisons in Canada (Lilhare et al. 2019, Wong et al. 2017), emphasizing the importance of choice of gridded meteorological products for precipitation estimation.
Selecting al dataset is more important when precipitation fields are required to address
local environmental issues such as atmospheric deposition of pollutants. State-of-thescience meteorological drivers such as WRF can downscale global and regional datasets to
much smaller domains. In addition to defining initial and boundary conditions, their implementation requires user-defined choices for nudging coefficients, physics parameterizations, nesting ratios, grid resolutions, etc. Several investigations of precipitation simulations in relation to these additional dependencies have been conducted. Li et al. (2014)
assessed the role of physical parameterization scheme and model resolution on simulation of summer rainfall over the south-eastern US. They found convective rainfall to be
more sensitive to planetary boundary layer (PBL) schemes than microphysics schemes,
also inferring that the choice of cumulus scheme is more important than increase in model
resolution. Hu et al. (2018) reached similar conclusions for the Great Plains region of
the US where differences were more profound with change in cumulus parameterization
schemes than other physics schemes. For snowfall over the northwestern Iberian Peninsula, Fernández-González et al. (2015) inferred better choices of microphysics and PBL
schemes. Whereas physical parameterization and hydrological downscaling sensitivities are widely recognized as integral to reconstruction of catchment climatologies, much
attention has been for short-term forecasts (Lee et al. 2015, Moya-Álvarez et al. 2020),
extreme events (Pontoppidan et al. 2017, Remesan et al. 2014, Sarmadi et al. 2019) and

136

parametric uncertainty (Tian et al. 2017, Yang et al. 2012). Little work has explored precipitation sensitivity to atmospheric datasets and computational physics schemes from a
common grid resolution.
In this chapter, numerical simulations of precipitation for a full year over a coastal valley
region of northern British Columbia (BC), Canada, are examined. Evaluations of outputs of dynamical downscaling with three atmospheric forcing datasets, and two PBL
schemes in WRF are conducted with the aim of establishing the appropriateness of precipitation fields for potential deposition modeling from various combinations of forcing
data and PBL parameterizations. This study therefore addresses the three following research questions: 1.) How well do individual simulations reproduce observed spatial
and temporal distribution of precipitation in the area? (2) Which forcing dataset(s) and
PBL scheme yield precipitation amounts most consistent with observations at individual
stations? (3) What is the uncertainty in the precipitation field that comes with the best
fit simulations? The answers to these questions are sought by evaluating model outputs
with several performance measures and analytical plots. The remainder of this chapter
is organized as follows. In section 6.2, precipitation measurement data, meteorological
model configurations and driving atmospheric datasets, and the data analyses methods
are described. Precipitation outputs from the simulations and comparisons to observations are presented in section 6.3. A discussion of dynamical downscaling sensitivities
appears in section 6.4. The last section (6.5), concludes the study.

137

6.2

Methods

6.2.1 Description of datasets
The selection of gridded products was based on their availability for the region at subdaily frequency. Sources providing data at longer time steps (e.g. monthly) or at very
coarse spatial gridding (> 32 km) were not considered. Another consideration was retrieval in formats compatible with the WRF code. Lastly, the evaluations were for two major categories of assimilated products, namely analyses and reanalysis datasets (https://
rda.ucar.edu/datasets/ds083.2/docs/Analysis.pdf). For these reasons, the North American Mesoscale Forecast System’s Analyses (NAM_ANL: Environmental Modeling Center
2017), the North American Regional Reanalysis (NARR: Mesinger et al. 2006) and the
fifth generation of the European Centre for Medium-Range Weather Forecasts (ECMWF)
atmospheric reanalyses of the global climate (ERA5: ECMWF 2019) were selected. Procedures for creation of each dataset are not within the scope of this chapter; however detailed descriptions can be found in the cited literature and in references for the overview
that is presented below.
The National Centers for Environmental Prediction (NCEP) uses WRF to produce different mesoscale forecasts products and analyses over North America. The 6-hourly analysis dataset on 12-km horizontal grids (NAM_ANL) that is used however has a spatial
coverage that excludes Canada north of 60 ◦ N. (https://www.emc.ncep.noaa.gov/mmb/
namgrids/g212.12km.jpg). Upgrades including transitioning from a 12-hour, to a 6-hour
data assimilation cycle with hourly updates was recently implemented for this product

138

(Environmental Modeling Center 2017).
The NARR is a 3-hourly product for North America developed by the NCEP. The NARR
system uses the Eta 32-km atmospheric model with 45 vertical layers and a three-dimensional
variational data assimilation approach. Precipitation is assimilated from regional hourly/6hourly multi-sensor (radar+gauges) precipitation analyses produced by 12 river forecast
centers. (Lin and Mitchell 2005, Mesinger et al. 2006). While 1/8 ◦ daily rain gauge data
analysis is used for the conterminous US, a 1◦ rain gauge analysis is used for Mexico and
Canada (Shafran et al. 2004).
ERA5 that has been used in previous chapters is a recent global reanalysis from ECMWF.
It is produced from high-resolution model forecasts on a regular grid, and a ten-member
ensemble, four dimensional variational data assimilation. Atmospheric data generated
with 137 model levels are interpolated to 37 pressure levels up to 0.01 hPa. ERA5 is available at hourly intervals and a horizontal resolution of 0.281 ◦ (31 km) (ECMWF 2019).

6.2.2 Modeling configurations and experiments
The WRF model (version 4.0) is used for simulating precipitation from the three atmospheric datasets for two PBL parameterizations. For turbulence parameterizations, the
MYJ and MYNN3 PBL schemes were selected because they best quantified ambient concentrations of air pollutants in the valley (refer to Chapter 3). The WRF domains are set
up with a nesting configuration in such a way that the innermost domains are identicallysized over the TKV (Table 6.1) and horizontal grid spacing is 1 km for all three datasets. To
accomplish this, downscaling ratios of 1:5 for NARR and ERA5, and 1:3 for NAM_ANL
139

were used. Grid size of parent domains were 25 km for NARR/ERA5 and 9 km for
NAM_ANL, hence intermediate domain sizes were 5 km and 3 km, respectively (Fig.
6.1). All domains are composed of 40 vertical pressure levels with the top level set at 50
hPa for NAM_ANL and ERA5, and 100 hPa for NARR.
The same physics options that were used in Chapter 2 (section 2.2.2) were deployed for
simulations, spanning 00:00 UTC 20 December 2016 and 00:00 UTC 2 January 2018. Similar to the previous procedure, each model run was initialized monthly, with model spinup of 1 day, and a further 1 day at the end of each month. The monthly overlap days,
including for the period 20–31 December 2016 were discarded when merging hourly output files and only outputs for 2017 were retrieved.
Table 6.1 Domain set-up and nesting with NAM_ANL (left) and ERA5/ NARR
(right)
Domains (dimensions)

NARR / ERA5

NAM_ANL

Parent (W-E x N-S)

100 x 100

100 x 100

Parent grid center

54.200 °N, 128.600 °W 54.200°N, 128.600°W

1st nested (W-E x N-S)

121 x 121

121 x 121

1st nested grid center

53.850 °N, 128.795°W

54.158°N, 128.671°W

2st nested (W-E x N-S)

101 x 121

100 x 121

2st nested grid center

54.223°N, 128.640°W

54.213°N, 128.695°W

140

Figure 6.1 Domain set-up and nesting with NAM_ANL (left) and ERA5/ NARR
(right)

6.2.3 Observational data and evaluation protocol
The WRF outputs are evaluated against station observations (Table 6.2) for 2017. Precipitation monitoring at the Lakelse station is as a part of atmospheric deposition sampling
and uses the OTT Pluvio2 gauge with wind shielding. The other stations use Meteorological Service of Canada (MSC) Type-B rain gauges that are unshielded, with a collection
rim height of 40 cm above the ground, and typical measurement uncertainty of ± 0.2 mm
per measurement.
Conventional statistical measures including for biases and errors, and Spearman’s rank
correlation coefficients for monthly accumulations (Table 6.3) are used to assess fitness
between model outputs and observations. Spatially-normalized plots are also used for
domain-wide model outputs, and station-paired observances of precipitation amounts.
Analyses are also performed with predictive scoring schemes. These are equitable threats

141

Table 6.2 Observational data locations for evaluation of precipitation simulations.
Gauge stations include those operated by Environment and Climate Change
Canada (ECCC) and BC Ministry of Forests, Lands, Natural Resource Operations,
and Rural Development (BCMFLNRORD)

Station

Latitude Longitude Elevation

Operating

Collection

(◦ N)

(◦ W)

(m)

Agency

frequency

Terrace

54.501

128.625

58.2

ECCC

Daily

Lakelse

54.377

128.576

111.0

Rio Tinto

Weekly

Riverpark

54.170

128.577

75.0

BCMFLNRORD

Hourly

Kitimat

54.054

128.634

128

ECCC

Daily

score (ETS), probability of detection (POD), the false-alarm ratio (FAR), and the frequency
bias index (FBI). POD is the rate of correct forecast of precipitation events (range 0–1 and
a perfect score of 1). FAR is the rate of false positives (range 0–1 and a perfect score of
0). ETS is the rate of correct event forecasts, adjusted for correct detections that would be
expected because of random chance (range -1/3 to 1, with perfect, and no skill scores of
1 and 0 respectively). The FBI gives the ratio of the estimated to observed precipitation
frequency (range 0 to ∞ and a perfect score of 1). These detection metrics are often used in
precipitation verification literature (e.g Peña-Arancibia et al. 2013, Schirmer and Jamieson
2015). Appendix H explains how these metrics are calculated.

142

Table 6.3 Fitness measures based on annual precipitation values for model outputs (M) and observations (O). In the formulae, rmi (roi ) are ranks of modeled
(observed) amount for each month, hence n = 12
Statistical measure

Formula

Perfect value

Bias (B)

M-O

0 mm

Error (E)

|Bias|

0 mm

Percent bias (PBIAS)

M-O/O × 100 %

0

Percent error (PE)

|M-O/O| × 100 %

0

n

6

Spearman’s correlation coefficient rk

6.3

1−

∑ ( r mi − r oi ) 2

i =1

n ( n2 −1)

1

Results and analyses

6.3.1 Quantitative biases for datasets and PBL schemes
Recorded precipitation totals in the valley in 2017 varied from 1168 mm at the Terrace
station in the north, to 2381 mm at the Kitimat station in the south; peak snow contributions (Fig. 6.2) of 25–40 % occured in March and November. Roughly one-third of
the total precipitation amounts are during the months of October and November. Model
outputs for total precipitation in 2017 at station locations, indicate PBIAS that are within

± 60 % of gauge measurements for all simulations (Table 6.4). The mean of PE (MPE)
from the various combinations of PBL schemes and forcing datasets range between 28
% and 44 %. Whereas simulations with the MYJ scheme produce mixed (positive and
negative) biases, all outputs for the MYNN3 scheme underestimate observations. Indeed,
143

Table 6.4 indicates that the MYJ scheme ameliorates overly dry conditions simulated by
the MYNN3 scheme. The MYNN3 scheme biases decreases as one moves inland from
the Kitimat station. Since the actual precipitation decrease away from the coast, the consistent drop in the MYNN3 biases indicates this scheme less represents the precipitation
gradient from coast to inland, compared to the MYJ scheme (Fig. 6.3). Domain-wide precipitation outputs amongst the three datasets mostly differ by factors of 0.7–1.5 for simulations with theMYNN3 PBL scheme, with greater differences for those of the MYJ PBL
scheme. Noteworthy is how the yield from the datasets are affected by the PBL schemes.
The NAM_ANL and ERA5 outputs are quite comparable and greater(less) than the NARR
output with the MYJ(MYNN3) PBL scheme. Remarkably, the greatest differences are over
the southern and western portions of the domain where precipitation estimates differ by
factors > 2. As these areas are coincident with the landfall of synoptic-scale storms from
the Pacific Ocean, the spatial plots imply that the choice of PBL schemes is fundamental
to frontal rain contributions of mid-latitude cyclones that are frequent in the region between October to March. With the MYNN3 scheme, diagnosis of boundary precipitation
fields is perhaps suppressed, and the skill of individual datasets seems less consequential
to model output with this choice of PBL scheme.
Direct spatial comparison of precipitation from the two PBL schemes (Fig. 6.4) show
much larger precipitation amounts for the MYJ scheme with NAM_ANL and ERA5. Differences in the capacity of either scheme to simulate precipitation from east-moving midlatitude cyclones are again alluded to in the NAM_ANL and ERA5 plots. In the case
of NAM_ANL for instance, the MYJ scheme precipitation is about twice as much as

144

Figure 6.2 2017 monthly observed precipitation (grey bars), alongside percentage
contribution of snow water equivalent.
Table 6.4 PBIAS values for 2017 precipitation simulation at station locations. MPE
is the mean of all PE per PBL scheme or dataset. Values in italics are the lowest
biases of evaluations at each station.
MYJ

MYNN3

Kitimat Riverpark Lakelse Terrace Kitimat Riverpark Lakelse Terrace MPE
NAM_ANL

+21

+36

+38

+30

-58

-46

-40

-40

37

ERA5

-21

-12

-4

-7

-57

-46

-43

-33

28

NARR

-55

-42

-36

-31

-53

-40

-36

-31

41

MPE

28

44

that from the MYNN3 scheme. For model forcing with NARR, the difference between
PBL schemes is marginal (∼ 0.9–1.1 multiplicative ratio). Recall from Table 6.4 that the
NAM_ANL and ERA5 outputs have more precipitation than NARR outputs. The resemblance of precipitation estimates from both PBL schemes for the NARR dataset, also
evident in similarity of negative PBIAS values across stations, thus suggests persistent
undercapture of the precipitation field over the valley with the NARR dataset.

145

146
Figure 6.3 Spatial plots of dataset-normalized annual precipitation estimates for
simulations with MYJ PBL scheme (top) and MYNN3 PBL scheme (bottom).
NAM_ANL/ERA5 means NAM_ANL output, normalized by ERA5 output, etc.

Figure 6.4 Spatial plots of PBL scheme-normalized (MYJ/MYNN3) 2017 precipitation estimates for each dataset. For NARR it is MYNN3/MYJ.

6.3.2 Spatio-temporal verifications
The spatial and temporal correspondence of precipitation estimates with observations are
further assessed for outputs from the MYJ PBL scheme whose MPE is smaller than that

147

of the MYNN3 scheme. Additional information is gained from examining monthly totals
and Fig. 6.5 shows that all three datasets represent the October peak precipitation to varying degrees; however, NARR does not depict the March and November accumulations
that are also the months of peak contributions from snow (refer to Fig. 6.2). In essence,
a main issue with the NARR output is poor attribution of snow amounts in the southern
part of the valley — virtually all the snowfall at the Kitimat station is unaccounted in the
NARR output. At the Kitimat station, less than 2 % of total precipitation for the NARR
simulation is attributed to snowfall (Table 6.5), in contrast to 12 %, 8.5 % and 12.5 % for
the station record, NAM_ANL and ERA5, respectively. The monthly NARR outputs however, are more positively correlated with station data than those of ERA5, perhaps due to
it producing less precipitation. Overall the best correlations are for NAM_ANL, exceeding 0.70 at all four stations.
Table 6.5 Contributions of snow water equivalent to total precipitation for observations (OBS) and those of simulations with various datasets and MYJ PBL
scheme.
Kitimat

Terrace

OBS NAM_ANL ERA5 NARR OBS NAM_ANL ERA5 NARR
Snow water equivalent (mm) 284.4

246.4

240.4

123.0

142.6

127.4

1875.8 1103.9 1168.0

1543.8

1089.2 803.9

Total precipitation (mm)

2380.7

2890.7

% snow contribution

12.0

8.5

12.8

42.7

1.8

3.7

9.2

11.7

Whereas month-by-month correlations can evaluate temporal fitness, pairwise comparisons of seasonal contributions to annual precipitation can be more insightful of precip148

85.8

10.7

Figure 6.5 The 2017 monthly series of observations (OBS) at gauge stations, and
simulations with the MYJ PBL scheme for different datasets.
itation dynamics. This is because precipitation events are more frequent and intense at
certain periods of the year, than others. For the purpose of evaluating model outputs, the
year-long simulation is partitioned into four seasons namely spring (March–May), summer (June–August), autumn (September–November) and winter (December–February).
Season-normalized proportions for all four seasons, at each station is presented in Fig. 6.6.
Changes in accumulated precipitation across seasons are best emulated by NAM_ANL as
it is often the closest to the measurement ratios. This correspondence is more for Kitimat
and Riverpark, than for the other two stations that are further inland, thus demonstrating
149

the best representation of the seasonality of precipitation in the wetter, southern portion
of the valley.

Figure 6.6 Pairwise seasons-normalized 2017 precipitation for observations (OBS)
at gauge locations, and simulations with MYJ PBL scheme and the various meteorological datasets.

Analysis is extended to assessing the representativeness of inherent spatial variability of
precipitation by tracking the proportionality of changes in model outputs. The stationnormalized plot of annual precipitation amounts for gauge measurements and model

150

outputs (Fig. 6.7) shows that spatial variability in the simulations is comparable to those
of measurements for the most part. Over longer separation distances between any two
gauges, modeled ratios tend towards being smaller than those of observations. The
greater disparities for Kitimat/Lakelse and Kitimat/Terrace in particular, may imply that
the rain shadow effect of coastal mountains that result in drier conditions in the northern
part of the valley is less in the simulations than in reality. However, it should be noted
that measurements are also uncertain. For instance, although 2017 precipitation records
at the Lakelse station were complete, this station measures weekly accumulations, and
measurement errors are difficult to identify at longer reporting frequencies. Further, all
the stations are within the valley channel and precipitation undercatch, which can be substantial, particularly, for snowfall and in windy conditions, may have resulted in higher
ratios than is correct. Nonetheless, the NARR ratios are consistently the most deviant
from measurement data, suggesting that NARR precipitation magnitudes are less suited
to downsaling in this region.

6.3.3 Predictive evaluations versus summary distributions
Evaluations of the uncertainity in precipitation simulations are often geared towards
improving the accuracy of short range forecasts, particularly for daily periods. Consequently, the use of conventional forecast indices has become widespread but such practice
may not be revealing of the credibility of model outputs for quantitative applications. For
the majority of precipitation events (days with accumulation ≥ 0.2 mm), forecast scores
among the datasets are comparable (Table 6.6), de-emphasizing the amount deficit of the

151

Figure 6.7 Pairwise stations-normalized 2017 precipitation for observations (OBS)
at gauge locations, and simulations with MYJ PBL scheme and the various meteorological datasets.
NARR outputs in comparison to those of NAM_ANL and ERA5. Predictive outcomes
for a smaller subset of precipitation events (daily amounts ≥ the mean for all days with
precipitation) are more differentiated and show the NARR forecast performs worse than
those of the other two datasets.
However, performance indicators are not always consistent. As an example, whereas
POD, ETS and FBI at the Kitimat station are more favorable for ERA5 than NARR, the
latter has a lower false alarm ratio. Poorer forecast quality with smaller data size misrepresents the adequacy of all three datasets. Indeed, timing accuracies, exemplified by
numerical forecast scores can be indecisive for choice of precipitation product where environmental consequence of absolute amounts is the end-purpose. For instance, the frequency of precipitation quantities ≥ the mean of daily events may be a small subset of all
events, yet could be much impactful for the removal of pollutants from the atmosphere
152

to the surface. For this reason, quantitative evaluation metrics that are relevant to atmospheric deposition are more ideal.
,

ERA5

NARR

NAM_ANL

NAM_ANL

0.79

0.11 0.18

0.14

0.54 0.40 0.43 0.95 0.99 0.92

Riverpark 0.85 0.82

0.78

0.16 0.22

0.21

0.49 0.38 0.37 1.01 1.05 0.99

Terrace

0.84 0.75

0.78

0.22 0.26

0.25

0.43 0.31 0.35 1.07 1.01 1.03

Kitimat

0.77 0.48

0.20

0.32 0.41

0.33

0.48 0.27 0.14 1.13 0.81 0.30

Riverpark 0.81 0.54

0.32

0.43 0.42

0.43

0.42 0.30 0.19 1.42 0.97 0.56

Terrace

0.28

0.49 0.55

0.64

0.37 0.27 0.13 1.54 1.30 0.78

NARR

NAM_ANL

0.85 0.82

ERA5

NARR

Kitimat

NARR

Station

ERA5

FBI (perfect =1)

ERA5

Target

POD (perfect =1) FAR (perfect =0) ETS (perfect =1)

NAM_ANL

Table 6.6 Predictive scores of daily precipitation for all events (≥ 0.2 mm) and
events ≥ mean daily amounts. Mean daily amounts are 11.7, 9.6 and 6.9 mm at
the Kitimat, Riverpark and Terrace stations, respectively.

All daily events
(amount ≥ 0.2mm)

≥ Mean daily
quantity
0.78 0.59

Figures 6.8 and 6.9 depict the fitness of datasets based on event frequencies. At monitoring stations, daily events ≥ the mean daily precipitation account for no less than 25 % and
70 % frequency and amount, respectively, of all events (Fig. 6.8), with the NAM_ANL and
ERA5 outputs reasonably matching these contributions. Fig. 6.9 also highlights the uncertainity of simulated precipitation. Apart from the 10th decile, the mean of each decile
from NARR does not exceed the annual mean of all events days in Kitimat. This is unlike
the other datasets where the means of the top 3-4 deciles are greater than the average of
station data. The other stations equally depict poor performance for NARR. Considering

153

that the observation quantiles are bounded by ERA5 and NAM_ANL, it is concluded that
both datasets provide more credible estimates of precipitation than NARR.

Figure 6.8 Categorical evaluation of simulated precipitation amounts by various
datasets with the MYJ PBL schemes with respect to station data (OBS) for all daily
events (days with precipitation amounts ≥ 0.2 mm), and daily events ≥ the mean
daily precipitation amount. The latter’s contributions in terms of frequency and
amount are also plotted.

6.4

Discussion

Across stations, the lowest biases in simulated annual precipitation ranged between 21
% underestimation to 3 % overestimation. This range is smaller than the -15 % to +25 %
uncertainty band across seasons reported by Wong et al. (2017) for the Pacific Maritime

154

Figure 6.9 Cumulative mean distributions of simulated and observed daily precipitation at validation stations. The amount at each decile is the mean of all values within that decile. Dashed vertical line is mean precipitation of daily events.
terrestrial ecozone of Canada. That study evaluated five gridded precipitation products
against precipitation-gauge station data but, was based on a regridding to a coarser resolution (0.5 ◦ ) for all compared datasets. No intermediary regional climate model was
used. Lilhare et al. (2019) reported annual precipitation biases ranging -8.2% to 13.5% for
five hydroclimatic datasets over the Lower Nelson River Basin in the Canadian Shield.
Their study was as well based on regridding to 10 km (∼ 0.1◦ ) spatial resolution with
155

bilinear interpolation (no dynamical downscaling), hence precluded the role of model
physics. Datasets in that study mostly exhibited positive biases, unlike results in the
present study. Aside from being over relatively flat terrain, Lilhare et al. (2019)’s study
was for a sub-arctic continental climate with annual observed precipitation of ∼ 500 mm
and peak precipitation during summer. Dissimilar geographies and methodologies could
therefore be the principal reasons for differences in uncertainty ranges between this study
and other Canadian investigation. Because model outputs over the TKV were from simulations at high horizontal resolution (∼ 0.01◦ ) with various combinations of meteorological forcing data and PBL schemes, the bias range in this study provides information about
WRF competency to reproduce point observations of precipitation in a coastal valley region influenced by mid-latitude cyclones originating over the Pacific Ocean.
With a sparse observation network and more dependency on parent model forecasts, uncertainties in physical parameterizations of atmospheric processes become prominent.
Overall and irrespective of dataset, simulations using the MYNN3 PBL scheme exhibited
larger and more negative biases than simulations using the MYJ scheme. WRF physics
parameterizations are highly integrated, with PBL schemes providing the interface between cumulus, microphysics and radiation schemes on one hand, and surface-layer and
land surface schemes on the other. While both tested PBL schemes used similar physics
settings, significant portions of parent and intermediate grids, at which spatial scales convective eddies are parameterized, consisted of water (the Pacific Ocean). The MYNN3
scheme was designed to improve several aspects of earlier Mellor-Yamada schemes such
as an insufficient growth of the convective boundary layer and underestimation of the

156

turbulent kinetic energy and the turbulent length scale (Nakanishi and Niino 2009), but
such concerns are perhaps more applicable to warm, dry climates. Moreover, much of the
precipitation in the TKV occurs during the months of October-March when east-going,
low pressure systems and frontal passages dominate weather. Convective periods are infrequent and precipitation that may be produced by this process, such as during summer
contributes little to total annual amount. Better model performance with the MYJ scheme
is similar to findings by Evans et al. (2011) for the east coast of Australia that experiences
heavy rain and strong onshore winds as occurs over the northern BC coast. In geographical areas with convective precipitation (e.g. Srinivas et al. 2018), MYNN-type schemes
have been reported to outperform the MYJ scheme for intense events. In the present
study, it is suspected that the lack of convection in the cold wet season suppresses precipitation simulation in the MYNN3 scheme much more than with the MYJ PBL scheme.
Implicit therefore, is that MYJ PBL scheme is a better fit with the new Thompson microphysics scheme that was deployed for all model runs (Thompson et al. 2008), since
non-convective precipitation in WRF is controlled by the microphysics scheme.
At all sites, the driving dataset that gave results closest to observed annual total precipitation was ERA5, followed by NAM_ANL. With model runs using the MYJ PBL scheme,
both datasets demonstrated a reasonable fit with precipitation distributions. Whereas
ERA5 produced moderately low biased outputs for the most part, NAM_ANL outputs
were biased high. Within the context of environmental change applications, such as modeling atmospheric deposition of air pollutants, these findings have important ramifications since precipitation fields are needed to drive relevant models. ERA5 has the poten-

157

tial of causing slightly less wet deposition than would occur since precipitation is biased
low. In contrast, NAM_ANL would project much higher rates of wet deposition; possibly with greater spatial and temporal fitness than ERA5. Perhaps using both datasets for
environmental modeling in the region would ensure that neither the quantitative accuracy of ERA5 nor the spatio-temporal performance of NAM_ANL would be lost. Such
an approach will allow for projections that are realistic. The NARR dataset least quantified precipitation. The assessment of NARR as significantly dry-biased agrees with Wong
et al. (2017) for their study over the whole of Canada. Nonetheless, in regions with much
lower precipitation than northwest BC (e.g. Lilhare et al. 2019), NARR has been reported
as slightly overestimating observations.
Because of differences in nesting required to downscale each dataset within WRF, also
considering the hydroclimatology of the TKV area, topographic resolution of the coarsest
grid might also have influenced the performance of the meteorological datasets. The raw
resolution of NAM_ANL is 12 km, and was more amenable to a downscaling ratio of 1:3
than the other datasets. This ratio, if used for ERA5 and NARR would have required
more nested grids, with additional computational cost and perhaps more errors accumulating in the finest domain. While the ratio of 1:5 as used for the reanalyses datasets is
permissible for precipitation downscaling (Liu et al. 2012), a ratio of 1:3 allowed for the
use of higher-resolved terrain data for NAM_ANL. With a 9-km grid for the largest domain in the NAM_ANL simulations, topographic data resolution was 5-arc-minute and
not 20-arc-minute that was used for the coarsest grids in ERA5/NARR simulations. For
the intermediate and TKV domains, topographic resolution was the same for all three

158

datasets. Since the north BC coast has steep elevation gradients and complex topography,
finer representation of surface terrain for the 9-km grid in NAM_ANL simulations may
have contributed to it having less bias from observational data than NARR. However, as
this study was not for an entirely continental setting, the benefit of finer-resolved topography could be small since a large portion of the coarsest grids and intermediate grids for
all three datasets is water (the Pacific Ocean) and not land.
Another plausible reason for differences in performance would be inherent quality of
datasets. Better spatio-temporal correspondence of NAM_ANL with observations than
for ERA5 and NARR was found despite that NAM_ANL is an analyses product, available
at 6-hourly interval. Although improvements are being made, such as implementing a
four-dimensional data assimilation system, and increasing grid and temporal resolutions
(e.g. ERA5), reanalysis products are still much reliant on the number of observations that
are ingested. Observation data sparsity particularly in remote regions translates to low
quality of atmospheric products for these areas. Over northwestern Canada, there are not
many weather stations—a status that impacts the quality of atmospheric data products
for the region. All datasets would benefit from more quality control of assimilated data,
including the treatment of missing observations. Dataset availability at higher temporal
frequencies would also be helpful as this most likely contributed to better performance of
ERA5 than NARR. Indeed overall poor performance with NARR, in spite of its specificity
for North America, suggests a need for further enhancement of its value to the Pacific
northwest. In NARR development, no data are assimilated over the oceans north of 43.5
◦ N, and there has been non-assimilation of gauge data over Canada after 2002 (Mesinger

159

et al. 2006), ostensibly causing the large discrepancies with gauge records.
This study compared model outputs using the same scale (hourly, 1-km2 grid-cells). An
aspect seldom contemplated in many model evaluation studies is whether the non-equivalence
of spatio-temporal resolutions of atmospheric datasets is addressed when they are compared to one another. Analyses and reanalyses products, and even gridded interpolated
records are often directy compared against one another, despite that they are mainly available at different space and/or temporal resolutions. The downscaling approach in this
study not only created a common spatio-temporal reference, but also used a fine horizontal resolution. Gauge measurements were considered point values valid over a limited
area. Hence, 1-km2 grid-cells were used to generate reasonably discrete outputs for the
TKV area, although expending significant computing resources. This recognition is particularly relevant when computational tools are intended for assessment of local environmental concerns, rather than for projections across continental/global climates.

6.5

Concluding remarks

The urgency of addressing the environmental impact (e.g. atmospheric deposition of air
pollutants) of industrial projects that are planned in the TKV required evaluating various
choices for initial and boundary meteorological forcing, and PBL schemes in WRF for precipitation simulation, where due to few surface weather observations, gridded outputs
will be input to other models. For combinations of three atmospheric datasets and two
PBL schemes, large (> 40 %) underestimation of precipitation was found for simulations

160

involving NARR as forcing data, or MYNN3 as a PBL scheme. Model configuration with
either NAM or ERA5 as atmospheric forcing data and MYJ as a PBL scheme performed
better. With this arrangement, the uncertainty in annual precipitation amount ranged between 40 % overestimation and 20 % underestimation. The differences between model
outputs not only inform the relative suitability of various modeling configurations to
provide realistic precipitation estimates, but also highlight the importance of continuous
improvement of individual components for optimal performance over Pacific Canada.
This could be through greater observational data assimilation into raw products, dataset
production at higher spatial and temporal resolutions, and verifying PBL schemes over
inhomogeneous terrain.

161

Chapter Seven
Modeling terrestrial ecosystems
exposure to incremental smelter sulfur
dioxide emissions and deposition in a
complex coastal valley airshed
Abstract
Predicting the impact of air pollution from industries is important for developing management strategies under changing emissions. This chapter assessed the effect of increased SO2 emissions from a large aluminum smelter in the Terrace-Kitimat valley, a
predominantly pristine airshed in northwestern British Columbia, Canada, on the natural environment using an atmospheric-chemistry model. CMAQ simulations with 1 km2
grid cells indicated at least 50 % increase in ambient SO2 concentrations due to rise from
previously 27 tonnes day−1 smelter emissions rate to 42 tonnes day −1 maximum permis-

162

sible discharge. The risk of harm from acidifying emissions on flora, and on soils from
new levels of SO2 exposure and deposition were evaluated based on critical level and critical acidity load values, respectively. Comparisons to baseline impacts estimated 50–88
% increase in aerial exceedance of limit for protection of lichen, and 37–67 % increase in
spatial exceedance of threshold for protection of soils. Cumulatively, 16–18 km2 of plant
habitat and 10–11 km2 of soil in an area contiguous with the smelter site will likely be
damaged by its SO2 emission under the latest regulation. Whereas these projections are
consistent with expectations of marginal spatial impact from tall stack emissions, it is imperative that soils are routinely sampled for phytotoxicity beyond the range outlined in
this study.

7.1

Introduction

Industrial air waste permitting by regulatory agencies often prioritize environment protection goals ahead of profit-making by businesses (Dietz 2003, Longley 2019). This is
because many production processes are emission-intensive and their environmental impact may linger, long after operations have ceased. For example, it is common knowledge
that some feedstock in metals manufacturing produce environmentally damaging emissions (Andrews and Lattanzio 2013, Cirtina et al. 2016, Habashi 2011). A precautionary
approach to airshed management thus requires regulating emissions from major sources
so that neither the resultant pollutant concentration in the atmosphere, nor deposition
to ground receptors are occurring at levels that cause significant harm to specified sensitive components of an ecosystem. Critical levels which are exposure limits above which
163

adverse effects of ambient concentrations on key biota are observed, and critical loads
which are deposition thresholds above which acidification and eutrophication will occur
in the long term (1-100 years), have been proposed for a range of environments (Coordination Centre for Effects 2017, Josipovic et al. 2011, WHO 2000). Originally addressing
issues pertaining to transboundary acidifying emissions impact on ecosystems, critical
levels and critical loads have become pivotal to managing local air emissions including in
Canada where they are being used to guide decisions concerning industrial development
in wilderness areas.
Chemical transport models are useful to determining compliance to critical levels and
loads over geographical areas since they are able to simulate ambient concentrations and
deposition of air pollutants at discrete locations far away from monitoring stations. Because they ingest meteorological and terrain data, in addition to implementing robust
physical and chemical interactions between multiple pollutants, these tools provide a
means to link spatial concentrations/depositions of pollutants to emissions (Brook et al.
2019, Kelly et al. 2018, Vivanco et al. 2018). This advantage is routinely exploited for indicating and projecting air pollution and acid deposition effects, including over complex
environments, such as deep valleys (e.g. Kelly et al. 2018).
In the past decade, air dispersion and deposition modeling in topographically-confined
areas of British Columbia began placing more emphasis on effects of aggregate emissions
from multiple sources (ESSA Technologies et al. 2014, Krzyzanowski 2010). Overly aggregated analyses however, may obscure contributions of large sources to potential concerns.
Environmental impact attribution to industrial emissions is an integral part of social ac164

countability particularly when they are present near or within natural resource-based
communities. This requirement becomes more compelling as pollutant emission levels
change. Identifying exposure and deposition fields of terrestrial ecosystems in response
to changes in emissions helps track the course of effects. It also facilitates designating environmental responsibility to polluters. Further, it can provide valuable information on
solutions, for instance, whether mitigations should be more at the level of sources or at
receptors. Therefore, while cumulative impacts assessment (Krzyzanowski 2010, Pickard
et al. 2019) remains important, specifying the impact of emissions change of single, large
sources is also pivotal in areas that host such sources.
This chapter assesses the magnitude and extent of the effect of recent aluminum smelter
emissions change (refer to Fig. 5.1 for its position), on the land ecosystem in the TKV.
The CMAQ model is used to simulate air concentrations of SO2 , and deposition of sulfur
(S). Attention is primarily on comparisons of model outputs to critical levels of exposure
of sensitive organisms to SO2 , and to critical loads of S, as an indicator of potential soil
acidification. In the next section, existing and incremental industrial emissions, as well as
the modeling approach are described. In Section 8.3, modeling results, including changes
in pollutant concentrations from baselines, and exceedance of natural resource protection
limits are presented. Section 8.4 provides a discussion of the results while section 8.5
concludes the chapter.

165

7.2

Modeling framework and procedure

The WRF-SMOKE-CMAQ modeling system consists of three main steps namely meteorological simulation, emissions modeling/ processing and air chemistry/deposition
simulation. Year 2017 meteorological forcing fields at 1 km horizontal resolution over
the TKV and surrounding areas were generated (see Chapter 6) from two datasets: the
North American Mesoscale Analyses (NAM_ANL) and ERA5. These were used to provide reasonable precipitation estimates. A similar approach to emissions modeling with
the SMOKE tool described in Chapter 3 was utilized for emissions processing, which in
this study, was to generate two sets of gridded emissions for each meteorological input
file. One relates to pre-existing emissions (that of ERA5 for emissions prior to aluminum
smelter upgrade, or the pre-modernization discharge limit was already derived in Chapter 3), while the other was for smelter emissions up to maximum, 42 tonnes day−1 , SO2
discharge limit (post-modernization period). Emissions from all existing sources in 2017
in the airshed, including for the boundary file at 3 km grid spacing (the grid size of the
intermediate domain of NAM_ANL input meteorology (Table 7.1) were from the BC portion of Canada-wide, SMOKE-ready inventory.
To account for incremental ship-related emissions, the increase in quantity was based on
scaling from total estimates (ESSA Technologies et al. 2014) in proportion to the projected
movement of an additional 23 cargo vessels for increased export of finished products
and import of raw materials. Actual allocation in space was according to mode of vessel
and tugboat activity, in addition to consideration of regulations on ship fuels in North

166

American waters, and pollution emission factors of ocean-going vessels (USEPA 2009).
For all pollutants including SO 2 , emissions from marine vessels traveling in the Douglas
Channel were modeled as 40 waterway positions from the marine terminal next to the
smelter down to the edge of the domain. Emissions at berth was fixed at half the total
of underway releases (that is, 1/3 of all maritime transportation related-emissions). The
spacing of positions was between 1–1.4 km with few clustered nearer the marine terminal
to imitate higher traffic. All emissions from the smelter site, at the marine terminal, and
for travel of vessels in Channel waters, were modeled as elevated point sources using
discharge parameters indicated in regulatory/government publications (Environmental
Appeal Board 2015, ESSA Technologies et al. 2014).
The final step was running CMAQ in a two-sequence nesting mode: the first to derive
boundary concentrations and the second, to generate air concentrations and deposition
fluxes over the TKV area. For each meteorology forcing, runs with, and without emissions
changes were performed for the period between 20th December 2016 to 2nd January 2018,
retaining only outputs for 2017 for the 1-km grids. Hourly concentrations of SO2 were retrieved. Total depositions (wet plus dry deposition) of S were calculated using in-built
formulas that add-up relevant species. Chapters 3 and 5 that reported satisfactory modeling of key acidifying air pollutants, and acidic wet depositions in the valley, respectively,
can be consulted for baseline concentrations and deposition fluxes with ERA5.

167

Table 7.1 Meteorological and emissions domain sizes and attributes for SMOKECMAQ modeling with ERA5 and NAM_ANL data. Meteorological simulations
were performed using the MYJ PBL scheme.
Type

Domains (dimensions)

ERA5

NAM_ANL

Parent grid size (km)

25

9

Grid (nesting) ratio

1:5

1:3

Parent (W-E × N-S)

100 x 100

100 x 100

Parent grid center

54.200 °N, 128.600 °W 54.200 °N, 128.600 °W

Meteorology
1st nested (W-E × N-S)

121 x 121

121 x 121

1st nested grid center

53.850 °N, 128.795°W

54.158 °N, 128.671 °W

2nd nested (W-E × N-S)

101 x 121

100 x 121

2nd nested grid center

54.223°N, 128.640°W

54.213 °N, 128.695 °W

Grid centers
Emissions

7.3

54.200 °N, 128.600°W

Boundary grid (W-E × N-S)

40 x 60

60 x 60

TKV grid (W-E × N-S)

36 x 106

36 x 106

Results and analyses

7.3.1 Baseline SO2 levels and changes
For assessing changes in ecosystem exposure to SO2 , baseline pollutant levels are required
and Fig. 7.1a shows modeled ambient concentrations for smelter SO2 emissions at the
previous limit of 27 tonnes day−1 . Spatial plots are for simulations using NAM_ANL
and ERA5 as sources of meteorological data, respectively. Except around the smelter site,

168

ambient SO2 levels are low. Highest annual average concentrations are within a plume
directly north of the smelter site, but beyond 15 km northward, concentrations fall to
less than 1 ppb, and less than 0.5 ppb by the halfway distance between Kitimat and Terrace. A southward plume barely exists, which could be expected since SO2 would readily
be diluted by more frequent and stronger winds around the shoreline than encountered
within the valley. The ERA5 simulation generates slightly higher concentrations than that
of NAM_ANL at longer distance away from the smelter. NAM_ANL simulates more precipitation that may wash-out more pollutants, hence lesser amounts that would linger in
the atmosphere. Nonetheless, both simulations are quite consistent in their indication of
pollutant patterns, namely, a small area of intense concentration (> 10 ppb), and onshore
emissions advection and confinement by valley’s west side-walls. Overall, modeled annual SO2 concentration ranged 0–23 ppb.
Emissions alter pollutant baseline and Fig.7.1b shows the percentage increase in ambient
SO2 due to smelter SO2 discharge permit amendment from 27 to 42 tonnes day−1 . It is
evident that there is substantial rise in relative terms in view of at least 50 % increase over
much of the valley area, although this in part would be due to pristine settings and preexisting low levels. Consequently, ambient concentration sensitivity to emissions change
is high at the present stage of industrial development in the valley. The greatest concentrations changes are near the smelter source, which rise by as much as 100 % or more.
Apart from slight differences in percentage rise over the middle and northern sections of
the valley, spatial SO2 concentration gradients for ERA5 and NAM_ANL simulations are
quite similar. For both, the relative rise in ambient SO2 extend throughout the valley area,

169

even though other anthropogenic emissions sources (e.g area sources) were unchanged.
This demonstrates how important SO2 emissions from a single facility is at present in the
airshed.

Figure 7.1 Average annual SO2 levels for (a) ambient concentrations at smelter’s
pre-modernization limit (b) relative increase in ambient concentration for discharge at present maximum threshold.

7.3.2 Critical level exceedances and mapping
While relative changes in pollutant levels are useful, absolute concentrations are important for quantifying the magnitude and intensity of exposure of environmental compo170

nents to ambient pollution. Persistent SO2 exposure of agricultural crops, and vegetation
is widely associated with decline in farm yield and forest health, respectively. For this reason, modeled SO2 concentrations for incremental emissions from the modernized smelter
and allied activity are referenced to critical levels (CLv) outlined by the World Health Organization (WHO 2000) for floral communities. This allows identifying and mapping the
spatial extent of excessive SO2 pollution that can be detrimental to vegetation. Because
accumulated temperature sum above +5 ◦ C is > 1000 ◦ C·days per year in the TKV, the SO2
concentrations thresholds of 7.2 ppb for protecting forests and natural vegetation, and 3.6
ppb for sensitive lichen were deemed applicable CLv. Exceedance is the concentration of
modeled SO2 above the CLv and spatial plots (for lichen alone) are displayed in Fig. 7.2.
As expected, SO2 CLv exceedances occur around the smelter location. For vegetation
and forest (Table 7.2), aerial SO2 exceedance is estimated to enclose an additional 2 km2
from the permitted maximum emissions during the pre-modernization period, alongside
1.4–1.6 ppb more SO2 exposure, or roughly a 20 % increase in intensity over the original affected areas. For lichen, an additional 6–7 km2 spatial exceedance area is modeled,
meaning a 50–88 % increase in affected area from the baseline. Exceedances in the original
areas are also estimated to increase by 1.3–1.4 ppb or about 30% on average. Westward,
exceedance spread appears curtailed by bounding valley walls which is quite close to the
smelter site. Instead, new exceedance areas are an elliptical area of elevated SO2 plume,
spanning 3 km south (along the shoreline) and 5 km north of the aluminum smelter. There
is some expansion eastward into lands across Kitimat River, although the general area on
the east side of the Douglas Channel is unaffected. In fact, even for pre-modernization

171

Figure 7.2 SO2 spatial exceedance for lichen in the Kitimat area at smelter’s premodernization emissions limit (27 tonnes day−1 ) and maximum threshold (42
tonnes day−1 ). Exceedance is modeled SO2 minus CLv.
SO2 emissions limit, there are no exceedances elsewhere in the domain, alluding to little
impact of other pollutant sources (e.g residential wood burning, transportation) to ambient SO2 levels. Thus SO2 exceedance in the TKV at present primarily depends on the
quantity of smelter SO2 emissions.

7.3.3 Critical load of acidity exceedances and mapping
Annual wet, and dry S deposition output from CMAQ are also totaled for the two periods
of smelter emissions, for calculating critical load of acidity (CLA) exceedances. Because
CLA relate more to effects of S deposition on ecosystem structure and functioning over

172

Table 7.2 Estimated aerial SO2 exceedance of critical levels of vegetation and
lichen exposures due to smelter emissions changes using ERA5 and NAM_ANL
datasets. Exceedance is modeled SO2 minus CLv.

Lichen

and Forest

Vegetation

NAM_ANL

3

7.4

2

0.6

8.8

5

5.5

ERA5

3

8.9

2

0.3

10.5

5

6.4

NAM_ANL

9

4.4

7

0.8

5.8

16

3.6

ERA5

12

3.9

6

0.7

5.2

18

3.7

periods of decades, they indicate the state of existing or required environmental protection in the long-term. In this regard, the critical load that is reported for the area is used.
The average CLA for terrestrial ecosystems in the TKV estimated by Williston et al. (2016)
from steady-state mass balance is equivalent to 29 kg S ha−1 yr−1 . The mass balance
models used as critical limits, base cation to aluminum ratios (Bc:Al) of 1, and 6 as critical
limits for the predominant coniferous forests growing on mineral soils (65 % of the area),
and deciduous forest tree species, respectively (ESSA Technologies et al. 2014, Williston
et al. 2016). Their calculation for the TKV took consideration of mixed vegetation at 1 km2
resolution further facilitating usage in the present study. This CLA value is subtracted
from modeled S deposition fields, thereby providing spatial extent of exceedances (Fig.
173

for cumulative area (ppb)

Average exceedance

area (km2 )

Cumulative exceedance

in previous area (ppb)

Average exceedance

in additional area (ppb)

Average exceedance

area (km2 )

exceedance (ppb)

Average

Total area (km2 )

forcing

Post modernization

Additional exceedance

Atmospheric Pre-modernization

7.3). Area-weighted exceedance magnitudes are also calculated (Table 7.3).

Figure 7.3 Modeled exceedances of CLA from SO2 emissions of aluminum
smelter in Kitimat at pre- and post-modernization rates with ERA5 and
NAM_ANL datasets. Exceedance is modeled sulfur deposition minus CLA

Patterns of CLA exceedance for the two emission periods are broadly similar to ambient
SO2 exceedance of critical limits in Fig. 8.3. However, modeled CLA exceedance areas are
smaller. In particular, the exceedance area does not extend as far south as seen for the SO2
exceedance. It is unclear why this is so, since frequent and large precipitation amounts
in the coastal part of the valley should cause more gaseous SO2 to be removed from the
atmosphere. This probably is an artefact of contouring different quantities (S versus SO2 ).
Nonetheless, CLA exceedance is projected to broaden by 3–4 km2 if the smelter emits SO2

174

Table 7.3 Estimated exceedances of CLA from SO2 emissions of aluminum
smelter in Kitimat at pre- and post-modernization rates with ERA5 and
NAM_ANL datasets.

for cumulative area (ha−1 yr−1 )

Average exceedance

area (km2 )

Cumulative exceedance

in previous area (ha−1 yr−1 )

Average exceedance

in additional area (ha −1 yr−1 )

area (km2 )

Average exceedance

Post modernization

Additional exceedance

exceedance (ha−1 yr−1 )

Average

Total area (km2 )

Pre-modernization

NAM_ANL

6

41.1

4

8.7

49.8

10

33.5

ERA5

8

37.9

3

10.6

48.0

11

37.8

at its increased level. The average exceedances of S deposition for the cumulative and the
pre-modernization areas are quite comparable. But as Table 7.3 also indicates, and analogous to SO2 critical limits, exceedance in the original areas will amplify. From model
outputs, exceedent deposition will increase by 8.7–10.1 kg S ha−1 yr−1 (∼ 21–27 % rise) in
the previous impact area. These increments between emission amounts point to greater
aerial impact of cumulative sulfur deposition nearer the smelter source.

7.4

Discussion

CMAQ model outputs found that a 56 % smelter source SO2 increase translated to ≥
50% increase in ambient levels over the valley. This similarity is expected, since ignor175

ing transformation and removal mechanisms, ambient concentrations are linearly related
to emission amount, and the smelter source is the dominant source in the airshed. The
highest modeled annual concentrations along a 70-km north-south valley transect following emissions increase (Fig. 7.4) is about 1 ppb but it is below this value for nearly the
whole valley length, comparable to levels at remote locations (> 100 km from major anthropogenic sources) in Western Canada (e.g. Hsu 2013). A greater increase in ambient
levels over the valley than over surrounding ridges attested to the role of steep valley
topography in limiting the smelter plume dispersal; however, concentration changes also
depend on the method of discharge. At the pre-modernization SO2 rate of 27 tonnes
day−1 in 2017, emissions at the smelter were collected and vented as pre-heated effluent
from tall stacks. Consequently, SO2 is dispersed over a wide area which perhaps would
not have been the case if source was at ground level. The attenuation of SO2 exposure risk
near the smelter is thus enhanced by long-range transport and deposition. Nonetheless,
the relative rise in ambient SO2 at as far as Terrace not only emphasize the role of valley
atmosphere in redistributing industrial emissions but also the potential of long-range effects from such sources.
Maximum permissible smelter emission was projected to yield SO2 concentrations 3.6–3.7
ppb above thresholds that are dangerous to sensitive lichens, in an area 16–18 km2 . At the
same time, the cumulative area for vegetation that is exposed to SO2 was 5 km2 , with an
average of 5.5–6.4 ppb above the critical level. Because the critical limit for lichen was
lower than for forest vegetation, non-exceedance for lichens assures the spatial range
of short-term protection of plants from ambient SO2 pollution. The expanded effects

176

Figure 7.4 Annual mean concentrations from smelter SO2 source at maximum
emissions rate of 42 tonnes day−1 along a north-south transect through the valley
(right).
area relative to previous exceedance however, suggests increased threat to their abundance. Regarded endemic to the Kitimat foreshore are rare lichens such as Cryptic paw
(Nephroma occultum), Norman (Tholurna dissimilis), and Old-growth specklebelly (Pseudocyphellaria rainierensis) (Besse 2010, Committee on the Status of Endangered Wildlife in
Canada 2006, United States Department of Agriculture 2013), and it is likely that they
will be extirpated in nascent exceedance areas. Such prospect is worrisome considering
the limited natural resilience for some species. Nephroma occultum for instance has poor
dispersal efficiency and is easily displaced by competitors (Committee on the Status of
Endangered Wildlife in Canada 2006). Hence, excessive SO2 concentration could pose an
extra burden for their survival in those areas of exceedance. Still, on evidence of modeled
exceedance, it appeared that pollution risk on lichen growth for areas east of Minette Bay
is low and the advantage of little human disturbance for the most part might sustain local presence. At this time however, it is not known how lichen distribution has changed
177

with industrial SO2 emission rates in the valley because no relevant field survey has been
conducted since more than 30 years.
In contrast, vegetation has been inspected for foliar injury on a biennial basis since 1970.
Inspections are usually along established transects, although conditions at several remote
sites have also been captured. The commonest observations in the vicinity of the smelter
have been leaf chlorosis, leaf notching and tissue necrosis but symptoms to date have
mingled with pests and pathogens attacks, as well as exposure to historically significant
hydrogen-fluoride emissions (Stantec 2015). Tissue concentration of sulfur in westernhemlock needles have also been analyzed at the same time interval as visual inspections.
Foliar sulfur content has slightly risen following increase in SO2 smelter emissions above
the previous 27 tonnes day−1 limit (Rio Tinto 2018). Since SO2 concentrations generally attenuate with distance from the smelter source, the nearest sampling sites would be
sulfurous. Specifically, 5 out of 8 vegetation sample locations having the highest sulfur
content in hemlock foliage in the Stantec report are within the gridded exceedance of SO2
exposure at pre-modernization emission limits. It thus can be expected that foliar S will
rise in vegetation that falls under the new SO2 exceedance area.
Acidification of terrestrial ecosystems, assuming the smelter emits SO2 at maximum permissible amount was estimated to comprise an area of 10–11 km2 centered around the
facility, with mean exceedance of 33.5–37.8 kg S ha−1 yr−1 . Estimated spatial exceedance
was dependent on a generalized critical load of acidity for the area which in turn relied
on calculations of base cation weathering rates, base cation uptake (removal) in harvested
biomass, critical acid neutralizing capacity, and several other predictor variables (ESSA
178

Technologies et al. 2014, Williston et al. 2016). Different methodological approaches in
the derivation of critical loads can result in different conclusions about soil acidification
status. For example, recent analyses across sites in the TKV (Levasseur et al. 2020) that
combined mineralogy-based and particle-size properties adapted to local factors for calculating base cation weathering rates, produced high critical loads and no exceedance
of S deposition. The critical load as developed by Williston et al. (2016) and used in the
present study derives from a widely-applied texture-based function (Sverdrup and Warfvinge 1995) for estimating surface area and weathering rates (United Nations Economic
Commission for Europe 2017, Whitfield et al. 2018). Therefore, even as acidification due
to new smelter SO2 emission is projected to increase over a limited area, deposition estimates in this study represent a worst-case prediction.
Although the approach to emissions release such as using tall stacks, high exit velocity,
etc. alleviates SO2 vegetal exposure in the immediate surrounding, surveillance for indirect pollution is still advisable. Observations (Stantec 2015) have found poor growth
of vegetation in proximity to the smelter site (within 1 to 2 km) suggesting plant toxicity
effect of soil acidification. This is unsurprising considering that the smelter has been in
continuous operation for decades; however, such outcomes can persist or become severe
with future operations. Indeed, that modeled average exceedance of S deposition was
more than 100 % the critical load, and extended as far north as 5 km from the facility,
foretells further soil deterioration. Whether significant ecological values will be lost if
projections are borne out, is unclear since the spatial footprint of adverse impact would
mainly be on lands pre-zoned for industrial use in Kitimat (see Fig. 1.3). Nonetheless, on

179

account of the possibility of about 1225 tonnes excess S in 35 years of future aluminum
production being received over an area that can contain 333,000–400,000 mature trees (see
Appendix I), remedial measures may well be sought much in advance of the smelter’s decommissioning.

7.5

Concluding remarks

The assessment of the effect of a 56 % increase in maximum SO2 emissions from a large
aluminum smelter in Kitimat, with an atmospheric-chemistry model projected 50–88 %
increase in aerial exceedance of limit for protection of lichen and 37–67 % increase in spatial exceedance of threshold for protection of soils. In total, 16–18 km2 of plant habitat
and 10–11 km2 of soil in an area contiguous with the smelter site will likely be damaged
by its SO2 emissions at the maximum permitted levels.
Smelter emission of nitrogen oxides (NOx ), though included in the modeling are minor,
and were not part of the analyses. However, several other industrial projects (e.g. liquefied natural gas facilities) are being developed in the valley, which when operational,
will emit large amounts of SO2 and NO x . These would add to total pollutant emissions
and the next chapter will examine their potential to further expand aerial exceedance of
critical levels and critical loads.

180

Chapter Eight
Quantifying incremental and cumulative
terrestrial ecosystems impacts of NOx
and SO2 emissions from LNG operations
in the Terrace-Kitimat valley of
northwestern British Columbia
Abstract
Natural resource development projects in British Columbia, Canada are subject to impact
assessments to protect health and the environment. This study modeled incremental and
cumulative terrestrial ecosystem effects of NOx and SO 2 emissions from two planned
LNG facilities in the Terrace-Kitimat valley, a predominantly pristine coastal airshed in
northwestern British Columbia. Emissions from these projects will cause at least 50 %

181

and 150 % rise in ambient SO2 and NO x , respectively; however only an additional 4 km2
will be exposed to SO2 concentrations that is directly harmful to vegetation. Cumulative
NOx concentrations are expected to remain below harmful levels. For indirect vegetation impacts via soil pollution, limited areal exceedance (≤ 5 km2 ) of nitrogen deposition
will barely be altered (0–1 km2 increase) by new emissions. Incremental SO2 discharge
translated to 3 km2 extra exceedance of critical load of acidity and 13–14 km2 in total. Total area-weighted sulfur exceedance ranged 29.7–35.0 kg ha−1 yr−1 . Setting target loads,
routine liming and ample distance offset of emission areas from sensitive habitats are
measures that can mitigate soil acidification and irreversible changes in vegetation composition.

8.1

Introduction

The quest by the Canadian province of British Columbia (BC) to be a major player in the
global energy market by adding value to natural resources has meant increased commercialization of the downstream petroleum sector. One strategy is to site liquefied natural
gas (LNG) processing and storage facilities at coastal locations for export overseas. LNG
is natural gas that has been condensed into liquid at approximately -162 ◦ C, near atmospheric pressure, for easy and safe maritime transport (Center for Liquefied Natural Gas
2020). LNG has fewer emissions than other conventional fuels such as diesel and gasoline, also serving a viable alternate energy source for industrial, domestic and commercial
uses (Center for Liquefied Natural Gas 2020, Smajla et al. 2019). Its production for export
is hinged on abundant shale gas reserves in the northeastern part of BC (BC Oil and Gas
182

Commission 2020), concomitant with high demand in the Asia-Pacific region (Aguilera
et al. 2014, BP 2020). Investment in the LNG business is expected to boost government
revenue and provide more jobs. Potential monetary benefits notwithstanding, the location of LNG projects in pristine areas has caused concerns about adverse environmental
effects. Other anxieties related to the LNG industry such as opposition from some First
Nations communities to the construction of pipelines (The Interior News 2020) have also
been reported.
Construction of new LNG infrastructure in BC falls under large-scale projects that are
subject to assessments and reviews by the government and the public through an environmental regulatory and permitting system. This process includes the evaluation of
likely effects on key-value components such as the atmosphere. LNG projects when operational, emit sulfur dioxide (SO2 ), nitrogen oxide (NO), nitrogen dioxide (NO2 ), and
other air pollutants. Exposure of plants to high levels of SO2 and NO x (NO2 + NO) can
cause direct injury to leaves (Gheorge and Ion 2011), reduce crop yield through disruption
of photosynthetic ability (Sun et al. 2016) and increase plant susceptibility to other environmental stresses, e.g. pests (Blande et al. 2014, Vacek and Matějka 2010). Further, SO2
and NO2 have acidifying properties which alongside other sulfur- and nitrogen-bearing
compounds, can be deposited to soils. Excessive sulfur and nitrogen in base-poor soils
can make them more acidic, causing leaching of nutrients and decline in forest productivity (Chumanová-Vávrová et al. 2015, Duan et al. 2016). Surplus soil nitrogen can cause
certain plants to proliferate at the expense of others, leading to the disappearance of native flora (Bobbink et al. 2010). While injury to plants from acute and chronic exposure

183

to acidifying air pollutants may be detected through field inspections, direct observation
of future scenarios is not possible. Addressing potential atmospheric pollution is complicated because the effects on ecological systems are lagged, making it difficult to ascertain
the spatial progression of damage.
In BC, regulatory atmospheric effects assessment of large development projects is conducted with the Lagrangian CALPUFF dispersion model (Sakiyama 2015). Although accepted as a screening modeling technique for long-range (≥ 50 km) transport, CALPUFF
is typically not recommended for setting progress goals for multiple sources of very reactive pollutants (USEPA 2015, 2017). CALPUFF uses fixed, uniform concentrations of important oxidants such as ozone and neutralizing agents such as ammonia, with an overly
simplified representation of secondary-formed chemical species (USEPA 2015). Methodological approaches so far are also less attentive to incremental impacts of industrial activities. Overly aggregated air dispersion modeling, whereby the contribution of an industrial sector is indistinct from pre-existing baselines (e.g ESSA Technologies et al. 2014,
2016), blur their indication of culpability for potential ecological harm. The foregoing not
only suggest the need for procedures that are more consonant with environmental accounting but also, to using tools that are reflective of the state-of-the-science in chemical
dispersion phenomena.
The goal of this chapter is to predict atmospheric deposition of acidifying pollutants
from LNG operations in Kitimat, British Columbia, Canada. It uses the CMAQ model
(https://www.epa.gov/cmaq) to simulate how terrestrial ecosystems exposure to acidifying pollutants may evolve as precursor emissions occur. This study therefore addresses
184

the following questions: 1.) To what additional extent will vegetation be exposed to harmful NOx and SO2 concentrations from LNG projects? 2.) How much direct vegetation
exposure to harmful NO x and SO2 concentrations will result from aggregate industrial
emissions? 3.) By how much will soil nitrogen enrichment and acidification change as
a result of the LNG industry? 4.) What aerial exceedances of critical loads of nitrogen
and sulfur deposition will arise from aggregate industrial emissions? The remainder of
this chapter is organized as follows. In section 8.2 , the numerical modeling approach,
together with the processing and gridding of anticipated LNG pollutant emissions, are
described. Analysis of model outputs, in the context of incremental and aggregate exceedance of wellness thresholds of key forest components are presented in section 8.3. A
discussion of land ecosystem modeling and management implications appears in section
8.4. Section 8.5 concludes the chapter.

8.2

LNG emissions and numerical modeling set-up

Deposition modeling with WRF-SMOKE-CMAQ, forced by separate meteorological datasets
(NAM_ANL and ERA5) follows the same procedure described in Chapter 7. The difference is in the processing of emissions from LNG operations for simulations with, and
without their inclusion. Two LNG projects whose realization are more proximate in time
than all others were considered. These are LNG Canada (currently under construction)
and Kitimat LNG (currently waiting a final investment decision) (Fig 8.1). Emission estimates of major air contaminants (NOx , SO 2 , VOC and PM10 ) for the two projects relied on
information from proponents’ descriptions and scoping reports (ESSA Technologies et al.
185

2014, Impact Assessment Agency of Canada 2019). Emissions were modeled under three
major categories: sources at production sites; hoteling carriers at terminals; and LNG carriers in transit (Fig 8.2). Combined pollutant emissions at main production sites of LNG
Canada and Kitimat LNG was proportioned based on tentative export capacities. Emissions at terminals were estimated based on arrivals and departures of 350 and 250 LNG
carriers (including assist tugs) annually for LNG Canada and Kitimat LNG, respectively.
The same expected maritime traffic was used to project emissions for carriers in transit.
For both projects, off-berth emissions were modeled as a series of point sources 1-1.4 km
apart (the smallest mean distance between adjacent emission grid cells) along the Douglas Channel up to the boundary of the modeling domain. While the number of discrete
points was 43 for LNG Canada, that of Kitimat LNG was 29; the difference being due
to the location of their respective terminals. These points were slightly more clustered
nearer the terminal locations to mimic higher maritime traffic emissions on getting closer
to them. All LNG industry-related emissions were processed as elevated point sources.
Stack parameters were obtained from scoping reports or analogous facilities.
Following emissions processing, CMAQ version 5.2 was run to generate air concentrations of NOx , SO2 and deposition fluxes. Simulations for contributions of LNG operations followed a brute-force approach, that is, with, and without LNG emissions. Model
runs were for the period between 20th December 2016 to 2nd January 2018, retaining
only outputs for 2017 for the 1-km grids. Hourly concentrations of NOx , and SO2 were
retrieved. Total depositions (wet plus dry deposition) of N and S were calculated using
in-built formulas in CMAQ that sum appropriate species.

186

Figure 8.1 Locations of two LNG projects in the Kitimat area. Blue markers are
export terminal sites, separate from LNG production sites (red markers).

187

Figure 8.2 Annual emission estimates of major air pollutants from proposed LNG
projects in Kitimat, including on-way shipping along the Douglas Channel.

8.3

Results and analyses

8.3.1 Relative changes in ambient NO x and SO 2 and exceedance of critical levels
Chemical dispersion and transport affect residual pollutant quantities and Fig. 8.3 shows
modeled changes in ambient concentrations of acidifying gases due to LNG emissions.

188

For NOx , a worse-case projection (ERA5) is at least 150 % increase from the industrial
baseline over much of the valley area although this in part would be due to largely to
pristine settings and pre-existing low levels. For example, within the valley, precisely
over Terrace that is more urbanized than Kitimat, increase is negligible (< 20%). Compared to NOx , the change in ambient SO 2 is lower (∼ 30–60 % increase) mainly because
of the existing emissions from an aluminum smelter in Kitimat. The greatest SO2 concentrations changes are projected to be over small areas just outside the valley—around
Kitamaat Village where levels would rise by as much as 100 %. These areas are contiguous with allied LNG activity such as shipping.
Comparison of modeled annual SO2 concentrations to the critical level (CLv : 3.6 ppb
annual average) of protection of lichen (WHO 2000) which is the most sensitive of floral
communities (Fig. 8.3, Table 8.1), shows increase in aerial exceedance by 4 km2 (22–25
% increase). The exceedance increment appears along a north-south axis that marginally
elongates the original elliptical area, which is to be expected since sideways spread of precursor plume by wind, particularly to the west is curtailed by steep topography. Although
industrial source plumes are additive, excedent SO2 does not encompass the Kitimat LNG
plant site, on account of smaller emitted quantity as well as dispersal by stronger winds
in the maritime channel. However, the intensity of SO2 exposure is amplified in the near
field of Canada LNG source. Whereas area-weighted exceedance with and without LNG
emissions are comparable, there is 0.6–0.7 ppb more SO2 exposure over the original exceedance areas with the LNG emissions.
Despite modeled increase, NOx ambient concentrations inclusive of LNG emissions are
189

Figure 8.3 NOx and SO 2 concentration changes relative to levels without contributions from LNG industry using the CMAQ modeling system driven by two
different meteorological datasets.
well below the critical level (15.6 ppb) for protection of vegetation (not shown). Highest modeled annual NO x concentrations in the vicinity of proposed LNG industries was
roughly 5 ppb. Top concentrations were also limited to few grid cells, pointing to anthropogenic release locations as the commonest sources of non-negligible ambient NOx .
Apart from the emitted quantity, shorter atmospheric lifetime of NOx compared to SO2

190

Figure 8.4 Exceedance (modeled concentration minus critical level) of lichen exposure to SO2 with, and without contributions from LNG industry emissions.
(Seinfeld and Pandis 2016), whereby the former undergo several chemical reactions probably contributes to NO x non–exceedance. NOx concentration is mediated by natural
ozone formation and destruction processes, including the formation of several atmospheric radicals. In view that the highest simulated concentrations were an order of magnitude less than CLv throughout the modeling domain, direct harm to vegetation arising
from exposure to NOx concentration should be of low concern.

191

Table 8.1 Estimated exceedances (Modeled SO2 minus 3.6 ppb) of lichen exposure
with and without LNG emissions using ERA5 and NAM_ANL datasets. The
cumulative area is previous area plus new area

NAM_ANL

16

3.6

4

0.7

4.2

20

3.5

ERA5

18

3.7

4

0.8

4.7

22

3.7

for cumulative area (ppb)

Average exceedance

area (km2 )

Cumulative exceedance

in previous area (ppb)

Average exceedance

in additional area (ppb)

area (km2 )

Average exceedance

With LNG emissions

Additional exceedance

exceedance (ppb)

Average

Total area (km2 )

Without LNG

8.3.2 Relative changes in nitrogen and sulfur deposition and exceedance
of critical loads
A portion of emitted NO x and SO2 will be deposited while being transported in the valley,
and Fig. 8.5 shows the relative changes in N and S deposition to the surface that is anticipated with LNG emissions. For both, the increase in deposition from pre-existing deposition amount is modest (20–50 % rise) within the valley but less over high elevations. This
distinction in space, which is analogous to patterns for precursor concentrations in Fig.
8.3, demonstrates the constraining of pollutant transport by valley side-walls. Emitted
SO2 will be deposited more in the valley channel than over the ridges. For S deposition,
only in the vicinity Kitimat LNG facility would the relative increase be large (≥ 150 %),
consistent with completely new industrial development in the area. Beyond this location
192

to the sout, deposition fluxes would barely be altered, suggesting that spatial deposition
would also be controled by onshore winds that weaken as they approach the valley.

Figure 8.5 Nitrogen (N) and sulfur (S) deposition changes relative to the loads
without contributions from LNG industry emissions

Evaluation of exceedance of critical load of nutrient nitrogen (CLNnut ) indicates that LNG
NOx emissions would scarcely result in new exceedances, even around project sites. The
CLNnut is an empirical deposition of 4 kg N ha−1 yr−1 below which semi-natural ter-

193

restrial habitats are protected. In the present, within ≤ 5 km2 combined area around
Terrace is the CLNnut exceeded, with an additional 0-1 km2 projected for new emissions
(Table 8.2). LNG operations probably would lead to a small increase in the magnitude
of exceedance (∼ 0.2–0.3 kg ha−1 yr−1 more N overall). Qualitatively, excess N deposition from LNG emissions project low impact, not only on account of negligible affected
area but also from an understanding that aerial exceedance, if any, would bear mostly
on lands already under urban use. Nitrogen deposition in the valley at present, pertains
to the Terrace area, and stems from very local sources. The major nitrogen-bearing emissions around Terrace are from road and rail transportation, rather than industrial sources.

Table 8.2 Estimated exceedances (modeled nitrogen deposition minus 4 kg ha−1
yr−1 ) of critical load of soil nutrient nitrogen with, and without LNG emissions
using ERA5 and NAM_ANL datasets.

for cumulative area (ha−1 yr−1 )

Average exceedance

area (km2 )

Cumulative exceedance

in previous area (ha−1 yr−1 )

Average exceedance

Average exceedance

area (km2 )

in additional area (ha −1 yr−1 )

With LNG emissions

Additional exceedance

exceedance (ha−1 yr−1 )

Average

Total area (km2 )

Without LNG

NAM_ANL

5

1.0

0

-

1.2

5

1.2

ERA5

5

0.7

1

1.0

1.0

6

0.8

194

LNG SO2 emissions on the other hand, are expected to broaden aerial exceedance of critical load of acidity (CLA) by 3 km2 or roughly 30% increase (Fig. 8.6, Table 8.3). The
baseline exceedance derives from emissions of a long-standing aluminum smelter nearby
Canada LNG, and the additional area is mainly west of the Kitimat River (Fig. 8.6), again
reflecting the deflection of air pollutant movement by steep topography. Nowhere else,
including at the Bish Cove site of Kitimat LNG is S deposition exceedance expected. Indeed, modeled exceedance of CLA does not extend as far south as in Fig. 8.4, despite
anticipated increased shipping activity and concomitant SO2 release. It is inferred that
only areas in very close proximity of the existing smelter SO2 source will be subject to
additional sulfur depsition from Canada LNG. The projection is that original CLA exceedances areas would be impacted by a further 4–6 kg ha−1 yr−1 of S due to LNG facilities.

8.4

Discussion

Atmospheric chemistry modeling projected as much as 50% and 150 % increase in ambient SO2 and NO x respectively in the Terrace-Kitimat valley due to emissions from LNG
projects. Along a north-south valley transect, these translate to maximum annual air concentrations of 5.2 ppb NOx and 1.2 ppb SO2 (Fig. 8.7) with LNG industry operations.
Projected NOx concentrations are less than the monitored Canada-wide and BC averages
of 7.8 and 9.2 ppb, respectively for NO2 , and much less than the 17 ppb national standard
for 2020 (ECCC 2018b). The highest modeled SO2 concentration is slightly more than
the 1.0 ppb Canada-wide and BC averages, but much less than the 5 ppb Canadian stan195

Figure 8.6 Spatial exceedance (modeled sulfur deposition minus 29 kg ha−1 yr−1 )
of CLA with, and without contributions from LNG industry using ERA5 and
NAM_ANL datasets.
dard (ECCC 2018b). The majority of modeled NOx and SO 2 concentrations, accounting
for LNG industry emissions, are comparable to ambient levels that have been reported
at locations 100-200 km from large industrial sources in Western Canada (e.g. at Fort
Chipewyan in Alberta, NOx : 0.7–2.0 ppb, SO2 :< 0.8 ppb (Cho et al. 2017, Hsu 2013)).
These suggest that LNG emissions in the immediate future would not result in violation
of national and provincial objectives for atmospheric quality. Unlike many regions however, the study area is geographically-constrained and recent increased industrialization,

196

Table 8.3 Estimated exceedances of critical load (29 kg ha−1 yr−1 ) of sulfur deposition and without LNG emissions using ERA5 and NAM_ANL datasets.

for cumulative area (ha−1 yr−1 )

Average exceedance

area (km2 )

Cumulative exceedance

in previous area (ha−1 yr−1 )

Average exceedance

in additional area (ha −1 yr−1 )

area (km2 )

Average exceedance

With LNG emissions

Additional exceedance

exceedance (ha−1 yr−1 )

Average

Total area (km2 )

Without LNG

NAM_ANL

10

33.5

3

5.0

37.1

13

29.7

ERA5

11

37.8

3

4.9

43.2

14

35.0

would require prolonged monitoring of pollutant concentrations for adaptive emissions
management.
Although the relative changes in ambient levels are greater than for SO2 , incremental
NOx emissions were less significant of vegetation exposure. It is not likely that spatial exceedance of critical limits of NOx concentrations for sensitive lichen will occur with LNG
operations, similar to projections by ESSA Technologies et al. (2014) and Williston et al.
(2016). Virtually no additional area will exceed the critical load of N. Current and projected N exceedances in the limited Terrace area were fractions of the critical load which
implied low increment in N accumulation. The assessment of little aerial impact of N
deposition was due separate areas of the present, and future sources of NOx . Future LNG

197

Figure 8.7 Projected annual mean NOx and SO2 concentrations due to cumulative
(total) emissions along a north-south transect through the TKV (right).
facilities would be 60–75 km from the present urban core (Terrace) of vehicular NOx emissions. Industrial point emissions dilute and decrease their ambient concentration with
longer travel distance. Consequently, the aggregate effect of NO x is less than it would
be if emission stacks were closer to the principal area source. For SO2 on the other hand,
new industrial source locations are nearer to an existing source of high emissions.The
Canada LNG plant site for example, will be within 2 km of Rio Tinto’s smelter that is
allowed up to 42 tonnes day −1 SO2 discharge, hence the modest extension of detrimental
concentrations and deposition. Therefore, from an environmental sustainability perspective, clustering industrial activity vis-a-vis siting them at disparate, hitherto undisturbed

198

locations, should be carefully considered in planning and proposing new projects.
Vegetation in an area of 20–22 km2 was projected to experience harmful exposure to total
SO2 emissions. Routine inspection of foliar injury alongside tissue analysis of S content in
hemlock needles in the valley, dating from several decades has continued to the present.
The trend in recent years has been a rise in foliar sulfur with increasing SO2 emissions
(Rio Tinto 2018), although the cause of higher occurrence of leaf chlorosis and necrosis on native plants in the vicinity of existing industrial source has often been unclear.
However, on the evidence that modeled aerial exceedance is co-extensive with frequent
identification of plant injury, direct SO2 exposure likely contributes to deleterious effects,
and will be a major determinant of vegetation health in areas near proposed facilities.
Further away (∼ 10 km) from project sites, there may be potential for enhanced sensitivity of conifers to pollution stress during needle elongation. Understory herbs are likely
to be more impacted than trees especially in the growing season (late April to September) when leaf broadening and drier conditions facilitate leaf surface accumulation and
intake. Adverse response would differ among floral components; however, except for the
west-side-valley wall area around the shoreline, sub-alpine vegetation is particularly not
threatened. The safety of high-elevation forest however, may not be absolute considering
that exceedances were based on concentrations in ambient air (the lowest model layer)
and not on in-cloud SO2 mixing ratios nor sulfate quantities in fog which are quite frequent in the Kitimat area.
Due to aggregate industrial emissions, an area of 13–14 km2 was predicted to exceed the
critical load of acidity for protection of forests on the predominantly mineral soils in the
199

study area. This estimate is within the range (1–28 km2 ) of areal exceedance projected by
a government-led study with the CALPUFF model (ESSA Technologies et al. 2014). The
CALPUFF study included emission scenarios from other projects (e.g. bitumen refinery,
hydropower station) that are no longer seriously advanced by proponents. Apart from
emission totals, differences in modeling technique and uncertainty in model outputs can
lead to different projections. For instance, the CALPUFF study predicted concentrations
that are approximately double (227 %) the measured SO2 concentrations. Further it assumed 80 ppb ozone concentration over the TKV, possibly resulting in high SO2 to sulfate
conversion and total sulfur deposition rates than would occur when using site-specific
ozone data. In the present study with CMAQ, gridded ozone mixing ratios are simulated
in-situ, but hardly exceeded 40 ppb that approximates observed spring season maximum
concentrations. CMAQ-predicted annual mean SO2 concentrations was roughly 50 %
overestimation of Kitimat monitoring data. The indication thus, is that CMAQ acid deposition modeling is reasonably precise, whilst providing oxidant chemistry that is preferred (USEPA 2017) for regulatory assessment of multi-air pollutant discharges.
How much restoration would be required for affected areas by acidification is as important as identifying terrestrial habitats at risk. Although cumulative areal CLA exceedance
is small, projected area-weighted acidity loading of 29.7–35.0 kg S ha−1 yr−1 is slightly
greater than the CLA itself (29.0 kg S ha−1 yr−1 ). This also, is about an order of magnitude greater than exceedances for forest soils reported for most parts of eastern North
America (Duarte et al. 2013, McNulty et al. 2007), where acidifying industrial emissions
has long been a major environmental challenge. Soils recovery from acidification has

200

been documented in regions that have experienced significant SO2 emissions reduction
(Lawrence et al. 2015), but regeneration of original plant communities has also been less
successful. Indeed, the danger of prolonged acidic deposition is the alteration of soil
chemistry ultimately resulting to conditions that are favourable to fewer species. Because
irreversible changes in habitat structure and composition can occur with new CLA exceedances, timely mitigation of acidic deposition is needed. Setting target loads of S and
N deposition is one step to prevent adverse changes to the natural environment. Counteracting base cation depletion through liming could also be beneficial in maintaining the
growth of native vegetation nearby LNG emission locations especially in view that acidifying intensity may span several future decades. With industrial development at pristine
valley sites, it is important that the environment, including vegetation resources are fully
profiled to ensure that future emissions are distant from the most valuable areas.

8.5

Concluding remarks

Estimates of incremental impacts of major sources of atmospheric pollutants are useful,
without which it will be difficult tracking the course of departures from environmental
baselines. Emissions from two LNG projects in Kitimat will cause at least 50% and 150 %
rise in ambient SO2 and NOx , respectively; however only an additional 4 km2 will be exposed to SO2 concentrations that are directly harmful to vegetation. An additional 3 km2
exceedance of critical load of acidity, with sulfur exceedance ranging 29.7–35.0 kg ha−1
yr−1 for a cumulative 13–14 km2 , suggested the importance of setting target loads and
routine liming to prevent soil acidification and irreversible changes in vegetation compo201

sition in the vicinity of new emissions. Cumulative NO x concentrations are expected to
remain below harmful levels; negligible (0–1 km2 ) increase in areal exceedance of nitrogen deposition is also anticipated. By defining aerial effects of acidifying LNG emissions
on the environment ahead of actual releases, decisioning on liability for damage, compensation, restoration, etc. can be enriched. The environmental accountability perspective of
mapping both incremental and cumulative impacts is therefore recommended for evaluations of future industrial emissions.

202

Chapter Nine
Summary and Conclusion
Optimization of a state-of-the-science platform required evaluating physics options and
input data to extend the usefulness of recent information on critical levels and loads of
atmospheric pollutants over an area with few ground measurements. The modeling for
the industrializing Terrace-Kitimat valley, a physiographically complex region of northwestern British Columbia, Canada, was conducted at 1 km horizontal resolution with the
WRF-SMOKE-CMAQ system. Investigations centered on representativeness of meteorological fields and air pollutant concentrations, evaluation of compliances to atmospheric
quality standards, and quantification of acidification impact of industrial emissions.

9.0.1 Recapitulation of research findings
A recap of findings of this research in the context of the questions in Chapter 1 is outlined:
1. What explains the behaviors of alternative PBL schemes in the WRF model for simulating the TKV’s surface meteorology?
Analyses of spatial differences among PBL schemes for surface variables suggested some
203

factors that could have led to varying representations of meteorological variables that are
of interest to atmospheric quality modeling. The coupling of PBL schemes to alternative surface-layer schemes which also differed in their diagnosis of variables in the lowest model layer such as skin temperature and friction velocity led to greater nighttime
temperatures with the TKE-based schemes (MYNN3, MYJ and UW) than the first–order
schemes. The inherent design for certain physical processes, for example moist convection and cloud-topped boundary layers with MYNN3 and UW PBL schemes was responsible for output of greater water vapor content with these schemes than those not specifically addressing these features e.g. the YSU and SH PBL schemes. Varying estimates of
PBL depth attested to distinct methods for diagnosis of mixing heights, which also hinged
on whether eddy transport across layers was modeled locally, nonlocally or in a hybrid
fashion by individual schemes. The YSU and SH PBL schemes output similar properties
apparently due to modeling at 1 km spatial resolution.
2. How does the choice of PBL scheme affect the simulation of PM2.5 , SO2 and NO2 concentrations and what reasons may account for differences in model outputs?
For modeling the above air contaminants, it was found that the choice of PBL parameterization influenced quantitative and qualitative agreement with observational data. The
top-ranked PBL schemes were MYNN3 for PM2.5 and MYJ for NO2 . Both schemes ranked
high for absolute SO2 levels, but the MYJ and ACM2 schemes qualitatively better emulated peak summertime diurnal concentrations near elevated point sources in Kitimat.
Greater nighttime SO2 concentrations with MYNN3 and the YSU PBL schemes, in less
agreement with station monitoring 8 km downwind of emissions from tall stacks, sug204

gested sustained pollutant mixing and downward transport within the nocturnal boundary layer due to their representations of nonlocal mass flux transfers between model layers. With these schemes, inland penetrations of pollutant plumes were farther than those
of ACM2, MYJ, and UW PBL schemes. For NO2 and PM2.5 that mainly discharged passively from fugitive, ground-level sources, hence are less accurately quantified than SO2
emissions, the MY- PBL schemes more reasonably reproduced peak season concentrations
than other schemes. It was concluded that for air pollution modeling in rugged, remote
areas, the mode of pollutant emissions is important for the choice of a PBL scheme.
3. Could quantile-based bias correction of CMAQ output improve usefulness for assessing regulatory compliance to air quality objectives?
Linear regression formulas based on quantiles-mean matching of model output to observational data reduced quantitative biases in PM2.5 modeling especially for peak autumn
and winter season concentrations. On average for year 2017, absolute errors in simulating
annual means and 98th percentiles of daily averages, were 11 % and 10 %, respectively
for bias-corrected outputs, compared to 45 % and 61 %, respectively for raw outputs.
Initial non-correspondence in risk management categorization of pollutant exposure was
corrected, thereby allowing for projections over unmonitored areas.
4. How well do simulations with various PBL parameterizations capture wet deposition
of acidifying ions and what are the baseline exceedances of sulfur and nitrogen deposition
in the TKV area?
The skill of the WRF-SMOKE-CMAQ modeling system for wet deposition of acidifying

205

ions varied with nearness of measurement locations to major anthropogenic source of
precursors and the quantities of individual species. Modeled quantities for sulfate and
ammonium were more matched with observations at the Lakelse Lake station where the
MYNN3, MYJ and YSU PBL schemes produced slight overestimations than at Haul Road
station that is on the fence line of industrial pollutant sources. Normalized biases tended
to be smaller for ammonium than for nitrate and sulfate that occurred in greater amounts.
In 2017, forest soils were estimated to be in exceedance of critical load of acidity in the
vicinity of the Rio Tinto smelter by 30.1–53.5 kg S ha −1 yr−1 . Exceedance of critical load
of nutrient nitrogen restricted to the Terrace area (≤ 7 km2 ) ranged between none and
0.71 kg ha −1 yr−1 .
5. What choices of atmospheric forcing data and PBL scheme in the WRF model can reproduce precipitation for the purpose of projecting atmospheric deposition over the TKV
and what is the uncertainty in the precipitation field with the best fit simulations?
The MYJ and MYNN3 PBL schemes that best quantified ambient concentrations of air pollutants in the valley alongside three coarse-resolution atmospheric datasets that are available in formats compatible with the WRF model were further assessed for how closely
their outputs emulate spatial and temporal distribution of precipitation. Underestimation by more than 40 % on average of total precipitation ranging 1170–2380 mm in 2017,
was found for simulations using either NARR for atmospheric forcing or MYNN3 as a
PBL scheme. The large precipitation undercapture with this combination likely stemmed
from little assimilation of observational data in the reanalyses and poor adaptation of
the MYNN3 PBL scheme for frontal precipitation that dominate the wet months. With
206

NAM_ANL and ERA5 as atmospheric forcing data and MYJ as the PBL scheme, the uncertainty in annual simulated precipitation amount ranged between 40 % overestimation
and 20 % underestimation of observational data.
6. What incremental and aggregate impacts could arise should the existing smelter at
Kitimat emit 42 tonnes day−1 that is the highest permissible rate?
An additional 15 tonnes daily SO2 emission was projected to result in new aerial exceedances of 6– 7 km2 and 3–4 km2 of the critical level of plant exposure and critical load
of acidity respectively. Exceedances of SO2 concentrations and sulfur deposition over
pre-existing impacted areas were predicted to rise by 21–27 % and 30–33 %, respectively.
Cumulatively, 16–18 km2 of plant habitat and 10–11 km2 of soil in an area contiguous with
the smelter site will likely be damaged by its SO2 emission under the latest regulation.
7. For proposed developments in the TKV,
(a) To what additional extent could vegetation be exposed to harmful NOx and SO2
concentrations from LNG projects?
(b) How much direct vegetation exposure to harmful NOx and SO2 concentrations could
result from aggregate industrial emissions?
(c) By how much could soil nitrogen enrichment and acidification change as a result of
the LNG industry?
(d) What aerial exceedances of critical loads of nitrogen and sulfur deposition will arise
from aggregate industrial emissions?

207

In the event that two LNG projects are built and in operation, an area of 4 km2 could further be exposed to SO2 concentrations that are directly harmful to vegetation, with an average exceedance of 0.7–0.8 ppb in the new areas and 17–27 % increased magnitude over
pre-existing impacted areas. For aggregate emissions, a 20-22 km2 area above SO2 critical
limit for lichen, with average exceedance of 3.5–3.7 ppb was predicted. Cumulative NO x
concentrations are expected to remain below harmful levels. For indirect vegetation impacts via soil pollution, pre-existing areal exceedance of nitrogen will barely increase (0–1
km2 ), with overall average exceedance of 0.8–1.2 kg ha−1 yr−1 . Additional 3 km2 new
aerial exceedance of critical load of acidity, for a cumulative 13–14 km2 area and 29.7–35.0
kg ha−1 yr−1 average exceedance was also predicted. These projections assume that new
LNG facilities will emit all NOx , SO2 and other air pollutants from elevated point sources.

9.0.2 Significance of study findings
By comparing model outputs with actual measurements, this study guides what configurations in the WRF meteorological driver are more appropriate for the northwest BC
coast that is dominated by sharp elevation gradients, and frequently humid conditions.
This importance is in view of errors that could arise from inexactness of emissions input,
terrain representation, and chemical reaction chains that are processed within CMAQ.
Especially for the major air pollutants (SO2 , NO2 , PM 2.5 ), the evaluations of whether the
CMAQ model captures observed quantities for various PBL parameterizations are useful. PBL schemes development and verification have mainly been for numerical weather
prediction; their realisms of chemical transport and dispersion seldom investigated. The

208

analyses of model outputs for separate runs and different peak seasons provide ample
information on their dependencies on the weather, type of pollutant source, and topography. For example, it was found for the various air pollutants that the UW PBL parameterization was most different from other PBL schemes due to these factors. This work
contributes significantly to atmospheric science literature since to the author’s knowledge, no prior modeling study of PBL schemes influence on air quality variables over
fjord valleys exists.
Model evaluations in this study can be helpful to CMAQ users in general, although the
use of a simple, comprehensible, post-processing technique to ameliorate the initial bias
in model output as was done for PM2.5 , has particular relevance for airshed managers.
In objectively adjusting modeled concentrations, with corrections spread to all grid cells,
a realistic classification of pollutant exposure was achieved. That demonstration, it is
believed, will benefit compliance assessment of pollutant concentrations especially if relying on the output from a single CMAQ configuration.
Projected ambient pollutant concentrations and acidifying deposition from this study
may provide a reference for changes in the atmospheric quality of similarly isolated areas
that are planned for industrialization. It however contributes lastingly through mapping
pre-existing levels of major pollutants in the TKV— an exercise that was not realized until this study. The TKV, like much of northern BC is sparsely monitored for air quality.
However, retrospective modeling with CMAQ allowed not only the quantification of preexisting pollutant levels from natural and anthropogenic sources, but also the estimation
of spatial exceedances of safe limits for the forest ecosytem. Future monitoring and mod209

eling data may be compared to them for changes in the spatial distribution of pollutants.
In this regard, the study’s findings are useful for effective planning and management of
air emissions in the TKV.

9.0.3 Directions for future research
Pollutant concentrations and deposition estimates were from the best fit simulations. Appropriate choice of model physics, however, does not preclude the need for continuous
improvement of initial and boundary source fields. Availability of atmospheric reanalyses at 1 km horizontal gridding for instance would apart from obviating downscaling
errors and enhancing the precision of numerical schemes, reduce computational costs.
Similar improvement for area source emissions inventorying would also facilitate spatiotemporal gridding with the SMOKE tool. With more refining and updating of atmospheric modeling systems, a top priority would be their validation in areas with geographical similarity to the TKV, to allow for more comparisons and creation of unified
benchmarks for high-resolution air quality modeling over complex orography.
High-resolution CMAQ simulations are fundamental to precise projections of future scenarios and the approach of mapping atmospheric deposition exceedances, with and without emission increase is applicable to areas where the permitting of industrial projects
are phased. Even for the TKV area itself, chemical transport models can also linked to
other models to investigate emissions effects on aquatic environment, public health, atmospheric visibility, etc. It is necessary that predictive modeling of big industrial projects
utilize updated science tools to enrich information about the extent and magnitude of

210

environmental impacts.

211

References
Aguilera, R. F., J. Inchauspe, and R. D. Ripple, 2014: The Asia Pacific natural gas market: Large enough for all?

Energy Policy, 65, 1–6, doi:10.1016/

j.enpol.2013.10.014.
Andrews, A., and R. K. Lattanzio, 2013: Petroleum coke: industry and environmental issues. R43263,

http://petcoke.dev.psiclientweb.com/wp-content/uploads/

2014/06/nps71-120213-07.pdf.
Appel, K. W., S. J. Roselle, R. C. Gilliam, and J. E. Pleim, 2010: Sensitivity of the
community multiscale air quality (CMAQ) model v4.7 results for the eastern United
States to MM5 and WRF meteorological drivers. Geosci. Model Dev., 3 (1), 169–188,
doi:10.5194/gmd-3-169-2010.
Appel, K. W., and Coauthors, 2018: Overview and evaluation of the Community
Multiscale Air Quality (CMAQ) modeling system version 5.2. 69–73, doi:10.1007/
978-3-319-57645-9_11.
Arakawa, A., J.-H. Jung, and C.-M. Wu, 2011: Toward unification of the multiscale

212

modeling of the atmosphere. Atmos. Chem. Phys, 11, 3731–3742, doi:10.5194/acp-113731-2011.
Atkinson, R. W., I. C. Mills, H. A. Walton, and H. R. Anderson, 2015: Fine particle
components and health—a systematic review and meta-analysis of epidemiological
time series studies of daily mortality and hospital admissions. J. Expo. Sci. Environ.
Epidemiol., 25 (2), 208–214, doi:10.1038/jes.2014.63.
Augustaitis, A., 2011: Impact of meteorological parameters on responses of
pine crown condition to acid deposition at aukštaitija national park. Balt.
For.,

(17), 205–214,

https://www.balticforestry.mi.lt/bf/PDF_Articles/2011-

17[2 ]/Augustaitis_201117(2)_205_214.pdf.
Avolio, E., S. Federico, M. Miglietta, T. L. Feudo, C. Calidonna, and A. Sempreviva,
2017: Sensitivity analysis of WRF model PBL schemes in simulating boundary-layer
variables in southern Italy: An experimental campaign. Atmos. Res., 192, 58–71, doi:
10.1016/j.atmosres.2017.04.003.
Balzarini, A., and Coauthors, 2014: Sensitivity analysis of PBL schemes by comparing WRF model and experimental data. Geosci. Model Dev. Discuss., 7, 6133–6171,
doi:10.5194/gmdd-7-6133-2014.
Banks, R. F., and J. M. Baldasano, 2016: Impact of WRF model PBL schemes on
air quality simulations over Catalonia, Spain. Sci. Total Environ., 572, 98–113, doi:
10.1016/j.scitotenv.2016.07.167.
BC Oil and Gas Commission, 2020: 2017 oil and gas reserves and production report.
213

BCMOECS,

2020:

Provincial

air

quality

objective

for

pm2.5 .

https:

//www2.gov.bc.ca/gov/content/environment/air-land-water/air/air-qualitymanagement/regulatory-framework/objectives-standards/pm2-5.
Beck, H. E., and Coauthors, 2017: Global-scale evaluation of 22 precipitation
datasets using gauge observations and hydrological modeling. Hydrol. Earth Syst.
Sci., 21 (12), 6201–6217, doi:10.5194/hess-21-6201-2017.
Belo-Pereira, M., E. Dutra, and P. Viterbo, 2011: Evaluation of global precipitation data sets over the Iberian peninsula. J. Geophys. Res., 116, D20 101, doi:
10.1029/2010jd015481.
Besse, W. J., 2010: Old-growth specklebelly lichen surveys – fia final report 2009/10.
41 pp, https://www.for.gov.bc.ca/hfd/library/FIA/2010/LBIP_6906008.pdf.
Blande, J. D., J. K. Holopainen, and U. Niinemets, 2014: Plant volatiles in polluted
atmospheres: stress responses and signal degradation. Plant Cell Environ., 37 (8),
1892–1904, doi:10.1111/pce.12352.
Bobbink, R., and Coauthors, 2010: Global assessment of nitrogen deposition effects
on terrestrial plant diversity: a synthesis. Ecol. Appl., (20), 30–59, doi:10.1890/081140.1, https://doi.org/10.1890/08-1140.1.
Boylan, J. W., and A. G. Russell, 2006: PM and light extinction model performance
metrics, goals, and criteria for three-dimensional air quality models. Atmos. Environ., 40 (26), 4946–4959, doi:10.1016/j.atmosenv.2005.09.087.

214

BP, 2020: Energy outlook: demand by fuel. https://www.bp.com/en/global/
corporate/energy-economics/energy-outlook/demand-by-region.html.
Bretherton, C. S., and S. Park, 2009: A new moist turbulence parameterization in the community atmosphere model. J. Climate, 22, 3422–3448, doi:10.1175/
2008JCLI2556.1.
Brook, J., S. Cober, M. Freemark, T. Harner, S. Li, J. Liggio, P. Makar, and B. Pauli,
2019: Advances in science and applications of air pollution monitoring: A case
study on oil sands monitoring targeting ecosystem protection. J. Air Waste Manage.
Assoc., 69 (6), 661–709, doi:10.1080/10962247.2019.1607689.
Camuffo, D., 2014: Atmospheric stability and pollutant dispersion. In: Microclimate
for Cultural Heritage. 245–282, doi:10.1016/B978-0-444-63296-8.00008-1.
Carolina Environmental Program, 2009: SMOKE v2.6 user’s manual. The University of North Carolina at Chapel Hill, https://www.cmascenter.org/smoke/
documentation/2.6/html/.
CCME, 2012:

Guidance document on air zone management.

https://

www.ccme.ca/files/Resources/air/aqms/pn_1481_gdazm_e.pdf.
CCME, 2014a: 2008-09 Acid Rain Strategy Progress Report. https://www.ccme.ca/
files/Resources/air/acid_rain/2008_09_ar_progress_rpt_pn1458_e.pdf.
CCME, 2014b: AQMS. http://www.ccme.ca/en/resources/air/aqms.html.

215

Center for Liquefied Natural Gas, 2020: What is lng? https://lngfacts.org/aboutlng/what-is-lng/.
Chang, J. C., and S. R. Hanna, 2004: Air quality model performance evaluation.
Meteorol. Atmos. Phys., 87 (1–3), 167–196, doi:10.1007/s00703-003-0070-7.
Cheng, F.-Y., S.-H. Chin, and T.-H. Liu, 2012: The role of boundary layer schemes in
meteorological and air quality simulations of the taiwan area. Atmos. Environ., 54,
714–727, doi:10.1016/j.atmosenv.2012.01.029.
Cheng, L., H. Sandhu, R. Angle, K. McDonald, and R. Myrick, 2000: Rural particulate matter in alberta, canada. Atmos. Environ., 34 (20), 3365–3372, doi:10.1016/
s1352-2310(00 )00123-0.
Cho, S., K. Vijayaraghavan, D. Spink, J. Jung, R. Morris, and R. Pauls, 2017:
Assessment of regional acidifying pollutants in the Athabasca Oil Sands area
under different emission scenarios. Atmos. Environ., 156, 160–168, doi:10.1016/
j.atmosenv.2017.02.038.
Chumanová-Vávrová, E., O. Cudlín, and P. Cudlín, 2015: Spatial and temporal
patterns of ground vegetation dominants in mountain spruce forests damaged by
sulphur air pollution (Giant mountains, Czech Republic). Boreal Environ. Res., 20,
620–636, https://helda.helsinki.fi/bitstream/10138/228299/1/ber20-5-620.pdf.
Cirtina, D., N. Ionescu, and L. Cirtina, 2016:

Environmental impact as-

sessment related to metallurgical industry activities. Metalurgija, 55, 482–484,
https://hrcak.srce.hr/153690.
216

Coats Jr., C. L., 2017: The BAMS/EDSS/Models-3 I/O API: User Manual. https:
//www.cmascenter.org/ioapi/documentation/all_versions/html/AA.html.
Cohen, A., S. M. Cavallo, M. C. Coniglio, and H. E. Brooks, 2015: A review of planetary boundary layer parameterization schemes and their sensitivity in simulating
southeastern U.S. cold season severe weather environments. Wea. Forecasting, 30,
591–612, doi:10.1175/WAF-D-14-00105.1.
Committee on the Status of Endangered Wildlife in Canada, 2006: Assessment
and update status report on the cryptic paw Nephroma occultum in Canada. 28
pp, https://www.sararegistry.gc.ca/virtual_sara/files/cosewic/sr_cryptic_paw_
lichen_e.pdf.
Coordination Centre for Effects, 2017: Manual for modelling and mapping critical loads & levels. Chapters 3 and 5, https://www.umweltbundesamt.de/en/
manual-for-modelling-mapping-critical-loads-levels.
Cosby, B. J., G. M. Hornberger, J. N. Galloway, and R. E. Wright, 1985: Time scales
of catchment acidification. a quantitative model for estimating freshwater acidification. Environ. Sci. Technol., 19 (12), 1144–1149, doi:10.1021/es00142a001.
Crouse, D. L., and Coauthors, 2012: Risk of nonaccidental and cardiovascular mortality in relation to long-term exposure to low concentrations of fine particulate
matter: A canadian national-level cohort study. Environ. Health Perspect., 120 (5),
708–714, doi:10.1289/ehp.1104049.

217

Dennis, R., and Coauthors, 2010: A framework for evaluating regional scale numerical photochemical modelling systems. Environ. Fluid Mech., 10, 471–489, doi:
10.1007/s10652-009-9163-2.
Dietz, T., 2003: The struggle to govern the commons. Science, 302 (5652), 1907–1912,
doi:10.1126/science.1091015.
Dise, N. B., and Coauthors, 2011: Nitrogen as a threat to european terrestrial biodiversity. The European Nitrogen Assessment, M. A. Sutton, C. M. Howard, J. W. Erisman, G. Billen, A. Bleeker, P. Grennfelt, H. van Grinsven, and B. Grizzetti, Eds.,
Cambridge University Press, 463–494, doi:10.1017/cbo9780511976988.023.
District of Kitimat, 2020: Current major projects. https://www.kitimat.ca/en/
business-and-development/current-major-projects.aspx.
Djalalova, I., L. D. Monache, and J. Wilczak, 2015: PM2.5 analog forecast and
kalman filter post-processing for the community multiscale air quality (CMAQ)
model. Atmos. Environ., 108, 76–87, doi:10.1016/j.atmosenv.2015.02.021.
Driscoll, C. T., and Coauthors, 2001: Acidic deposition in the northeastern United
States: Sources and inputs, ecosystem effects, and management strategies. BioScience, 51 (3), 180–198, doi:10.1641/0006-3568(2001)051[0180:aditnu]2.0.co;2.
Du, E., M. E. Fenn, W. D. Vries, and Y. S. Ok, 2019: Atmospheric nitrogen deposition
to global forests: Status, impacts and management options. Environ. Pollut., 250,
1044–1048, doi:10.1016/j.envpol.2019.04.014.

218

Duan, L., Q. Yu, Q. Zhang, Z. Wang, Y. Pan, T. Larssen, J. Tang, and J. Mulder,
2016: Acid deposition in Asia: Emissions, deposition, and ecosystem effects. Atmos.
Environ., 146, 55–69, doi:10.1016/j.atmosenv.2016.07.018.
Duarte, N., L. H. Pardo, and M. J. Robin-Abbott, 2013: Susceptibility of forests in the
northeastern USA to nitrogen and sulfur deposition: Critical load exceedance and
forest health. Water Air Soil Pollut., 224 (2), 1355, doi:10.1007/s11270-012-1355-6.
ECCC, 1991: The potential of soils and bedrock to reduce the acidity of atmospheric
deposition. http://ftp.geogratis.gc.ca/pub/nrcan_rncan/raster/atlas_5_ed/eng/
environment/ecology/mcr4157.jpg.
ECCC, 2018a:
port 2016).

Canada–United States Air Quality Agreement (Progress re-

https://www.canada.ca/en/environment-climate-change/services/

air-pollution/publications/canada-united-states-air-quality-report-2016.html.
ECCC, 2018b:

Canadian environmental sustainability indicators:

Air qual-

ity. https://www.canada.ca/content/dam/eccc/documents/pdf/cesindicators/
air-quality/air-quality-en.pdf.
ECCC, 2019:

Canadian climate normals 1981-2010 station data.

http:

//climate.weather.gc.ca/climate_normals/results_1981_2010_e.html?stnID=
402&autofwd=1.
ECCC, 2020: Air Pollutants Emissions Inventory online search. https://pollutionwaste.canada.ca/air-emission-inventory/.

219

ECMWF, 2019:

ERA5 data documentation.

https://confluence.ecmwf.int/

display/CKB/ERA5datadocumentation.
Emery, C., Z. Liu, A. G. Russell, M. T. Odman, G. Yarwood, and N. Kumar,
2017: Recommendations on statistics and benchmarks to assess photochemical
model performance. J. Air Waste Manage. Assoc., 67 (5), 582–598, doi:10.1080/
10962247.2016.1265027.
Emery, C. A., E. Tai, and G. Yarwood, 2001:

Enhanced meteorologi-

cal modeling and performance evaluation for two Texas ozone episodes.
https://www.tceq.texas.gov/assets/public/implementation/air/am/contracts/
reports/mm/EnhancedMetModelingAndPerformanceEvaluation.pdf.
Environmental Appeal Board, 2015: Decisions: Emily Toews; Elisabeth Stannus v. Director, Environmental Management Act (Rio Tinto Alcan Inc., Third
Party). 1–2, 58–84 pp,

http://www.eab.gov.bc.ca/ema/2013ema007g_010g_

corrigendum_2.pdf.
Environmental Modeling Center, 2017: Mesoscale model/analysis systems faq.
https://www.emc.ncep.noaa.gov/mmb/research/FAQ-meso.html.
ESSA Technologies, J. Laurence, Limnotek, Risk Sciences International, Trent University, and Trinity Consultants, 2014: Kitimat airshed emissions effects assessment.
205 pp, https://www2.gov.bc.ca/assets/gov/environment/air-land-water/air/
reports-pub/2014-04-25_essa-qa_kitimat_airshed.pdf.

220

ESSA Technologies, J. Laurence, Risk Sciences International, Trent University, and
Trinity Consultants, 2016: Supplementary report to the Prince Rupert Airshed
study. 98 pp, https://www2.gov.bc.ca/assets/gov/environment/air-land-water/
air/reports-pub/pr-airshed-study-report-summ.pdf.
ESSA Technologies,

Risk Sciences International,

and Trinity Consultants,

2015: Pm2.5 supplement to the Kitimat Airshed emissions effects assessment.
https://www2.gov.bc.ca/assets/gov/environment/air-land-water/air/reportspub/2015-03-31_essa-pm2-5_suppl-kaeea.pdf, 19 pp.
Evans, J. P., M. Ekström, and F. Ji, 2011: Evaluating the performance of a WRF
physics ensemble over south-east Australia. Climate Dyn., 39 (6), 1241–1258, doi:
10.1007/s00382-011-1244-5.
Fernández-González, S., F. Valero, J. L. Sánchez, E. Gascón, L. López, E. GarcíaOrtega, and A. Merino, 2015: Numerical simulations of snowfall events: Sensitivity
analysis of physical parameterizations. J. Geophys. Res. Atmos., 120, 10 130–10 148,
doi:10.1002/2015jd023793.
Food and Agricultural Organization, 1989: Arid zone forestry - A guide for field technicians. Table 4.1, http://www.fao.org/3/T0122E/t0122e08.htm.
García-Díez, M., J. Fernández, L. Fita, and C. Yagüe, 2013: Seasonal dependence of
WRF model biases and sensitivity to PBL schemes over. Europe. Quart. J. Roy. Meteor.
Soc., 139, 501–514, doi:10.1002/qj.1976.

221

Gerboles, M., and H. I. Reuter, 2010: Estimation of the measurement uncertainty
of ambient air pollution datasets using geostatistical analysis. 38 pp,

http://

publications.jrc.ec.europa.eu/repository/bitstream/JRC59441/lbna24475enc.pdf.
Gheorge, I. F., and B. Ion, 2011: The effects of air pollutants on vegetation and the
role of vegetation in reducing atmospheric pollution. In: The Impact of Air Pollution on Health, Economy, Environment and Agricultural sources. 241–280, doi:
10.5772/17660.
Gilliam, F., 2019: Excess nitrogen in temperate forest ecosystems decreases herbaceous layer diversity and shifts control from soil to canopy structure. Forests, 10, 66,
doi:10.3390/f10010066.
Grell, G. A., and S. R. Freitas, 2014: A scale and aerosol aware stochastic convective parameterization for weather and air quality modeling. Atmos. Chem. Phys., 14,
5233–5250, doi:10.5194/acp-14-5233-2014.
Gunwani, P., and M. Mohan, 2017: Sensitivity of WRF model estimates to various PBL parameterizations in different climatic zones over India. Atmos. Res., 194,
43–65, doi:10.1016/j.atmosres.2017.04.026.
Habashi, F., 2011: Pollution problems in the metallurgical industry: A review. J.
Mining Environ., 2 (1), 17–26, doi:10.22044/jme.2012.16.
Hariprasad, K. B. R. R., C. V. Srinivas, A. B. Singh, S. V. B. Rao, R. Baskaran, and
B. Venkatraman, 2014: Numerical simulation and intercomparison of boundary

222

layer structure with different PBL schemes in WRF using experimental observations at a tropical site. Atmos. Res., 145, 27–44, doi:10.1016/j.atmosres.2014.03.023.
Hedley, M., and D. L. Singleton, 1997: Evaluation of an air quality simulation of
the lower fraser valley-i. meteorology. Atmos. Environ., 31, 1605–1615, doi:10.1016/
S1352-2310(96.
Hong, S.-Y., Y. Noh, and J. Dudhia, 2006: A new vertical diffusion package with
an explicit treatment of entrainment processes. Mon. Wea. Rev., 134, 2318–2341, doi:
10.1175/MWR3199.1.
Hsu, Y.-M., 2013: Trends in passively-measured ozone, nitrogen dioxide and sulfur
dioxide concentrations in the Athabasca oil sands region of Alberta, Canada. Aerosol
Air Qual. Res., 13 (5), 1448–1463, doi:10.4209/aaqr.2012.08.0224.
Hu, J., J. Chen, Q. Ying, and H. Zhang, 2016: One-year simulation of ozone and particulate matter in China using WRF/CMAQ modeling system. Atmos. Chem. Phys.,
16 (16), 10 333–10 350, doi:10.5194/acp-16-10333-2016.
Hu, X.-M., J. Nielsen-Gammon, and F. Zhang, 2010: Evaluation of three planetary
boundary layer schemes in the WRF model. J. Appl. Meteor. Climatol., 49, 1831–1843,
doi:10.1175/2010JAMC2432.1.
Hu, X.-M., M. Xue, R. A. McPherson, E. Martin, D. H. Rosendahl, and L. Qiao,
2018: Precipitation dynamical downscaling over the Great Plains. J. Adv. Model.
Earth Syst., 10 (2), 421–447, doi:10.1002/2017ms001154.

223

Hu, Y., N. Bellaloui, M. Tigabu, J. Wang, J. Diao, K. Wang, R. Yang, and G. Sun,
2015: Gaseous NO2 effects on stomatal behavior, photosynthesis and respiration
of hybrid poplar leaves. Acta Physiol. Plant, 37, 39, doi:10.1007/s11738-014-1749-8,
https://doi.org/10.1007/s11738-014-1749-8.
Iacono, M. J., J. S. Delamere, E. J. Mlawer, M. W. Shephard, S. A. Clough, and
W. D. Collins, 2008: Radiative forcing by long-lived greenhouse gases: Calculations with the aer radiative transfer models. J. Geophys. Res., 113, D13 103, doi:
10.1029/2008JD009944 |.
Impact Assessment Agency of Canada, 2019: Kitimat LNG expansion poject description summary. https://iaac-aeic.gc.ca/050/documents/p80180/130805E.pdf.
Janjić, Z. I., 1994: The step-mountain Eta coordinate model: Further developments
of the convection, viscous sublayer, and turbulence closure schemes. Mon. Wea. Rev.,
122, 927–945, doi:10.1175/1520-0493.
Janjić, Z. I., 2002: Nonsingular implementation of the Mellor-Yamada level 2.5
scheme in the NCEP meso model. 437, NCEP Office Note, 61 pp,

https://

repository.library.noaa.gov/view/noaa/11409.
Jiménez, P. A., J. Dudhia, J. F. Gonzalez-Rouco, J. Navarro, J. P. Montavez, and
E. García-Bustamante, 2012: A revised scheme for the WRF surface layer formulation. Mon. Wea. Rev., 140, 898–918, doi:10.1175/MWR-D-11-00056.1.
Josipovic, M., H. J. Annegarn, M. A. Kneen, J. J. Pienaar, and S. J. Piketh, 2011:
Atmospheric dry and wet deposition of sulphur and nitrogen species and assess224

ment of critical loads of acidic deposition exceedance in South Africa. S. Afr. J. Sci.,
107 (3/4), ID 478, doi:10.4102/sajs.v107i3/4.478.
Kelly, J. T., and Coauthors, 2018: Modeling NH4 NO3 over the San Joaquin Valley
during the 2013 DISCOVER-AQ campaign. J. Geophys. Res. Atmos., 123 (9), 4727–
4745, doi:10.1029/2018jd028290.
Klock, R., and J. Mullock, 2000:

The Weather of British Columbia:

Graphic

Area Forecast 31. 61–73 pp., http://www.navcanada.ca/EN/media/Publications/
LocalAreaWeatherManuals/LAWM-BC-EN.pdf.
Kota, S. H., and Coauthors, 2018: Year-long simulation of gaseous and particulate air pollutants in India. Atmos. Environ., 180, 244–255, doi:10.1016/
j.atmosenv.2018.03.003.
Krzyzanowski, J., 2010: Approaching cumulative effects through air pollution modelling. Water Air Soil Pollut., 214 (1-4), 253–273, doi:10.1007/s11270-010-0421-1.
Kurt, O. K., J. Zhang, and K. E. Pinkerton, 2016: Pulmonary health effects of air
pollution. Curr. Opin. Pulm. Med., 22, 138–43, doi:10.1097/MCP.0000000000000248.
Lange, O. S., 2003: Regional weather guides: northern mainland coast. In: Living
with weather along the British Columbia Coast: the veil of chaos. Environment Canada,
174–187 pp.
Lawrence, G. B., P. W. Hazlett, I. J. Fernandez, R. Ouimet, S. W. Bailey, W. C. Shortle,
K. T. Smith, and M. R. Antidormi, 2015: Declining acidic deposition begins reversal

225

of forest-soil acidification in the northeastern U.S. and eastern Canada. Environ. Sci.
Technol., 49 (22), 13 103–13 111, doi:10.1021/acs.est.5b02904.
Lee, J., H. H. Shin, S.-Y. Hong, P. A. Jiménez, J. Dudhia, and J. Hong, 2015: Impacts
of subgrid-scale orography parameterization on simulated surface layer wind and
monsoonal precipitation in the high-resolution WRF model. J. Geophys. Res. Atmos.,
120 (2), 644–653, doi:10.1002/2014jd022747.
Legge, A. H., and S. V. Krupa, 2002: Effects of sulphur dioxide. In: Air Pollution and
Plant Life (second edition). Wiley, Chichester, 135–162 pp.
Levasseur, P. A., S. A. Watmough, J. Aherne, C. J. Whitfield, and M. C. Eimers,
2020: Estimating mineral surface area and base cation weathering rates of spodosols
under forest in British Columbia, Canada. Geoderma Reg., 20, e00 247, doi:10.1016/
j.geodrs.2019.e00247.
Li, L., W. Li, and J. Jin, 2014: Improvements in WRF simulation skills of southeastern United States summer rainfall: Physical parameterization and horizontal
resolution. Climate Dyn., 43 (7-8), 2077–2091, doi:10.1007/s00382-013-2031-2.
Li, P., 2005: Box-cox transformations: An overview. https://www.ime.usp.br/
~abe/lista/pdfm9cJKUmFZp.pdf.
Li, T., H. Wang, T. Zhao, M. Xue, Y. Wang, H. Che, and C. Jiang, 2016: The impacts
of different PBL schemes on the simulation of PM2.5 during severe haze episodes in
the Jing-Jin-Ji region and its surroundings in China. Adv. in Meteorol., ID 6295 878,
15 pp, doi:10.1155/2016/6295878.
226

Li, Y., M. C. Barth, and A. L. Steiner, 2019: Comparing turbulent mixing of atmospheric oxidants across model scales. Atmos. Environ., 199, 88–101, doi:10.1016/
j.atmosenv.2018.11.004.
Lilhare, R., S. J. Déry, S. Pokorny, T. A. Stadnyk, and K. A. Koenig, 2019:
Intercomparison of multiple hydroclimatic datasets across the Lower Nelson
River Basin, Manitoba, Canada. Atmos. Ocean, 57 (4), 262–278, doi:10.1080/
07055900.2019.1638226.
Lin, Y., and K. E. Mitchell, 2005: The NCEP stage II/IV hourly precipitation
analyses: Development and applications. http://citeseerx.ist.psu.edu/viewdoc/
download?doi=10.1.1.182.2080&rep =rep1&type=pdf.
Liu, J., M. Bray, and D. Han, 2012: Sensitivity of the weather research and forecasting (WRF) model to downscaling ratios and storm types in rainfall simulation.
Hydrol. Process., 26 (20), 3012–3031, doi:10.1002/hyp.8247.
Longley, H., 2019: Conflicting interests: Development politics and the environmental regulation of the Alberta oil sands industry, 1970–1980. Environ. Hist. Camb., doi:
10.3197/096734019x15463432086919.
Lyu, B., Y. Zhang, and Y. Hu, 2017: Improving PM2.5 air quality model forecasts
in china using a bias-correction framework. Atmosphere, 8 (12), 147, doi:10.3390/
atmos8080147.
Makar, P. A., and Coauthors, 2018: Estimates of exceedances of critical loads for

227

acidifying deposition in Alberta and Saskatchewan. Atmos. Chem. Phys., 18 (13),
9897–9927, doi:10.5194/acp-18-9897-2018.
McKendry, I. G., 2006:

Background concentrations of PM2.5 and ozone in

British Columbia, Canada. http://citeseerx.ist.psu.edu/viewdoc/download?doi=
10.1.1.510.4465&rep =rep1&type=pdf.
McNulty, S. G., E. C. Cohen, J. A. M. Myers, T. J. Sullivan, and H. Li, 2007: Estimates of critical acid loads and exceedances for forest soils across the conterminous
United States. Environ. Pollut., 149 (3), 281–292, doi:10.1016/j.envpol.2007.05.025.
Meidinger, D., 2018: Terrestrial Vegetation of British Columbia. Lab for Advanced Spatial Analysis, Department of Geography, University of British Columbia, Vancouver, https://ibis.geog.ubc.ca/biodiversity/eflora/TerrestrialVegetation.html.
Mesinger, F., and Coauthors, 2006: North American Regional Reanalysis. Bull. Amer.
Meteor. Soc., 87 (3), 343–360, doi:10.1175/bams-87-3-343.
Miao, J.-F., and Coauthors, 2007: Evaluation of MM5 mesoscale model at local scale
for air quality applications over the Swedish west coast: Influence of PBL and LSM
parameterizations. Meteorol. Atmos. Phys., 99 (1-2), 77–103, doi:10.1007/s00703-0070267-2.
Misenis, C., and Y. Zhang, 2010: An examination of sensitivity of WRF/Chem predictions to physical parameterizations, horizontal grid spacing, and nesting options. Atmos. Res., 97, 315–334, doi:10.1016/j.atmosres.2010.04.005.

228

Mohan, M., and M. Gupta, 2018:

Sensitivity of PBL parameterizations on

PM1 0 and ozone simulation using chemical transport model WRF-Chem over
a sub-tropical urban airshed in India. Atmos. Environ., 185, 53–63, doi:10.1016/
j.atmosenv.2018.04.054.
Moran, M. D., and Coauthors, 2008: Predicted acid deposition critical-load exceedances across canada from a one-year simulation with a regional particulatematter model. 20 pp, http://ams.confex.com/ams/pdfpapers/132916.pdf.
Moya-Álvarez, A. S., R. Estevan, S. Kumar, J. L. F. Rojas, J. J. Ticse, D. MartínezCastro, and Y. Silva, 2020:

Influence of PBL parameterization schemes in

WRF_ARW model on short-range precipitation's forecasts in the complex orography of Peruvian Central Andes. Atmos. Res., 233, 104 708, doi:10.1016/
j.atmosres.2019.104708.
Mues, A., A. Laurer, A. Lupascu, M. Rupakheti, F. Kuik, and M. G. Lawrence,
2018: WRF and WRF-Chem v3.5.1 simulations of meteorology and black carbon
concentrations in the Kathmandu Valley. Geosci. Model Dev, 11, 2067–2091, doi:
10.5194/gmd-11-2067-2018.
Nakanishi, M., and H. Niino, 2009: Development of an improved turbulence closure model for the atmospheric boundary layer. J. Meteor. Soc. Japan, 87, 895–912,
doi:10.2151/jmsj.87.895.
Nasr, M., M. Castonguay, J. Ogilvie, B. A. Raymond, and P. A. Arp, 2010: Modelling
and mapping critical loads and exceedances for the georgia basin, british columbia
229

using a zero base-cation depletion criterion. J. Limnol., 69, 181–192, doi:10.4081/
jlimnol.2010.s1.181.
Neal, L., P. Agnew, S. Moseley, C. Ordóñez, N. Savage, and M. Tilbee, 2014: Application of a statistical post-processing technique to a gridded, operational, air quality
forecast. Atmos. Environ., 98, 385–393, doi:10.1016/j.atmosenv.2014.09.004.
Nielsen-Gammon, J. W., X. M. Hu, F. Q. Zhang, and J. E. Pleim, 2010: Evaluation of
planetary boundary layer scheme sensitivities for the purpose of parameter estimation. Mon. Weather Rev., 138, 3400–3417, doi:10.1175/2010MWR3292.1.
Olson, J. B., J. S. Kenyon, W. M. Angevine, J. M. Brown, M. Pagowski, and K. Sušelj,
2019: A description of the Mynn-Edmf scheme and the coupling to other components in WRF-ARW. NOAA Technical Memorandum OAR GSD, 61, 37 pp, doi:
10.25923/n9wm-be49.
Otte, T. L., and J. E. Pleim, 2010: The meteorology-chemistry interface processor (MCIP) for the CMAQ modeling system: updates through MCIPv3.4.1. Geosci.
Model Dev., 3 (1), 243–256, doi:10.5194/gmd-3-243-2010.
Pardo, L. H., 2010: Approaches for estimating critical loads of N and S deposition
for forest ecosystems on U.S. federal lands. U.S. Department of Agriculture, Forest
Service, https://doi.org/10.2737/NRS-GTR-71, 25 pp.
Parra, R., 2018: Performance studies of planetary boundary layer schemes in WRFChem for the Andean region of southern Ecuador. Atmos. Pollut. Res., 9 (3), 411–428,
doi:10.1016/j.apr.2017.11.011.
230

Peña-Arancibia, J. L., A. I. J. M. van Dijk, L. J. Renzullo, and M. Mulligan, 2013:
Evaluation of precipitation estimation accuracy in reanalyses, satellite products,
and an ensemble method for regions in Australia and South and East Asia. J. Hydrometeorol., 14 (4), 1323–1333, doi:10.1175/jhm-d-12-0132.1.
Pérez, C., P. Jiménez, O. Jorba, M. Sicard, and J. M. Baldasano, 2006: Influence of the
PBL scheme on high-resolution photochemical simulations in an urban coastal area
over the western Mediterranean. Atmos. Environ., 40 (27), 5274–5297, doi:10.1016/
j.atmosenv.2006.04.039.
Pernigotti, D., M. Gerboles, C. Belis, and P. Thunis, 2013: Model quality objectives
based on measurement uncertainty. part II: NO2 and PM10. Atmos. Environ., 79,
869–878, doi:10.1016/j.atmosenv.2013.07.045.
Pickard, D., P. de la Cueva Bueno, E. Olson, and C. Semmens, 2019: Evaluation of cumulative effects assessment methodologies for marine shipping.
https://www.tc.gc.ca/en/services/marine/documents/EVALUATION_OF_
CUMULATIVE_EFFECTS_ASSESSMENT_METHODOLOGIES_FOR_MARINE_
SHIPPING_EN-2019-08-26.pdf.
Pleim, J. E., 2007: A combined local and nonlocal closure model for the atmospheric
boundary-layer. part i: model description and testing. J. Appl. Meteorol. Climatol., 46,
1383–1395, doi:10.1175/JAM2539.1.
Pojar, J., K. Klinka, and D. A. Demarchi, 1991: Ecosystems of British Columbia. B.C.

231

Ministry of Forests, Victoria, 95–111 pp, https://www.for.gov.bc.ca/hfd/pubs/
docs/Srs/Srs06/chap6.pdf.
Pontoppidan, M., J. Reuder, S. Mayer, and E. W. Kolstad, 2017: Downscaling an intense precipitation event in complex terrain: the importance of high grid resolution.
Tellus A, 69 (1), 1271 561, doi:10.1080/16000870.2016.1271561.
Porter, P. S., S. T. Rao, C. Hogrefe, E. Gego, and R. Mathur, 2015: Methods for
reducing biases and errors in regional photochemical model outputs for use in
emission reduction and exposure assessments. Atmos. Environ., 112, 178–188, doi:
10.1016/j.atmosenv.2015.04.039.
Qiao, X., and Coauthors, 2015: Modeling dry and wet deposition of sulfate, nitrate,
and ammonium ions in Jiuzhaigou National Nature Reserve, China using a sourceoriented CMAQ model: Part I. base case model results. Sci. Total Environ., 532, 831–
839, doi:10.1016/j.scitotenv.2015.05.108.
Queen, A., and Y. Zhang, 2008: Examining the sensitivity of MM5–CMAQ predictions to explicit microphysics schemes and horizontal grid resolutions, part II—PM
concentrations and wet deposition predictions. Atmos. Environ., 42 (16), 3856–3868,
doi:10.1016/j.atmosenv.2007.12.066.
Rakesh, P. T., R. Venkatesan, and C. V. Srinivas, 2013: Formulation of TKE based
empirical diffusivity relations from turbulence measurements and incorporation in
a lagrangian particle dispersion model. Environ. Fluid Mech., 13 (4), 353–369, doi:
10.1007/s10652-013-9273-8.
232

Reboredo, B., R. Arasa, and B. Codina, 2015: Evaluating sensitivity to different options and parameterizations of a coupled air quality modelling system over Bogotá, Colombia. Part I: WRF model configuration. Open J. Air Pollut., 4, 47–64, doi:
10.4236/ojap.2015.42006.
Remesan, R., T. Bellerby, I. Holman, and L. Frostick, 2014: WRF model sensitivity to
choice of parameterization: a study of the ‘York Flood 1999’. Theor. Appl. Climatol.,
122 (1-2), 229–247, doi:10.1007/s00704-014-1282-0.
Rife, D. L., C. A. Davis, and Y. Liu, 2004: Predictability of low-level winds by
mesoscale meteorological models. Mon. Weather Rev., 132, 2553–2569, doi:10.1175/
MWR2801.1.
Rio Tinto, 2013:

So2 Technical Assessment volume 2. 450 pp,

https:

//www.riotinto.com/can/search/documents#main-search_e=0&main-search_
sxatags =bcworks%2Cenglish&canada-search_e=20.
Rio Tinto, 2016: Rio tinto’s modernised kitimat aluminium smelter reaches full production. https://www.riotinto.com/news/releases/Kitimat-smelter-reaches-fullproduction.
Rio Tinto,

2018:

Annual environmental report. 43–45 pp,

https://

www.riotinto.com/search/documents#main-search_e =0&main-search_sxatags=
bcworks.
Sakiyama, S., 2015: Guidelines for air quality dispersion modeling in British
Columbia. https://www.bcogc.ca/node/13339/download.
233

Sarmadi, F., Y. Huang, G. Thompson, S. T. Siems, and M. J. Manton, 2019: Simulations of orographic precipitation in the Snowy Mountains of southeastern Australia.
Atmos. Res., 219, 183–199, doi:10.1016/j.atmosres.2019.01.002.
Schirmer, M., and B. Jamieson, 2015: Verification of analysed and forecasted winter
precipitation in complex terrain. Cryosphere, 9 (2), 587–601, doi:10.5194/tc-9-5872015.
Seinfeld, J. H., and S. N. Pandis, 2016: Atmospheric chemistry and physics : from air
pollution to climate change. 3rd ed., John Wiley & Sons, Hoboken, New Jersey.
Shafran, P., J. Woollen, W. Ebisuzaki, W. Shi, Y. Fan, R. Grumbine, and M. Fennessy,
2004: Observational data used for assimilation in the NCEP North American Regional Reanalysis. https://ams.confex.com/ams/84Annual/techprogram/paper_
71689.htm.
Shin, H. H., and S.-Y. Hong, 2015: Representation of the subgrid-scale turbulent
transport in convective boundary layers at gray-zone resolutions. Mon. Weather
Rev., 143, 250–271, doi:10.1175/MWR-D-14-00116.1.
Skamarock, W. C., and Coauthors, 2019: A description of the advanced research
WRF version 4. NCAR Tech. Note NCAR/TN-556+STR. 145, doi:10.5065/1dfh6p97.
Smajla, I., D. K. Sedlar, B. Drljača, and L. Jukić, 2019: Fuel switch to LNG in heavy
truck traffic. Energies, 12 (3), 515, doi:10.3390/en12030515.

234

Soil Classification Working Group, 1998: The Canadian System of Soil Classification
(third edition). Agriculture and Agri–Food Canada. Publication 1646, 187 pp, http:
//sis.agr.gc.ca/cansis/publications/manuals/1998-cssc-ed3/cssc3_manual.pdf.
Solazzo, E., and S. Galmarini, 2016:

Error apportionment for atmospheric

chemistry-transport models — a new approach to model evaluation. Atmos. Chem.
Phys., 16 (10), 6263–6283, doi:10.5194/acp-16-6263-2016.
Srinivas, C., V. Yesubabu, D. H. Prasad, K. H. Prasad, M. Greeshma, R. Baskaran,
and B. Venkatraman, 2018: Simulation of an extreme heavy rainfall event over
Chennai, India using WRF: Sensitivity to grid resolution and boundary layer
physics. Atmos. Res., 210, 66–82, doi:10.1016/j.atmosres.2018.04.014.
Srinivas, C. V., K. B. R. R. H. Prasad, C. V. Naidu, R. Baskaran, and B. Venkatraman,
2015: Sensitivity analysis of atmospheric dispersion simulations by FLEXPART to
the WRF-simulated meteorological predictions in a coastal environment. Pure Appl.
Geophys., 173 (2), 675–700, doi:10.1007/s00024-015-1104-z.
Stantec, 2015: 2014 vegetation inspection, monitoring and assessment program. 17
pp.
Statistics Canada, 2018: Census profile, 2016 census. http://www12.statcan.gc.ca/
census-recensement/index-eng.cfm?GEOCODE=59#keystats.
Sullivan, T. J., 2000: Aquatic Effects of Acidic Deposition. CRC Press, doi:10.1201/
9780429187186.

235

Sun, J., H. Hu, Y. Li, L. Wang, Q. Zhou, and X. Huang, 2016: Effects and mechanism
of acid rain on plant chloroplast ATP synthase. Environ. Sci. Pollut. Res., 23 (18),
18 296–18 306, doi:10.1007/s11356-016-7016-3.
Sun, Q., C. Miao, Q. Duan, H. Ashouri, S. Sorooshian, and K.-L. Hsu, 2018: A review
of global precipitation data sets: Data sources, estimation, and intercomparisons.
Rev. Geophys., 56 (1), 79–107, doi:10.1002/2017rg000574.
Sverdrup, H., and P. Warfvinge, 1995: Critical loads of acidity for Swedish forest
ecosystems. Ecol. Bull., 44, 75–89, https://www.jstor.org/stable/20113151?seq=
2#metadata_info_tab_contents.
Swall, J. L., and K. M. Foley, 2009: The impact of spatial correlation and incommensurability on model evaluation. Atmos. Environ., 43 (6), 1204–1217, doi:
10.1016/j.atmosenv.2008.10.057.
Syrakov, D., M. Prodanova, E. Georgieva, I. Etropolska, and K. Slavov, 2016: Simulation of European air quality by WRF–CMAQ models using AQMEII-2 infrastructure. J. Comput. Appl. Math., 293, 232–245, doi:10.1016/j.cam.2015.01.032.
Talbot, C., E. Bou-Zeid, and J. Smith, 2012: Nested mesoscale large-eddy simulations with WRF: performance in real test cases. J. Hydrometeorol., 13, 1421–1441,
doi:10.1175/JHM-D-11-048.1.
Tessum, C. W., J. D. Hill, and J. D. Marshall, 2015: Twelve-month, 12 km resolution
North American WRF-Chem v3.4 air quality simulation: performance evaluation.
Geosci. Model Dev., 8 (4), 957–973, doi:10.5194/gmd-8-957-2015.
236

Tewari, M. F., and Coauthors, 2004: Implementation and verification of the unified
NOAH land surface model in the WRF model. Am. Meteorol. Soc., 11–15, https:
//ams.confex.com/ams/84Annual/techprogram/paper_69061.htm.
The Interior News, 2020: UPDATE: 11 arrested in day 3 of injunction enforcement
against Wet’suwet’enl.

https://www.interior-news.com/news/rcmp-continue-

to-enforce-injunction-against-wetsuweten-opposing-coastal-gaslink/.
Thompson, G., P. R. Field, R. M. Rasmussen, and W. D. Hall, 2008: Explicit forecasts
of winter precipitation using an improved bulk microphysics scheme. part ii: implementation of a new snow parameterization. Mon. Weather Rev., 136, 5095–5115,
doi:10.1175/2008MWR2387.1.
Thunis, P., A. Pederzoli, and D. Pernigotti, 2012: Performance criteria to evaluate air quality modeling applications. Atmos. Environ., 59, 476–482, doi:10.1016/
j.atmosenv.2012.05.043.
Tian, J., J. Liu, D. Yan, C. Li, and F. Yu, 2017: Numerical rainfall simulation with
different spatial and temporal evenness by using a WRF multiphysics ensemble.
Nat. Hazards Earth Syst. Sci., 17 (4), 563–579, doi:10.5194/nhess-17-563-2017.
United Nations Economic Commission for Europe, 2017: Manual on methodologies and criteria for modelling and mapping critical loads and levels and air pollution effects, risks and trends. Chapter 3, https://www.umweltbundesamt.de/
sites/default/files/medien/4292/dokumente/ch3-mapman-2017-10.pdf.

237

United States Department of Agriculture, 2013: Conservation assessments for 5
species of lichens. 54–64, https://www.fs.fed.us/r6/sfpnw/issssp/documents2/
ca-li-5-lichens-2013-04.pdf.
USEPA, 1996: Air quality criteria for particulate matter. https://nepis.epa.gov/
Exe/ZyPDF.cgi/20008MAJ.PDF?Dockey=20008MAJ.PDF, ePA/600/P-95/001aFcF.
USEPA, 2009: Current methodologies in preparing mobile source port-related
emission inventories.

https://www.epa.gov/sites/production/files/2016-06/

documents/2009-port-inventory-guidance.pdf.
USEPA, 2015: Interagency workgroup on air quality modeling phase 3 summary report: Long range transport and air quality related values. https://www3.epa.gov/
ttn/scram/11thmodconf/IWAQM3_LRT_Report-07152015.pdf.
USEPA, 2016: List of designated reference and equivalent methods. https://
www3.epa.gov/ttnamti1/files/ambient/criteria/AMTICListDec2016-2.pdf.
USEPA, 2017: Revisions to the guideline on air quality models: Enhancements to
the AERMOD dispersion modeling system and incorporation of approaches to address ozone and fine particulate matter. https://www.govinfo.gov/content/pkg/
FR-2017-01-17/pdf/2016-31747.pdf.
Vacek, S., and K. Matějka, 2010:

Health status of forest stands on perma-

nent research plots in the Krkonoše. Mts. J. Forest Sci., 56, 555–569,
www.cabdirect.org/cabdirect/FullTextPDF/2011/20113038271.pdf.
238

https://

Veira, A., P. L. Jackson, B. Ainslie, and D. Fudge, 2013: Assessment of background particulate matter concentrations in small cities and rural locations—prince
george, canada. J. Air Waste Manage. Assoc., 63 (7), 773–787, doi:10.1080/
10962247.2013.789091.
Vingarzan, R., 2007: Ambient particulate matter concentrations in Canada and
background levels. Environment Canada, 20 pp,

http://publications.gc.ca/

collections/collection_2007/ec/En56-215-2007E.pdf.
Vivanco, M. G., and Coauthors, 2018: Modeled deposition of nitrogen and sulfur
in Europe estimated by 14 air quality model systems: evaluation, effects of changes
in emissions and implications for habitat protection. Atmos. Chem. Phys., 18 (14),
10 199–10 218, doi:10.5194/acp-18-10199-2018.
Vorobeichik, E. L., M. R. Trubina, E. V. Khantemirova, and I. E. Bergman, 2014:
Long-term dynamic of forest vegetation after reduction of copper smelter emissions. Russ. J. Ecol., 45, 498–162, doi:10.1134/S1067413614060150.
Whitfield, C. J., J. N. Phelan, J. Buckley, C. M. Clark, S. Guthrie, and J. A. Lynch,
2018: Estimating base cation weathering rates in the USA: Challenges of uncertain
soil mineralogy and specific surface area with applications of the PROFILE model.
Water Air Soil Pollut., 229 (3), 15 pp, doi:10.1007/s11270-018-3691-7.
WHO, 2000: Air Quality Guidelines for Europe (second edition). WHO Regional Publications, European Series, 91, 219–246 pp,

http://www.euro.who.int/__data/

assets/pdf_file/0006/123099/AQG2ndEd_12ozonelevel.pdf?ua=1.
239

WHO, 2016: Ambient air pollution: a global assessment of exposure and burden of disease.
ttps://www.who.int/phe/publications/air-pollution-global-assessment/en/.
Williams, J. J., S. H. Chung, A. M. Johansen, B. K. Lamb, J. K. Vaughan, and M. Beutel, 2017: Evaluation of atmospheric nitrogen deposition model performance in
the context of U.S. critical load assessments. Atmos. Environ., 150, 244–255, doi:
10.1016/j.atmosenv.2016.11.051.
Williston, P. J., and Coauthors, 2016: Critical levels and loads and the regulation of
industrial emissions in northwest British Columbia, Canada. Atmos. Environ., 146,
311–323, doi:10.1016/j.atmosenv.2016.08.058.
Willmott, C., S. Robeson, and K. Matsuura, 2012: A refined index of model performance. Int. J. Climatol., 32, 2088–2094, doi:10.1002/joc.2419.
Wong, J. S., S. Razavi, B. R. Bonsal, H. S. Wheater, and Z. E. Asong, 2017: Intercomparison of daily precipitation products for large-scale hydro-climatic applications over Canada. Hydrol. Earth Syst. Sci., 21 (4), 2163–2185, doi:10.5194/hess-212163-2017.
Wyngaard, J., 2004: Toward numerical modeling in the “terra incognita. J. Atmos.
Sci, 61, 1816–1826, doi:10.1175/1520-0469.
Xie, B., J. Fung, A. Chan, and A. Lau, 2012: Evaluation of nonlocal and local planetary boundary layer schemes in the WRF model. J. Geophys. Res, 117, 12 103, doi:
10.1029/2011JD017080.

240

Yang, B., Y. Qian, G. Lin, R. Leung, and Y. Zhang, 2012: Some issues in uncertainty
quantification and parameter tuning: a case study of convective parameterization
scheme in the WRF regional climate model. Atmos. Chem. Phys., 12 (5), 2409–2427,
doi:10.5194/acp-12-2409-2012.
Yarwood, G., S. Rao, M. Yocke, and G. Whitten, 2005: Updates to the carbon bond
chemical mechanism: CB05 final report to the USEPA. http://www.camx.com/
publ/pdfs/CB05_Final_Report_120805.pdf.
Zhang, H., Z. Pu, and X. Zhang, 2013: Examination of errors in near-surface temperature and wind from WRF numerical simulations in regions of complex terrain.
Wea. Forecasting, 28, 893–914, doi:10.1175/WAF-D-12-00109.1.
Zhang, L., R. Vet, J. M. O'Brien, C. Mihele, Z. Liang, and A. Wiebe, 2009: Dry deposition of individual nitrogen species at eight Canadian rural sites. J. Geophys. Res.,
114, D02 301, doi:10.1029/2008jd010640.

241

Appendices

Appendix A
Selection of meteorological year for
model simulations
To find a meteorological year that is representative of the study area, a ranking of differences between annual records for wind speed and precipitation in the TKV is presented.
Wind speed is a major determinant of pollutant transport and dilution, while precipitation, which exceeds 2000 mm per annum in the Kitimat area, is a principal factor for the
deposition of atmospheric contaminants. The ranking and selection of a meteorological
year, thus, was based on these variables.
Due to data availability for model evaluations, time span is years 2006–2018. Records
are restricted to stations with longer and more complete archives. Mean wind speeds
are hourly measurements while total precipitation are monthly sums of daily precipitation (rain + snow). See https://envistaweb.env.gov.bc.ca/DynamicTable2.aspx?G_ID=
327 and http://climate.weather.gc.ca/historical_data/search_historic_data_e.html for historical data.
The computations are in two steps. First, pairwise absolute differences in monthly (Jan-

243

uary - December) values between all year pairs are calculated. Next, the mean of the
differences for each year is calculated and ranked. Lower values are ranked higher (see
Table A.1). Year 2017 is well ranked across sites for wind speed and is in top half of rank-

Precipitation (mm)

Wind speed (m s−1 )

Table A.1 The averages of differences between monthly means of wind speed
and precipitation for each year (2006 to 2018) and their rankings. The less the
difference, the better the ranking as a representative year. The empty values are
instances of considerable missing data
2006

2007

2008

2009

2010

2011

2012

2013

2014

2015

2016

2017

2018

3.90

4.39

4.01

3.89

4.47

3.92

4.81

4.35

4.37

4.56

5.34

3.58

3.78

4th

9th

6th

3rd

10th

5th

12th

7th

8th

11th

13th

1st

2nd

4.58

5.14

4.46

4.28

4.24

4.41

4.64

4.78

4.89

4.96

6.79

3.92

3.98

7th

12th

6th

4th

3rd

5th

8th

9th

10th

11th

13th

1st

2nd

Terrace

6.79

7.54

6.79

8.43

7.34

8.06

7.04

8.45

8.43

7.96

8.61

6.98

7.48

Airport

1st

7th

1st

10th

5th

9th

4th

12th

10th

8th

13th

3rd

6th

Kitimat

1343

1132

1035

1032

1019

1315

1281

-

1101

997

1213

1076

1079

(Town site)

12th

8th

4th

3rd

2nd

11th

10th

-

7th

1st

9th

5th

6th

-

568

559

499

518

625

549

602

580

548

700

524

557

-

8th

7th

1st

2nd

11th

5th

10th

9th

4th

12th

3rd

6th

Haul road

Whitesail

Terrace

ings for precipitation. Consequently, this year is selected as the meteorological year.
Figure A.1 shows the profiles of this year’s values, and those of the averages for the 2006–
2018 period. Overall, the Kitimat 2017 data has -0.05 m s−1 , + 25.4 mm and - 0.8 ◦ C for
mean wind speed, total precipitation and monthly air temperature, respectively, relative
to the 2006–2018 values.

244

Figure A.1 Observed monthly average wind speed and precipitation in 2017 versus the averages for the 2006–2018 period. Terrace wind data is from the Airport
that is 8 km south of the city
245

Appendix B
WRF physics settings and model heights
Table B.1 The under-listed physics and dynamics settings with option numbers
(in WRF version 4.0) were used for generating meteorological fields.
Physics/dynamics

Option name

Option number

Land surface scheme

Noah

2

Microphysics

Thompson

8

Long- and short-wave radiation

RRTMG

4

Cumulus

Grell-Freitas

3

Diffusion option

Full

2

Eddy coefficient option

Horizontal Smagorinsky

4

246

Table B.2 Layer top heights (H1, H2, . . . , H39) for WRF simulations
H1

24.9 m

H11

1799.3 m

H21

7699.1 m

H31

14,628.7 m

H2

81.7 m

H12

2197.8 m

H22

8414.1 m

H32

15,320.0 m

H3

153.8 m

H13

2641.7 m

H23

9117.5 m

H33

16,010.7 m

H4

245.0 m

H14

3126.9 m

H24

9810.3 m

H34

16,700.8 m

H5

359.6 m

H15

3653.3 m

H25

10,493.9 m

H35

17,390.0 m

H6

502.2 m

H16

4224.1 m

H26

11,174.4 m

H36

18,079.0 m

H7

677.7 m

H17

4842.6 m

H27

11,859.9 m

H37

18,766.8 m

H8

891.1 m

H18

5512.2 m

H28

12,551.5 m

H38

19,455.4 m

H9

1146.9 m

H19

6233.1 m

H29

13,244.6 m

H39

20,145.0 m

H10

1449.0 m

H20

6233.1 m

H30

8414.1 m

247

Appendix C
Formulae for statistical measures
Below are the formulas for the various statistical measures that are used in this dissertation: Mean bias (MB) =

1 n
( Mi − Oi );
n i∑
=1

Mean fractional bias (MFB) =

Normalized mean bias (NMB) =

MB
O

M−O
0.5( M + O)

Normalized mean square error (NMSE)=
Root mean square error (RMSE)=

( Mi − Oi )2
MO

( Mi − O i ) 2

Pearson’s correlation coefficient (r) =

Modified index of agreement (IOA) =

∑( Mi − M)(Oi − O)
∑( Mi − M)2 ∑(Oi − O)2

n



∑ | Mi − Oi |



i

1 − =1n
,





2 ∑ |Oi − O|


if

n

n

i =1

i =1

n

n

i =1

i =1

∑ | Mi − Oi | ≤ ∑ |Oi − O|

i =1

n



2
∑ |Oi − O|



i
=1


− 1,

n



 ∑ | Mi − Oi |

if

∑ | Mi − Oi | > ∑ |Oi − O|

i =1

Normalised standard deviation (NSD) =

Standard deviation of modeled values
σm
=
Standard deviation of observation data σo

Where Mi and Oi are modeled and observed values (paired) respectively, n is the number
of pairs and terms with overbars are mean values.

248

Appendix D
Verification of atmospheric sounding

Figure D.1 SkewT plot of simulations and observation for air temperature (solid
lines), dew point (dashed lines), and wind for times in winter (top) and summer (bottom). Observation data are from the Annette Island radiosonde station,
Alaska (latitude 55.03 ◦ N, longitude 131.57 ◦ W)

249

Appendix E
Summer and winter days comparisons of
air temperature and wind speed to
seasonal averages
Figure E.1a shows average hourly temperatures for July 22, 2017 and January 21, 2017,
minus average hourly temperatures in summmer (left) and winter (right), respectively.
The Light-gray portions between dark-gray backgrounds are daytime periods. The average daily temperature at Whitesail (Terrace) in summer is 15.4 ◦ C (16.1 ◦ C ). The average
daily temperature at Whitesail (Terrace) in winter is -1.8◦ C (-2.0 ◦ C ).
Figure E.1b shows seasonal windroses (S) versus those of the specified days (D) for daytime and nighttime periods. Marked on each plot is wind speed in m s−1 . Notice their
overall correspondence for wind direction.

250

Figure E.1 Winterday (January 21, 2017) and summerday (July 22, 2017 ) comparisons to observed seasonal averages of air temperature and wind.

251

Appendix F
Performance statistics for simulation of
surface meteorological variables

252

Table F.1 Statistical performance of simulations of air temperature at a diurnal
time scale for each season. Values in bold indicate that criteria (refer to Table 2.3)
are met, and italicized means a value is top-ranked for a station. An instance
where more than two schemes are joint-best is not ranked. Fitness of individual schemes are assessed from counts of top-ranked and satisfactory evaluations
(bolded and italicized). Nanakwa, Whitesail, Onion Lake and Terrace stations are
N, W, O and T, respectively.
spring
Measure

summer

autumn

winter

PBL scheme
N

W

O

T

N

W

O

T

N

W

O

T

N

W

O

T

MYJ

-0.47

-1.89

-1.54

-2.37

-0.96

-0.35

-0.32

-0.41

1.14

0.64

3.80

0.93

1.99

0.18

0.33

0.55

MYNN3

0.06

-1.65

-1.28

-2.19

-0.70

-0.39

-0.42

-0.53

1.16

0.99

4.13

1.18

2.17

0.50

0.75

0.89

UW

0.04

-1.71

-1.36

-2.23

-0.66

-0.18

-0.18

-0.27

1.18

0.93

4.10

1.17

2.12

0.32

0.48

0.69

ACM2

-0.03

-1.92

-1.72

-2.38

-0.44

-0.40

-0.46

-0.47

1.15

0.65

3.65

0.93

1.63

-0.05

0.08

0.24

YSU

-0.13

-2.06

-1.86

-2.49

-0.50

-0.51

-0.56

-0.58

1.06

0.45

3.52

0.80

1.47

-0.28

-0.07

0.09

SH

-0.12

-2.05

-1.88

-2.49

-0.50

-0.51

-0.57

-0.58

1.07

0.45

3.49

0.81

1.48

-0.27

-0.08

0.09

MYJ

1.35

2.25

1.92

2.63

1.72

0.94

1.22

1.19

1.44

1.09

3.82

1.20

2.08

0.77

0.58

0.77

MYNN3

1.25

2.17

1.84

2.53

1.54

1.09

1.40

1.32

1.44

1.33

4.15

1.42

2.24

0.94

0.91

1.06

UW

1.27

2.12

1.81

2.53

1.57

0.94

1.24

1.16

1.48

1.27

4.13

1.39

2.21

0.81

0.68

0.88

ACM2

1.25

2.31

2.13

2.66

1.52

1.05

1.38

1.27

1.48

1.22

3.76

1.28

1.74

0.96

0.55

0.81

YSU

1.27

2.41

2.26

2.77

1.52

1.13

1.44

1.33

1.40

1.16

3.64

1.20

1.58

1.03

0.61

0.85

SH

1.28

2.42

2.27

2.76

1.54

1.12

1.44

1.34

1.41

1.16

3.62

1.20

1.59

1.05

0.65

0.84

MYJ

0.06

0.96

0.94

0.92

-0.10

0.97

0.94

0.93

-0.22

0.90

0.94

0.90

-0.08

0.89

0.96

0.87

MYNN3

0.06

0.97

0.95

0.93

-0.07

0.98

0.95

0.94

-0.19

0.91

0.94

0.90

-0.10

0.89

0.96

0.87

UW

0.04

0.96

0.94

0.92

-0.12

0.97

0.94

0.93

-0.22

0.90

0.93

0.89

-0.13

0.88

0.95

0.86

ACM2

0.12

0.92

0.88

0.89

-0.11

0.95

0.91

0.91

-0.24

0.84

0.87

0.85

-0.05

0.71

0.88

0.71

YSU

0.10

0.91

0.86

0.88

-0.09

0.94

0.90

0.91

-0.21

0.83

0.86

0.84

0.01

0.70

0.86

0.66

SH

0.09

0.91

0.86

0.88

-0.11

0.94

0.90

0.91

-0.22

0.83

0.86

0.85

0.00

0.69

0.85

0.67

MYJ

0.43

0.61

0.65

0.45

0.25

0.84

0.80

0.81

-0.05

0.70

-0.42

0.63

-0.56

0.70

0.75

0.62

MYNN3

0.47

0.65

0.67

0.49

0.32

0.81

0.77

0.78

-0.06

0.63

-0.47

0.56

-0.59

0.62

0.61

0.46

UW

0.47

0.65

0.67

0.48

0.30

0.83

0.80

0.80

-0.08

0.65

-0.47

0.57

-0.58

0.68

0.72

0.56

ACM2

0.47

0.59

0.59

0.44

0.30

0.82

0.78

0.78

-0.05

0.70

-0.40

0.64

-0.46

0.67

0.77

0.65

YSU

0.46

0.57

0.56

0.42

0.31

0.80

0.77

0.77

0.02

0.73

-0.38

0.67

-0.40

0.64

0.75

0.62

SH

0.45

0.57

0.56

0.42

0.30

0.80

0.76

0.77

0.01

0.73

-0.37

0.67

-0.40

0.64

0.74

0.63

MB (K)

RMSE (K)

r

IOA

253

Table F.2 Performance statistics for specific humidity at diurnal time scale for
each season are given below. Values in bold indicate that criteria (refer to Table 2.3) are met and italicized means a value is top-ranked for each station. An
instance where more than two schemes are joint-best is not ranked. Fitness of
individual schemes are assessed from counts of top-ranked and satisfactory evaluations (bold and italicized). Onion Lake and Terrace stations are O and T, respectively.
spring

summer

autumn

winter

O

O

O

spring

summer

autumn

winter

O

O

O

T

O

T

PBL scheme
O

T

T

T

T

T

T

MB (g kg−1 )

r

MYJ

-0.47

-0.19

-0.27

-0.05

-0.87

-0.61

-0.37

-0.05

0.79

0.82

0.71

0.76

0.81

0.90

0.84

0.85

MYNN3

-0.43

-0.16

-0.19

0.03

-0.86

-0.6

-0.35

-0.03

0.78

0.81

0.67

0.70

0.81

0.90

0.84

0.85

UW

-0.54

-0.25

-0.43

-0.18

-0.94

-0.67

-0.37

-0.05

0.78

0.81

0.71

0.76

0.81

0.90

0.84

0.85

ACM2

-0.71

-0.39

-0.66

-0.41

-1.09

-0.82

-0.48

-0.14

0.77

0.81

0.75

0.79

0.77

0.89

0.82

0.84

YSU

-0.69

-0.41

-0.68

-0.48

-1.01

-0.82

-0.48

-0.14

0.79

0.82

0.74

0.78

0.78

0.89

0.84

0.84

SH

-0.65

-0.33

-0.53

-0.27

-1.06

-0.78

-0.5

-0.14

0.76

0.81

0.72

0.77

0.78

0.90

0.83

0.84

RMSE (g kg−1 )

IOA

MYJ

1.06

0.91

1.25

1.07

1.26

1.17

0.89

0.79

0.56

0.66

0.60

0.65

0.54

0.73

0.72

0.76

MYNN3

1.06

0.93

1.34

1.20

1.26

1.16

0.88

0.79

0.56

0.66

0.57

0.62

0.55

0.74

0.72

0.76

UW

1.08

0.91

1.27

1.06

1.32

1.19

0.90

0.79

0.54

0.66

0.58

0.65

0.52

0.73

0.71

0.76

ACM2

1.02

0.96

1.27

1.03

1.48

1.31

1.01

0.85

0.48

0.63

0.56

0.64

0.46

0.70

0.67

0.74

YSU

1.13

0.94

1.28

1.07

1.40

1.31

0.95

0.83

0.50

0.64

0.55

0.63

0.49

0.69

0.69

0.75

SH

1.19

0.96

1.29

1.05

1.46

1.27

1.01

0.85

0.50

0.64

0.57

0.65

0.47

0.71

0.68

0.74

254

Table F.3 Wind direction bias per season (in ◦ ) using all hourly values. Values >0 ◦
signify clockwise bias. Values in bold indicate that being within ± 10 ◦ is achieve
and italicized means a value is top-ranked for each station. An instance where
more than two schemes are joint-best is not ranked. Fitness of individual schemes
are assessed from counts of top-ranked and satisfactory evaluations (bold and
italicized). Nanakwa, Whitesail, Onion Lake and Terrace stations are N, W, O
and T, respectively.
spring

summer

autumn

winter

PBL scheme
N

W

O

T

N

W

O

T

N

W

O

T

N

W

O

T

MYJ

10.5

16.9

-31.8

10.9

3.7

20.5

-23.7

0.4

2.6

20.4

-40.1

22.9

1.9

17.0

-43.8

26.8

MYNN3

5.7

18.3

-38.6

9.2

4.8

18.8

-29.8

-4.9

3.4

20.7

-43.2

17.8

0.1

14.6

-48.8

25.3

UW

5.2

16.2

-33.2 11.2

3.5

19.1

-25.3

-2.3

0.6

21.8

-37.8

20.5

-1.6

17.2

-45.6

22.4

ACM2

3.2

21.7

-26.5 15.2

3.9

21.8

-19.4

0.9

1.4

24.5

-34.9

21.1

-10.0

20.9

-43.8

19.1

YSU

5.2

20.6

-33.9

11.6

6.3

21.6

-25.0

-0.8

1.2

21.5

-40.7

19.1

-9.1

17.0

-52.3

17.6

SH

5.9

19.6

-32.8

11.2

7.0

21.6

-24.9

-1.6

1.7

21.0

-40.4

19.2

-9.4

17

-53.5

18.4

255

Table F.4 Performance statistics for wind speed at diurnal time scale for each
season is presented in the table below. Values in bold indicate that criteria (refer
to Table 2.3) are met and italicized means a value is top-ranked for a station.
An instance where more than two schemes are joint-best is not ranked. Fitness
of individual schemes are assessed from counts of top-ranked and satisfactory
evaluations (bold and italicized). Nanakwa, Whitesail, Onion Lake and Terrace
stations are N, W, O and T, respectively.

MB (m s −1 )

RMSE (m s−1 )

PBL

spring

summer

scheme

N

W

O

T

N

MYJ

1.39

0.98

0.71

2.64

0.46

MYNN3

1.53

0.87

0.50

2.37

UW

1.64

0.99

0.68

ACM2

1.98

0.99

YSU

1.91

SH

autumn

W

winter

O

T

N

W

O

T

N

W

O

T

1.4

0.26

2.43

1.21

1.63

1.85

3.02

0.81

1.32

1.84

3.67

0.56

1.14

-0.14

1.88

1.21

1.57

1.82

2.80

0.83

1.29

1.82

3.50

2.60

0.71

1.50

0.34

2.43

1.24

1.67

1.92

3.01

0.79

1.25

1.77

3.55

0.50

2.45

0.83

0.88

-0.13

1.91

1.77

1.47

1.67

2.78

1.61

1.75

1.97

3.73

0.88

0.41

2.31

0.77

0.76

-0.21

1.79

1.72

1.31

1.58

2.63

1.65

1.63

1.98

3.53

1.95

0.89

0.45

2.31

0.88

0.78

-0.18

1.85

1.77

1.33

1.59

2.69

1.66

1.67

2.02

3.52

MYJ

1.48

1.00

0.94

2.65

1.06

1.43

0.72

2.44

1.27

1.65

1.86

3.03

1.01

1.35

1.90

3.68

MYNN3

1.58

0.91

0.97

2.41

0.91

1.17

0.97

1.91

1.28

1.59

1.82

2.81

1.01

1.31

1.87

3.51

UW

1.74

1.01

0.92

2.62

1.18

1.53

0.79

2.43

1.29

1.71

1.94

3.01

0.98

1.28

1.85

3.55

ACM2

2.02

1.06

1.01

2.50

1.04

0.97

1.16

1.94

1.85

1.48

1.68

2.79

1.73

1.78

2.05

3.74

YSU

1.95

0.95

0.94

2.36

0.99

0.84

1.10

1.81

1.80

1.32

1.59

2.64

1.76

1.66

2.06

3.54

SH

2.00

0.97

0.91

2.36

1.11

0.84

1.08

1.87

1.84

1.34

1.60

2.69

1.77

1.70

2.09

3.53

MYJ

0.35

0.95

0.95

0.89

-0.01

0.96

0.98

0.98

0.24

0.94

0.91

0.74

-0.44

0.51

0.56

-0.09

MYNN3

0.56

0.95

0.94

0.53

0.19

0.93

0.97

0.94

0.04

0.94

0.93

0.54

-0.48

0.44

0.59

0.13

UW

0.33

0.95

0.97

0.87

0.01

0.95

0.98

0.99

0.37

0.94

0.91

0.75

-0.55

0.46

0.34

0.22

ACM2

0.54

0.90

0.94

0.13

0.44

0.94

0.95

0.92

-0.53

0.90

0.88

0.16

-0.51

-0.14

0.02

0.13

YSU

0.60

0.92

0.95

0.42

0.43

0.92

0.98

0.97

-0.63

0.88

0.87

0.33

-0.59

-0.03

0.24

0.06

SH

0.50

0.90

0.96

0.56

0.3

0.94

0.98

0.98

-0.50

0.89

0.88

0.54

-0.55

-0.10

0.33

0.14

MYJ

-0.38

0.16

0.61

-0.64

0.30

-0.04

0.78

-0.51

-0.45

-0.68

-0.68

-0.88

-0.46

-0.77

-0.51

-0.96

MYNN3

-0.44

0.25

0.58

-0.6

0.35

0.15

0.69

-0.37

-0.45

-0.67

-0.68

-0.87

-0.46

-0.76

-0.50

-0.96

UW

-0.47

0.15

0.61

-0.64

0.22

-0.10

0.75

-0.51

-0.46

-0.69

-0.69

-0.88

-0.44

-0.76

-0.49

-0.96

ACM2

-0.56

0.16

0.56

-0.61

0.28

0.34

0.62

-0.38

-0.62

-0.65

-0.65

-0.87

-0.69

-0.83

-0.54

-0.96

YSU

-0.55

0.25

0.59

-0.59

0.31

0.43

0.65

-0.34

-0.61

-0.60

-0.63

-0.86

-0.70

-0.81

-0.54

-0.96

SH

-0.56

0.24

0.61

-0.59

0.23

0.42

0.65

-0.36

-0.62

-0.61

-0.63

-0.86

-0.70

-0.82

-0.55

-0.96

r

IOA

256

Appendix G
Centreline profile profiles of key WRF
output variables

257

Figure G.1 Below are are profiles of WRF diagnosed variables along-valley and
through the coastal inlet (see Fig. 2.1). PBLH is planetary boundary-layer height,
CLDB is cloud base height, both in meters above terrain height. CFRAC is cloud
fraction.

258

Appendix H
Formulae for indices of precipitation
prediction
Precipitation forecast scores are from counts of true and false responses for predictions
and observation in the form of a contingency table: Such that:
Table H.1 Contingency table for precipitation events forecast
Modeled
True (T)

False (F)

True (T)

TT

TF

False (F)

FT

FF

Observation

Probability of detection (POD) =

True positives
TT
=
True positives + False negatives TT + TF

Frequency alarm ratio (FAR) =

False positives
FT
=
True positives + False positives TT + FT

Equitable threats score (ETS) =

True positives -C
TT - C
=
True positives + False positives + False negatives TT + TF + FT

where C =

(TT + TF)(TT + FT )
(TT + TF)(FT + FF )

Frequency bias index (FBI) =

True positives + False positives TT + FT
=
True positives + False negatives TT + TF
259

Appendix I
The implication of sulfur exceedance on
vegetated land
From Table 7.3, average S deposition exceedance is ∼ 35 kg ha yr −1
1 km2 = 100 ha
A 10 km2 area (the lower estimate) would recieve 10 x 100 ha x 35

kg
kg
= 35000
= 35
ha yr
yr

tonnes yr−1
Volume 2, page 14 of the technical assessment of Rio Tinto’s application for ammendment
of its SO2 emission permit (Rio Tinto 2013), notes 35 years of future aluminum production,
hence total excess surfur deposition of:
35 yr x 35 tonnes yr−1 = 1225 tonnes.
For a 6 m distance between trees, and 5-6m between tree rows (Food and Agricultural
Organization 1989), I ha can contain 333–400 trees.
Therefore, a 10 km2 area (1000 ha) can contain 1000x the above or 333,000–400,000 trees.

260