DISPERSION MODELLING DURING PARTICULATE MATTER EPISODE
EVENTS IN GOLDEN, BRITISH COLUMBIA

by

Tyler Abel
B.Sc., The University of British Columbia, 2003

THESIS SUBMITTED IN PARTIAL FULFILLMENT OF
THE REQUIREMENTS FOR THE DEGREE OF
MASTER OF SCIENCE
IN
NATURAL RESOURCES AND ENVIRONMENTAL STUDIES

THE UNIVERSITY OF NORTHERN BRITISH COLUMBIA
APRIL 2011

© Tyler Abel, 2011

1+1

Library and Archives
Canada

Bibliotheque et
Archives Canada

Published Heritage
Branch

Direction du
Patrimoine de I'edition

395 Wellington Street
Ottawa ON K1A0N4
Canada

395, rue Wellington
Ottawa ON K1A 0N4
Canada
Your file Votre reference
ISBN: 978-0-494-87568-1
Our file Notre reference
ISBN: 978-0-494-87568-1

NOTICE:

AVIS:

The author has granted a non­
exclusive license allowing Library and
Archives Canada to reproduce,
publish, archive, preserve, conserve,
communicate to the public by
telecommunication or on the Internet,
loan, distrbute and sell theses
worldwide, for commercial or non­
commercial purposes, in microform,
paper, electronic and/or any other
formats.

L'auteur a accorde une licence non exclusive
permettant a la Bibliotheque et Archives
Canada de reproduire, publier, archiver,
sauvegarder, conserver, transmettre au public
par telecommunication ou par I'lnternet, preter,
distribuer et vendre des theses partout dans le
monde, a des fins commerciales ou autres, sur
support microforme, papier, electronique et/ou
autres formats.

The author retains copyright
ownership and moral rights in this
thesis. Neither the thesis nor
substantial extracts from it may be
printed or otherwise reproduced
without the author's permission.

L'auteur conserve la propriete du droit d'auteur
et des droits moraux qui protege cette these. Ni
la these ni des extraits substantiels de celle-ci
ne doivent etre imprimes ou autrement
reproduits sans son autorisation.

In compliance with the Canadian
Privacy Act some supporting forms
may have been removed from this
thesis.

Conformement a la loi canadienne sur la
protection de la vie privee, quelques
formulaires secondaires ont ete enleves de
cette these.

While these forms may be included
in the document page count, their
removal does not represent any loss
of content from the thesis.

Bien que ces formulaires aient inclus dans
la pagination, il n'y aura aucun contenu
manquant.

Canada

ABSTRACT
The CALPUFF modeling system was used to investigate two episodes of high particulate
matter (PM) during December 2005 and February 2006. During this time, Golden was a
British Columbia Ministry of Environment (BC MOE) intensive observation site for air quality
research specific to PM.

Observations from 4 meteorological stations were used to

characterize the winds and dispersion parameters within CALMET. Emission rates were
determined from the existing Golden Emissions Inventory and receptor modelling
commissioned by the BC MOE. Statistical comparison of model predicted and observed PM
concentrations show that model performance compares well to similar CALPUFF
studies at two of the air quality monitoring stations in Golden.

The source

apportionment of the CALPUFF results identified the major contributors to degraded
air quality levels during the two episodes under investigation as space heating, road
dust and, intermittently, Louisiana Pacific operations.

i

TABLE OF CONTENTS

LIST OF TABLES

IV

LIST OF FIGURES

VII

ACKNOWLEDGEMENTS

XII

LIST OF ABBREVIATIONS

XIII

1

INTRODUCTION

1.1
1.2
1.3
1.4
2

OVERVIEW
DISPERSION MODELING STUDY RATIONALE
RESEARCH RATIONALE, QUESTIONS AND GOALS
FORMAT OF THESIS

STUDY BACKGROUND

2.1
2.2
2.3
2.4
2.5

PARTICULATE MATTER
PARTICULATE MATTER IN GOLDEN
METEOROLOGY IN GOLDEN
DISPERSION IN COMPLEX TERRAIN
AIR POLLUTANT DISPERSION MODELLING

2.5.1
2.5.2
2.5.3
2.5.4
2.5.5
3

MODELLING SYSTEM AND DATA COLLECTION

3.1

CALPUFF MODELING SYSTEM

3.1.1
3.1.2
3.1.3

3.2
4

CALMET
CALPUFF.
CALPOST

DATA COLLECTION IN GOLDEN

EPISODE ANALYSIS

4.1

EPISODES MODELED

4.1.1
4.1.2

4.2

4.3

Episode 1 (December 7-14, 2005)
Episode 2 (February 8-23, 2005)

COMPARISON OF WIND CHARACTERISTICS DURING EPISODIC AND NON-EPISODIC
PERIODS

4.2.1
4.2.2
4.2.3
5

Development of Dispersion Models
Plume Models
Advanced Models
CALPUFF Modeling System Development
CALPUFF Case Studies

Wind Speed
Wind Direction
Wind Rose

EPISODE ANALYSIS DISCUSSION

CALMET

5.1

SETUP OF CALMET

5.1.1
5.1.2
5.1.3
5.1,3a
5.1.3b

Modeling Domain
Geophysical and Land Use Data
Meteorological Modeling Data
Surface Station Data
Upper Air Data

1
1
2
5
8
9

9
12
14
17
22
24
26
29
32
35
38

38
38
41
42

42
46

47
47
49

50
51
56
59

62
65

65
65
68
72
72
73

ii

5.1,3c
5.1.4
5.1.5

5.2
5.3
6

Weighting Factors for Surface Stations and Upper Air Stations Used in CALMET.
Calmet Model Options
Calmet Evaluation Methods

CALMET EVALUATION RESULTS
CALMET DISCUSSION

CALPUFF

6.1
6.2
6.3
6.4

CALPUFF SETUP
RECEPTORS
CALPUFF MODEL OPTIONS
ESTIMATING EMISSION DATA

6.4.1
Point Sources
6.4.1a Louisiana Pacific Engineered Wood Products Ltd.
6.4.2
Line Sources
6.4.3
Area Sources

6.5
6.6
6.7
6.8
6.9
6.10
7

77
79
81

85
97
104

106
106
107
108
109
110
115
118

EMISSION RATE ADJUSTM ENTS
EXAMPLE OF CALPUFF EVENING WIND FLOW AND HIGH PM EVENT
CALPUFF EVALUATION PROCEDURES
EPISODE 1 MODEL PERFORMANCE
EPISODE 2 MODEL PERFORMANCE
OVERALL MODEL PERFORMANCE

128
131
138
142
151
159

PARTICULATE MATTER ANALYSIS AND SOURCE APPORTIONMENT

166

7.1
7.2

PARTICULATE MATTER AND METEOROLOGICAL ANALYSIS
SOURCE APPORTIONMENT

7.2.1
7.2.2

7.3

INDIVIDUAL SOURCE RESULTS

7.3.1
7.3.2
7.3.3
7.3.4
7.3.5
7.3.6
7.3.7
7.3.8

7.4
7.5

Episode 1 Source Apportionment
Episode 2 Source Apportionment
LP Emissions
Highway Emissions
Railway Emissions
Space Heating Emissions
Road Dust
Local Traffic Emissions
Construction Emissions
Burning Emissions

COMPARISON WITH RECEPTOR MODELING
DISCUSSION

166
173
173
176

179
179
179
180
181
181
182
182
183

184
187

8

MODELING SUMMARY

194

9

REFERENCES

198

iii

LIST OF TABLES

Table 2-1

Provincial 24-hour Average PM10 air quality objective

11

Table 2-2

Provincial PM2.5 ambient air quality objectives

12

Table 4-1
Comparison of episode 1 average hourly PM concentrations and PM ratios
(PM2.S:PM10) with hourly PM concentrations and ratios from the Winter season (2005 2006)
48
Table 4-2
Comparison of Episode 2 average hourly PM concentrations and PM ratios
(PM25:PM10) with hourly PM concentrations and ratios from the Winter season (2005 2006)
49
Table 5-1

Conversion of B.C. Land Use Codes to CALMET Codes

Table 5-2 Default
Parameters
Table 5-3

70

CALMET Land Use Categories and Associated Geophysical
70

Summary of the four meteorological stations used in CALMET modelling

73

Table 5-4
Weighting Factors for surface stations and upper air station used in
determining CALMET winds in the 10 modeled layers
79
Table 5-5

CALMET wind field model options

80

Table 5-6 Episode 1 CALMET "full" run wind speed statistics. Sp and S0 are the wind
speed predicted and observed. CTp = standard deviation of predicted; CT0 = standard
deviation of observed; ME = mean error; MAE = mean absolute error; MAX = maximum
error; MED = median error; MIN = minimum error; RMSE = root mean square error; FB
= fractional bias, and; CORR = correlation coefficient
86
Table 5-7
Episode 1 CALMET "full" run wind direction statistics. Dp and D0 are the wind
direction predicted and observed. % 10°is the percentage of winds within 10 degrees
of the observed winds. % 45° is the percentage of w inds within 45 degrees of the
observed winds. Refer to Table 5-5 for other parameter definitions
86
Table 5-8 Episode 1 wind speed statistics for CALMET using the "leave one-out" method.
Sp and S0 are the wind speed predicted and observed. Refer to Table 5-5 for other
parameter definitions
87
Table 5-9 Episode 1 wind direction statistics for CALMET using the "leave one-out"
method. Dp and D0 are the wind direction predicted and observed. % 10° is the
percentage of winds within 10 degrees of the observed winds. % 45° is the percentage
of winds within 45 degrees of the observed winds. Refer to Table 5-5 for other
parameter definitions
87
Table 5-10
Episode 2 CALMET "full" run wind speed statistics. Sp and S0 are the wind
speed predicted and observed. Refer to Table 5-5 for other parameter definitions
88
Table 5-11
Episode 2 CALMET "full" run wind direction statistics. Dp and D0 are the wind
direction predicted and observed. % 10°is the percentage of winds within 10 degrees

IV

of the observed winds. % 45° is the percentage of w inds within 45 degrees of the
observed winds. Refer to Table 5-5 for other parameter definitions
88
Table 5-12
Episode 2 wind speed statistics for CALMET using the "leave one-out"
method. Sp and S0 are the wind speed predicted and observed. Refer to Table 5-5 for
other parameter definitions
89
Table 5-13
Episode 2 wind direction statistics for CALMET using the "leave one-out"
method. Dp and D0 are the wind direction predicted and observed. % 10° is the
percentage of winds within 10 degrees of the observed winds. % 45° is the percentage
of winds within 45 degrees of the observed winds. Refer to Table 5-5 for other
parameter definitions
89
Table 5-14

Pasquil Gifford Stability Classes

96

Table 6-1

CALPUFF model options selected

107

Table 6-2
Summary table of LP point emissions. E1 = episode 1, E2 = episode 2, ER =
emission rate
113
Table 6-3 Summary of highway segment emissions. The curved road path was closely
followed using multiple straight line segments. Note: the southern most segment starts
at the southern most point of Highway 95 in the modeling domain and the northern most
point ends at the Town of Golden where it intersects with Highway 1. The eastern most
point of Highway 1 starts at the eastern edge of the domain and the western most point
ends at the western edge of the domain
116
Table 6-4 Summary of rail line emissions. The curved railway path was followed as
closely as possible using multiple straight-line segments. The southern, northern,
eastern and western most points refer to the point where the railway meets the
respective edge of the modeling domain
118
Table 6-5
Area source emission rates for Episodes 1 and 2. All areas were assumed to
emit at the same rate
120
Table 6-6

Diurnal variation of space heating emissions by hour of day

123

Table 6-7

Diurnal variation of road dust emissions by hour of day

125

Table 6-8

Summary of railway yard area and emission rates

127

Table 6-9
Summary of LP yard area and emission rates. Split into north and south to
define more accurate area source polygon
128
Table 6-10
Emission inventory (El), receptor modeling and CALPUFF model predicted
percentages (based on the adjusted El emission rates) of total estimated PM in each
case is provided for (a) episode 1, and (b) episode 2. Emission inventory results are
from those at the TOWN station. LP is used as a comparison because the emission
estimates were the most reliable and complete. CALPUFF results have been adjusted
by comparison with the receptor modeling
131
Table 6-11
PM levels, mixing heights, wind speed and wind direction at the TOWN
station surrounding the high PM event described in Figure 6-5
138

v

Table 6-12
Average observed and predicted values at all three monitoring stations
during Episode 1
142
Table 6-13
Normalised Root Mean Square Error (NRMSE) values for the entire first
episode. 148
Table 6-14
Normalised Root Mean Square Error (NRMSE) values for the top ten
observed values during the first episode
149
Table 6-15

CALPUFF model performance measures for PM10 during Episode 1

150

Table 6-16

CALPUFF model performance measures for PM2 5 during Episode 1

150

Table 6-17
Average observed and predicted values at all three monitoring stations
during Episode 2
151
Table 6-18
episode

Normalised Root Mean Square Error (NRMSE) values for the entire second
157

Table 6-19
Normalised Root Mean Square Error (NRMSE) values for the top ten
observed values during the second episode
158
Table 6-20

CALPUFF model performance measures for PM10 during Episode 2

159

Table 6-21

CALPUFF model performance measures for PM2 5 during Episode 2

159

Table 7-1 Correlation coefficients between predicted meteorological parameters and
predicted PM levels during both episodes. Wspd = wind speed, Mix hgt = mixing height,
Temp = temperature, Mon-obu = Monin-Obukhov length
171
Table 7-2
Multiple regression results for wind speed (WS), temperature (T) and mixing
height (Mix Hgt) as compared to predicted PM10 and PM2.5 levels during each episode...
171
Table 7-3
LP emission source contributions to predicted PM levels at the CPR, TOWN
and GOLF receptors. For comparative purposes the total PM loading for all sources
during the episode (|ag/m3) is also included
179
Table 7-4
Highway emission source contributions to predicted PM levels at the CPR,
TOWN and GOLF receptors. For comparative purposes the total PM loading for all
sources during the episode (|ig/m3) is also included
180
Table 7-5
Railway emission source contributions to predicted PM levels at the CPR,
TOWN and GOLF receptors. For comparative purposes the total PM loading for all
sources during the episode (ng/m3) is also included
180
Table 7-6
Space heating emission source contributions to predicted PM levels at the
CPR, TOWN and GOLF receptors. For comparative purposes the total PM loading for
all sources during the episode (ng/m3) is also included
181
Table 7-7
Road dust emission source contributions to predicted PM levels at the CPR,
TOWN and GOLF receptors. For comparative purposes the total PM loading for all
sources during the episode (ng/m3) is also included
182

vi

Table 7-8
Local traffic emission source contributions to predicted PM levels at the CPR,
TOWN and GOLF receptors. For comparative purposes the total PM loading for all
sources during the episode (ng/m3) is also included
182
Table 7-9 Construction emission source contributions to predicted PM levels at the CPR,
TOWN and GOLF receptors. For comparative purposes the total PM loading for all
sources during the episode (ng/m3) is also included
183
Table 7-10
Burning emission source contributions to predicted PM levels at the CPR,
TOWN and GOLF receptors. For comparative purposes the total PM loading for all
sources during the episode (ng/m3) is also included
184

LIST OF FIGURES

Figure 1-1
The Northern Columbia River Basin. Golden is located at the junction of
Highway 1 (Transcanada Highway) and Highway 95. Perspective view is intended to
show that Golden is located in a deep valley surrounded by steep valley walls. Inset
indicates Golden's location in BC (Map courtesy of Google Earth™)
2
Figure 1-2
The Golden airshed. Indicated by the horizontal blue lines. Map courtesy of
BC MOE
4
Figure 1-3
Wind rose for Golden including all winds recorded from January 2001
through April 2005 (top). Contour map showing valley alignment for comparison with
wind rose. 6
Figure 2-1
2005)

Annual average PM10 and PM2.5 concentrations in BC communities (200313

Figure 2-2
(a) Proportion of days exceeding 25 |ig/m3, the BCMOE 24-hour objectives
for PM10 (50 ng/m3) and 100 ng/m3 from 2003-2005 in Golden, BC (b) Proportion of
days exceeding 15 |ig/m3 (lowest observable effects threshold) and the numeric value
of the Canada Wide Standard for PM25 (30 ng/m3, 24 hour average) from 2003-2005 in
Golden, BC
13
Figure 2-3
Map of Golden showing the location of the 3 within town PM and
meteorological monitoring locations: the golf course north of the city; the CPR site south
of the city and; the town site or central hub. Also shown is the location of the
meteorological station at the airport, the extracted point used in the CALMET evaluation
and the location of Lousiana Pacific Ltd. (LP)
16
Figure 2-4
Diagram of a typical Gaussian plume from an elevated point source. Note
the normal distribution of pollutant dispersal horizontally and vertically (from Oke, 1987).
27
Figure 3-1 A schematic overview of the creation of wind fields in CALMET (adapted from
Scire et al. 2000)
40

vii

Figure 3-2 SCINTEC FAS64 phased-array Doplar sodar system installed near the GOLF
station
44
Figure 3-3 Map of the six HOBO H8 Pro temperature / RH loggers (HOBOs) at 798 metres
(1), 839 metres (2), 960 metres (3), 1036 metres (4), 1081 metres (5), and 1223 metres
(6) along the road to the Kickinghorse Mountain Ski Resort to the west of Golden
45
Figure 4-1
Episode 1 profile of observed PM10 concentrations (ng/m3) and observed
PM2.5 concentrations (ng/m3) at the TOWN station
48
Figure 4-2
Episode 2 profile of observed PM10 concentrations (|ag/m3) and observed
PM2.5 concentrations (|ig/m3) at the TOWN station
50
Figure 4-3 Frequency distribution of wind speed classes (m/s) for annual, winter season,
the month of December, and the month of February with those observed during episode
1 and episode 2 at the Town station. Calm winds are those with a wind speed less than
0.5 m/s. Distributions are calculated from data ranging from 2001-2005. The Town
station is used because it has been operational for the longest period
51
Figure 4-4 Diurnal variation of wind speed by hour of day during each episode modelled at
the Town Station. Other stations showed similar trends
52
Figure 4-5 Hourly PM concentrations and wind speed for Episode 1 at (a) Town station;
(b) CPR station; (c) Golf station
54
Figure 4-6 Hourly PM concentrations and wind speed for Episode 2 at (a) Town station;
(b) CPR station; (c) Golf station
55
Figure 4-7 Hourly PM concentrations and wind direction for Episode 1 at (a) Town station;
(b) CPR station; (c) Golf station
57
Figure 4-8 Hourly PM concentrations and wind direction for Episode 2 at (a) Town station;
(b) CPR station; (c) Golf station
59
Figure 4-9
Wind rose for Golden including (a) all winds recorded from January 2001
through December 2005; (b) all winds recorded in the winter season (November February) from 2001-2005; (c) all winds recorded during Episode 1; (d) all wind
recorded during Episode 2. The wind rose displays the distribution of wind speed
frequency by wind direction. The dashed circles represent the percentage winds in
each direction and the color scale corresponds to the wind speed. Each arm of the
wind rose measures the incremental contribution of each wind speed class to the
percentage of winds in that direction
61
Figure 5-1 Map showing the approximate modeling domain extent around the Town of
Golden. 66
Figure 5-2 Land Use Category Map for the Modeling Domain.
numbers on color scale correspond to those defined in
Figure 5-3

Land use category
71

Twice Daily Upper Air Profiles for (a) Episode 1 and (b) Episode 2

76

Figure 5-4

CALMET surface (10m) wind field for 18:00, December 11, 2005

91

Figure 5-5

CALMET Level 4 (120m) wind field for 18:00, December 11, 2005

92

Figure 5-6

CALMET Level 8 (1250m) wind field for 18:00, December 11, 2005

93

Figure 5-7 Wind speed frequency distribution for the three surface stations and the
extracted point for (a) Episode 1, (b) Episode 2
94
Figure 5-8 Diurnal variation in modeled wind speeds at the extracted point during each
episode
94
Figure 5-9 Predicted wind rose at the extracted point for (a) episode 1, and (b) episode 2,
(c) TOWN station Episode 1 and (d) GOLF station Episode 2
95
Figure 5-10
24-hour mean temperatures for the three surface stations and the extracted
point for (a) Episode 1, (b) Episode 2
96
Figure 5-11
Frequency distribution of predicted stability class at the three surface stations
and extracted point for (a) Episode 1, (b) Episode 2. Stability Class 1 = very unstable;
Stability Class 2 = unstable; Stability Class 3 = slightly unstable; Stability Class 4 =
neutral; Stability Class 5 = slightly stable; Stability Class 6 = stable
97
Figure 6-1
Aerial photo showing source (SRC) locations and building (BLD) dimensions
at Louisiana Engineered Wood Products Ltd. Site in Golden BC
112
Figure 6-2

Location of four areas defined for area source emissions in CALPUFF

120

Figure 6-3 Summary of Golden resident's answers to the question "when do you add
wood?" to their woodstoves asked in the BCMOE woodstove survey (BCMOE 2004)
122
Figure 6-4
2007)

Factor breakdown of PMF analysis for PM25 in Golden (Evans and Jeong
130

Figure 6-5 (a-e) Hourly CALPUFF ground level PM10 concentrations (|ig/m3) from 18:00 to
22:00 on December 10, 2005, representing the entire modeling domain, showing wind
vectors and PM10 concentrations for 5 consecutive hours. Note the increase in PM as
the wind speed slows and the air stagnates around the Town of Golden located at the
convergence of the valley to the east and the main river valley
137
Figure 6-6 Time series plots showing model predicted and observed PM10 and PM25 for
episode 1 at (a) CPR station, (b) Town station, (c) Golf station. Also included is (d) a
time series of mixing height and wind speeds at the Town station
146
Figure 6-7 Scatterplots of predicted versus observed concentrations, paired in location
and time at the CPR station for (a) PM10> and (b) PM25 for Episode 1. The outer black
lines bound predicted concentrations within a factor of 2 of observed values
147
Figure 6-8 Scatterplots of predicted versus observed concentrations, paired in location
and time at the TOWN station for (a) PMi0, and (b) PM2 5 for Episode 1. The outer black
bound predicted concentrations within a factor of 2 of observed values
147
Figure 6-9 Scatterplots of predicted versus observed concentrations, paired in location
and time at the GOLF station for (a) PMi0, and (b) PM2.5 for Episode 1. The outer black
lines bound predicted concentrations within a factor of 2 of observed values
148

IX

Figure 6-10
Fractional bias: Episode 1 using entire episode (most stringent) US EPA
method. The inner box represents acceptable model performance measures (US EPA
1992)
149
Figure 6-11
Time series plots showing model predicted and observed PM10 and PM25 for
episode 2 at (a) CPR station, (b) Town station, (c) Golf station. Also included is (d) a
time series of mixing height and wind speeds at the Town station during episode 2...155
Figure 6-12
Scatterplots of predicted versus observed concentrations, paired in location
and time at the CPR station for (a) PM10, and (b) PM25 for Episode 2. The outer black
bound predicted concentrations within a factor of 2 of observed values
156
Figure 6-13
Scatterplots of predicted versus observed concentrations, paired in location
and time at the TOWN station for (a) PM10, and (b) PM25 for Episode 2. The outer black
bound predicted concentrations within a factor of 2 of observed values
156
Figure 6-14
Scatterplots of predicted versus observed concentrations, paired in location
and time at the GOLF station for (a) PM10, and (b) PM25 for Episode 2. The outer black
bound predicted concentrations within a factor of 2 of observed values
157
Figure 6-15
Fractional bias: Episode 2 using entire episode (most stringent) USEPA
method. The inner box represents acceptable model performance measures (USEPA
1992)
158
Figure 6-16
Hourly PM concentrations and wind direction for Episode 1 at (a) Town
station; (b) CPR station; (c) Golf station. Graphs to the left shows the measurements
observed during each Episode 1 and the graph on the right shows the model predicted
concentrations by wind direction
164
Figure 6-17
Hourly PM concentrations and wind direction for Episode 2 at (a) Town
station; (b) CPR station; (c) Golf station. Graphs to the left shows the measurements
observed during each Episode 2 and the graph on the right shows the model predicted
concentrations by wind direction
165
Figure 7-1 Modeled wind speed and observed particulate matter concentrations at the
TOWN station for (a) Episode 1; (b) Episode 2. The wind speed and PM averages are
from all days in each episode
168
Figure 7-2 Modeled mixing height and observed particulate matter concentrations at the
TOWN station for (a) Episode 1; (b) Episode 2, The mixing height and PM averages
are from all days in each episode
169
Figure 7-3 PM10 source contribution to the entire first episode. RDUST = road dust;
CONS = construction; BURN = burning; TROAD = town road traffic; SPCHT = space
heating; LP = Lousiana Pacific Wood Products Ltd.; Rail = railway; HWAY = highway....
174
Figure 7-4
PMi0 source contribution to the top 25 hourly PM concentrations during the
first episode
175
Figure 7-5

PM25 source contribution to the entire first episode

175

Figure 7-6 PM25 source contribution top 25 hourly PM concentrations during the first
episode
176

x

Figure 7-7

PM10 source contributions to the entire second episode

177

Figure 7-8 PM10 source contribution to the top 25 hourly PM concentrations during the
second episode
177
Figure 7-9

PM25 source contributions to the entire second episode

178

Figure 7-10 PM25 source contribution to top 25 hourly PM concentrations during the
second episode
178
Figure 7-11 Source apportionment comparison between seasonal receptor modeling,
receptor modeling specific to the episode and CALPUFF predicted percentages of total
estimated PM2.5 is provided for (a) episode 1, and (b) episode 2
186

xi

ACKNOWLEDGEMENTS
The author would like to thank first and foremost his supervisor Dr. Peter L. Jackson
for guidance throughout this long journey. Also, his committee, Dr. Margot Mandy,
Dr. Stephan Dery and Steve Sakiyama, who provided great feedback to (finally)
accomplish a finished product. Many thanks to Denny Straussfogel for his help with
the fieldwork and great conversation along the long road to Golden. Special thanks
to Paul Willis of the BC MOE for his words of encouragement and who inspired me
to do this thesis in the first place, the Town of Golden Air Quality Committee and
Mike Brygger from LP.
Funding for this project was provided by the Clean Air Research Fund (CARF), the
British Columbia Ministry of Environment (BC MOE) and NSERC Discovery Grant
research funding via Dr. Peter L. Jackson. All contributors are greatly appreciated!

xii

LIST OF ABBREVIATIONS
Definition

Acronym
o

Standard deviation

ACE

Air Contaminants Emission Project

AGL

Above Ground Level

BC MOE

British Columbia Ministry of Environment

BLD

Building

BPIP

Building Profile Input Program

BURN

Burning Emissions

CONS

Construction Emissions

CORR

Correlation Coefficient

CPR

Canadian Pacific Railway; Site of Monitoring Station in Golden

D

Willmott's Index of Agreement

El

Emissions Inventory

FAC2

Fraction of Predictions within a Factor of 2 of Observed Values

FB

Fractional Bias

FGE

Fractional Gross Error

GIS

Geographic Information System

GOLF

Golf Course Monitoring Station in Golden

km

Kilometre

LP

Louisiana Pacific Engineered Wood Products Limited

LTRFC

Local Traffic Emissions

NRMSE

Normalized Root Mean Square Error

M

metre

m/s

Meters per Second

MAE

Mean Absolute Error

MAX ERR

Maximum Error

ME

Mean Error

MED ERR

Median Error

MG

Geometric Mean Bias

MIN ERR

Minimum Error

Mix Hgt

Mixing Height

Mon-obu

Monin Obukhov Length

MSE

Mean Square Error

MM5

Fifth-Generation NCAR / Penn State Mesoscale Model

OB

Observed Measurement

PM

Particulate Matter

PM10

Particulate Matter smaller than 10 microns in size

PM .5

Particulate Matter smaller than 2.5 microns in size

PR

Model Predictions

RDUST

Road Dust Emissions

RUC

National Oceanic and Atmospheric Association's National
Center for Environmental Prediction Rapid Update Cycle Model

R1

Point at which the influence of a surface meteorological station
is equal to 50 percent

Rmax,

Maximum radius of influence of surface meteorological stations
in CALMET

SPCHT

Space Heating Emissions

SRC

Source

SODAR

Sonic Detection and Ranging

TEMP

Temperature

Terrad

Distance to which a terrain feature in CALMET can affect 3-D
wind flows

TOWN

Townsite Monitoring Station in Golden

Mg/m3

Micrograms per Meter Cubed

US EPA

United States Environmental Protection Agency

UTM

Universal Transverse Mercator Coordinate System

VKmTs

Vehicle Kilometres Travelled

Wdir

Wind Direction

WHO

World Health Organization

Wspd

Wind Speed

2

1

INTRODUCTION

1.1

OVERVIEW
The public, industry and government are becoming increasingly aware of the

detrimental health, economic and environmental effects associated with airborne
pollutants.

Growing concerns have pushed the government, communities and

academia towards furthering their understanding of all processes involved in the
release, dispersion of, and exposure to, airborne pollutants (British Columbia
Ministry of Environment 2002).
To better understand these issues, an intensive particulate matter (PM)
speciation study conducted by the British Columbia Ministry of Environment (BC
MOE) was carried out in the Town of Golden, British Columbia. Golden is a small
community of approximately 5000 residents situated in the northern Columbia River
Basin, bordered by the Purcell Mountains to the west and the Rocky Mountains to
the east (Figure 1-1). To compliment the BC MOE speciation study in Golden, this
study used the CALPUFF modeling system to model the dispersion of PM in the
local airshed. The CALPUFF modelling system is a diagnostic air dispersion model
that utilizes meteorological conditions and wind fields calculated from the
meteorological model, CALMET, to estimate pollutant concentrations in an air
dispersion model called CALPUFF.

1

1.2

DISPERSION MODELING STUDY RATIONALE
The BCMOE identified the following main goals for the Golden speciation study:
1. Ensure federal, provincial, and local air quality goals for the study are
achieved.
2. Identify anthropogenic pollutants found in the Golden airshed so that an
Airshed Management Plan can be optimized.
3. Provide information to allow community planners to determine the most
efficient and economically viable emission reduction options.
4. Establish guidelines for future provincial PM speciation studies.

Figure 1-1

The Northern Columbia River Basin. Golden is located at the junction
of Highway 1 (Transcanada Highway) and Highway 95. Perspective
view is intended to show that Golden is located in a deep valley
surrounded by steep valley walls. Inset indicates Golden's location in
BC (Map courtesy of Google Earth™).

2

In an effort to provide a tool for future airshed management in Golden, this
study used local meteorological and pollutant emission data collected in Golden to
model the dispersion of PM in the Golden Airshed (Figure 1-2).
An airshed is defined here as the extent where topography and meteorology
have a significant effect on hindering the dispersion of air pollutants away from the
area. The airshed extent in Figure 1-2 is bounded by the Rocky Mountains to the
east and the Purcell Mountains to the west.

The winds in the Golden airshed

predominantly flow along the the northeast to southwest orientation of the valley.
The airshed north and south boundaries are based on the dispersion of potential
pollutants in the airshed given the predominant wind pattern.
The data collection activities in Golden during the speciation study provided
an opportunity to evaluate the CALPUFF modeling system in complex, mountainous
terrain.

Results from the dispersion model, can be used as a valuable tool for

constructing a local air quality management plan (AQMP). It is the aim of this study
to model as accurately as possible episodes of high PM. An AQMP for Golden will
not solely focus on episodes of high PM. Such a plan will account for all pollutants
in the airshed and take into account the acute and chronic impacts of degraded air
quality on the community.

Modeling PM episodes will provide insight into the

meteorological and emission conditions that give rise to such episodes. In a local
AQMP, this information could be used to identify when acute impacts from the major
pollutant of concern in Golden, PM, are likely to occur and be used to develop
strategies to reduce community exposure during these times.
3

yOHO

NAT P

KO

GOLDEN

5

1:650,000

0

5 Kilometers

MoTHWY. Transportation Centreline
Network. 120K
=

Highway • Unclassified
Road • linciassiied

SRMB. Annotation, 12M

Ministry of Land.
Water and Air Prokrcnon
Gotten Region
March 30. 3003

SRMB, Annotation, 1:2M
EPBmch, Air Quality Areas ot
*

Interest
r' ->-1

Figure 1-2

Air Quality Area of Interest

The Golden airshed.
courtesy of BC MOE.

Lrifcn&H
uumto

Indicated by the horizontal blue lines.

Map

4

1.3

RESEARCH RATIONALE, QUESTIONS AND GOALS
Winds in Golden tend to be terrain forced flows following the valley alignment

(Figure 1-3). At low wind speeds, dispersion is expected to be influenced by the
diurnal mountain wind system described in Section 2.4. Scire and Robe (1997)
showed that improvements made to CALMET, the meteorological processor for the
CALPUFF dispersion model, improved estimation of wind fields, wind channeling
and mountain valley flows at a complex terrain site in the Columbia River Valley,
when compared to other US EPA regulatory models,.

It was anticipated that

CALMET would be able to estimate both the main valley terrain-induced flows and
the diurnal mountain wind system to model episodes of PM in Golden. Modeling
conditions of low wind speeds and proper diurnal mountain wind system effects is
vital. It is expected that periods of stagnation will be associated with the increase of
PM during episodes.
Episodes of high PM will be modeled.

CALPUFF will be used to investigate

the meteorological, emission and dispersal processes that contribute to the
degradation of air quality and the creation and sustenance of PM episode events.
Two episodes with high concentrations of PM10 and PM2 5as well as different ratios
of PM2.5:PM10 will be modeled for this study.

This will allow for comparison of

dispersion and emission processes during each episode.
Receptor modeling results from BC MOE determined the source contribution
to seasonal PM episodes (Evans and Jeong 2007). The present study will apportion
sources that are contributing to two specific PM episodes by modeling individual
point, area and line sources in CALPUFF.
5

• >7.0 m/s
• S.O • 7.0 m/s

NE

• 3.0 • 5.0 nVs
•1.5 -3.0 m/s

W;

• 0.5-1.5 m/s
• 0.2 • 0.5 m/s

tJTSW

,SE

Calm (<*0.2 m/s) • '

1.0%

4*

Elevation (m)
2800
2700
2600
2500
2400
2300
2200
2100
2000
1900
1800
1700
1600
1500
1400
1300
1200
1100
1000
900
800
700

I

0

2

4

6

Kilometres

Figure 1-3

Wind rose for Golden including all winds recorded from January 2001
through April 2005 (top). Contour map showing valley alignment for
comparison with wind rose.

6

A comparison of the modeled results with those from the receptor modeling will be
used to evaluate CALPUFF's effectiveness as a source apportionment tool.
Along with identifying the source contributions, CALPUFF will be used to investigate
boundary layer or emission processes that lead to:

• spatial differences in PM. Why does source x have a greater contribution
to receptor y as opposed to receptor z in another location?
• temporal differences in PM. Why does source x contribute more to ground
level concentrations when high PM concentrations are recorded as
opposed to when low PM concentrations are observed?
Meeting the study goals entails answering: What are the general dispersal
mechanism and emission processes that contribute to episodes of PM in Golden? It
is hypothesized that evaluation of observational and modeled data will reveal that
episodes are associated with low wind speeds and meteorological conditions that
give rise to periods of stagnation.

Diurnal variation in emission rates may also

contribute to elevated PM levels.
CALPUFF modeling in Golden will evaluate the model's performance in complex
topography. Of interest is CALPUFF's ability to model the dispersal and emission
conditions and accurately predict PM concentrations during the episodes. Previous
studies found that CALPUFF performs well in complex terrain (Scire and Robe 1997,
Barna and Grimson 2002). In Golden, it is expected that CALPUFF will be able to
accurately predict the observed PM concentrations. The meteorological network set
up during the study period allows for evaluation of CALMET in predicting observed
7

surface winds; and CALPUFF in simulating dispersion of PM. Previous BC MOE
CALPUFF modeling efforts in Golden were hindered by using upper air data
collected more than 400 kilometres away, in Kelowna.

The sodar system and

surface HOBO data in combination with upper air data were expected to provide
more accurate local upper air data.
To review, this study sought to answer the following research questions:
•

What are the dispersal mechanisms and emission processes that contribute
to episodes of PM in Golden?

•

Can these conditions be modeled in CALPUFF with the existing emissions
information

to

accurately

represent

PM concentrations

during

these

episodes?
•

Will CALPUFF predicted PM source apportionments compare well with
receptor modeling conducted by BCMOE (Evans and Jeong 2007)?

1.4

FORMAT OF THESIS
This thesis is organized by first presenting the major concepts covered in the

study. This is followed by a description of the methods and modeling system. Next,
a review and analysis of the episodes modeled in the study is provided.

The

CALMET/CALPUFF setup and evaluation is presented in the subsequent two
chapters.

Finally, CALPUFF source apportionment results and analysis are

examined.

8

2

STUDY BACKGROUND

2.1

PARTICULATE MATTER
Airborne PM has been associated with a wide range of adverse human health

effects (Vedal 1995).

Specifically, elderly people who may have compromised

cardiovascular function and younger children with asthma have been recognized as
"at risk" sectors of the population (Caton and Bates 2002, Enstrom 2005). However,
there is epidemiologic evidence that chronic effects from even low level exposure to
PM have the potential to affect all members of the population (Vedal 1997). Over
the short-term, exposure to PM leads to school absences, extreme discomfort for
those with respiratory diseases, and increased hospital admission rates due to
recurrences of respiratory and cardiac conditions (British Columbia Provincial Health
Officer 2003). Long term effects have been associated with depressed lung function
in children, increased cases of bronchitis, increased risk of lung cancer, and
increased mortality due to respiratory and cardiac conditions (British Columbia
Provincial Health Officer 2003, Ovadnevaite et al. 2002). Epidemiological evidence
indicates that Golden and other BC interior communities should be concerned about
high PM levels (British Columbia Provincial Health Officer 2003) and PM will be the
focus of this modeling study.
Specifically this modeling study will focus on episodic (short-term) increases
in airborne PM. A number of studies have coupled the sharp, short-term increases
during episodes with significant changes in lung function and respiratory illness
(Ostor, 1993; Dockery et al. 1996). In extreme cases the high PM episodes can
result in immediate mortality with the US EPA estimating that the pollution-induced
9

spike in the death rate can range from 2% to 8% for every 50 pg/m3 increase in PM
levels (US EPA, 1996). In addition, the World Health Organization (WHO) estimates
that pollution episodes account for 7% - 10% of respiratory illness in children and
0.6% - 1.1% of deaths (WHO, 1994).
As the health effects of airborne PM were identified, it became apparent that
smaller particles (less than 10 microns) have more significant health impacts as they
can be inhaled more deeply into the lungs (British Columbia Provincial Health officer
2003). Specifically, particles with an aerodynamic diameter of less than 10 microns
(and greater than 2.5 microns) have been shown to be capable of penetrating
through the natural protective barriers of the human respiratory system leading to
respiratory and cardiac disease (Ghose et al. 2005). Studies have indicated that the
even smaller particles (less than 2.5 microns) have the ability to penetrate even
further into the lungs causing more severe health effects (Kaur et al. 2005).

In

addition, these smallest particles are more easily able to penetrate indoors, be
transported over long distances, and remain in the air for prolonged periods, as
compared to larger particles (Pope 2000). Smaller particles of PM are generally
associated with fossil fuel combustion from mobile sources or industrial boilers
(Jansen et al. 2005, Win Lee et al. 2004).
It has been established that the size of airborne PM can serve to distinguish
the origin, composition and associated health effects of particles. Coarse particles,
defined as those particles with an aerodynamic diameter greater than 2.5 microns,
are derived from soil and earth's crustal materials (Pope 2000).

PM with an

aerodynamic diameter of less than 10 microns (PM10) are capable of penetrating into
10

the lungs. Therefore, measuring PM10 concentrations is of interest to health officials
as it will quantify soil and crustal material particulates as well as other emitted
particles that affect human health. Fine particulates, defined as those particles with
an aerodynamic diameter less than 2.5 microns (PM2.5), are primarily derived from
combustion during processes such as transportation, manufacturing and power
generation (Pope 2000), but can also include finer dust particles. Measurement of
PM2.5, a subset of PM10, is of great interest in determining the effects of these
specific, generally anthropogenic, combustion processes on human health because
of the noted ability of these smaller particles to penetrate deeper into the lungs
causing more severe health effects than larger particles.
In an effort to reduce PM concentrations in British Columbia communities, the
provincial government has instituted a province-wide 24-hour PM10 objective that is
applicable to the Town of Golden. Table 2-1 outlines the definition of the objective
as well as PM concentration targets.

Table 2-1

Provincial 24-hour Average PM10 air quality objective.

National Maximum
Acceptable Level

Provides adequate protection against
adverse effects on human health,
vegetation and animals. Usually set as an
intermediate objective for all existing
discharges to reach within a specified
50 |ig/m
time period, and as an immediate
objective for existing discharges which
may be increased in quantity or altered in
quality as a result of process expansion or
modification

11

The objectives set for PM2 5 in British Columbia were released in April 2009 and are
summarized in Table 2-2.

Table 2-2

Provincial PM2.s ambient air quality objectives

Air Quality Objective
Averaging Period
24-hour
25 |ig/ma *
8 |ig/m3
Annual
th
* based on annual 98 percentile value

In addition to adverse health effects, PM is of concern in communities, such as
Golden, that rely on tourism. Particles in the air scatter light, causing a degradation
of visibility (Watson 2002). Golden, a community that promotes itself as a pristine
natural environment for outdoor activities ranging from skiing to eco-tourism has a
great interest in having clear vistas, free of haze and smog, for its many visitors to
enjoy the area.

2.2

PARTICULATE MATTER IN GOLDEN
This study focuses on dispersion modeling of PM in the Golden airshed.

Specifically, two fractions of the total PM were investigated, PMi0 and PM2.5. Golden
had the highest average concentration of PM10 in BC communities from 2003-2005
(Figure 2-1). PM25 concentrations were also very high when compared to other BC
communities (Figure 2-1).
Steep valley walls, mountain-valley winds and low wind speeds can make
Golden susceptible to wintertime inversions and stagnation events throughout the
year (Burkholder 2005). Local monitoring reveals that episodes of elevated levels of
PM are typical during the winter months (Figure 2-2). Episodes greatly contribute to

12

the high annual averages observed in Golden and as mentioned previously are of
concern to the health of the residents of Golden, especially children, the elderly and
those with respiratory illnesses who are more easily affected by higher PM levels.

25

• Golden
• Prince George

20

• Quesnel
• Chilliwack

CO

ia.1 5
••e

n

• Vancouver

10

o.

PM10

Figure 2-1

PM2.5

Annual average PM10 and PM2.5 concentrations in BC communities
(2003-2005).

(a)

(b)
0> I00pg/m3
Q 50 -100 |j^m3

• >30 ng/m3
• 1S - 30 utfm3

25 • 50 pg/m3

Figure 2-2

(a) Proportion of days exceeding 25 (jg/m3, the BCMOE 24-hour
objectives for PM10 (50 ng/m3) and 100 (jg/nr from 2003-2005 in Golden,
BC (b) Proportion of days exceeding 15 ng/m3 (lowest observable
effects threshold) and the numeric value of the Canada Wide Standard
for PM2.5 (30 yg/m3,24 hour average) from 2003-2005 in Golden, BC.

13

Although Golden does not have a large number of industries, it consistently
ranks among the highest mean annual concentrations of PM in the province.
Golden was chosen as an ideal site for the BC MOE speciation study for several
reasons that also make it a challenging and interesting airshed to model with
CALPUFF:
1. The Town's topography (i.e. valley bottom), climate and meteorological
characteristics are typical of many communities in the interior of BC.
2. The Town's historical PM10 and PM25 levels are well above the provincial
average.
3. The Town's predominant emission sources (woodstoves, open burning,
transportation, rail yards, wood processing, and road dust) are typical for
communities in the interior of BC.
4. There is strong community support for the development of an air quality
management plan in Golden.
(Golden

Source

Apportionment

Study:

Implementation

and

Planning

Document May 10, 2004)

2.3

METEOROLOGY IN GOLDEN
Dispersion models attempt to accurately estimate the meteorological conditions

that disperse airborne pollutants. Meteorological conditions influence ground level
PM concentrations in Golden and other communities in numerous ways.

The

release of PM from fugitive dust sources such as roadways is affected by rain or
snowfall that can suppress or even stop the emissions from leaving the surface.

14

There is a greater level of emissions from residential heating during the winter or on
colder fall and spring days. Dust from wind erosion increases when winds are high.
Following the release of airborne PM from the source, the dispersal pattern is
determined by the meteorological conditions.

The stability of the atmosphere,

determined largely by the vertical temperature gradient, enhances or restricts the
ability of pollutants to mix with the air above. High wind speeds can quickly disperse
and dilute a source's emissions. Conversely low wind speeds may allow for the
build-up of pollutants around a source and a rapid increase in PM concentrations.
The creation of temperature inversions and conditions of low wind speeds are
enhanced by mountainous topography, serving to trap pollutants closer to the valley
bottom where communities are typically located. Thus, dispersion models attempt to
incorporate meteorological, terrain and land-use data to predict the stability, wind
flows and turbulence of the atmosphere.

This can be especially difficult when

dealing with complex terrain.
Typically, meteorology is monitored only at the Golden airport (AIRPORT)
and the Golden Townsite (TOWN).

However, an additional two BC MOE

meteorological stations were added to the GOLF and CPR stations during the
speciation study period (Figure 2-3).

15

5687

5686

5685
E
CI)

c 5684

!c
•e

o

2 5683
iZJ
5682

5681

5680

IMWVAR
498

499

500

501

502

503

504

505

506

507

508

UTM Easting (km)
Figure 2-3

Map of Golden showing the location of the 3 within town PM and
meteorological monitoring locations: the golf course north of the city; the CPR
site south of the city and; the town site or central hub. Also shown is the
location of the meteorological station at the airport, the extracted point used in
the CALMET evaluation and the location of Lousiana Pacific Ltd. (LP)

16

2.4

DISPERSION IN COMPLEX TERRAIN
Golden is located in a deep valley surrounded by steep valley walls and

tributary valleys immediately east and near the north end of the main valley (Figure
1 -1). Dispersion in Golden was expected to be heavily influenced by the complex
topography surrounding the Town.

The following section will explain the basic

processes involved with dispersion in complex terrain.
There are two general types of winds associated with mountainous terrain:
terrain-forced flows and diurnal mountain winds. Terrain-forced flows are produced
when large-scale synoptic winds are modified or channeled by complex terrain
(Whiteman 2000).

Diurnal mountain winds result from the temperature contrasts

that form within the mountains or between the mountains, when the elevated land
surface heats and cools more rapidly, due to solar and infrared radiation, than the
adjacent atmosphere.

This results in horizontal pressure differences that create

thermally driven circulations (Whiteman 2000). The combination of these two types
of winds largely influences dispersion in complex terrain (Arya 1999).
Terrain-forced flows result as winds encounter mountain barriers. As the air
flow approaches the mountain it can either be carried over, around, be forced
through gaps or blocked by the mountain. Three factors will determine the path that
a particular air flow will take when approaching a mountain barrier: the stability of
the air, the wind speed, and the topography of the terrain (Whiteman 2000).
In boundary layer meteorology, atmospheric stability can be classified into
three basic stability types: unstable, neutral and stable. The classification of stability

depends on the rate of change of temperature with height above ground, and
whether the air is saturated or not. In simple terms an unstable boundary layer
typically occurs on hot days, where the air temperature near the surface is much
greater than the temperature aloft.

Any vertical motion is enhanced since the

surface air rises rapidly (it is less dense that the surrounding air), and mixes.

A

neutral atmosphere means that there is no enhancement or suppression of vertical
motion induced because of density differences - in other words, the air temperature
of vertically displaced air is the same as the air temperature of the surrounding air at
a given height. Unstable and neutral air masses can be carried over mountainous
terrain because they are not resistant to vertical motion. Finally, a stable atmosphere
means that vertical motions are suppressed.

An example of stable conditions

occurs during inversions, where cold air at lower elevations will remain near the
surface because it is more dense than the air above.
This means that stable air masses are more resistant to lifting. Stable air is
generally blocked, forced through terrain gaps or undergoes splitting as it
approached mountainous terrain.

Blocked stable air can be trapped on the

windward side of the mountain. Splitting of the stable air mass occurs along the
dividing streamline determined largely by the speed of the air mass, the surface
roughness, the air's stability and the shape of the terrain.

Air above the dividing

streamline flows over the mountain and air below the dividing streamline flows
around the mountain or is blocked (Whiteman 2000). Barriers of flow can also
produce eddies that re-circulate air on either the windward or leeward side of the
mountain (Arya 1999). Mountainous terrain can also greatly reduce wind speeds in
18

deep valleys that are often protected from the prevailing winds by the topography
(Oke 1997), certainly a phenomena likely to affect dispersion in Golden.
One challenge of a dispersion model is the ability to properly predict whether
pollutants being dispersed towards mountains will be blocked, go over or around, be
split, or be re-circulated by the barrier. CALPUFF accomplishes this through a puffsplitting scheme incorporated in the model (Scire et al. 2000).
Diurnal mountain winds are thermally driven circulations that are common in
mountainous terrain.

Wind generally flows upslope, upvalley and from low lying

areas to the mountain in the daytime.
(Whiteman 2000).

Reverse flows are seen during the night

These types of flows are produced by horizontal temperature

differences causing winds near the surface to blow from areas with lower
temperatures and higher pressures to areas with higher temperatures and lower
pressures (Curry and Webster 1999). The influence of diurnal mountain winds is
strongest when skies are clear and large-scale winds are weak (Whiteman 2000). In
Golden, blocking of large-scale winds by the mountainous terrain may mean diumal
mountain winds are quite influential in the dispersion of pollutants.
There are three types of wind flows that define the diurnal mountain wind
system in Golden:
1. The slope wind system
2. The along-valley wind system
3. The cross-valley wind system
(Whiteman 2000)
19

The slope wind system produces the upslope and downslope winds in
mountainous areas. It is driven by the horizontal pressure gradient between the air
near the slope and the air at the same elevation nearer to the center of the valley.
The horizontal pressure gradient exists because of

horizontal temperature

differences caused by the diurnal heating and cooling of the slope surface
(Whiteman 2000). The along-valley wind system is driven by pressure difference
that occurs when the higher elevations in the valley generally become colder during
the nighttime and warmer during the daytime compared with the air at the same
altitude elsewhere in the valley. This causes an upvalley wind during the day and a
downvalley wind during the night (Whiteman 2000). Cross-valley winds occur when
there is a temperature difference between the two sidewalls of the valley (Whiteman
2000). These winds are more prominent during the day when the sidewalls are
unequally heated by the sun. All of these flows are closed circulation flows, meaning
they have an associated recirculation air flow higher above the surface.
A combination of diurnal mountain winds, terrain-forced flows and the surface
heat budget contribute to a daily pattern of winds that influence the dispersion of
pollutants in complex terrain.

This study will focus on short time periods during

episodes of high PM concentrations. Episodes usually last for several days at the
most. Therefore, it is necessary to review the diurnal patterns of winds, as these
winds are likely important in the creation, sustenance and degradation of PM
episode events in Golden (Burkholder 2005).
During the daytime, the mixing height is quite high in the valley and may often
be coupled with the atmosphere above the valley (Arya 1999), especially during
20

summer with prominent upslope and upvalley winds and convective atmospheric
conditions result in good dispersal. During the evening transition phase the slope
wind system reverses and flows down the slope. This usually causes cold air to
drain down the slopes and into the valley bottom and leads to the build up of an
inversion (Whiteman 2000). During an inversion, dispersion of emissions released
below the level of the inversion is poor as the mixing height is low, limiting vertical
motion in the atmosphere.

Down-slope flow and the resultant temperature

difference in the valley lead to a reversal of the along-valley wind during the evening
transition period (Whiteman 2000). During the night, downslope winds continue to
contribute to the inversion and the down-valley winds blow within the inversion layer
providing for minimal dispersion of pollutants out of the inversion layer (Whiteman
2000).

During the morning, typically the reversal of both slope and along-valley

flows and the convective currents from the ground combine to breakup the inversion,
and thus improve dispersion (Whiteman 2000). This is a description of typical daily
conditions that would be realized under surface high pressure systems with weak
synoptic pressure gradients resulting in calm, clear conditions. In Golden, episodes
of high PM are usually associated with a combination of high emissions and
stagnant conditions that prevent the breakup of the nighttime inversions (Burkholder
2005). In this case, the night-time inversions tend to persist and mixing heights
remain low for much of, or the entire day if the solar heating is weak causing
pollutants to remain trapped near the surface and the population of the town.

21

2.5

AIR POLLUTANT DISPERSION MODELLING
An atmospheric dispersion model provides a mathematical simulation of the

physics and chemistry governing the transport, dispersion and transformation of
pollutants in the atmosphere. A dispersion model is capable of estimating air
pollution concentrations given information about the pollutant emissions and nature
of the atmosphere (NIWA 2004). Air quality modeling allows researchers to simulate
and assess air quality issues without the expense associated with extensive air
quality monitoring networks.

Air quality monitoring networks are necessary to

validate dispersion modeling, however, the models can provide specific quantitative
estimates of pollutant levels at many points over a wide geographical area (Scire
and Godfrey 2005).

The expense associated with air quality monitoring restricts

measurements of ambient pollutant concentrations to only a few locations in a
region.
Receptors are points in a modeling domain where concentrations of pollutants
are estimated from the emissions and meteorological information supplied.
Receptors can be placed anywhere within the modeling domain allowing
researchers to provide more accurate estimates of pollutant concentrations in
complex terrain. This allows modelling to investigate the influence of geophysical
and meteorological factors on pollutant concentrations in much more detail than
would ever be possible with traditional monitoring techniques.

With a validated

model researchers can use the modeling domain as a numerical laboratory where
they can conduct experiments that would not be feasible in the real world.

An

accurate representation of how pollutants disperse near populated areas is useful in:
22

•

assessing compliance of air emissions with air quality guidelines, criteria and
standards;

•

planning new industrial facilities;

•

determining appropriate industrial stack heights;

•

managing existing emissions in an airshed;

•

designing ambient air quality monitoring networks;

•

identifying the contributors to existing air pollution problems;

•

evaluating air quality policy and pollution mitigation strategies;

•

forecasting pollution episodes;

•

assessing the risks of and planning for the management of rare events such
as accidental hazardous substance releases;

•

estimating the influence of geophysical factors on dispersion;

•

running 'numerical laboratories' for scientific research involving experiments
that would otherwise be too costly in the real world and;

•

saving cost and time over monitoring - modeling costs are a fraction of
monitoring costs and a simulation of annual or multi-year periods may only
take a few weeks to assess. (NIWA 2004)
Achieving an accurate picture of air quality at a local level typically requires a

model capable of correctly representing dispersion conditions, emissions and
pollutant chemistry in complex terrain as well as during periods of stagnation. This
23

stems from the fact that episodes of high pollutant concentrations are generally
related to a combination of topographical effects and stagnant atmospheric
conditions.

With health and environmental consequences of elevated pollutant

levels becoming increasingly better understood (Vedal 1995), modeling these
conditions is of interest at a community health level.

One advanced dispersion

model capable of modeling dispersion in a non-steady state, complex terrain
environment is the CALPUFF modeling system developed by EarthTech Inc. for the
United States Environmental Protection Agency (USEPA) (Scire et al. 1999).
CALPUFF will be used to model dispersion of PM in Golden. The following is a
review of the current status of dispersion modeling with a specific emphasis on
modeling during periods of high pollutant concentration and assessment of
CALPUFF in a variety of applications.

2.5.1 Development of Dispersion Models
Most early mathematical models were based on the behaviour of plumes in
neutral atmospheric conditions (Venkatram 1988). It was thought that under neutral
conditions the structure of the planetary boundary layer could be defined through a
series of theoretical assumptions based on the fundamental laws describing
dispersion of windborne material presented by Pasquil (1961). However, it became
apparent through experimentation that description of dispersion based on these
theoretical assumptions did not always hold true (Wyngaard 1988). It has been a
blend of experimentation and refinement of theory that has influenced model
development over the past few decades. For short-range dispersion especially, it
became evident that accurate description of constantly changing parameters, such
24

as mixing heights, wind speed, wind direction, as well as phenomena such as the
formation and breakup of temperature inversions, within the boundary layer were
necessary to model the movement of air pollutants.

It also became clear that

modeling these parameters over complex terrain was necessary as populated areas
commonly lie in valleys and mountainous, non-flat terrain. Meteorological conditions
vital to dispersion in complex terrain are not defined well by models designed for,
and tested in, flat terrain (Schnelle and Dey 2000).
The dispersion of pollutants in the surface boundary layer over homogeneous
conditions has been captured relatively well by dispersion models such as ISC
(USEPA), AUSPLUME (Environmental Protection Authority (Australia) 2000), and
AERMOD (Cimorelli et al. 2005). However, flows and dispersal characteristics over
complex topography have not been modeled well.

Past models developed to

describe dispersion over relatively flat terrain had difficulty simulating dispersion over
complex terrain. Modeling difficulties in mountainous terrain are compounded by the
fact these areas are prone to stagnant atmospheric conditions and terrain induced
local circulation (Triantafyllou 2002).

Thus, a model capable of mathematically

including these phenomena has been the goal of leading air quality agencies such
as the US EPA.
Most modern air pollution models are computer programs that calculate the
pollutant concentration of the sources in an area using input information that is more
detailed than, but relies on similar concepts, as early models. These include:

•

pollutant emission rate
25

•

characteristics of the emission source (e.g. temperature, velocity,
elevation above ground)

•

local topography

•

meteorology of the area

•

ambient or background concentrations of pollutant. (NIWA 2004)

2.5.2 Plume Models
Currently, the most commonly used dispersion models are steady-state
Gaussian-plume models (e.g. Ausplume, ISC). These are based on a mathematical
approximation of plume behaviour and are the easiest models to use. They
incorporate a simplistic description of the dispersion process, and some fundamental
assumptions are made that may not accurately reflect reality. These assumptions
state primarily that as pollutants are dispersed from a source, they tend to follow a
normal (Gaussian) distribution in the horizontal and vertical direction (Figure 2-4).
However, even with these limitations, this type of model can provide reasonable
results when used appropriately, for example in regulatory applications (Isakov
2004).
Gaussian-plume models have been useful in regulatory application (Brechler
2000), but have been shown to produce unreliable results in complex terrain and
under stagnant conditions (Oettl et al. 2001, Abdul-Wahab 2004).

Adaptations

allowing for non-Gaussian dispersion in the vertical direction during convective
conditions has allowed for certain Gaussian-plume models (e.g. AERMOD) to
capture maximum concentrations of pollutants reasonably well in complex terrain
(US EPA 2003). However, formulas guiding strict Gaussian-plume models are
26

Figure 2-4

Diagram of a typical Gaussian plume from an elevated point source.
Note the normal distribution of pollutant dispersal horizontally and
vertically (from Oke, 1987).

based on "steady-state" conditions, meaning the formulas calculate concentrations
for each hour based on an emission rate and meteorological conditions that are
assumed to be uniform across the modeling domain (Sakiyama et al. 2005).
Conditions from hour to hour can change in Gaussian models. However, each hour
is treated independently of other hours, with each hour's emissions dispersing in a
straight-line trajectory to the edge of the modelling domain.

Although these

assumptions work reasonably well under relatively steady conditions, they do not
work well when there are large changes in meteorological conditions from hour to
hour (Schnelle and Dey 2000). Gaussian plume models assume that a pollutant
plume will have its highest concentration on a straight-line trajectory in both the
horizontal and vertical direction downwind of the emission source (Whiteman 2000).
The diffusion of pollutants from this straight-line maxima is dependent on dispersion
27

coefficients usually determined over flat homogeneous terrain and follow a normal
(Gaussian distribution) outwards from the maxima (Whiteman 2000) These
assumptions place temporal and spatial limitations on Gaussian models that inhibit
their ability to model dispersion accurately in complex terrain, because of their
inability to handle spatially varying wind flows, flows from tributary valleys, terraininduced flows or diurnal mountain winds.

Gaussian models do provide some

conveniences that make them appropriate for more simplistic modeling situations.
The simple approach means that the models do not require significant computer
resources, are user-friendly, have simple meteorological data requirements and
were designed to give conservative results (NIWA 2004).

However, without the

ability to model complex convective and stable conditions, Gaussian models cannot
effectively model causality effects (i.e. the transport time required for pollutants to
reach receptors), dispersion at low wind speeds, dispersion in complex terrain,
changes in atmospheric conditions, and processes, such as inversion break-up,
fumigation and diurnal recycling of pollutants, varying meteorological conditions,
changing pollutant dispersion coefficients, and emissions from previous hours that
affect subsequent hourly concentrations of pollutants (NIWA 2004). Still, Gaussian
type models have proven useful in providing a simple model of atmospheric
dispersion. They allow an adequate description of the boundary layer processes
affecting dispersion under homogeneous conditions.
For the most part, the less-sophisticated Gaussian plume models are used
primarily for regulatory applications. Modeling for regulatory applications refers to
those activities concerned with assessing compliance of emissions with ambient air
28

quality guidelines, criteria and standards.

In general, Gaussian dispersion is

adequate enough to model for regulatory purposes because regulations focus on the
maximum concentrations independent of where or when they occur rather than
specific levels of pollutant concentrations across the modelling domain.

2.5.3 Advanced Models
New dispersion coefficients, vertical wind shear and plume behaviour in the
boundary layer have been incorporated into a new generation of dispersion models
capable of better estimating dispersion in complex terrain. Since many communities
and their industries are located in complex terrain, the need for understanding and
modeling boundary layer processes under these conditions is apparent (Sakiyama et
al. 2005). For this reason, field studies over recent years have focused on improving
estimates of dispersion in complex terrain (e.g. Triantafyllou 2002, Dosio et al. 2001)
so more accurate estimates of important parameters such as dispersion coefficients,
vertical wind shear and plume behaviour can be incorporated into these new
models.
Better ways of describing the spatially varying turbulence and diffusion
characteristics within the atmosphere have been, and continue to be, developed (eg.
Bellasio 2005, Moraes 2005). The new generation dispersion models are capable of
modeling steady-state conditions as well as non-steady state conditions (Schnelle
and Dey 2000). They allow for modeling of variable and curved plume trajectories,
variable meteorological conditions, retention of previous hour emissions, and better
handling of low and zero wind speeds (Scire and Godfrey 2005). The models include
terrain steering effects based on wind variation due to topography and surface
29

properties balanced by an overland surface energy budget, as well as, splitting of
plumes around and over hills, allowing for better representation of dispersion
processes in complex terrain (Scire and Godfrey 2005).
The three most common types of advanced dispersion models are particle,
puff, and grid point models. Particle models (e.g. KSP, Strimaitis 1995) represent
pollutant releases by a steady stream of particles which are advected and diffused
by the modeled wind fields.

Puff models (e.g. CALPUFF, Scire et al. 1999)

represent pollutant releases as a series of puffs which are also advected and
diffused by the modeled meteorology. Grid point models (e.g. CALGRID, Yamartino
et al. 1992) represent pollutant distribution by concentrations and chemical
transformations on a three-dimensional grid of points (NIWA 2004). The models
depend on a specific meteorological processor to determine the meteorological
parameters affecting dispersion.

Each of the three model-types vary in their

computational efficiency and application (NIWA 2004).
An important difference between advanced models and Gaussian-plume
models is the calculation of gridded wind fields, derived from vertical and horizontal
winds combined with kinematic terrain effects and diurnal mountain winds in a
divergence minimization scheme, that are then used to transport pollutants.
Advanced models require three-dimensional meteorological fields because they do
not rely on the assumption of steady state winds across the modeling domain (Scire
and Godfrey 2005).

The detailed wind fields produced by the meteorological

processor associated with an advanced dispersion model allows for more 'realistic'
outcomes. In many cases the meteorological data required for these models are
30

limited or not available.

The limitations of these models must be understood and

modelers must take care in choosing the appropriate level of sophistication for the
meteorological data available (Schnelle and Dey 2000). Even with quality
meteorological input data, modeling efforts in complex terrain and under stagnant
conditions can be difficult, despite the use of more advanced models. During these
times low speed wind flows affecting dispersion are difficult to define. This is a result
of a number of factors including anemometer thresholds (typically anemometer
thresholds do not allow recording of wind speed of less than 0.5 m/s), using hourly
averaged meteorological parameters and model algorithms that breakdown under
low or zero wind speed conditions (i.e. CALPUFF mixing height calculation during
the night).
This study will apply a dispersion model to short episodes of high PM
concentration. The more advanced, new generation models are primarily concerned
with modeling of non-steady state conditions, allowing for a more accurate
description of the dispersion of pollutants from source to receptor. These models
are designed to describe atmospheric dispersion phenomena such as terrain
steering effects, inversion formation and breakup, and fumigation that allows the
user to assess pollutant concentrations in complex terrain and varying dispersal,
emission and chemical transformation conditions (Schmitz 2005).

The advanced

model selected for use in this study is the CALPUFF modeling system (Scire et al.
1999). The features of CALPUFF that suit its application in episode analysis are
described below.

31

2.5.4 CALPUFF Modeling System Development
The CALPUFF modeling system from the US EPA (Scire et al. 1999) has
been developed with the ability to model dispersion in a non-steady state, complex
terrain environment. The original design specifications for the model demonstrate
the intent to improve the spatial and temporal complexity of previous Gaussian-type
models:
•

the capability to treat time-varying point and area sources;

•

suitability for modeling domains from tens of meters to hundreds of kilometers
from a source;

•

predictions for averaging times ranging from one-hour to one year;

•

applicability to inert pollutants and those subject to linear removal and
chemical conversion mechanisms and;

•

applicability for rough or complex terrain situations (Scire et al. 2000)
The CALPUFF modelling system was originally developed by Earth Tech Inc.

under contract from the California Air Resources Board (CARB) (Scire et al. 2000).
The research team developed a modeling system that contained a meteorological
modeling package, CALMET, to define the wind fields affecting dispersion. This
meteorological information is then used to drive a Gaussian puff dispersion model
(CALPUFF) on a gridded modeling domain to deal with a variety of modeling
applications, such as complex terrain, fumigation, wet and dry deposition and
building downwash (Scire et al. 2000). A peer-review of the model recommended
the use of the CALPUFF modelling system in its designed application of long-range
transport assessments and near-field applications (Allwine et al. 1998).
32

The

meteorological

model,

CALMET,

utilizes

meteorological

and

topographical data commonly available for most regions to develop hourly wind and
temperature fields on a three-dimensional, gridded modeling domain (Scire et al
2000). Scire and Robe (1997) showed that CALMET was able to reproduce key
elements of wind flow in the Columbia River Valley, a complex terrain site.
The outputs from CALMET are used to drive the transport and dispersion
model, CALPUFF, which advects and diffuses puffs of material emitted from sources
based on the temporal and spatial variation in the meteorological fields. The three
dimensional, gridded modeling domain allows for spatially and temporally varying
conditions rather than the traditional Gaussian plume model that assumes uniformity
over the entire hour for the plume.
Extensive hourly meteorological input parameters including wind speed and
direction, temperature, cloud cover, surface pressure, relative humidity, and
precipitation, as well as twice daily upper air radiosonde data, are required as inputs.
Meteorological data reliability and availability within the modeling domain of interest
is important to assess before CALPUFF is used (Scire and Godfrey 2005). The
meteorological observations required to drive the CALMET algorithms is designed to
provide a more realistic representation of wind fields and atmospheric processes
used to drive dispersion within the modeling domain. However, this could be viewed
as a stumbling block in proceeding with CALPUFF modeling in rural areas where
such meteorological data are not regularly available. For these instances, CALMET
has the option to utilize outputs from prognostic mesoscale models (e.g. RUC, MM5,
or MC2 model output data) that are able to predict many of the parameters from
33

limited data (McEwen 2003, Pielke 1998, Robe and Scire 1998). This option will not
be used in the present study which focuses on using the available surface and upper
air data in Golden as the only input data.
The model's effectiveness for local episode analysis applications stems from
its ability to detail the major processes affecting transport, diffusion, and deposition
of pollutants (Allwine et al. 1998). The transport and dispersion of pollutants follows
the three-dimensional

wind fields generated in CALMET.

Other gridded

meteorological fields including surface friction velocity, convective velocity scale,
Monin-Obukhov length, mixing height, air temperature, and precipitation rate are
computed using an overland boundary layer energy balance proposed by Holtslag
and van Ulden (1983).

Meteorological fields are determined by both upper air

observations and an option for vertical extrapolation of surface observations using
similarity theory during steady state conditions and probability density fluctuation
during convective conditions.

The model also retains previous hours' emissions,

allowing pollutants to remain on the modeling grid and continue to be advected,
diffused and detected as part of subsequent hours' pollutant concentrations. The
retention of puffs from hour to hour also allows CALPUFF to model calm hours by
simulating stagnant puffs. During hours with a zero wind speed, stagnant puffs on
the modeling grid are not dispersed via advection, but may still undergo turbulencerelated dispersion (Scire 2000). Furthermore, even if the measured wind speed is
zero, CALPUFF accounts for other possible flow components (i.e. slope flows)
during these stagnant modeling periods. The numerous algorithms used to describe
dispersion processes in CALPUFF have been reasonably well assessed on an
34

individual basis (Allwine et al. 1998). It is still necessary to evaluate how the model
components perform in the Golden airshed. Evaluation of the CALPUFF modeling
system is one of the goals of this study and will include a separate evaluation of
CALMET and CALPUFF. Evaluating the two model components separately will help
to discern what part of the disagreement between the model and observations has
its roots in CALMET and what disagreement is due to CALPUFF.

2.5.5 CALPUFF Case Studies
Evaluation of CALPUFF in local scale episode analysis applications has not
been extensive, but a review will provide evidence for its validity in these situations.
The following examples will also demonstrate the diversity in applications of
CALPUFF. CALPUFF has also been used in regulatory applications (Elbir 2003,
Irwin et al. 2005), however that will not be the focus of the following assessments.
Barna and Grimson (2002) applied the CALPUFF modeling system to a
wintertime PM pollution episode. In their evaluation of CALPUFF's ability to predict
wintertime PM10 concentrations during an episode, they used the fractional gross
error and fractional bias, along with the index of agreement and correlation
coefficients to compare modeled results with observed PM10 levels.
Barna and Grimson (2002) found highly encouraging results with statistical
performance tests of CALPUFF reporting good agreement between model predicted
and observed PM concentrations.

Differences between observed and predicted

values are generally explained by peak concentrations being modeled at receptors
close to, but not exactly at, the observed monitoring location. This suggests the

35

model is accurately predicting the particulate concentrations, but is misrepresenting
the location of the maxima.
CALPUFF was used to analyze a two-day wind episode in the Mexico City
Basin in 1997 (Villasenor et al. 2003). The goal of the study was to simulate PM10
emissions from a wind-blown dust event. Wind-blown dust from agricultural lands as
well as river and lake beds was thought to be an important source of PM in the city.
The wind episode was carefully chosen so modelers could examine processes
creating, maintaining and terminating the PM10 episode. Throughout the episode,
CALPUFF agreed reasonably well with observed values and provided strong
evidence that the hypothesis, stating that the episode arose from dust originating
from wind erosion, was correct.
Levy et al. (2002) used CALPUFF to evaluate the impacts of power plant
emissions on residents of Illinois.

Their analysis included an assessment of the

model's sensitivity to various input parameters. Estimates based on some of the
most influential parameters used in CALPUFF such as background concentrations,
deposition, chemical mechanisms, and size of the receptor region were found to be
moderately insensitive, with the sensitivity varying based on the pollutant under
question (Levy et al. 2002).

According to the model creator (Joe Scire, personal

communication, 2005) sensitivity tests should focus on influential CALMET
parameters (e.g. R1; R2; Rmaxl; Rmax2; Terrad, and; Bias). See Chapter 5 for a
complete description of these CALMET parameters that largely determine the radius
of influence of terrain and surface or upper air meteorological observations on the
final wind fields used in CALPUFF (Scire and Godfrey 2005).
36

Levy and Spengler (2002) present how CALPUFF can be coupled with
epidemiological evidence regarding the health benefits of reduced emission
strategies. Levy and Spengler (2002) found that emission controls on two power
plants in Illinois lead to a reduction of approximately 70 premature deaths per year.
This reduction is associated with an overall 2% reduction in ambient secondary PM
concentrations. The calculations of mortality are based on a concentration-response
estimation derived from several epidemiological studies conducted by the American
Cancer Society and Adventist Health among others.

This type of estimation is

beyond the scope of the present study, but it does highlight application of CALPUFF
to emission reduction strategies.
It can be seen from the variety of studies presented that CALPUFF is diverse
in its application.

However, the repeated assessment of CALPUFF in similar

applications is needed to test its performance against numerous data sets. Using
CALPUFF in Golden will provide another evaluation of its performance in complex
terrain.

37

3

MODELLING SYSTEM AND DATA COLLECTION

3.1

CALPUFF MODELING SYSTEM
The CALPUFF modeling system consists of three main programs and

accompanying pre- and post-processors.

The first component is the CALMET

meteorological model; the second, the CALPUFF dispersion model; and the third,
the CALPOST post-processor.

There are also a number of pre-processors that

prepare and merge a wide variety of data formats into CALMET input data. The
modeling system is contained within a newly developed graphical user interface
called CALPUFF PRO 5, but can also be used by editing the input files in a text
editor. CALMET, CALPUFF, and CALPOST are briefly described below.

3.1.1 CALMET
CALMET consists of

a diagnostic wind field module and a micro-

meteorological module for overland boundary layers.
The CALMET diagnostic, 'mass consistent' modeling approach constructs
gridded wind fields that are consistent with available terrain, land use and
meteorological data, while also satisfying the governing equation for conservation of
mass (Cox et al. 2005).

Mass consistent wind field models represent a good

compromise between accuracy and computational efficiency (Arena et al. 1997).
Difficulties arise when using the mass consistent interpolation of three dimensional
wind fields.

The wind fields must be adjusted by minimizing divergence in the

horizontal and vertical components.

However, the horizontal wind velocity

component is much greater than the vertical velocity component. In CALMET, then,
it is necessary to define a fixed vertical velocity field based on the input surface
38

characteristics (micro-meteorological module in CALMET) and adjust the horizontal
wind components to minimize divergence in each grid cell (wind field module in
CALMET).
A schematic overview of how CALMET determines the wind fields is provided
in Figure 3-1. In the first step, either observational data or prognostic model outputs
(e.g. Fifth-Generation NCAR / Penn State Mesoscale Model (MM5) or the National
Oceanic

and

Atmospheric

Administrations's

(NOAA)

National

Center

for

Environmental Prediction Rapid Update Cycle (RUC) Model) are used to create an
initial guess field in CALMET. At the edge of the domain the model applies an initial
guess wind field.

There is an option to include a surface station outside of the

modeling domain which the model will consider in setting boundary conditions. In
Golden, there are no nearby surface stations that would be representative of the
edge conditions, so no observed data was used to initialize the edge of the domain.
The model then uses this initial-guess wind field and adjusts it for kinematic effects
of terrain, slope flows, and terrain blocking to produce a Step 1 wind field. The Step
1 wind field is then used in an objective analysis procedure that utilizes
observational surface and upper air data to produce the final wind field to be used in
CALPUFF.
In Step 1, the initial-guess field is used along with the geophysical data to
compute the terrain steering effects. Thermodynamic blocking effects of terrain on
the wind flow are parameterized in terms of the local Froude number (Scire et al
2000.) Slope flows are then computed based on the shooting flow parameterization
of Mahrt (1982).

The slope flow is characterized in terms of the terrain slope,
39

distance to the crest and local sensible heat flux. The final output of CALMET's Step
1 wind field is a gridded field of 3-D wind components.

Observations

Setup initial guess
field

prognostic model output
used an "initial guess
field"

Compute terrain
steering effects
Minimize divergence

Step 1 wind field
created

Observations

Perform objective
analysis procedure

Step 2 final wind
field created

Figure 3-1

A schematic overview of the creation of wind fields in CALMET (adapted from
Scire et al. 2000).

The second step is then performed that introduces observational data into the
Step 1 wind field through an objective analysis procedure.

An inverse-distance

squared interpolation scheme is used which weighs observational data more heavily
in the vicinity of the observational station, while the Step 1 wind fields are weighted
heavily in domain areas lacking observational data. The user has control over the
radius of influence of observational data. Choosing the radius of influence correctly
allows an appropriate balance of observational data and the Step 1 wind fields

40

computed by CALMET to determine the final wind fields. The final wind fields are
used as the dispersal winds in the CALPUFF dispersion model.
The diagnostic nature of the CALMET model allows for minimal data
requirements based on routinely collected surface and upper air meteorological
observations. Although this allows easier application of the model, the hourly data
lacks sufficient horizontal, vertical and temporal resolution to accurately depict
atmospheric processes over the entire modeling domain.

In areas where data

resolution or data completeness is lacking, CALMET must rely heavily on
interpolation techniques to estimate wind, temperature and turbulence fields. This
inherently reduces the overall accuracy of the modelled meteorology and eventually
the dispersion of the pollutants in CALPUFF.
CALMET's micro-meteorological module calculates surface friction velocity,
convective velocity scale, Monin-Obukhov length, mixing height, Pasquill Gifford
Turner (PGT) stability class, air temperature, and precipitation rate for each grid by
using the overland boundary layer energy balance method of Holtslag and van
Ulden (1983). This module describes the convective and mechanical turbulence in
the boundary layer which ultimately determines the vertical extent of dispersion. The
micro-meteorological module relies on both meteorological observations and
assumptions of geophysical parameters based on land use type.

3.1.2 CALPUFF
CALPUFF is a puff dispersion model capable of handling multiple-layers
within the atmosphere and multiple pollutant species (Scire et al. 2000). The

41

CALPUFF model involves more complicated and comprehensive simulation
processes compared to the conventional steady-state, single-layer and singlespecies guassian models. The non-steady state model allows it to handle time and
space varying meteorological and emission conditions. Expanded capabilities
include non steady-state effects (spatial heterogeneity, causality, fumigation, etc.),
complex terrain algorithms, calm and low wind speed conditions, flexible source
variability

options,

chemical

transformation,

and

differential

advection

and

dispersion.
CALPUFF uses CALMET's final wind field to disperse point, line and area
emission sources throughout the gridded modeling domain. The puff model
represents a continuous emission source, such as a plume, as a number of discrete
packets of pollutant material. The puffs are released and evolve in size according to
Gaussian- like diffusion and dispersion, but the multiple puffs allow changes in
meteorological and emission conditions to be captured.

3.1.3 CALPOST
CALPOST includes a number of useful features to process the data output by
CALPUFF.

Users define the averaging period and CALPOST processes the

concentration of PM (or other pollutants modeled) at receptor sites. Data can be
processed as gridded fields for mapping emissions or as data values for time series
and statistical analysis.

3.2

DATA COLLECTION IN GOLDEN
Most of the observed data were collected by the BC MOE. The BC MOE had 3

monitoring locations in Golden:

the golf course north of the city (GOLF); the
42

Canadian Pacific Railway (CPR) site south of the city; and a downtown location near
the hospital (TOWN) (Figure 2-3). At each of the locations, the Ministry collected
hourly meteorological data. In addition each location continuously monitored PM10
and PM2.5 levels.

Meteorological data were also collected at the Golden Airport

(AIRPORT) by Environment Canada (Figure 2-3).
Additional equipment was added to supplement the data collection efforts of
the BC MOE and Environment Canada. A SCINTEC FAS64 phased-array Doppler
sodar system (SODAR) was operated at the golf course location (Figure 3-2). The
SODAR system is capable of resolving the vertical wind profile by measuring the
scattering of sound waves by atmospheric turbulence. The wind speed and direction
is found by the Doppler effect. The SODAR transmits and receives beams of sound
in different directions. By monitoring the change in frequency of the back-scattered
sound, the wind speed in the direction of the beam can be found. Wind moving
toward the antenna will decrease the wavelength of the reflected sound wave and
wind moving away from the antenna will increase the wavelength of the reflected
sound wave. The SODAR monitors this change in frequency (wavelength) of the
reflected sound wave and calculates the wind speed in the particular direction of the
transmitted sound. By transmitting beams in different direction the SODAR resolves
the three-dimensional wind fields.
The SODAR system must be installed in an area clear of obstructions (i.e. tall
buildings and trees)

in the immediate area that would result in scattering of the

sound waves. The system was installed beside the GOLF station that was free of

43

large obstructions arid had the available operational power requirements (Figure 3-

2).

Figure 3-2

SCINTEC FAS64 phased-array Dopiar sodar system installed near the GOLF
station.

Obtaining a local vertical temperature profile in the lowest portion of the
atmosphere was important as temperature lapse rates calculated from upper air
stations hundreds of kilometres away would not allow for accurate prediction of
mixing heights and stability classes. In order to capture a vertical temperature
profile, HOBO H8 Pro temperature / RH loggers (HOBOs) were enclosed in a
radiation shield and mounted on trees 1.5 metres above the ground. Six HOBOs
were installed at elevations of 798 metres (1), 839 metres (2), 960 metres (3), 1036
metres (4), 1081 metres (5), and 1223 metres (6) (Figure 3-3).

Locations were
44

selected on the mountain side to the west of Golden at safely accessible locations
along the road to the Kickinghorse Mountain Ski Resort.

5687 ;

5686

5685

2 5683

5682

5681

5680 MmXMv

498

499

500

501

502

503

504

505

506

507

508

UTM Easting (km)
Figure 3-3

Map of the six HOBO H8 Pro temperature / RH loggers (HOBOs) at 798 metres
(1), 839 metres (2), 960 metres (3), 1036 metres (4), 1081 metres (5), and 1223
metres (6) along the road to the Kickinghorse Mountain Ski Resort to the west
of Golden.

Methods regarding the use of the collected air quality and meteorological data
are provided in Chapter 5.

45

4

EPISODE ANALYSIS
This modeling study focused on episodes of high PM. Generally, episodes are

defined as prolonged periods of elevated PM ambient concentrations above a
defined threshold level.

In British Columbia episodes are typically defined as

periods with 24 hr PM10 averages above 50 |jg nf3, the objective for BC (Table 2-1),
for at least 3 days.
During the 2005-2006 winter, PM concentrations were not as high compared to
previous Golden winters. Therefore, episodes were chosen based on increases in
PM10 and PM25 concentrations compared to the 2005-2006 winter season mean
data, and from inferences made regarding the ratio of PM2.5 to PM10 (PM ratio). A
low PM ratio indicates that most of the PM10 measured is larger than 2.5 microns in
diameter. Therefore the emission sources creating the PM are more likely to be
crustal or fugitive dust sources that create larger particles. A high PM ratio indicates
that most of the PM10 measured is smaller than 2.5 microns in diameter.

The

emissions sources creating the PM in this high PM ratio period are more likely to be
combustion sources that create finer particles.
An episode reasoned to be representative of a typical "wintertime" episode was
chosen (December 7, 2005 - December 14, 2005. A second episode reasoned to
be representative of a typical "spring-time" episode was chosen as well (February 8
- 23, 2005). Modeling an episode of each type allowed for comparison between the
emission and meteorological characteristics associated with

each episode.

46

Episodes were chosen to correspond to periods when reliable data were available
from all meteorological stations, the SODAR and the PM monitors.

4.1

EPISODES MODELED
This section summarizes PM concentrations during each episode modeled and

the rationale behind choosing these specific episodes.

4.1.1 Episode 1 (December 7-14, 2005)
This period exhibits higher concentrations of PM in comparison to the rest of
the winter monitoring season.

There are several factors that contribute to this

episode being labeled as typical of wintertime.

The increased PMi0 levels are

associated with an increase in the PM2.5 fraction and high PM Ratio (Table 4-1). It
was hypothesized that high PM levels during this episode were due to combustion
sources (i.e. woodstoves) resulting in the PM2.5 increase.

Spatially, the highest

levels of PM10 and PM2.5 occur at the TOWN location, followed by CPR and the
GOLF location. This supports the hypothesis that residential heating concentrated in
the downtown core would be a significant contributor to episode 1. Temperatures
were low during the episode, average daily minimum temperature of -9.7°C
compared to normal daily minimum of -6.1 °C in the rest of December, so many
residents would likely be using wood heat appliances during this episode. Figure 41 shows a consistent diurnal pattern of high PM concentrations during the late
afternoon and into the evening at the TOWN station.

47

Table 4-1

Comparison of episode 1 average hourly PM concentrations and PM ratios
(PM2.5:PM10) with hourly PM concentrations and ratios from the Winter season
(2005 - 2006).

PM25 (M9/m3)

PM10 (pg/m3)
Episode 1 Winter

Monitor
CPR

Average
Maximum
Minimum
Std Dev
Average
Maximum
Minimum
Std Dev
Average
Maximum
Minimum
Std Dev

TOWN

Golf

16.0
37.0
3.0
7.9
20.0
78.0
0.0
11.2
11.7
24.0
4.0
4.2

Episode 1

12.3
414.6
0.0
15.0
17.1
293.5
0.0
22.8
7.6
43.4
0.0
4.2

10.3
27.0
1.0
6.1
13.3
64.0
0.0
10.1
7.0
18.0
0.0
3.6

PM Ratio
Episode 1 Winter

Winter
5.4
32.0
0.0
5.1
7.4
64.0
0.0
7.8
2.8
33.1
0.0
3.0

0.6
0.8
0.1
0.1
0.7
1.0
0.0
0.2
0.6
0.9
0.0
0.2

0.4
1.0
0.0
0.2
0.5
1.0
0.0
0.3
0.3
1.0
0.0
0.2

90

80

PM10
70

PM2.5

60

50

40

30

20

10

0
o'O

Figure 4-1

©

D
©

o-

v>

Episode 1 profile of observed PM10 concentrations (Mg/m3) and observed PM2.s
concentrations (pg/m3) at the TOWN station.

48

4.1.2 Episode 2 (February 8-23, 2005)
This period also exhibits higher levels of PM in comparison with the rest of the
monitoring period. This episode is labelled as a springtime episode because it
occurs in February and has a lower PM ratio in comparison to both episode 1 and
the rest of the monitoring period (Table 4-2).

It was hypothesized that high PM

levels during this episode were due to emission of crustal materials (i.e. road dust,
fugitive dust emissions) related to the spring thaw. However, temperatures remained
low, averaging -8.CTC, during this period as well, so it is hypothesized that there will
still be a significant residential space-heating contribution to the emissions at the
modeled receptors.

An episode closer to the spring season would have been

preferable, however data were not available. Figure 4-2 shows the same consistent
diurnal pattern of high PM concentrations during the late afternoon and into the
evening as was exhibited in Episode 1.
Table 4-2

Comparison of Episode 2 average hourly PM concentrations and PM ratios
(PM25:PM10) with hourly PM concentrations and ratios from the Winter season
(2005 - 2006).
PM10 (ug/m3)
Episode 2 Winter

Monitor
CPR

TOWN

Golf

Average
Maximum
Minimum
Std Dev
Average
Maximum
Minimum
Std Dev
Average
Maximum
Minimum
Std Dev

26.6
122.9
1.0
18.5
39.2
180.5
0.0
28.7
7.3
38.6
0.0
4.3

12.3
414.6
0.0
15.0
17.1
293.5
0.0
22.8
7.6
43.4
0.0
4.2

PM2.5 (Mg/m3)
Episode 2 Winter
7.3
24.3
0.0
5.7
10.1
49.3
0.0
8.3
1.7
10.9
0.0
2.2

5.4
32.0
0.0
5.1
7.4
64.0
0.0
7.8
2.8
33.1
0.0
3.0

PM Ratio
Episode 2 Winter
0.3
0.7
0.0
0.2
0.3
1.0
0.0
0.2
0.2
0.6
0.0
0.2

0.4
1.0
0.0
0.2
0.5
1.0
0.0
0.3
0.3
1.0
0.0
0.2

49

200

PM10 :
PM2.5

Figure 4-2

Episode 2 profile of observed PM10 concentrations (|jg/m3) and observed PM2.5
concentrations (ng/m3) at the TOWN station.

4.2 COMPARISON OF WIND CHARACTERISTICS DURING EPISODIC AND NON-EPISODIC
PERIODS
Prior to modeling, an analysis of the observational data available during the
episodes was performed. Episodes usually persist when the air is stagnant. The
dispersal and emission processes contributing to the two episodes of higher PM
were investigated by comparing available meteorological data during the episodes
with data typical of Golden on an annual, seasonal and monthly basis.

This

comparison helped to create hypotheses about the wind conditions contributing to
the particular episodes under investigation. These hypotheses will be tested further
by modeling the episodes in CALPUFF.

50

4.2.1 Wind Speed
PM is dispersed and diluted by wind. Low wind speeds will result in higher
concentrations of pollutant as they are emitted into smaller volumes of air. Higher
wind speeds help to mix and dilute the pollutant into larger volumes of air thus
lowering the concentration.

Typically, episodes occur during periods of stable

atmospheric conditions when low wind speeds are common.
Figure 4-3 shows that both episodes occur during periods with a high
frequency of calm (<0.5 m/s) and low wind speed when compared on an annual,
seasonal and monthly basis.

Episode 2 has a greater frequency of calm winds

compared with episode 1. Each episode shows wind speeds that are consistenly
low with very little diurnal variation (Figure 4-4).

• Annual
• Winter
• December
• February
• Episode 1
• Episode 2

> 40

Calm

0.5-1

1 -1.5

2-3

3-4

4-6

>6

Wind Speed Class (m/s)
Figure 4-3

Frequency distribution of wind speed classes (m/s) for annual, winter season,
the month of December, and the month of February with those observed during
episode 1 and episode 2 at the Town station. Calm winds are those with a wind
speed less than 0.5 m/s. Distributions are calculated from data ranging from
2001-2005. The Town station is used because it has been operational for the
longest period.

51

1.6

1.4
|1.2

I1

• Episode 1

CO 0.8

•o
£ 0.6

Episode 2

§ 0.4
S
0.2 -

0I
0

5

,

r

,

10

15

20

2

Hour of Day

Figure 4-4

Diurnal variation of wind speed by hour of day during each episode modelled
at the Town Station. Other stations showed similar trends.

Figures 4-5 and 4-6 demonstrate that higher concentrations of both PM10 and
PM2.5 occur during hours with lower mean wind speeds. At all three stations wind
speed is negatively correlated to PM concentrations for both episode 1 and 2. This
trend is not as evident at the GOLF station as the other two stations. This may be
because the GOLF station is further away from the emission sources in Golden or
that at higher wind speeds a background PM source (i.e. windblown fugitive dust)
affects the GOLF station. At the other stations, periods of low wind speeds during
both episodes lead to the periods of highest PM concentration.

52

(a)
90
<CO 80

• PM10
• PM2.5

E

o> 70
3
c
o
2 60

o 50
o

••

c
o
0
40

1
E

30

« 20
10

» I•

1
1.5

0.5

•
2

m

1

,
2.5

•
3.5

Wind Speed (m/s)

(b)

40

<to
o>

35

• PM10
• PM2.5

:: •
•

JL

C 30

o

*.•»

E 25
c
0)
o
c 20

**

»» •
*••
•• ••

o
o

••

• • • »
•

15

».»»•...
10

t:

8.

•*

•

••••
• ••*

turn

5
•

0 4
0

• •

• •• §i£+i •
• f••i •t •l • + i• • i
••

•

0.5

•

1.5

:
• • •
•

•

2

•
•

•

•

• •
• •

•

•••

•

• i *
•

2.5

3.5

Wind Speed (m/s)

53

(C)
30.00

r• PM10

to
<
E 25.00

PM2.5

^3>

.2 20.00
•

2
C

•

s

•

o
o

•
••

••

c 15.00

»

k.

4)

rto
E 10.00
0)
ra
3
O
5.00
r
<0

••
•*
•»

»»
»»

•
••

*

Q.

•
•
•
•• • •
•
•
•
••
• •• • •
•
•• •
•• •
• »• • •
•
• •* •• • • •
« • •• • •
•
••• » » » •
» *• • • • •
• *
» • » • » •• • • •
•» » • • • • * • m • •
•• » • • • • • • • •
• ••
•• * • •
» •• • * * • • •
••• •• •• ••
•• •
• •
•••

•
•

•

•
•

•

•••

• ••
••
•
»
•• •
•
••

•

•
•

• a

•
•

•

•

0.00

0.5

1.5

2

2.5

3.5

Wind Speed (m/s)
Figure 4-5

Hourly PM concentrations and wind speed for Episode 1 at (a) Town station; (b)
CPR station; (c) Golf station.

(a)
200

« 180
E
"9) 160
a

• PM10
• PM2.5

|140
|120

8

••

g 100

0
1 80

•

•

•*•

*

iE

**:
1.5

.•
2

2.5

3.5

Wind Speed (m/s)

54

(b)

• PM10
• PM2.5

•

•

>

•
••

•

.t ifs *

I'**
llll
0

0.5

1

1.5

2

2.5

3.5

4.5

Wind Speed (m/s)
(C)
45
CO
<

40

• PM10

"9>
3 35 -I

• PM2.5

E
c

o

30

8 25
S

0 20
1

E 15

1
3 10
U
1
(S
a 5
0

:
* : M * •
t

iiliiiilii i I i i i : i
= I •

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

Wind Speed (m/s)
Figure 4-6

Hourly PM concentrations and wind speed for Episode 2 at (a) Town station; (b)
CPR station; (c) Golf station.

55

4.2.2 Wind Direction
Variation in wind direction will disperse emissions in different directions and
influence source-specific PM concentrations at the ambient monitors. For example,
a north wind would likely increase Louisiana Pacific's (Figure 2-3) contribution to PM
concentrations at the Town and CPR monitors, while decreasing its contribution at
the Golf monitor. Southerly winds would likely result in the opposite.
Figures 4-7 and 4-8 show that high concentrations of PM are observed during
both the southerly and northerly valley wind flow. An exception occurs at the Town
station during the second episode where higher PM10 concentrations are exhibited
with winds from the North.

(a)

? 80

• PM10

•

E

• PM2.5

•

§ 50

•

c

•
•

0

45

90

135

180

225

270

315

360

Wind direction (degrees)

56

(b)

40
S

• PM10
• PM2.5

*

•S 30
|
S 25
c
s

•
•

• •

• •

§ 20

u
t
*
<B
•
•
M
•
S 15 i4» •
E
0)
a 10 - • • 3
O
t
c
« 3
a
b.

45

••
••• ••

•• • • •
r •
•

•

x•

* j«?.

• %I . W t , ? •
t. •«* . • % .
I I I #
^ 1 4
.
!- *.
*
/ . WW-/.-.
.
•
• •
m mm m m
m

90

135

180

225

•

* mmJm %
.
+ I + +
+
' .—J* .

v
rt.—.W
m
MI

m

270

m

m

315

m

360

Wind direction (degrees)
(C)

30

• PM10
• PM2.5

CO
S 25
O)
c
0

1

20

••
•
•
•
S 15
•

1

•» ••

10

•

*

a
"5
0
1
a
a.

45

90

135

180

225

270

315

360

Wind direction (degrees)
Figure 4-7

Hourly PM concentrations and wind direction for Episode 1 at (a) Town station;
(b) CPR station; (c) Golf station.

57

(a)

200 -i

• PM10
• PM2.5

to 180
<

E

l3i 160
3
c 140
o
'1 120 -I
c

8

O 100
0

1
E

3
O
f(0

D.

•
••

80

•
%•

60

•

40 20

0

•

I

•

/
•

••

J

% • 4
>V»H

!

%
0

45

90

135

180

225

270

315

360

Wind direction (degrees)

(b)
• PM10
PM2.5

135

180

225

360

Wind direction (m/s)

58

(C)

• PM10
• PM2.5

•

•

••

•

.*• ;

yAOt

'

••

•
90

135

•
•
180

<» *
f' -a

/

B-

225

•Mf
• •_ • d" .
270

315

360

Wind direction (degrees)

Figure 4-8

Hourly PM concentrations and wind direction for Episode 2 at (a) Town station;
(b) CPR station; (c) Golf station.

4.2.3 Wind Rose
A wind rose shows the distribution of wind speed frequency by wind direction.
All

wind

roses

(Figure

4-9)

show

the

terrain-forced

flows

along

the

Northwest/Southeast valley alignment. The annual flow shows a higher percentage
of winds from the Northwest (Figure 4-9a). More southeasterly winds occur as during
the winter (Figure 4-9b). Episode 1 (Figure 4-9c), shows the strongest influence of
southeasterly winds when compared to the other wind roses. This is different than
the winds displayed for the annual and winter when winds from the northwest are
more prominent. Episode 2 shows more typical flows with more northwesterly winds
(Figure 4-9d).

59

(a)

• >7.0 m/s
• 5.0 - 7.0 mfe
• 3.0 - 5.0 m/8
• 1.5 - 3.0 m/s
• 0.5 -1.5 m/6
• 0.2 - 0.5 m/s

Calm (<=0.2 m/s)

• >7.0 m/8

13.0 -5.0 m/s
•1.5-3.0 m/s
• 0.5 -1.5 m/s
•0.2-0.5 m/s
Calm (<=0.2 m/s)

Calm (<=0.2 m/s)

N

NW.-

•>7.0 m/s
•5.0-7.0 m/s

•NE

•3.0 - 5.0 m/s
•1.5 - 3.0 m/s

Wi

•0.5-1.5 m/s
•0.2-0.5 m/s

Calm (<=0.2 m/s) =

2.7%

SW .

Figure 4-9

Wind rose for Golden including (a) all winds recorded from January 2001 through
December 2005; (b) all winds recorded in the winter season (November - February) from
2001-2005; (c) all winds recorded during Episode 1; (d) all wind recorded during
Episode 2. The wind rose displays the distribution of wind speed frequency by wind
direction. The dashed circles represent the percentage winds in each direction and the
color scale corresponds to the wind speed. Each arm of the wind rose measures the
incremental contribution of each wind speed class to the percentage of winds in that
direction.

61

4.3

EPISODE ANALYSIS DISCUSSION
The two episodes modeled using CALPUFF exhibit periods of high PM10 and

PM2.5 concentrations. During the episodes, pollutant concentration will be estimated

using emission rates, wind characteristics and atmospheric conditions such as,
stability, mixing heights and vertical temperature profile. Before analyzing the model
output it is helpful to decipher any trends from the monitored data available in
Golden.
The episodes occur during times of calm to light wind conditions in Golden.
The synoptic, terrain-forced winds are weak and high surface pressures are
observed during both episodes. Wind speeds appear to be an indicator of a potential
rise in PM concentrations. During both episodes, overall mean wind speed is lower
and periods of higher particulate concentrations are generally coupled with low
hourly wind speed. High PM concentrations are observed primarily during typical
valley aligned wind directions from the northwest and southeast.
When large-scale synoptic winds are weak, pollutants can become trapped in
the local circulation and accumulate over time (Arya 1999). During these periods the
local diurnal mountain winds largely determine the dispersion of pollutants in the
valley. Indeed, the diurnal patterns exhibited during both episodes (Figures 4-1 and
4-2) would indicate this association. The dampened effects of the daytime diurnal
circulation during the winter episodes studied can result in even lower wind speeds
than periods during the summer and can contribute to the build-up of pollutants in
the valley.

62

During both episodes, the higher PM concentrations were recorded in the late
afternoon and evening.

This likely occurred because of a combination of

meteorological and emissions related processes. Downslope flows during this part
of the day can bring emissions toward the valley floor, where PM monitors in Golden
are located. In addition, theory indicates along-valley flow is typically slowing and
reversing at this time (Whiteman 2000). However, it could also be the case at low
wind speeds that the reversal in wind direction may not have much of an effect and
dispersion is mostly hindered by the further decrease in wind speed typical of these
hours of the day. This is supported by the diurnal variation exhibited during the
episodes that indicates lower wind speeds in the afternoon and into the evening
(Figure 4-4). Emissions also change during the late afternoon and early evening,
Emissions associated with traffic increases as residents drive home from work or
school and space heating emissions increase as residents burn wood, oil or gas to
heat their homes upon arrival.
Wind from the northwest and southeast resulted in nearly all occurrences of
high PM concentrations (Figure 4-7 and 4-8).

Winds from these directions are

expected given the terrain-forced flows along the valley alignment. A trend cannot
be discerned that shows higher PM concentrations from either the northerly or
southerly flow.

The wind roses (Figure 4-9) revealed that winds in Golden are

channeled over the main valley corridor. The modelled winds may show that the
diurnal variation of weak winds along the valley could mean that winds that disperse
pollutants down-valley at night are simply returned by the reversal of winds in the
morning leading to the build-up of pollutants in the Town (Whiteman 2000).
63

In addition to low horizontal wind speeds, episodes typically occur during
times of low mixing height, low convective activity, inversions and atmospheric
stability (Malek et al. 2006).

These atmospheric conditions are associated with

capping and suppression of vertical motion in the boundary layer, reducing the
dispersion of pollutants (Arya 1999).

Since direct measurements of these

atmospheric characteristics are not available in most areas, the results of validated
CALMET modeling will allow for the investigation of the CALMET predictions of
these parameters during the episodes.
Modeling the episodes in CALPUFF will now allow for further analysis using
the CALMET predicted wind fields and atmospheric parameters causing the
episodes. CALMET will calculate the winds and other meteorological fields by using
the observed winds in Golden as a basis.

Ultimately, CALPUFF will use the

predicted winds, calculated mixing heights, stability and other atmospheric
parameters to predict PM concentrations in Golden.

Following the modeling,

stability, mixing height and vertical wind velocity will be available as predicted model
variables. This will allow for the isolation of these parameters in analysis of each
episode's dispersal characteristics.

64

5

CALMET
This chapter describes the setup of the model and evaluation of the

meteorological processor CALMET during the two PM episodes outlined (Section
4.1). Winds and other meteorological parameters output by CALMET will then be
used in the dispersion portion of the modeling system, CALPUFF.

5.1

SETUP OF CALMET
CALMET used digital terrain and land use data along with data from four

surface stations and a hybrid upper air station combining data from locally operated
SODAR and more distant radiosonde measurements. The setup of the model is
described here.

5.1.1 Modeling Domain
An air dispersion modeling domain must include all relevant emission sources
and encompass an area large enough to allow the meteorological patterns of the
area to develop.

The CALPUFF modeling domain chosen for this study

encompasses a 35 by 40 kilometer area centered at the Town of Golden (Figure 51). The boundaries include the entire east-west extent of the Columbia River Valley
to allow CALMET to establish the main valley flow as well as any tributary flows from
the Kicking Horse and Blaebarry Passes.

Inclusion of the Purcell and Rocky

Mountain Ranges to the west and east respectively allowed for the incorporation of
steering effects from the complex terrain surrounding the Town of Golden (Figure 51).

As described earlier, these mountain ranges act as barrier to dispersion of

pollutants out of the valley and influence the dominant wind pattern in Golden,
providing the basis for the airshed extent described in Figure 1-2.

The modelling
65

domain encompasses all of the sources deemed to be significant in adding to the
PM loading in the Town of Golden with enough additional terrain modeled to account
for the meteorological influences affecting dispersion in the airshed. The modelling
domain also includes all significant receptors, namely the populated areas in the
Golden airshed.

0

Figure 5-1

5

10
15
Kilometres

20

Map showing the approximate modeling domain extent around the Town of
Golden.

A larger modeling domain would make it difficult to choose appropriate
distances for critical CALPUFF parameters such as R1, Rmax, and Terrad as these
are held constant across the modeling domain.

For example, effects of terrain
66

outside of the main valley may become too influential if included in the modeling
domain because Terrad, the parameter that controls the terrain effects on wind
fields, is set uniformly across the domain. In addition, the lack of accurate emission
estimates and meteorological stations from the surrounding areas of Golden would
likely reduce the accuracy of the model if the domain were expanded further.
Computer processing time would also increase.
To properly characterize the terrain, the model was run with a grid spacing of
0.25 km.

The same grid resolution was used by Scire and Robe (2000) in an

application of CALPUFF in the Columbia River Valley. CALMET determines a wind
vector for each hour in each grid cell.

In addition to the 250 metre-spacing of

receptors in the modeling domain, discrete receptors were located at the ambient
monitoring locations (CPR, TOWN, and GOLF) in Golden (Figure 2-3). A CALPUFF
receptor is placed in the centre of each grid cell. Each receptor represents a point at
which CALPUFF calculates the concentration of PM. In total 22,343 gridded and
discrete receptors are located in the modeling domain. The gridded receptors allow
for visual display of the wind fields and dispersion results on a 3-D map created in
Surfer™, the mapping program associated with CALPUFF. The statistical evaluation
will focus on the three discrete receptors where continuous meteorological and PM
data were collected by the BCMOE during the field study.

By only looking at

episodes lasting several days, the computer run time for this resolution is
appropriate. Grid spacing less than this may not be appropriate considering the
resolution of the terrain and land use data sets that were available.

67

CALMET also requires the vertical extent of the modeling domain to be
defined. These are defined in layers. Ten vertical layers were defined. A more
detailed description of the vertical layers is provided in Section 5.1,3c.

5.1.2 Geophysical and Land Use Data
CALMET relies on geophysical data and meteorological data to compute wind
fields and dispersion characteristics. Geophysical data entered in CALMET are
essential in determining the final wind fields through the internal model calculation of
the terrain steering effects. Geophysical data includes terrain elevations, land use
categories, surface roughness length (optional), albedo (optional), bowen ratio
(optional), soil heat flux (optional), anthropogenic heat flux (optional), and leaf area
index (optional). In CALMET this information is used to determine the components
of the overland surface energy budget used to conserve the energy throughout the
modeling system and steer the initial wind fields (Scire et al. 2000). For this study,
all of the optional data listed above were estimated by CALMET using the required
data of terrain elevations and land use categories. CALMET estimates the optional
parameters using a series of algorithms and default values assigned for each of the
land use categories.

The file produced is a CALMET data file including gridded

elevations, land use categories, and the optional surface parameters mentioned
above.
Terrain Data were downloaded from the EarthTech, Inc. website, via the
United States Geological Survey (USGS) in the form of Shuttle Radar Topography
Mission data (SRTM3) (http://www.src.com/datasets/SRTM_lnfo_Page). These data
have a resolution of approximately 90 metres in horizontal extent and 10 metres in
68

the vertical extent because they are sampled at 3 arc-seconds. 3 arc-seconds at the
equator corresponds roughly to 90 meters in horizontal extent. In the downloadable
format, SRTM3 data can be processed using a pre-processor called Terrel included
in the CALPUFF PRO 5 graphical user interface.

Terrel transforms the terrain

SRTM3 data into a useable gridded terrain field format for use in CALMET (Scire et
al. 2000).
Land use/land cover data were also available from the Earthtech, Inc.
website. Here, the Global Land Cover Characterization (GLCC) Database can be
accessed (http://www.src.com/datasets/GLCC_lnfo_Page.html).

Unfortunately, the

resolution provided by the GLCC database was not sufficient for the modeling in
Golden.

More detailed landuse data were obtained from the British Columbia

landuse database (resolution = 250m) and converted into CALMET landuse codes
using the conversion recommended in the BC Air Dispersion Modelling Guidelines
(2008) (Table 5-1). The converted land use categories are mapped in Figure 5-2.
The geophysical pre-processor for CALMET, MAKEGEO, computes area weighted
values for each model grid cell based on the amount of area each land use category
covers in the grid cell. MAKEGEO then assigns an arithmetic average of the default
geophysical parameters associated with each land use category (Table 5-2).
CALMET defines a land use category for perennial snow and ice, but does not allow
for recognition of seasonal changes in snow cover. Land use categories were not
changed to reflect that some areas of the Columbia River Valley are snow covered
during the two episodes modelled. Certainly this is a limitation of the methods used
to characterize the land use categories, as changes in the snow cover would result
69

in a change of the default geophysical parameters (Table 5-2) assigned to each grid
cell and would affect the calculated calculated overland energy budget model in
CALMET.
Table 5-1
B.C. Land
Class Code

Conversion of B.C. Land Use Codes to CALMET Codes
B.C. Land Use Category

CALMET Code

CALMET Land Use Category

*>

Agriculture

-20

Agricultural Land -Irrigated

3

Barren Surfaces

70

Barren Land

4

Fresh Water

50

Water

5

Muung

70

Barren Land

6

Old Forest

40

Forest Land

7

Recently Logged

30

Rangeland

8

Recreational Activities

40

Forest Land

9

ResidentialAgriculture
Mixtures

20

Agricultural Land

10

Selectively Logged

40

Forest Land

11

Urban

10

Urban

12

Wetlands

60

Wetland

13

Young Forest

40

Forest Land

Source: BC Air Dispersion Modelling Guidelines (2008)

The land use data along with the SRTM3 data from Terrel were combined in
the geophysical input data file for CALMET called MAKEGEO.dat. This file includes
elevations and fractional land use cover for each grid in the modeling domain.

70

Table 5-2

Default CALMET Land Use Categories and Associated Geophysical Parameters
Sarf*c«

I'A TVM

D^r.Dncc

RcuzfaMtt >'nj;

Aibtdo

:o

U:tao or 3'xit up Land

10

0 IS

:o

Ajr.c-irual Laiud - L'mu:»J

o;<

0 15

-:c

Ajr.culruai Laad •

o:<

0 15

Pjasclaad

oc<

i0

Focevt Liod

l0

o i:

?:

>cuLI \ V m« Body

: oo:

Bay* atd Esrur.**

:or.

o i:
c i:

targ* Ware Body

C CO'.

o t;
o i:

55
60

'•Vetbad

10

a:

Foctsted We: jod

]0

62

>.*coforaited Wttiitd

CI

•0

BrrecLard

0C5

0 3-

SO

T'.tadra

10

0 3C

P0

P«rcss^l Soon- oc lc«

20

0 *0

SCUTA RJOO

Soil K*ar

Txx Piling

Acthropcgexc
H»ar Flux <Wan

o;

0:
o:
o:
10
10
10

o:
o:

o:
o:
o:

o:

O) 5685000
C

£
S
Z

5680000

5675000-

5670000-

5665000-

Figure 5-2

0

* 0

:o

:o
:o
:o
:o

io
o :<

Land Use

E

}

oc
o:
o:

5700000

5690000

iO

o:

Source: CALMET User Guide (Scire 2008)

5695000-

Leaf A:«j

:o
:o

5.1.3 Meteorological Modeling Data
Along with the geophysical data, CALMET requires meteorological data from
several sources to generate the wind fields used to drive dispersion in CALPUFF.
CALMET utilizes a number of pre-processors to incorporate surface meteorological
data and upper air data.
5.1.3a

Surface Station Data

CALMET requires, at a minimum, surface observations of wind speed, wind
direction, temperature, cloud cover, ceiling height, surface pressure, precipitation
rates and type, and relative humidity from at least one of the four surface stations
within the modeling domain for every hour modeled.
The four surface meteorological stations used in the model were located at
CPR south of the Town, Lady Grey School in downtown Golden, the airport, and the
Golden Golf Course north of the Town (Figure 2-3). Data for each of the parameters
were available from at least one of the four surface stations for each hour of both
episodes (Table 5-3). Gaps in the data at any one station were limited. The data set
was as complete as can be expected.
Precipitation type was determined by coding the available precipitation data
from the airport station. Precipitation rates were not determined, but effects of wet
deposition are not likely to have a major influence during the episodes chosen for
this study as conditions were quite dry with only one hour of precipitation recorded
during Episode 1 and eight hours of precipitation recorded during Episode 2.

72

Cloud cover, ceiling height and pressure data were available from the airport
station for daytime hours only.

For overnight data, observations from the next

closest airport in Revelstoke was used. Cloud cover and ceiling height were taken
directly from Revelstoke. Pressure was adjusted using the hydrostatic equation to
account for the difference in elevation between the Golden airport and the
Revelstoke airport.
Table 5-3

Summary of the four meteorological stations used in CALMET modelling

Site
Name

Elev ASL
(m)

Meteorological sensors

Site
UTM
east (m)

Site UTM
North (m)

CPR
Town
Golf
Course
Airport

857
773
795

Temp, Wind, RH
Temp, Wind, RH
Temp, Wind, RH, Sodar
Wind
Temp,
Wind,
RH,
pressure, cloud cover,
ceiling
height,
precipitation categories

504.8252
501.9755
499.5356

5679.853
5682.693
5686.801

Wind Sensor
height AGL
(m)
10
10
10

502.3239

5683.188

10

5.1.3b

785

Upper Air Data

The minimum upper air data requirements for CALMET are twice-daily
radiosonde releases with observed vertical profiles of wind speed, wind direction,
temperature, pressure and elevation.

These vertical profiles are used in

combination with surface station data to predict wind fields and temperature lapse
rates in the layers aloft.
Previous modeling attempts in Golden were hampered by the lack of local
upper air data. To measure more locally appropriate upper air data, a SCINTEC
FAS64 phased-array Doplar sodar system was operated at the golf course location
73

(Figure 2-3). The SODAR system resolved wind speed and direction for the lower
part of the boundary layer to approximately 500 metres above ground level by
measuring the scattering of sound waves by atmospheric turbulence.
Obtaining a local vertical temperature profile in the lowest portion of the
atmosphere was important as temperature lapse rates calculated from upper air
stations hundreds of kilometres away would not allow for accurate prediction of
mixing heights and stability classes. In order to capture a vertical temperature
profile, six HOBOs were installed on the mountain side to the west of Golden (Figure
3-3). It is noted here that the temperature near the surface from the HOBOs is not
ideal as these measurements would be different than upper air in mid-valley,
however it was the best available option. The SODAR winds and HOBO temperature
profiles were combined with the twice daily radiosonde releases from the upper air
station in Prince George to meet the upper air requirements for CALMET. Prince
George was used as the upper air station as opposed Kelowna, the closest upper air
station, as the climate normals more closely matched those in Golden. The twicedaily upper air profiles used during the two episodes modeled in this study are
shown in Figure 5-3.
The combination of data from the HOBOs and SODAR system along with the Prince
George upper air data reveal profiles that indicate persistent temperature inversions
during both episodes. It has been hypothesized that wintertime inversions lead to
episodes of high particulate matter in Golden (Burkholder 2005). The inversions
depicted in the combined upper air data set may be a result of the differences in
temperature from the highest station of local HOBO temperature transect and the
74

next temperature recorded by the upper air data in Prince George. Since there were
no adjustments applied to meld the data sets together, the step from the local data
sets collected (SODAR wind data and HOBO temperature data) to the Prince
George upper air data could compromise the integrity of the combined upper air data
set if there is an artificial step change in temperature or wind introduced. Thus, it
could be that the combination of data used for the upper air data is not realistic of
the situation in Golden, however due to the limitations of the field monitoring with the
SODAR and HOBOs and the objectives of this study to use the available
meteorological data as input to CALMET, the upper air profiles in Figure 5-3 were
used.

75

(a)
7000

Dec.7 4:00
DeC-7 16:00
Dec.8 4:00
Dec.8 16:00
Dec.9 4:00
Dec.9 16:00

-? 4000

—I— Dec.10 4:00
— Dec.tO 16:00
— Dec. 11 4:00

Ui

Dec.11 16:00

3000

Dec.12 4:00
Dec.12 16:00
Dec.13 4:00
Dec.13 16:00
Dec.14 4:00

• Dec.14 16:00

235

250

240

255

260

280

265

285

Temperature (K)

(b)
- Feb.8 4:00

7000

-Feb.8 16:00
Feb.9 4:00
-Feb.9 16:00
-Feb.10 4:00
-Feb.10 16:00
-Feb.11 4:00
-Feb.11 16:00
-Feb.12 4:00
Feb.12 16:00
Feb.13 4:00
Feb.13 16:00
Feb.14 4:00

4000

Feb.14 16:00,
Feb.15 4:00
Feb.15 16:00
UJ

3000

-Feb.16 4:00
-Feb.16 16:00
-Feb.17 4:00
- Feb.17 16:00
Feb.18 4:00
Feb.18 16:00
-Feb.19 4:00
-Feb.19 16:00
Feb.20 4:00
-Feb .20 16:00
-Feb.214:00
- Feb.21 16:00

230

235

240

245

250

255

260

265

270

280

Temperature (K)

- Feb.22 4:00
- Feb.22 16:00
- Feb.23 4:00

Figure 5-3

Twice Daily Upper Air Profiles for (a) Episode 1 and (b) Episode 2

76

5.1.3c

Weighting Factors for Surface Stations and Upper Air Stations
Used in CALMET

The model was run with 10 vertical layers. The height above ground of each
of these layers is outlined in Table 5-2. Higher vertical resolution was used near the
surface as these layers are more important in the immediate dispersal of pollutants
in the boundary layer and ground-level pollutant concentrations that would affect the
population of Golden. Higher in the valley and above the valley it is less likely that
winds will be affecting the ground-level pollutant concentrations, especially during
episodes when mixing heights typically remain low.
As recommended by the US EPA, the CALMET option to extrapolate surface
data using similarity theory was chosen.

Using the observed data and similarity

theory, this method allows the influence of surface wind speed and direction to
extend into the other vertical layers.

In this study, winds at the surface were

determined based on surface observations and the CALMET initial guess wind field.
Winds in the layers aloft were derived from a combination of extrapolated surface
wind data and upper air data based on the biasing option in CALMET. Biases are
also outlined in Table 5-4.
The BIAS option in CALMET determines the extent to which the surface
station and upper air station data influences each vertical layer in the domain.

A

BIAS of zero equally weights the influence of upper air station winds and surface
station in the inverse distance squared interpolation of the initial guess field.

A

negative bias reduces the weight of the upper air station winds (i.e. a BIAS = -0.5

77

reduces the weight of upper air wind by 50%, while a BIAS = -1 reduces the weight
of upper air wind to zero). A positive bias reduces the weight of surface station wind
(i.e. a BIAS = +0.5 reduces the weight of surface station wind by 50%, while a BIAS
= +1 reduces the weight of surface station wind to zero).
For this study, BIAS was largely determined by the reliability of the SODAR
data incorporated into the upper air data file. Recall that the SODAR was used to
resolve wind speed and direction for the lower part of the boundary layer to
approximately 500 metres above ground level. BIAS values were set based on the
availability and reliability of the SODAR data captured. BIAS in the first two layers
were weighted completely towards surface winds (BIAS = -1) because the SODAR
data was unreliable at these elevations and the radiosonde data was from well
outside the modeling domain.

Modeled predictions improved with the two lower

layers being completely weighted to surface station extrapolation.

In layers 3-6,

where SODAR data were consistently available, the biases set to take advantage of
this local data by equally weighting surface observations and upper air data from the
SODAR in these layers (BIAS =0).
Sensitivity tests showed little difference in wind vectors when bias values
were adjusted for the upper-most layers. Layers 7 and 8 represent vertical layers
that are still below the top of the valley, thus the BIAS was set to be weighted slightly
to the surface station data (BIAS = -0.5).

Layers 9 and 10 are above the top of the

valley so these layers were weighted slightly to the upper air data (BIAS = 0.5)

78

Height at Top of Layer
(m)

Bias value

1

20

-1

2

40

-1

3

80

0

4

160

0

5

320

0

6

560

0

7

1000

8

1500

9

2200

0.5

10

3000

0.5

o o
cn cn

Vertical Layer

1

Weighting Factors for surface stations and upper air station used in
determining CALMET winds in the 10 modeled layers.

l

Table 5-4

5.1.4 Calmet Model Options
There are a number of user-specified input switches and options that
determine how CALMET handles terrain effects, interpolation of observational input
data, and other data to determine final wind fields. The options used and the
reasoning behind each choice are described in this section. Where deemed
appropriate, the recommended US EPA default parameters were used. Other model
parameters were chosen based on iterative testing of CALMET using the
meteorological, terrain and landuse data described above.

Final CALMET

parameters are summarized in Table 5-5.
As describe in Section 3-1, CALMET uses an inverse-distance squared
interpolation scheme to introduce observational data into the Step 1 wind field. The
interpolation scheme allows observational data to be heavily weighted in the vicinity
79

of the observational station. The wind field created in Step 1 dominates in regions
farther away or outside the radius of influence of the observational station.
Parameters R1 and R2 specify the weighting given to the observational data in the
surface layer and in the layers aloft, respectively.

R1 refers to the distance in

kilometers from an observational station at which the observation and the initialguess wind field produced by the diagnostic wind field module in CALMET are
weighted equally (Scire et al. 2000).

RMAX values are used to exclude

observational data from interpolation if the distance from the observational station to
a particular grid point exceeds the user-specified maximum radius of influence.
RMAX1, RMAX2 specify the parameter for the surface layer and layers aloft,
respectively (Scire et al. 2000).
Iterative testing of these CALMET parameters showed that to maintain valley
flow in the complex terrain R1 and R2 values that allowed observational data to
dominate between the TOWN and CPR stations were most appropriate. R1 was
chosen as slightly less than the distance between these two stations. Combined
with an RMax value of 5 km the wind fields balanced the observational dominated
valley flow with CALMET computed kinematic effects of terrain and slope flows.

80

Table 5-5

CALMET wind field model options

Parameter

Option Selected

US
Default

Froude number Adjustment Effects Calculated?

Yes

Yes

Kinematic Effects Computed?

Yes*

No

Surface Wind Observations Extrapolated to
Upper Layers?

Yes

Yes

Surface Winds Extrapolated even if Calm?

Yes

Yes

Maximum Radius of Influence over Land in the 5 km
Surface Layer (RMAX1)

No

Maximum Radius of Influence over Land Aloft 8 km
(RMAX2)

No

Radius of
(TERRAD)

Features 2 km

No

Relative Weighting of the First Guess Field and 4 km
Observations in the Surface Layer (R1)

No

Relative Weighting of the First Guess Field and 6 km
Observations in the Layers Aloft (R2)

No

Influence

of

Terrain

EPA

*Model runs with and without kinematic effects were performed with little variation between the two.
It is recommended that kinematic effects not be computed as it may cause "odd" model results.
Since there was little variation between the model runs, the computation of kinematic effects did not
cause "odd" results in this modelling exercise, so the results with kinematic effect turned on were
used.

5.1.5 Calmet Evaluation Methods
The evaluation of dispersion modeling studies focus on the model's ability to
provide accurate pollutant concentration fields as a result of the complete process of
emission release, transport by wind and turbulent diffusion in the atmosphere (Cox
2005). The evaluation here isolates a single element of the overall process, the
ability of the model to construct valid wind fields.

81

Visual assessment using scatter plots and the software mapping package
SURFER allowed for an effective initial gauge of CALMET's performance. Model
parameters (Table 5-2) were adjusted in subsequent model runs until the most
visually representative wind fields were observed in relation to the modelled terrain
and the observed wind data from the three stations in Golden. Common statistical
methods used in the evaluation of diagnostic wind models such as CALMET were
also used to evaluate model performance as suggested in Cox et al. (2005) and are
presented here.
The statistical evaluation in this study included the Mean Error (ME) and the
Mean Absolute Error (MAE). ME is the average difference between observed and
predicted values. The MAE is the average of the absolute differences between the
observed and predicted values.

A perfect model has a ME and MAE of zero.

Another common accuracy measure is the mean square error (MSE) which is similar
to the MAE except differences are squared making it more sensitive to larger errors
and outliers (Wilks 2006). The RMSE (the square root of the MSE) is reported here
as it has the same physical dimensions as the predictions and observations (Wilks
2006). The statistical distributions of the errors are further quantified by the minimum
(MIN ERR), median (MED ERR) and maximum error (MAX ERR).
Bias is sometimes used to assess wind speeds, but is not an accuracy
measurement and only reveals whether the model is typically over- or underpredicting the observed measurements (OB) (Wilks 2006). In this case, it is more
useful to determine the accuracy of the model predictions (PR). One can use
fractional bias (Fb) as an indicator of model accuracy:
82

Fb = 2(OB-PR)/(OB+PR).
Fb is bounded by +/- 2 with the extremes meaning no agreement between the
modeled and observed values. Fb indicates to what degree the model is over or
under-predicting. Negative values of Fb indicate the model is over-predicting and
positive values indicate under-prediction.

Fb of +/- 1.00 corresponds to model

prediction within a factor of 3 of observed values, whereas a value of +/- 0.67
indicates predictions are within a factor of 2 of observed values. The Fb statistic is
less reliable if the model is grossly over- or under-predicting observed values as it is
bounded so that higher levels of over or under prediction result in little change in the
statistic (Wilks 2006).
From the initial analysis it was determined that wind speed may have a large
influence on the dispersion in Golden (Section 4-2). As a measure of the accuracy
of wind speed predictions, the correlation coefficient (CORR) between calculated
and observed wind speeds was calculated.
In order to test the model predictions versus the observed measurements
several model runs were evaluated for each episode. First, the ability of CALMET to
produce wind fields consistent with the surface station input data was tested by
comparing observed values with a "full" CALMET run including all surface station
data as input data. This comparison simply ensures that surface station wind field
remains the same in the model output.

Next, one station's observed data was

excluded and the model run using the three other surface station data. Following
the model run, predicted data were processed from the grid cell containing the

excluded observational station and compared to the observed data from that station.
Three model runs were made leaving out one of the CPR, TOWN or GOLF stations
each time. This "leave one-out" method allowed investigation of how accurately
CALMET would capture the winds at each station's location without the data from
that station being used as input data. This is an indication of how CALMET is
predicting winds away from input stations. The model runs were repeated for both
episodes 1 and 2. Statistics for both the "full" and leave one out methods are
provided based on the observed and predicted data sets at all three stations
individually, and as a complete data set (ALL). Separate analysis of episode 1 and 2
are provided for comparison.
Another method was employed to determine how accurately CALMET was
predicting the wind fields and other meteorological parameters at points other than
those representing the input stations. This method was used by Ostermann and
Schutte (2004) in a modeling effort for Levelton Consultants in Williams Lake, BC.
Using the "full" CALMET run a point was extracted to compare predicted
meteorological parameters with the surface stations wind data, temperature data
and predicted stability classes. The point extracted was located in the Columbia
River Valley halfway between the northern GOLF station and the southern CPR
station (see Figure 2-3). If data at the extracted point were comparable to the other
stations it would allow for more confidence that the main valley flow was being
captured reasonably well during the "full" CALMET run.

84

5.2 CALMET EVALUATION RESULTS
As expected, CALMET was able to very accurately reproduce the input
observed data during the "full" CALMET run (Tables 5-5, 5-6, 5-9 and 5-10) since it
is essentially reproducing its input data..

Wind speed shows small MAE and Fb

statistics at all stations for Episode 1 and 2. Fb shows that the model is
underpredicting wind speeds during the second episode. In addition the correlation
coefficients are approaching 1 for all stations indicating good model performance.
Wind direction also shows very low MAE. With the exception of the CPR station
during Episode 2, all stations predict over 86% of wind directions, for the "full"
CALMET run, within 10 degrees of the observed values.

85

Table 5-7

Episode 1 CALMET "full" run wind speed statistics. Sp and S0 are the wind speed predicted and observed. CTp = standard
deviation of predicted; a0 = standard deviation of observed; ME = mean error; MAE = mean absolute error; MAX = maximum
error; MED = median error; MIN = minimum error; RMSE = root mean square error; FB = fractional bias, and; CORR =
correlation coefficient.
So

ap

o0

ME

MAE

ERR
MAX

MED

MIN

RMSE

FB

CORR

0.98

1.01

0.82

0.84

0.03

0.05

0.96

0.03

0.00

0.10

0.03

0.995

TOWN

1.03

1.05

0.63

0.63

0.02

0.04

0.20

0.03

0.00

0.05

0.02

.0,997

GOLF

0.97

0.97

0.63

0.64

0.01

0.01

0.18

0.01

0.00

0.02

0.01

0.999

ALL

0.99

1.01

0.70

0.71

0.02

0.03

0.96

0.02

0.00

0.06

0.02

0.996

Station

sP

CPR

Episode 1 CALMET "full" run wind direction statistics. Dp and D„ are the wind direction predicted and observed. %
10° is the percentage of winds within 10 degrees of the observed winds. % 45° is the percentage of win ds within 45
degrees of the observed winds. Refer to Table 5-5 for other parameter definitions.
ERR
MAE MAX MED

MIN RMSE

0s
O0

Table 5-6

% 45°

206.13 206.11 3.98 58.82 2.52

0.02

6.94

91.667

99.48

TOWN 213.24 210.56 1.58 15.53 1.20

0.00

2.24

99.479

100.00

GOLF 213.42 212.07 1.23 36.29 0.76

0.00

3.00

99.479

100.00

ALL

0.00

4.55

96.875

99.83

Station
CPR

Dp

D0

210.93 209.58 2.25 58.82 1.25

86

Table 5-8

Table 5-9

Episode 1 wind speed statistics for CALMET using the "leave one-out" method. Sp and SQ are the wind speed
predicted and observed. Refer to Table 5-5 for other parameter definitions.
Station

sp

So

ap

O0

ME

MAE

ERR
MAX

MED

MIN

RMSE

FB

CORR

CPR

0.98

1.01

0.60

0.84

0.04

0.45

2.61

0.27

0.00

0.68

0.04

0.603

TOWN

0.92

1.05

0.75

0.63

0.13

0.42

2.58

0.27

0.00

0.62

0.14

0.632

GOLF

0.89

0.97

0.61

0.64

0.09

0.35

1.78

0.26

0.00

0.50

0.09

0.700

ALL

0.93

1.01

0.66

0.71

0.09

0.41

2.61

0.27

0.00

0.60

0.09

0.618

Episode 1 wind direction statistics for CALMET using the "leave one-out" method. Dp and D0 are the wind direction
predicted and observed. % 10° is the percentage of winds within 10 degrees of the observed winds. % 45° is the
percentage of winds within 45 degrees of the observed winds. Refer to Table 5-5 for other parameter definitions.
Station
CPR

ERR
MAE MAX MED

% 10°

% 45°

206.13 206.11 3.98 58.82 2.52 0.02 6.94 91.667

99.48

Dp

D0

TOWN 213.24 210.56

MIN RMSE

1.58 15.53 1.20 0.00 2.24 99.479 100.00

GOLF

213.42 212.07 1.23 36.29 0.76 0.00

3.00

ALL

210.93 209.58 2.25 58.82 1.25 0.00

4.55 96.875

99.479 100.00
99.83

87

Table 5-10

Table 5-11

Episode 2 CALMET "full" run wind speed statistics. Sp and SQ are the wind speed predicted and observed. Refer to
Table 5-5 for other parameter definitions.

Station

sP

So

Cp

°0

ME

MAE

ERR
MAX

CPR

0.66

0.69

0.75

0.78

-0.03

0.06

0.50

0.04

0.00

0.08

0.05

0.996

TOWN

0.91

0.93

0.57

0.58

-0.03

0.07

0.63

0.07

0.00

0.11

0.03

0.982

GOLF

0.68

0.68

0.37

0.37

0.00

0.01

0.11

0.01

0.00

0.02

0.01

0.999

ALL

0.75

0.77

0.60

0.61

0.02

0.05

0.63

0.02

0.00

0.08

0.03

MED

MIN

RMSE

FB

CORR

Episode 2 CALMET "full" run wind direction statistics. Dp and D0 are the wind direction predicted and observed. %
10° is the percentage of winds within 10 degrees of the observed winds. % 45° is the percentage of win ds within 45
degrees of the observed winds. Refer to Table 5-5 for other parameter definitions.

Station
CPR

MAX

ERR
MED

MIN

RMSE %10° % 45°

198.80 186.66 11.90 154.30

4.50

0.01

25.14

69.79

98.70
99.74

Dp

D0

MAE

TOWN 303.05 303.66

4.28

141.27

4.28

0.00

13.34

91.93

GOLF

275.59 265.27

1.69

13.83

1.23

0.00

2.55

98.70 100.00

ALL

231.93 251.86

5.93

154.30

1.56

0.00

16.50

86.81

99.48

88

Table 5-12

Table 5-13

Episode 2 wind speed statistics for CALMET using the "leave one-out" method. Sp and S0 are the wind speed
predicted and observed. Refer to Table 5-5 for other parameter definitions.

Station

sp

So

ap

°0

ME

MAE

ERR
MAX

MED

MIN

RMSE

FB

CORR

CPR

0.94

0.69

0.64

0.78

0.25

0.61

2.24

0.47

0.00

0.78

-0.30

0.466

TOWN

0.78

0.93

0.97

0.58

-0.15

0.69

3.98

0.69

0.00

0.95

0.18

0.345

GOLF

0.77

0.68

0.60

0.37

0.09

0.32

2.20

0.23

0.00

0.48

-0.12

0.627

ALL

0.83

0.77

0.76

0.61

-0.06

0.54

3.98

0.37

0.00

0.76

-0.08

Episode 2 wind direction statistics for CALMET using the "leave one-out" method. Dp and D0 are the wind direction
predicted and observed. % 10° is the percentage of winds within 10 degrees of the observed winds. % 45° is the
percentage of winds within 45 degrees of the observed winds. Refer to Table 5-5 for other parameter definitions.

ERR
MED

MIN

RMSE % 10° % 45°

300.25 186.66 79.81 179.84 68.62

0.23

97.50

9.38

72.66

TOWN 193.25 303.66 64.42 179.70 64.42

0.02

84.51

12.50

77.34

GOLF

298.08 265.27 41.23 177.27 28.58

0.02

56.73

14.58

83.07

ALL

263.86 251.86 61.68 179.84 41.60

0.00

81.38

12.15

77.69

Station
CPR

Dp

D0

MAE

MAX

89

Using the leave one out method, the performance measures for CALMET
deteriorate in comparison to the 'lull" CALMET run as expected (Tables 5-7, 5-8, 511 and 5-12). Episode 1 performance measures are much better than those during
episode 2. Episode 1 wind speed shows low Fb for all stations, indicating that
speeds are well within a factor of 2 of observed wind speeds. Fb is positive during
Episode 1 indicating underestimation of wind speed, while during Episode 2 the
overall Fb is negative indicating overestimation of wind speed.

The correlation

coefficients are also very reasonable with a low at the CPR station of 0.603 and a
high at the GOLF station of 0.7. Wind direction predictions are within 10 degrees of
the observed values only 19 % of the time during episode 1 and within 45 degrees
64% of the time.

Episode 2 wind speed shows higher Fb and the correlation

coefficients are lower. Wind direction predictions are within 10 degrees of observed
values only 12% of the time, but within 45 degrees 78% of the time.
On a station-by-station basis the GOLF station performed the best.

The

correlation coefficient for wind speed was the highest at this station as well as the
highest percentage of winds within 10 and 45 degrees of the observed values.
To display CALMET's predictions of wind flows around Golden, a snapshot of
calculated wind vectors for level 1, 4 and 8 is provided (Figure 5-4 - 5-6). Near the
surface the wind vectors are more variable in direction and speed (Figure 5-4). In
general, the flow is terrain-forced along the valley. Slope flows steer vectors on the
mountainsides and CALMET also shows some easterly flow coming out of the
Kicking Horse Pass. At 120 m, wind vectors are calculated using equal weighting
from surface station and upper air station data (Figure 5-5). The wind speeds are
90

more uniform and direction is similar to that at the surface. At 1250 m, wind vectors
are nearly uniform in speed and direction across the entire domain (Figure 5-6).

Dec 11,2005
18:00 LST(UTC-0700)
117.2W
5705

5Sl '
5 5685

Terrain (m)

500

505

UTM East (km)

Figure 5-4

CALMET surface (10m) wind field for 18:00, December 11, 2005

91

Dec 11,2005
18:00 LST(UTC-0700)
117.2

14500
3900
3700
13500
3300
3100
2900
2700
2500
2300
: 2100
1900
1700
1500
1300
1100
900
700
1500
|300

5 5685

] 100
'-1

Terrain (m)

490

495

500

505

510

515

520

UTM East (km)

Figure 5-5

CALMET Level 4 (120m) wind field for 18:00, December 11, 2005

92

Dec 11, 2005
18:00 LST(UTC-0700)

k

^
KwNikVv

I

3900
4500
3700
3500
3300
3100
2900
2700
2500
2300
2100
1900
1700
1500
1300
1100
900
700
500
300
100

5 5685

-1

Terrain (m)

490

495

500

505

510

515

520

UTM East (km)

Figure 5-6

CALMET Level 8 (1250m) wind field for 18:00, December 11, 2005

The frequency distribution of the surface winds predicted at the extracted
point (point location shown in Figure 2-3) are comparable to at least one input
station's observed frequency in all wind speed classes (Figure 5-7).

There is a

reasonable distribution of wind speed classes at the extracted point with no grossly
under or overestimated wind speeds.

The modeled winds show the expected

diurnal variation in wind speed (Figure 5-8).

93

(a)

(b)

70%-

• Extracted
• CPR

^ 50%

• TOVW

• TOWN

•Gap

50%40%-

3. 30°*
20%-

<0.5

0.5- 1.5

1.5-3.0

3.0-5.0

5.0-7.0

>7.0

4)5

05- 1.5

Wind Speed Class (m/s)

Figure 5-7

1.5-3.0

30-5.0

5 0-7 0

>70

Wind Sp*«d CI«M(m/«)

Wind speed frequency distribution for the three surface stations and the
extracted point for (a) Episode 1, (b) Episode 2.

w 1.4

• Episode 1
Episode 2

' 0.8

10

15

20

25

Hour of Day

Figure 5-8

Diurnal variation in modeled wind speeds at the extracted point during each
episode.

Wind rose for the extracted point during episode 1 (Figure 5-9a) shows good
agreement with the wind rose from observed values reported at the TOWN station

94

(Figure 5-9c).

During episode 2, the wind rose shows good agreement with the

GOLF station observed surface wind rose (Figure 5-9d).

This reveals northerly

winds for almost the entire episode (Figure 5-9b).
(a)

(b)

•>7.0 ml*
DS.0 • 7.0 mft
•3.0-50FT*
•1.8 • 30irtt
•OB-tSmrt
•OJ-Oinrt

i)«

Figure 5-9

#.!%

Predicted wind rose at the extracted point for (a) episode 1, and (b) episode 2,
(c) TOWN station Episode 1 and (d) GOLF station Episode 2.

During both episodes, the extracted point has predicted temperatures that are
consistent with the observed values at the three surface stations (Figure 5-10).

95

(a)

(b)

Extracted
CPR
TOWN

Figure 5-10

—•—Extracted
• CPA
TOWN
GOLF

24-hour mean temperatures for the three surface stations and the extracted
point for (a) Episode 1, (b) Episode 2.

CALMET uses the Turner method to calculate Pasquil Gifford stability
classes. The CALMET stability classes are defined in Table 5-14.
Table 5-14

Pasquil Gifford Stability Classes

Stability Class
A
B
C
D
E
F

Atmospheric Conditions
Very Unstable
Unstable
Slightly Unstable
Neutral
Slightly Stable
Stable

Figure 5-11 shows the frequency distribution of the predicted stability classes
at each station and the extracted point. The extracted point and the three stations
predict nearly the same frequency of each stability class during both episodes.

(a)

(b)

60% 1

|

- ——
.Extracted]

tcfir

• Extracted
BCp„

40%

a TOWN

MB

50% '

I

•
40% a*
g 30% •
3

V

e

20% •

10% 0% +-

i

.GOLr

11

•
I

••

IIIL.II iilin II
Stability C)as«

Figure 5-11

°TOWN

Stability Claaa

Frequency distribution of predicted stability class at the three surface stations
and extracted point for (a) Episode 1, (b) Episode 2. Stability Class 1 = very
unstable; Stability Class 2 = unstable; Stability Class 3 = slightly unstable;
Stability Class 4 = neutral; Stability Class 5 = slightly stable; Stability Class 6 =
stable

5.3 CALMET DISCUSSION
CALMET was able to reproduce the observed surface station input data very
accurately during the full CALMET run. This evaluation was simply a check to make
sure the input data were being reasonably maintained by the model following the
interpolation scheme. Surface observations are used in CALMET as an initial guess
field. The initial guess field is adjusted for kinematic effects of terrain, slope flows
and three dimensional divergence minimization to produce a Step 1 wind field (Scire
et al. 2000). The second step in the CALMET procedure re-introduces observational
data through an inverse-distance squared interpolation scheme which weighs
observational data heavily in the vicinity of the observational station and the Step 1
wind field more heavily away from the surface stations. The resulting wind field is
then subject to smoothing and divergence minimization to produce the final wind
field (Scire et al. 2000)

97

The evaluation here compares well with a test of CALMET by Cox et al. (2005)
and with tests of other diagnostic wind field models in complex terrain reported by
Ratto et al. (1994) and Ross et al. (1988).

The mean absolute errors for wind

speeds in those tests ranged from approximately 0.2 - 0.4 m/s when using the full
complement of surface station data. The mean absolute errors for both episodes in
the "full" CALMET run are lower than these values, partly an artifact of low mean
wind speed. Cox et al. (2005) reported absolute wind speed errors of less than 10%
of the mean wind speed, values that are comparable to this study. Wind direction
errors were less than 10 degrees 87% of the time (Cox et al. 2005), in comparison
with an average of over 90% of predictions within 10 degrees in this study.
Cox et al. (2005) and Ross et al. (1988) provide a comparison to the leave
one out method. In these studies a larger number of stations were used in a full run
and a smaller number of input stations were used to evaluate diagnostic wind field
model performance. Mean wind speed errors in these studies range from 0.4 to 1.4
m/s. CALMET shows mean errors of less than 0.1 m/s in Golden that is well within
this range. The mean wind speed errors, which measures the difference between
observations and predictions regardless of whether it is positive or negative,
indicates a slight over-prediction of the observed values at the surface stations. Cox
et al. (2005) reported a mean absolute wind speed error that accounts for
approximately 32% of the mean observed wind speed.

In Golden, the CALMET

mean absolute wind speed error represents 41% of the mean wind speed during
episode 1 and 77% of the mean wind speed in episode 2. As mentioned previously,
diagnostic wind field models can have difficulty predicting low wind speeds that can

98

be attributed to the model interpolation techniques and the threshold of the
anemometer used to measure wind speed (i.e. wind speeds less than 0.5 m/s are
recorded as zero are compared to predicted wind speeds that are above 0.5 m/s,
thus adding 0.5 m/s to the absolute error, which at low wind speeds accounts for a
greater percentage of the mean wind speed).
Predicted wind direction is reported as within 20 degrees of observed values
42% of the time by Cox et al. (2005). In Golden, Episode 1 shows 40% percent of
predicted winds within 20 degrees, while Episode 2 shows only 24%. CALMET's
prediction of wind direction is not as good in Golden compared with the complex
terrain modelling exercise by Cox et al. (2005) during episode 2.
Previous studies by Cox et al. (1998, 2000, and 2005) showed that diagnostic
wind models typically predict both wind speed and direction more accurately during
non-stable atmospheric conditions. This was not the case in Golden. In comparison
to the periods modeled by Cox, the entire episodes modeled here would be
considered to have very low wind speeds. Perhaps because winds are relatively light
throughout the episodes, there is no real difference in the model accuracy
depending on stability. Examination of the stability classes shows that indeed there
is never a "very unstable" atmosphere reached during Episode 1. Almost 76% of the
time winds are being predicted under neutral or stable conditions during the episode
1 and 62% during Episode 2 which could explain the model's poorer performance.
Therefore, it is more appropriate to compare the results in Golden with those during
stable periods in Cox et al. (2005).

99

Cox et al. (2005) reported wind speed mean absolute errors of 1.8 m/s or
45% of the observed wind speed during stable conditions. This is more comparable
to the mean absolute errors reported for episode 1. Cox et (2005) also saw a wind
direction mean absolute error of 39 degrees during stable periods, which is
comparable to the errors reported in Episode 1. Even in comparison with the times
of stability, Episode 2 does not compare well with the Cox study.
The CPR and TOWN stations do not perform as well in the "leave one-out
method" because they are influenced by each other's observed wind data used in
determining the final wind fields in CALMET.

They are also influenced by the

observed data at the AIRPORT station. As mentioned before, R1 was maximized
between the CPR and TOWN station so that each stations radius of influence is
nearly overlapping. (The AIRPORT station could not be tested in the leave one out
method because it was the only station with observed measurements for RH,
pressure, cloud cover, ceiling height and precipitation categories.)

The GOLF

station, which is further away from the radius of influence of the TOWN, CPR and
AIRPORT stations, performs better in the leave one-out method. It appears that
CALMET is performing well at the GOLF station, despite its location nearer to the
mountainside which would be considered more complex terrain given the influence
of the slope flows at this location.
R1 was chosen as 1.3 km because CALPUFF predictions (Chapter 6)
improve with higher values of R1. Higher R1 maintains the along-valley flow around
Golden because it increases the influence of CPR and TOWN measurements and
decreases the influence of the model computed terrain effects and slope flows.
100

However, this results in the CPR and TOWN stations performing poorly in the leave
one out method. There appears to be a trade-off between improving statistically the
performance of stations in the CALMET evaluation and improving predicted PM
concentrations in CALPUFF. The performance of CALMET is hampered by defining
R1 universally for all stations in CALMET. The option to specify R and Rmax values
on a station by station basis could possibly improve the wind fields.
With large values for R1 and Rmax the modeled winds, especially those near
the Town of Golden are being influenced mainly by the surface station data and
therefore winds in and surrounding Golden typically follow the valley alignment as
expected (Figure 5-2). The exception is wind channelled through the Kicking Horse
pass to the east of Golden.
When CALPUFF was run the CALMET wind fields were calculated using data
from all stations. When using all stations in the CALMET run, it is obvious that the
winds will be accurate in the grid cell containing the station as we saw in the
statistics from 'lull" CALMET runs. However, what is not clear is how accurate the
winds are in other grid cells. Thus, the extracted point comparison and leave one
out analysis were performed. It is assumed that the winds do not change much
between the extracted point and the input stations. This assumption is based on
observed values at all stations that typically show flows are along valley from the
northeast or southwest. When compared to the input stations, Figures 5-6 - 5-10
reveal that CALMET is predicting wind speed and direction, temperature, and
stability class nearly as well at the extracted point as it does nearer to the input

101

stations. This is also confirmed by a visual assessment of wind vectors in Figures 53 - 5-5.
It

was hypothesized that the late afternoon and evening high PM

concentrations occurring during the episodes may be caused by an along-valley
wind flow slowing and reversing. It was suggested that this transition, along with
downslope flows and increased emissions, could lead to the build-up of PM during
this time of day. CALMET does capture along-valley flow slowing and a slight
reversal at times of high PM concentrations. During the day, winds are typically from
the south. In the evening, the winds slow and, in some cases, show a reversal and
come from the north. This is consistent with winter daytime and night-time wind rose
for Golden (Burkholder 2005). The variation in wind direction is best captured by
observing the CALPUFF plumes shown in Chapter 6. The diurnal variation can also
be observed in the modeled wind speeds (Figure 5-7).
Although the performance of CALMET was poorer than other comparable
studies using the leave one out method, visually the surface wind fields look to be as
expected when viewed as hourly vectors as in Figure 5-3. In addition the extracted
point shows reasonable agreement with observed and predicted values during the
"full" CALMET run. CALMET also seems to be capturing some of the important
valley wind flows (i.e. slope flows and along-valley flows) that have been
hypothesized to lead to increased PM concentrations. Even though these model
options and parameters produce poor results during the leave one out method, it is
the best performance of the model for the Golden modeling study using the data

102

provided. The model's ability to predict final PM concentrations using the predicted
CALMET wind fields will now be tested.

103

6

CALPUFF
PM concentrations in Golden consistently rank among the highest annual

averages in the province for both PM2.5 and PM10 (Figure 2-1). Episodes of high PM
occur more frequently during the winter and early spring (Figure 2-2). Following
CALMET optimization, the CALPUFF dispersion model was used to investigate
ambient air quality during two particulate matter episodes in the winter of 2005-2006
(Section 4-1).

Similar studies have been conducted in many locations, including

New Zealand (Barna and Gimson 2004), Beijing (Song et al. 2006) and Athens
(Assimakopoulos 2005).
Emission sources were modeled individually and CALPUFF's performance in
estimating PM concentrations was evaluated. The model was then used to apportion
sources contributing to the episodes (Chapter 7). The source apportionment was
compared to receptor modeling completed by the BC MOE in Golden. Emission
rates were determined based on the Golden Emissions Inventory (Abel et al. 2006)
and adjustments were made using the results of receptor modeling conducted by
BCMOE (Evans and Jeong 2007).
Winter episodes of high PM in Golden were hypothesized to occur because of
increases in emissions due to combustion sources as well as stagnant dispersal
conditions. During episode 1, the PM ratio increases significantly in comparison to
the winter mean PM ratio (Table 4-1). An increase in PM ratio indicates a rise in the
PM2.5 portion of PM10, indicating more contribution from combustion sources. Wood-

smoke from space heating in Golden is a significant source of PM during the winter
months. As residents do not have access to natural gas, many people burn wood to
104

heat their homes.

Incomplete combustion from burning wood in a fireplace or

woodstove results in PM2.5 emissions. Other combustion sources in Golden include,
but are not limited to, emissions from boilers at Louisiana Pacific, agricultural
burning, back yard burning, and emissions from prescribed burns in the forests
surrounding the Town.
Episodes during the early spring may be attributed to increased releases of
road dust and crustal materials due to the spring thaw. As the temperature increases
and roads dry out, road traction material that has built up over the winter is made
airborne by vehicles on the highways and streets in and around Golden.

Wind

blown fugitive dust sources will also increase as temperatures rise and snow melts
from the surface.

During episode 2, the PM ratio decreases significantly in

comparison to the winter mean and episode 1 PM ratio (Table 4-2). A decrease in
the PM ratio indicates a rise in particles greater than 2.5 microns. These particles
are largely comprised of crustal material emissions.
The steep valley walls and weak terrain forced local wind flows means
dispersion may largely depend on the diurnal mountain wind system (Section 2-4)
As previously discussed, low wind speeds, shallow mixing layers, neutral or stable
atmospheric conditions and recirculation in this system may result in the build-up of
pollutants near the bottom of the Columbia river valley. CALMET captures some
aspects of the diurnal mountain wind system and the available emissions information
was used in CALPUFF in an attempt to create the best possible model for use in
airshed management decisions.

105

This chapter explains the model setup, an example of CALPUFF output
during episode 1, and an evaluation of the model's performance.

6.1 CALPUFF SETUP
CALPUFF is a puff dispersion model capable of handling multiple-layers within
the atmosphere and multiple pollutant species (Scire et al. 2000). The non-steady
state nature of the model allows it to handle time and space varying meteorological
and emission conditions.
CALPUFF uses CALMET's final wind field to disperse point, line and area
emission sources within the gridded modeling domain. The puff model represents a
continuous emission source, such as a plume, as a number of discrete packets of
pollutant material. The puffs are released and evolve in size according to Gaussian­
like diffusion and dispersion, but the multiple puffs allow changes in meteorological
and emission conditions to be captured.

The meteorological processes were

determined by modeling in CALMET (Chapter 5). CALPUFF depends largely on the
emission estimates and dispersion options selected by the modeler. The following
sections describe emission source characterization and the model options selected
for this study.

6.2 RECEPTORS
The modeling domain contained a 35 by 40 km grid of receptors spaced 250 m
apart. Discrete receptors were located at the PM monitoring locations in Golden
(Figure 2-3) to allow for comparison between model estimations and observed
ambient PM concentrations. Each receptor point represents a location at which
model estimated PM concentrations are calculated.

106

6.3 CALPUFF MODEL OPTIONS
CALPUFF model options used in this study are provided in Table 6-1. Model
options follow the USEPA default recommendation unless otherwise stated. Model
options were chosen based on recommendations from the BC Dispersion Modeling
Guidelines (BC MOE 2008).
Table 6-1

CALPUFF model options selected

Parameter

Option Selected

Terrain Adjustment Method

Partial Plume Path Yes
Penetration

Transitional Plume Rise

Modeled

Yes

Stack Tip Downwash

Modeled

Yes

Vertical Wind Shear above Stack Top

Not Modeled

Yes

Chemical Mechanism

Not Modeled

No

Wet Removal

Not Modeled

No

Dry Deposition

Modeled

Yes

Method Used to Compute Dispersion
Coefficients

internally calculated No
using
micrometeorological
variables

Partial Plume Penetration of Elevated
Inversion

Modeled

Yes

Minimum Wind Speed Allowed for Non-Calm
Conditions

0.5 m/s

Yes

USEPA
Default

Chemical transformation was not calculated using MESOPUFF II, the
chemical transformation module included with CALPUFF, because during the winter
months secondary particulates would not be a significant source in Golden. The wet
removal option was not used because the appropriate precipitation information was

107

not available.

The wet removal of PM during the episodes was likely negligible

during the episodes modeled as precipitation occurred very infrequently during both
episodes. Computation of the dispersion coefficients was performed internally by
CALPUFF as recommended in the British Columbia Air Dispersion Modeling
Guidelines (BC MOE 2008)

6.4 ESTIMATING EMISSION DATA
Estimation of PM emission rates, along with source characteristics such as
emission velocity, temperature and height, combined with the meteorology are
essential to air dispersion modeling. Emissions from point, area, volume and line
sources are reported in airshed emission inventories. An emissions inventory (El)
for Golden provided the basis for the emissions input into CALPUFF (Abel et al.
2006). It was not the focus of this study to reiterate the methods used to develop
that inventory.

The El contained estimates of annual emissions from all known

significant sources in the modeling domain. Unfortunately, the El did not provide
spatial information beyond the point source estimations and small area sources such
as mill and rail yards. Most area and line sources are simply an estimate based on a
provincial emission inventory database, Air Contaminant Emissions (ACE) Project
(MWLAP 2001). This GIS database allows for extraction of local airshed emission
estimates based on a variety of factors such as landuse, area and population
density. Details on this database can be found in Glen and Wakelin (1998), Gibson
(1998), and Fam (1998). Inherent uncertainties in the El are discussed by Abel et al.
(2006).

108

From the El, PM emissions data were compiled for point, line and area
sources in the Golden Airshed. Point sources include those that can be attributed to
a single, fixed emission point such as a smokestack. Line sources, such as railway
or vehicle traffic, emit along a fixed line. Area sources include groups of point
sources that cannot be assessed on an individual basis, such as backyard burning
and residential space heating. For most of the modeled sources, the El contained
the best available emissions data.

Updated emissions information gathered is

described below. It was not the intention of this study to build an entirely new El, but
rather attempt to model using the emission estimates in the completed El.
It was necessary to transform the estimates from the El into specific emission
rates and identify or estimate the key source characteristics. An emission inventory
catalogs the total emission from each source for the entire airshed. Point source
locations are easily identified and source characteristics are measurable making
model input straightforward. However, in the El area sources are not given spatial
boundaries and source characteristics need to be estimated. With a lack of
information on area source characteristics and spatial bounds it becomes necessary
to estimate polygon areas that would be emitting the PM identified in the El. As the
El was not temporally resolved, annual emission estimates for each source from the
El were applied as emission rates using the specific methods outlined below.

6.4.1 Point Sources
Point sources are emission sources that are released from a stack or specific
point. Emissions from point sources are modeled as puffs of pollutant released from
a stack at a specified height, speed and temperature. The emission rate determines

109

the rate at which the puffs are released.

CALPUFF requires detailed source

characteristics such as the location of the emitting stack, the base elevation, stack
height, stack diameter, and the exit velocity/temperature of the discharge containing
PM. The only point sources modeled in Golden were from the operations at
Lousiana Pacific Engineered Wood Products Ltd (LP) (Figure 3-2).

6.4.1a Louisiana Pacific Engineered Wood Products Ltd.

Site Description
LP is located immediately north of the downtown core of Golden. The UTM
locations of the point sources as well as building locations and dimensions were
provided by LP's environmental engineer, Mike Brygger.

Sixteen separate point

sources were identified as emitting from the LP site north of the town centre in
Golden.

These include a hog boiler, cyclones, plywood dryer emissions and

plywood press emissions. Stack information was compiled for each source including
the location of the emitting stack, the base elevation, stack height, stack diameter,
exit velocity of the discharge containing PM and exit temperature (Table 6-2).
The option for CALPUFF to calculate the effects of building downwash at LP
was selected. Since buildings and structures can affect the dispersion of plumes
due to wake effects, the objects within close proximity to the stacks at LP were
incorporated into the modelling. Any buildings and structures with the potential to
cause downwash effects were selected based on the criteria in the US EPA Building
Profile Input Program (BPIP). Building dimensions and locations were provided by
LP (Figure 6-1).

This information was processed using BPIP to produce the

necessary array of 36 direction-specific building widths and heights for flow vectors
110

from 10 degrees to 360 degrees in 10 degree increments.

CALPUFF uses this

estimate of building dimensions to model the building downwash effects from each
stack emitting at the LP site depending on the wind direction at the time of the
emitted puffs.

Emission Rates
Emission rates were also provided by LP and were estimated using published
emission factors and production totals as an estimate of operations during each
episode. The emission rates for PM10 and PM2.5 specified for Episode 1 and 2 are
listed in Table 6-2. LP operations typically run 24 hours a day therefore, emissions
were assumed constant for each episode and the option to vary emissions in
CALPUFF was not used.

111

Figure 6-1

Aerial photo showing source (SRC) locations and building (BLD) dimensions at
Louisiana Engineered Wood Products Ltd. Site in Golden BC.

112

Table 6-2

Summary table of LP point emissions. E1 = episode 1, E2 = episode 2, ER ; emission rate

Emission Rates (g/s)

Stack
Exit
. .. Diameter
Velocity
. .
(m)
(m/sj

Height
i™\
(m)

Temp
<ve\
(K)

Hog Boiler (SRC1)

21.34

446.15

11.90

Plywood Waste Cyclone
(SRC2)

15.54

293.15

Chipper
(SRC3)

Cyclone

14.33

Powerhouse Dust Cyclone
(SRC4)

w

E1 pMo

E1 PM2.5

E2 PM10

E2 PM2.5

1.63

1.76E-01

1.15E-01

1.82E-01

1.19E-01

5.10

1.83

4.58E-01

2.29E-01

4.04E-01

2.02E-01

283.15

5.10

1.22

1.61E-01

8.06E-02

1.61E-01

8.06E-02

9.75

278.15

6.10

0.76

7.56E-02

3.76E-02

7.56E-02

3.76E-02

LVL Baghouse (SRC5)

3.05

293.15

39.70

0.76

8.19E-01

0.00E+00 8.19E-01

0.00E+00

Sanderdust
(SRC6)

2.44

293.15

35.94

0.91

1.08E+00 0.00E+00 1.08E+00 0.00E+00

Dryer 3 Cooling (SRC7)

9.75

306.85

21.75

0.37

8.04E-02

2.63E-02

8.04E-02

5.01E-02

Dryer 3 Heating (SRC8)

9.75

438.15

16.50

0.17

1.53E-01

5.01E-02

1.53E-01

2.63E-02

Plywood
(SRC9)

Trim

Cycline

15.85

293.15

6.09

0.91

1.07E-01

5.38E-02

1.07E-01

5.38E-02

Dryer 4
(SRC10)

Bypass

North

13.41

369.65

13.58

0.81

9.83E-02

3.22E-02

9.83E-02

3.22E-02

Dryer 4
(SRC10)

Bypass

Mid

13.41

448.65

3.10

0.81

9.83E-02

3.22E-02

9.83E-02

3.22E-02

South

13.41

448.65

10.04

0.81

2.01E-01

6.60E-02

2.01E-01

6.60E-02

Source

Dryer

4

Fines

Baghouse

Bypass

113

(SRC10)
Dryer 4 Cooling (SRC13)

12.8

306.85

0.19

1.25

8.06E-03

2.64E-03

8.06E-03

2.64E-03

Dryer 4 Cooling (SRC14)

12.8

306.85

0.19

1.25

8.06E-03

2.64E-03

8.06E-03

2.64E-03

Plywood
(SRC15)

9.75

298.15

0.27

3.05

1.13E-02

0.00E+00 1.13E-02

0.00E+00

6.1

298.15

8.14

2.44

4.54E-02

0.00E+00 4.91E-02

0.00E+00

Press

Vent

LVL Press Vent (SRC16)

114

6.4.2 Line Sources
Line sources are those sources that emit along a fixed line. Emissions are
modeled as puffs of pollutant released at equally distributed points along the line
segment. Traffic emissions from Highway 95 and the Transcanada Highway, as well
as railway locomotive emissions were mapped as line sources for CALPUFF
modeling.

The highway segments followed closely Highway 1 and Highway 95

indicated in Figure 4.2. The railway segments run alongside the highways.

6.4.2a Highway Traffic Emissions
Highway 95 and the Transcanada Highway were mapped as line sources from the
BCMOE's

air

emissions

map

(accessed

electronically

at:

http://srmapps.gov.bc.ca/apps/aei/). The emissions from the emissions inventory are
derived from the US EPA model M0BILE6. Motor vehicle emissions in M0BILE6
are estimated as the product of the number of vehicle kilometers traveled (VkmT) in
the airshed distributed into vehicle categories and a corresponding categorical
emission factor determined by the model. Highway VkmTs were estimated from
National Parks traffic data near Golden. The annual emissions from highway vehicle
sources from the emissions inventory were apportioned to each line segment based
upon the length of the road. Each highway line source segment (following paths of
Highway 95 and Highway 1 in Figure 5-1) was assigned an emission rate in metric
tons per year (Table 6-3). The emission rates were adjusted diurnally in the same
manner as local traffic and road dust. (Section 6.3.3b)
considered

as

line

sources

as

they

are

Local roads were not

generally

confined

to

the

residential/commercial areas of the Town of Golden. It was determined that given
1 1 5

the emissions information available local traffic emissions would be modelled as an
area source.
Table 6-3

Summary of highway segment emissions. The curved road path was closely
followed using multiple straight line segments. Note: the southern most
segment starts at the southern most point of Highway 95 in the modeling
domain and the northern most point ends at the Town of Golden where it
intersects with Highway 1. The eastern most point of Highway 1 starts at the
eastern edge of the domain and the western most point ends at the western
edge of the domain.
Road (Segments)
Highway 95
1 (Southern Most)
2
3
4
5
6
7
8
9
10
11 (Northern Most)

Distance (m) |
1416
1192
3479
1064
2330
1926
2432
1797
705
835
1097

Highway 1
12 (Eastern Most)
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39 (Western Most)

1959
2791
608
452
513
450
770
627
2239
1390
2636
2224
1537
1468
1052
384
302
899
1090
1423
5174
514
2102
4094
7459
2447
1295
2546

Yearly Emissions (tons)
PM2.5
PM10
0.28
0.24
0.24
0.20
0.70
0.59
0.21
0.18
0.47
0.39
0.33
0.39
0.49
0.41
0.36
0.30
0.14
0.12
0.17
0.14
0.22
0.19

0.39
0.56
0.12
0.09
0.10
0.09
0.16
0.13
0.45
0.28
0.53
0.45
0.31
0.30
0.21
0.08
0.06
0.18
0.22
0.29
1.04
0.10
0.42
0.82
1.50
0.49
0.26
0.51

0.33
0.47
0.10
0.08
0.09
0.08
0.13
0.11
0.38
0.23
0.45
0.38
0.26
0.25
0.18
0.06
0.05
0.15
0.18
0.24
0.87
0.09
0.36
0.69
1.26
0.41
0.22
0.43

116

6.4.2b Railway Locomotive Emissions
Railway lines in the modeling domain were mapped as line sources from BC MOE's
air emissions map. The annual emissions from railway locomotive sources were
apportioned to each line segment based upon the length of the track, assigning each
rail line source segment an emissions rate in metric tons per year (Table 6-4).
Emissions from the rail switching yard in Golden were estimated as an area source
(Section 6.4.3f)

117

Table 6-4

Summary of rail line emissions. The curved railway path was followed as
closely as possible using multiple straight-line segments. The southern,
northern, eastern and western most points refer to the point where the railway
meets the respective edge of the modeling domain.
Yearly Emissions (tons)
Railway (Segments)
1 (Southern Most)
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23 (Northern Most)

Distance (m) |
1620
8448
1229
2110
712
924
1195
621
1328
1315
1907
2816
944
3582
2935
495
1613
4397
2977
497
5690
2089
1165

24 (Eastern Most)
25
26
27
28
29
30
31
32 (Western Most)

3063
3956
1053
2042
3901
3589
1104
1536
2973

PM10

PM2.5

0.23
1.18
0.17
0.29
0.10
0.13
0.17
0.09
0.19
0.18
0.27
0.39
0.13
0.50
0.41
0.07
0.23
0.61
0.42
0.07
0.80
0.29
0.16

0.21
1.08
0.16
0.27
0.09
0.12
0.15
0.08
0.17
0.17
0.24
0.36
0.12
0.46
0.38
0.06
0.21
0.56
0.38
0.06
0.73
0.27
0.15

0.43
0.55
0.15
0.29
0.55
0.50
0.15
0.21
0.42

0.39
0.51
0.13
0.26
0.50
0.46
0.14
0.20
0.38

6.4.3 Area Sources
Area sources emit from an area rather than a distinct point or line. Generally,
they are a compilation of point sources too numerous to characterize individually. In
CALPUFF, area sources are represented by puffs released from multiple,
equidistant points within a defined polygon area. Each area is assigned an emission
rate.

118

From a local landuse map provided by the Town of Golden areas of the town
were appropriately chosen that represented the populated areas of the Town that
would emit each area source. The Town was divided into four polygons of the major
populated areas: South Town; North Town; Kicking Horse; Transcanada.

These

areas were assigned emission rates for each of the area sources described below
(Figure 6-2, Table 6-5).
Modeling during short episodes requires that diurnal variations in emission
rates be characterized to achieve an accurate picture of the emission scheme. Initial
modeling of area sources without diurnal variation resulted in area source estimates
that far exceeded observed PM levels. It is difficult, however, to estimate the diurnal
variation of PM.

Certainly, PM10 and PM2.5 ambient levels have distinct patterns

corresponding to the time of the day (Figure 4-1 and 4-2).

It is difficult to

characterize whether meteorology or changes in emissions cause the consistent rise
and fall of PM levels. Emission from vehicles rise during the times immediately
before and after the "workday" and space heating emissions would also be higher in
the morning and evenings when residents are more likely to be at home.
Conversely, wind speeds and mixing heights are generally higher during the day and
this may account for the drop in PM concentrations during the day. Likely, it is a mix
of the two, emission changes and changes in meteorology, which leads to the
distinct diurnal pattern seen in Golden.

With CALPUFF it is possible to model

diurnal variation in emission rates and the methods below describe the scaling
factors used for the appropriate area sources.

119

0
Figure 6-2

1
2
Kilometres

3

Location of four areas defined for area source emissions in CALPUFF.

Table 6-5
Area source emission rates for Episodes 1 and 2. All areas were assumed to
emit at the same rate.

Source
SPCHT
LTRFC
CONS
BURN
RDUST

Emission Rates (tons/m2/year)
E1 PM10 E1 PM2.5 E2 PM10 E2 PM2.5
9.74E-06 9.74E-06
2.25E-06
2.25E-06
1.26E-06
9.93E-07
1.26E-06 9.93E-07
2.55E-06
1.31E-06
2.55E-06
1.31E-06
1.00E-06
9.10E-07
1.00E-06 9.10E-07
4.90E-06
1.04E-06
3.92E-06
8.32E-07

Note: SPCHT = Space Heating, LTRFC = Local Traffic, CONS = Construction,
BURN =Burning, RDUST = Road Dust

120

6.4.3a Space heating emissions (SPCHT)

Space heating emissions encompass PM emitted from appliances used to
heat homes and businesses. This includes emissions from appliances burning all
types of fuels. Most space heating PM emissions in Golden come from woodburning appliances. Space heating emission rates were originally estimated using
the data provided in the El. The El estimates were determined through a woodstove
survey conducted by BC MOE (2004). Initial model runs showed the emission rate
estimates to be much too high. Rates were adjusted based upon receptor modeling
conducted in Golden (see Section 6-3 for details on rate adjustments). Final space
heating emission rates are provided in Table 6-5.
Space heating has a distinct diurnal pattern. Typically, homes are heated in
the mornings and evenings when residents are more likely to be at home. During the
province-wide woodstove survey, residents were asked when they add wood to their
woodstoves. Figure 6-4 displays the data specific to Golden.
Since space heating PM emissions in Golden are almost solely based on
woodstove emissions, rates were diurnally adjusted according to the results of the
survey question (Table 6-6).

121

When do users add wood?

c

4>
U
a>
a.

<o

9>-

\v

&

<*•

jv

&

.<5
<b

<0-

<o

4

\

\

9>-

J>

<o

J? ^
/ J

Time of day
Figure 6-3

Summary of Golden resident's answers to the question "when do you add
wood?" to their woodstoves asked in the BCMOE woodstove survey (BClVlOE
2004).

122

Table 6-6

Diurnal variation of space heating emissions by hour of day.
Hour of the Day
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23

Diurnal Scaling Factor
0.536
0.536
0.536
0.536
0.536
0.536
2.238
2.238
2.238
0.318
0.318
0.318
0.493
0.493
0.493
1.060
1.060
1.060
1.409
1.409
1.409
1.409
1.409
1.409

6.4.3b Road dust emissions (RDUST)

The El estimated road dust as a major source of PM. Road dust results from
the grinding of granular material deposited on road surfaces, usually from road
traction material, road construction activities and material trackout from gravel side
roads. It is made airborne by passing vehicles and wind.
As mentioned in the El, accurate estimates of road dust emissions in
inventories are quite difficult to achieve and have a high potential to be erroneous. It
was determined that differences in road characteristics and the absence of local silttesting of dirt, dust and other debris on local highways and roads would severely
123

limit the accuracy of modeling road dust emissions as line sources. Instead, road
dust emissions were modelled as a source emitting from the local polygon area
sources defined above (Figure 6-2). Road dust emissions are not constant and will
generally be suppressed by rainfall and even further limited by snow and ice on
roadways. However, since the proper precipitation input information for CALPUFF
was not available and precipitation occurred briefly on only 3 days during the
episodes, wet removal was not modeled.
A SENES consultant study (SENES 2000) stated that fugitive dust sources
would likely not travel large distances from the roadway under most atmospheric
conditions. Therefore road dust emissions from highway sources outside Golden
were not captured in the El emission estimates and were subsequently not modeled
during this study.
Road dust emission rates (Table 6-5) were varied by diurnal scaling factors
(Table 6-7). Since there were no traffic data available from Golden, data from a
Prince George air quality project (personal comm. John Spagnol) was used as an
estimate. Prince George is a community in the interior of BC that is largely supported
by the wood products industry and work traffic likely follows the shift work used in
this industry. Golden and Prince George also have a large, busy highway running
through their respective communities. Although the populations are different it was
assumed for modeling purposes that the diurnal traffic patterns in Golden were
similar to those quantified for Prince George.

124

Table 6-7

Diurnal variation of road dust emissions by hour of day.

Hour of the Day
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23

Diurnal Scaling Factor
0.22
0.17
0.14
0.17
0.29
0.55
0.86
1.06
1.27
1.27
1.42
1.61
1.68
1.73
1.80
1.92
1.85
1.58
1.20
0.98
0.79
0.62
0.48
0.34

6.4.3c Local traffic emissions (LTRFC)

Local traffic emissions are from local vehicles of all sizes traveling within the
Town of Golden.

The El separated the local traffic emissions from the highway

traffic emissions. Emission estimates in the El were determined by using the US
EPA MOBILE 6 program for mobile source emissions. In CALPUFF, local traffic
emissions were modeled as an area source in the populated areas of Golden rather
than multiple line sources mapping the entire town road system. Specific information
on traffic counts, speed and vehicle types would be necessary to provide more
detailed modeling of local traffic emissions on a specific street-by-street basis. The
area source emission rates based on the estimates provided in the emissions
125

inventory for local traffic are described in Table 6-5. The emission rates from the
emissions inventory are derived from the US EPA model MOBILE6 in the same
manner as described for highway line emissions (Section 6.3.2a). Local traffic
emissions were varied diurnally using the same emission rate scaling factors as the
road dust emissions (Table 6-7).
6.4.3d Construction emissions
Emissions from construction operations in Golden are based on estimates in
the emissions inventory related to residential building permits and commercial
project valuation. Emission rates for this source were determined based on the
annual PM emissions in the emission inventory (Table 6-5). Construction emissions
were not modeled during the night-time, non-working hours (6 p.m - 6 a.m).
6.4.3e Burning emissions
Burning emissions include estimated emissions from backyard burning.

A

major concern of residents in Golden is the effect of prescribed burning operations in
the Columbia River Valley. These operations are not included in the CALPUFF
modeling because during the episodes the BC Ministry of Forests permitted no
prescribed burns. Miscellaneous burning emission rates were determined from the
emissions inventory (Table 6-5).
6.4.3f Railway yard emissions
In addition to the railway line emissions, the El estimated yard emissions
specific to the large CPR switching yard to the south of Golden. These emissions
126

were modeled as an area source emitting from the area of the CPR switching yard.
El estimated emissions were apportioned equally based on the area (m2) of the
switching yard. It was assumed that each area of the switching yard emits at the
same rate. The area dimensions and emission rates are described in Table 6-8.
Table 6-8

Summary of railway yard area and emission rates.
Polygon Dimensions

Northwest
X
Y
501.92
5681.98

Railway Yard
PM10 Emission Rate
(tons/m2/year)
PM2.5 Emission Rate
(tons/m2/year)

Southwest
X
Y
503.47 5681.27

Northeast
X
Y
502.25 5682.04

Southeast
X
Y
503.27 5681.52

9.25E-06
8.55E-06

6.4.3g LP Yard Emissions

Yard emissions from LP result from the operation of heavy duty machinery
and trucks. PM emissions from this source would largely be fugitive dust emissions.
LP yard emission estimates were provided by the plant environmental manager,
Mike Brygger.

They were modeled as an area source emitting from the area

covered by LP operations in Golden. The area dimensions and emission rates are
described in Table 6-9.

127

Table 6-9

Summary of LP yard area and emission rates. Split into north and south to
define more accurate area source polygon.
Polygon Dimensions

LP Yard
North
South
PM10 Emission Rate
(tons/m2/year)
PM2.5 Emission Rate
(tons/m2/year)

6.5

Northwest
Southwest
Northeast
Southeast
X
YXYXYXY
500.93
5685.26 500.77 5684.70 501.16 5685.32 501.34 5684.98
500.77
5684.67 501.18 5683.96 501.36 5684.95 501.70 5684.27

9.40E-05
1.54E-05

EMISSION RATE ADJUSTMENTS
Preliminary model runs indicated that the model was drastically over-predicting

PM concentrations. However, the model was reasonable at predicting the timing of
highs and lows in the observed PM concentrations. This suggested that
meteorological parameters set by CALMET captured a reasonable dispersal pattern.
Indeed, sensitivity tests of CALMET parameters yielded little improvement in the
predicted concentrations. It was hypothesized that the emission rate estimates used
in the original run, based upon the El, were inaccurate. New estimates of emission
rates were necessary to improve the model accuracy.
There are several reasons why the El may not be accurate or adequate for
modeling purposes. El's are estimates of emissions usually derived from emissions
factors and some measure of the source's emitting activity.

In many cases the

emission factors may not be applicable to the conditions in the area or there may not
be an accurate way of measuring the source's activity.

Therefore, emission

inventories have inherent uncertainty based on the methods used to obtain
estimates, which is further explained in the El.
128

Other information was available to attempt to create more accurate emission
rates for the model. Postive Matrix Factorization (PMF) analysis, a type of receptor
modeling was performed on PM2.5 samples collected during the period of November
11, 2005 through August 15, 2006 in Golden. The analysis was performed by the
Southern Ontario Centre for Atmospheric Aerosol Research at the University of
Toronto.

Receptor modeling uses collected samples from a monitoring site to

distinguish relative contributions of different emission sources. This method uses
temporal variation in the speciation of PM2.5, marker elements and meteorological
data to identify specific emission sources (Evans and Jeong 2007).
The solution from the receptor modeling in Golden revealed seven factors
contributing to PM. The factors identified were: Road Salt, Sulphate, Residential
Wood Burning, Wood Processing, Crustal Material, Traffic and Residential Winter
Heating. Additionally, the analysis provided a factor breakdown of the average PM2.5
concentrations in each season (Figure 6-4).
The preliminary CALPUFF runs determined that two area sources, space
heating and road dust were unreasonably high, severely inflating overall predicted
PM concentrations. The receptor modeling provided much lower averages for the
daily contributions from road dust and space heating. To achieve more accurate
emission rates, the information in Figure 6-4 regarding daily contributions to the
average PM2.5 concentrations was used to reduce the original emission rates. It was
assumed that the winter heating factor and the residential wood burning factor would
represent the space heating emissions and that the road dust would be represented
by the crustal factor as suggested by Evans and Jeong (2007).

Predicted daily
129

average concentrations were determined from the modeled TOWN receptor that
represented the location used to collect the receptor modeling data. The original
emission rates were then reduced based on the percent reduction between the
predicted daily average concentrations from preliminary model runs and those
determined from the receptor modeling. The receptor modeling data were taken
from the episode specific days where data were collected for PMF analysis. Road
dust emission rates were reduced by 91% and 93% for episode 1 and 2,
respectively. Space heating emission rates were reduced by 65% and 92% for
episode 1 and 2, respectively. These seemingly large reductions reflect the
uncertainty and lack of seasonality in the El estimates for these sources.

20
• Winter Heating

18

• Traffic
• Crustal

16

• Wood Processing

el

a

14

• Wood Burning

.•

• Sulphate

1-

• Road Salt
'5

10

Winter
Figure 6-4

Spring

Summer

Fall

Factor breakdown of PMF analysis for PM2 5 in Golden (Evans and Jeong 2007).

130

Table 6-10 shows that in general the results of CALPUFF with the two new
emission rates were better than the results with the original emission rates derived
from the El, when compared to the receptor modeling.
Table 6-10

Emission inventory (El), receptor modeling and CALPUFF model predicted
percentages (based on the adjusted El emission rates) of total estimated PM in
each case is provided for (a) episode 1, and (b) episode 2. Emission inventory
results are from those at the TOWN station. LP is used as a comparison
because the emission estimates were the most reliable and complete.
CALPUFF results have been adjusted by comparison with the receptor
modeling.

(a)

6.6

Emission Source
Space heating
Road dust
LP

El
45.1
20.1
9.6

Receptor
75.0
8.3
2.8

CALPUFF
62.5
7.2
1.5

Emission Source
Space heating
Road dust
LP

El
45.1
20.1
9.6

Receptor
48.0
17.0
19.0

CALPUFF
35.9
12.6
3.3

EXAMPLE OF CALPUFF EVENING WIND FLOW AND HIGH PM EVENT
Modeled

meteorological

conditions

that

typically

concentrations in the evening are displayed in Figure 6-5.

lead

to

high

PM

In addition to wind

vectors and PM concentrations other modeled meteorological conditions are
provided (Table 6-11) to show the major factors affecting dispersion in this
CALPUFF model run. In the diagrams, Golden is located at the convergence of the
valley to the east and the main river valley. The wind vectors are proportional in size
to the wind speed. Note the increase in PM as the wind speed across the grid slows,
the air stagnates and the flow reverses. The wind speed modeled at the TOWN
station (Table 6-11 remains quite low, but it is evident that winds are slowing across

the modeling domain. As the wind speed slows, the wind direction changes from
southerly to northerly at the TOWN station (Table 6-11). During this time, the PM10
levels rise. Following the rise in PM, the mixing height remains low and the winds
come from the north throughout the night maintaining the high PM concentrations
until the morning.

132

(a)
Dec 10, 2005
18:00 LST(UTC-0700)
Wind, Cone (ug/m"3)
117.2W

117.1Wi

117i0W

i 116.9V

116.8

5705

14500
13900
13700
3500
3300
13100
12900
2700
J2500
]2300
-i2100
] 1900
1700
1500
1300
1100
J900
J700
1500
(300
'100

5 5685

'-1

Terrain (m)

490

495

500

505

510

515

520

UTM East (km)

133

(b)
Dec 10,2005
19:00 LST(UTC-0700)
Wind, Cone (ug/m"3)
117.2W

14500
[3900
13700
13500
13300
13100
J 2900
12700
2500
' 2300
^2100
11900
1700
1500
1300
1100
900
1700
500
300
J100

5 5685

'-1

Terrain (m)

500

505

UTM East (km)

134

(C)
Dec 10, 2005
20:00 LST(UTC-0700)
Wind, Cone (ug/m"3)
117.1W1

11

5700
[4500
[3900
[3700
[3500
3300
13100
2900
12700
12500
12300
^ 2100
1900
1700
1500
1300
1100
]900
1700
500
|300
j 100

5695

5690

•co
5685

5680

5675

'-1

Terrain (m)
5670

490

495

500

505

510

515

520

UTM East (km)

135

(d)
Dec 10, 2005
21:00 LST(UTC-0700)
Wind, Cone (ug/m"3)
117.1Wi

117LOW

1 116.9W

i

116.8W

5700
4500
13900
13700
3500
13300
|3100
J 2900
j2700
12500
J2300
^2100
j 1900
1700
1500
1300
1100

5695

5690
-C

i

tl
o

z

2 5685

5680

1900
I700
1500
1300

5675

J 100
••-1

Terrain (m)
5670

490

495

500

505

510

515

520

UTM East (km)

136

(e)
Dec 10,2005
22:00 LST(UTC-0700)
Wind, Cone (ug/m**3)
117.2W

117.1W i

117LOW

M16.9W

116.8W

5705

5700
14500
13900
13700
13500
13300
13100
J 2900
2700
2500

5695

5690

J2300
12100
j 1900
1700
1500
1300
1100

•C

Co
Z

5685

5680-

1900
700
1500
300
J100

J

5675

'-1

Terrain (m)
5670

UTM East (km)

Figure 6-5

(a-e) Hourly CALPUFF ground level PM10 concentrations (ng/m3) from 18:00 to
22:00 on December 10, 2005, representing the entire modeling domain,
showing wind vectors and PM10 concentrations for 5 consecutive hours. Note
the increase in PM as the wind speed slows and the air stagnates around the
Town of Golden located at the convergence of the valley to the east and the
main river valley.

137

Table 6-11

PM levels, mixing heights, wind speed and wind direction at the TOWN station
surrounding the high PM event described in Figure 6-5.
pm 2 ,

Time
12/10/05 12:00
12/10/05 13:00
12/10/05 14:00
12/10/05 15:00
12/10/05 16:00
12/10/05 17:00
12/10/05 18:00
12/10/05 19:00
12/10/05 20:00
12/10/05 21:00
12/10/05 22:00
12/10/05 23:00
12/11/05 0:00
12/11/05 1:00
12/11/05 2:00
12/11/05 3:00
12/11/05 4:00
12/11/05 5:00
12/11/05 6:00

PM10 (pg/m3)
3.7
10.2
17.7
24.6
38.8
10.8
8.2
42.2
70.0
97.8
90.6
45.9
68.2
38.1
33.6
43.1
41.6
37.3
19.0

(MC|/m )

1.7
5.3
8.26
14.6
23.4
9.7
6.9
40.0
59.4
78.7
74.8
42.3
54.6
27.5
26.9
33.9
27.3
26.6
19.3

Mix hgt
(m)
500
353
378
358
29
40
47
27
27
26
27
31
27
40
27
27
33
26
31

Wspd (m/s)
0.85
0.49
0.58
0.48
0.65
0.95
1.25
0.51
0.46
0.36
0.48
0.68
0.56
0.77
0.49
0.49
0.66
0.29
0.68

Wdir
168
185
229
172
168
161
159
338
346
282
127
309
338
341
146
341
339
298
156

6.7 CALPUFF EVALUATION PROCEDURES
This section provides a statistical comparison of the model predicted PM with
the observed PM data collected at all three monitoring locations in Golden. It is
important to test the model's ability to predict PM concentrations to assess the
validity of the model as a tool for airshed management. The ability of the model to
predict the highest levels of PM concentrations is given special consideration.
In order to assess the performance of the model during each episode the
predicted

PM

concentrations

were

compared

with

the

observed

ambient

concentrations in Golden using a variety of approaches. Chang and Hanna (2004)
suggested that there is no single best method of assessing air quality model

138

performance. They recommended that a suite of different performance measures be
used.
The suite of performance measures used in our evaluation were used in a
recent evaluation of CALPUFF by Song et al. (2006) as suggested by Willmott
(1982) and Seigneur et al. (2000). These were applied in Barna and Gimson (2002),
Zhang et al. (2004) and Assimakopoulos (2005).

Our evaluation also contains

methods used in another BC modeling study by Levelton Consultants Ltd. in
Williams Lake (Schutte et al. 2005)
First, a direct comparison of the hourly and daily averages was performed.
Scatter plots were also created to gauge initial model performance and to visually
indicate when the modeled predictions are within a factor of two of the observed PM
values (Ross 1991).
Next, as a primary indicator of model performance the dimensionless
Normalized Root Mean Square Error (NRMSE) was calculated at each monitor for
the entire episode:
NRMSE = V(AVG(PR)-AVG(OB))2/(MAX(OBS)-MIN(PR))2
In addition, NRMSE was used to evaluate the top 10 predicted concentrations in an
effort to evaluate how the model predicted the peak PM concentrations during the
episode. A low NRMSE value indicates good model performance.
The fractional bias (Fb) was calculated for each monitoring station:
Fb = 2(OB-PR)/(OB+PR).
139

Fb is a performance measure used to determine if a model meets the minimum
performance standards of the USEPA (USEPA 1992).

As mentioned previously

there is no standard measure of "good" model performance. Fb was chosen in this
study as it is an accepted and often performed statistical evaluation of a model with
procedures outlined and endorsed by the USEPA.
In contrast to the application of the Fb statistic in the CALMET evaluation
(Section 5-2), the USEPA provides a procedure where Fb is calculated twice at each
monitoring station. First, Fb was calculated for the average, where OB and PR
represent the hourly concentrations of PM.

Secondly, Fb was calculated for the

standard deviations, where OB and PR represent the standard deviations of the
hourly concentrations of PM. The number of hourly concentrations included in the
test determines the stringency. It can be performed on all paired values during the
entire episode (most stringent), can assess only the higher values (top 25) of the
episode pair-wise (less stringent) and can also compare the highest 25 values un­
paired (least stringent).

All levels of stringency were tested, however the results

from only the most stringent Fb test are provided.
The two Fb statistics were plotted with bias of average on the x-axis and bias
of standard deviation on the y-axis. The closer the fractional bias is to the centre of
the plot (zero), the fewer tendencies it has towards bias. An acceptable model
(USEPA 1992) has Fb statistics within +/- 0.67 representing an over/under prediction
within a factor of two.

140

To compare model performance with other CALPUFF and dispersion
modeling studies a number of additional statistical performance measures were
applied. This included:
Fractional gross error (FGE) = (OB-PR) / Q.5(OB +PR)
Statistical measure of perfect model
Range of this statistic from other studies

0.0
0.29 - 0.99

Geometric mean bias (MG) = exp(ln OB - In PR)
Statistical measure of perfect model
Range of this statistic from other studies

1.0
1.01 - 1.83

Correlation coefficient (CORR) = (OB-AVG(OB))(PR-AVG(PR) / aOB * aPR
Statistical measure of perfect model
Range of this statistic from other studies

1.0
0.40 - 0.82

Willmott's index of agreement (D) =
1.0 - (OB-PR)2/((OB-AVG(OB))+(PRAVG(OB)))2
Statistical measure of perfect model
Range of this statistic from other studies

1.0
0.08 - 0.89

Fraction of predictions within factor of 2 of observations (FAC2)
Statistical measure of perfect model
Range of this statistic from other studies

100%
52% - 98%

Statistical tests were performed to test both hourly and daily (24-hour) model
predictions.
141

6.8

EPISODE 1 MODEL PERFORMANCE
The time series plots (Figure 6-6) show the detailed model output, however

investigation of the model performance measures is needed. Averaged over the
entire first episode, the model under-predicted at the CPR and GOLF stations and
over-predicted at the TOWN station for both PM10 and PM2.5 (Table 6-12). The CPR
station shows the best agreement averaged over the entire episode for both PM10
and PM .52

Table 6-12

Average observed and predicted values at all three monitoring stations during
Episode 1.

Pollutant

Value

Concentration (pcj/m3)
CPR
Town
Golf

PM2.5

Observed
Predicted

10.90
10.71

14.23
21.64

6.96
2.46

PM10

Observed
Predicted

16.78
14.87

21.10
30.61

11.79
3.68

The average over the entire episode does not show how the model is
capturing the daily or hourly observed concentrations. Since this study is interested
in short periods of elevated levels of PM, the focus is on the ability of CALPUFF to
predict hourly concentrations.
Scatterplots of predicted and observed PM concentrations are provided
(Figure 6-7 - 6-9). The CPR station has a more balanced scatter, with some grossly
over- and under-estimated hourly concentrations. The TOWN station shows much
fewer under-predictions, but has many more over-estimated hourly concentrations,
while the GOLF station shows the opposite.
142

(a)

PM10 modeled
PM10 observed
PM2.5 modeled
*—PM2.5 observed

jft

U
s
o*

<o

p
p
<\/
<o
•$?

£

o
p

p'

<o

o
P
P'
<C

§
<V
ip

#
£

£•

4y

o
p
o'
<o

<§>

<P
5

«y
&

§
<V
<o

p
p
o'
io

§
<V
<o

03

& &

£
<V

o
p
o'
<o
5
;s
£y

§
<V
<p

p
p
p'

io
£
<r
<v

143

(b)

PM10 modeled
PM10 observed
PM2.5 modeled
—*—PM2.5 observed

o 60

§

o"
£
%
&

<§>
<V

<§>
o'
<o
*•»

§

£v
M?
§

$

§

o'

#
$

$

<V

M?
s

§

o<p
<v

o
<v
<o
<V
iv

§

O'
<o
03

§

<V

M?
£
03
<v

§

o'
M?
$

$

§
Sf
M?
§
v
>
iv

§
<o
<v

144

(c)

30 n

PM10 modeled
PM10 observed

25

PM2.5 modeled
—x— PM2.5 observed

V

n

j

v

A

/

i

n

\ -K

i \

'i
i ki

A
\

vv W '

O 10

r . •

s&

/

.1.

;/ ;*

I

i

ni

Lvl
VMJf/

\/* ,y\f

§

£

§

§
<v

O"
<p
<v

£

<v'

X?
§

$>

8

§

§
o"

o"
M?
$

<p

$

M?

j£>

§
<V

&

&

8

O'
OD

<V
<o

§

O"
<p
V

§
<v
M?
$

8

o"
<o
*•»»
<V

145

(d)

Mixing Height
Wind Speed

Figure 6-6

Time series plots showing model predicted and observed PM10 and PM25 for episode 1 at (a) CPR station, (b) Town
station, (c) Golf station. Also included is (d) a time series of mixing height and wind speeds at the Town station.

146

(a)

(b)

30

40

Observed PM10 (MO^ri*)

Figure 6-7

Observed PU2.5 (paAri)

Scatterplots of predicted versus observed concentrations, paired in location
and time at the CPR station for (a) PM10, and (b) PM25 for Episode 1. The outer
black lines bound predicted concentrations within a factor of 2 of observed
values.

(a)

(b)

120 -

40

60

80

100

Observed PM10 (pgftn*)

Figure 6-8

120

140

160

ObMrvtd PM2.5(iigftir)

Scatterplots of predicted versus observed concentrations, paired in location
and time at the TOWN station for (a) PM10, and (b) PM25 for Episode 1. The
outer black bound predicted concentrations within a factor of 2 of observed
values.

147

(a)

(b)

15

20

Observed PMlO(jjg/'ms)

Figure 6-9

Observed PM2.5

Scatterplots of predicted versus observed concentrations, paired in location
and time at the GOLF station for (a) PM10, and (b) PM25 for Episode 1. The outer
black lines bound predicted concentrations within a factor of 2 of observed
values.

The NRMSE ranges between stations with the lowest values occurring at the
TOWN station for both PM10 and PM2.5 (Table 6-13). PM10 errors are larger at the
TOWN and GOLF stations and slightly less at the CPR station. In general, errors
increase when the daily averages are compared. The NRMSE increases when only
the top ten observed concentrations are considered (Table 6-14).
Table 6-13

Normalised Root Mean Square Error (NRMSE) values for the entire first
episode.

Pollutant

PM2.S
PM10

Averaging
Period
1-hour
24-hour
1-hour
24-hour

CPR
0.474
0.404
0.415
0.424

NRMSE
Town
0.298
0.549
0.321
0.634

Golf
0.368
0.756
0.486
1.061

148

Table 6-14

Normalised Root Mean Square Error (NRMSE) values for the top ten observed
values during the first episode.

Pollutant

Averaaina
Period

PM2S

1-hour

PM10

1-hour

CPR

NRMSE
Town

GOLF

0.572

0.459

0.590

0,636

0.434

0.586

Fractional Bias is a performance measure used to determine if a model meets
the minimum performance standard by the USEPA (USEPA 1992). By USEPA
standards the model was performing adequately at the CPR station for all measures
(PMio-1hr, PM2.5-1hr, PMi0-24hr, PM2.5-24hr) and the TOWN station for two

measures (PMio-24hr, PM2.5-24hr) (Figure 6-10).

The model did not perform

adequately for any measures at the GOLF station.

- Series2
•

PM10 - 24hr

• PM2.5 - 24hr

o

• PM10- 1hr

H
<
>
UJ
Q
O
<E
<
Q
Z
<

I—
(0
u.
O
«
<
S

1.34

•

-2

PM2.5-1hr

-1.34

-1.34 J-

-2.G1
BIAS OF AVCTAGE

Figure 6-10

Fractional bias: Episode 1 using entire episode (most stringent) US EPA
method. The inner box represents acceptable model performance measures
(US EPA 1992).

149

Additional performance measures described in section 6.7 are provided
(Table 6-15, 6-16).

These will be evaluated comparatively with other studies

(Section 6.10).
Table 6-15

CALPUFF model performance measures for PM10 during Episode 1.

1-hour

FGE
MG
R
D
FAC2
24-hour

R
D
FAC2

Table 6-16

CPR
0.12
1.13
0.21
0.87
46

TOWN
0.37
0.69
0.59
0.87
56

GOLF
1.05
3.20
0.24
0.75
21

CPR
0.34
0.98
83

TOWN
0.69
0.96
100

GOLF
0.20
0.77
33

CALPUFF model performance measures for PM25 during Episode 1.

1-hour

FGE
MG
R
D
FAC2
24-hour

R
D
FAC2

CPR
0.02
1.02
0.15
0.84
40

TOWN
0.41
0.66
0.60
0.86
59

GOLF
0.96
2.83
0.36
0.81
22

CPR
0.31
0.97
67

TOWN
0.77
0.96
83

GOLF
0.60
0.82
17

150

6.9

EPISODE 2 MODEL PERFORMANCE
The time series plots (Figure 6-11) show the detailed model output, however

investigation of the model performance measures during episode 2 is needed.
Averaged over the entire second episode, the model over-predicted at the CPR and
TOWN stations and under-predicted at the GOLF station for PM25 (Table 6-17).
PM10 concentrations were under-estimated at all three stations. The CPR station
and the TOWN station shows the best agreement averaged over the entire episode
for PM10 and PM2.5, respectively.
Scatterplots of predicted and observed PM concentrations are provided
(Figure 6-12 - 6-14). The results reconfirm the predictions shown in the overall
episode averages. There are still many times when the model is grossly over- or
under-estimating the observed values.
Table 6-17

Average observed and predicted values at all three monitoring stations during
Episode 2.

Pollutant

Value

PM .5

Monitor
Predicted
Monitor
Predicted

2

PM10

Concentration (pp/m3)
CPR
Town
Golf
6.91
9.96
1.76
13.87
0.28
8.09
23.95
35.36
7.31
16.00
28.10
1.05

151

(a)
140

PM10 modeled
120

PM10 observed
PM2.5 modeled

100

PM2.5 observed

E
CD
c
o
+*

S

4^

c

d)
o
c
o

o

I
O O O O O O O
o o o o o o o
o' (\j o" <V 6' <v" o"
to to
Go *°* to
J P O _ 0 0 0 0 0 0

< y

O

A /

O

s> i/ p i/
# ^ ^ ^
O
*-•

A /

O

^

O

O O O O O O O O O O O O O O O O O O O C 5
o o o o o o o o o o o o o o o o o o o O
<v' 6" cV 6" <v o' «v o' «v o' <V o' <v" c> <V c>" <V 6' <V o'
*""• CD
Co
CO to
^ Co ^ CO
CO
CO to
*- Co
to o to
o to
o
. 0 0 . S 5 0 0 o o o o o o o o o o o o o o
^

O

A

/

O

A

/

O

A

/

P ^ P ^ ^ ^ ^ ^ P V P
^ ^ <v PV ^
^
^ £ ^
i~i>:
>»
»-;
»»c

O

< y p

^

n

°

^

P

P

^

P ^ P
p i> p i) P #V P
^ # cT ^ ^ 4>U ^
^ 4> C^
»>.
>c^
»<
<\T
cV

152

(b)

PM10 modeled
PM10 observed
PM2.5 rrjpdeled
sr 140

PM2.5 observed

a. 120
s 100
d>

80

O

60

1

P

p° S P° 8 f S s° 8 I 8 § S P « 9° « ? S ° 8
,Q
^
S' o ^/' o <V o cy- 6 cy- o' cv' o' cv' <6" cv' o'

S § 8 S S §
cv" 6' <v o' cv' o"

0/

<V

153

(C)
45

PM10 modeled

40

|-PM10 observed

35
E
o>

PM2.5 modeled
PM2.5 observed

30

•L

O
S

25

•4-1

c
©
o
c
o

"ft

o
Mi:-'!/ :]

p

O

p

O

p

O

p

O'
cy P
cy O'
cv' O"
cp
_
ip .
ip to
,S> to
O to
Ci
cp
O to

O

<\i

O"
Co
O to

C\j

p

O

p

O

p

O

p

O

p

O

p

Q

OO

p

o

o

O'
cv" o'
co
^
co
Q too top to

O

cV

P'
<t>

p

o

OO

OO

o

p

O

o

O

p

O

o

O

p

OO

o

p

OO

o

o

o

c

p

p

cy P
cv' P
cy
O'
cy
O'
cy
O"
<v
O'
*- to
co to
""to
^ to
co
P to
P
co
O to
O to
O to
O
O ,p
OO
O ,£>
OO
OO
OO

•0/
C5 -0/ P ^W O A/v p <y p A/
-0/ _ 4V ^
v
a>
^
of
^ cy
o>v ^
o7 ^
~S A7 ;S
^
CV V cv ^
\v <^v ^
\v <v
<y cy
cv
, <y ^
S
*
C
V
v
c
v
cy
cy
<v ^ ^
o fy s A/ p fy oVcy .p cy ^ «sy ^
s> iy © <y -P iv vO
^ 3 v
•$> cy ;Sr
>
& ,o £ ^>v ^
£ < o '^
P c li'
o 'to- P / >V ' P/Vc s^- $ CM
- ' J®
Cv
o cv ^
>--v ' ' ry>
cy
&
^
cv .
o
+**
C\j
C\j

$

154

(d)

3500 N

3.5

— Mixing Height
•—Wind Speed

40
g

g>
"a>

x

o>

c

x

i

Figure 6-11

TJ
$
OT
TJ
c

Time series plots showing model predicted and observed PM10 and PM2.5 for episode 2 at (a) CPR station, (b) Town
station, (c) Golf station. Also included is (d) a time series of mixing height and wind speeds at the Town station
during episode 2.
155

(a)

(b)

Observed PM10 (pg/ms)

Figure 6-12

(a)

Observed PM2.5 (jjgfan

Scatterplots of predicted versus observed concentrations, paired in location
and time at the CPR station for (a) PM10, and (b) PM2 5 for Episode 2. The outer
black bound predicted concentrations within a factor of 2 of observed values.

(b)

Observed PM2.S (pg/m*)

Figure 6-13

Scatterplots of predicted versus observed concentrations, paired in location
and time at the TOWN station for (a) PM10, and (b) PM25 for Episode 2. The
outer black bound predicted concentrations within a factor of 2 of observed
values.

156

(a)

(b)

C 20

Observed PM2.S (ug/rn

Obs*rv«d PM10 (tjg/m®)

Figure 6-14

Scatterplots of predicted versus observed concentrations, paired in location
and time at the GOLF station for (a) PM10, and (b) PM2 5 for Episode 2. The outer
black bound predicted concentrations within a factor of 2 of observed values.

The NRMSE shows a range between stations with the lowest values
occurring at the TOWN station for PM2.5 and at the CPR station for PM10 (Table 618). In general, errors are higher for PM2.5 in the hourly comparison. Errors also
increase when the daily averages are compared. The NRMSE increases when only
the top ten observed concentrations are considered (Table 6-19).
Table 6-18

Normalised Root Mean Square Error (NRMSE) values for the entire second
episode.

Pollutant
PM .5
2

PM10

Averaging
Period
1-hour
24-hour
1-hour
24-hour

CPR
0.340
0.576
0.181
0.454

NRMSE
Town
Golf
0.285
0.239
0.820
1.677
0.187
0.198
0.303
2.282

157

Table 6-19

Normalised Root Mean Square Error (NRMSE) values for the top ten observed
values during the second episode.

Pollutant
PM?S
PM10

Averaging
Period
1-hour
1-hour

NRMSE
Town
0.521
0.664

CPR
0.713
0.627

Golf
0.799
0.488

By USEPA standards regarding Fb the model was performing adequately at
the TOWN station for all measures (PMi0-1hr, PM2.5-1hr, PM10-24hr, PM2.5-24hr) and
the CPR station for three measures (PM10-1hr, PM2.5-1hr, PM2.5-24hr) (Figure 6-15).
The model did not perform adequately for any measures at the GOLF station.

Series2

<

A

PM10- 24hr

•

FM2.5 - 24hr

•

PM10 - 1hr

•

PM2.5-1hr

1.34

A

-&SJ-

s

A

Q
O
<r

<
o
z
< -2 01
</>
u.

o
in

•

• •

&QQ-1.34

-0 67

• 0 DO

0. 57

1.34

-QSf-

<

m
-1.34

BIAS OF AVERAGE

Figure 6-15

Fractional bias: Episode 2 using entire episode (most stringent) USEPA
method. The inner box represents acceptable model performance measures
(USEPA 1992).

158

Additional performance measures described in section 6.7 are provided
(Table 6-20, 6-21).

These will be evaluated comparatively with other studies

(Section 6.10).
Table 6-20

Table 6-21

CALPUFF model performance measures for PM10 during Episode 2.

1-hour
FGE
MG
R
D
FAC2

CPR
0.40
1.50
0.18
0.84
44

TOWN
0.23
1.26
0.15
0.86
50

GOLF
1.50
6.98
0.27
0.61
5

24-hour
R
D
FAC2

CPR
-0.56
0.92
69

TOWN
-0.20
0.97
75

GOLF
0.31
-1.62
0

CALPUFF model performance measures for PM2.5 during Episode 2.

1-hour
FGE
MG
R
D
FAC2

CPR
-0.16
0.85
0.25
0.84
45

TOWN
0.33
0.72
0.24
0.82
45

GOLF
1.45
6.28
0.21
0.58
15

24-hour
CORR
D
FAC2

CPR
-0.41
0.95
75

TOWN
-0.11
0.89
88

GOLF
0.63
-3.21
0

6.10 OVERALL MODEL PERFORMANCE
Before examining the CALPUFF results it was necessary to evaluate the
model's performance. Assessing the validity of the model predictions helps put into
context the following chapters analyzing CALPUFF's characterization of PM
episodes in Golden.

159

As mentioned in Section 4.8, the Fractional Bias (Fb) is a performance
measure used to determine if a model meets the minimum performance standard by
the US EPA (US EPA 1992). By US EPA standards the model is performing
adequately during both episodes in estimating hourly PM10 and PM2.5, as well as
daily PM2.5, concentrations at the CPR station.

At the TOWN station the model

performs adequately for only the daily estimations of PM10 and PM2.5 concentrations.
The model is far below acceptable standards for all estimations at the GOLF station.
The other performance measures are presented as a means to compare with
other studies. Song et al. (2006) reported FGE values between 0.29 and 0.59,
stating that these values indicated an acceptable degree of model agreement with
daily average observational data. FGE values for hourly predicted concentrations in
our study compared favorably with these values at the CPR and TOWN stations, but
high FGE values are recorded at the GOLF station.

MG values for Song et al.

(2006) ranged from 1.01 to 1.83 comparing well with values at the CPR and TOWN
stations that ranged from 0.66 to 1.50.

Song et al. (2006) reported high correlation

coefficients ranging from 0.74 to 0.82. CALPUFF predictions in Golden do not show
nearly as high correlation coefficients ranging from 0.15 to 0.60. Song et al. (2006)
reported Willmott's index of agreement (D) values ranging from 0.83 to 0.89. Our
study compared well at the CPR and TOWN stations with values ranging from 0.82
to 0.87. Finally, Song et al. (2006) reported percentage of predictions within a factor
of 2 between 68% and 98%. Based on hourly concentrations our model predicted
within a factor of 2 of observations only 40% to 59% of the time.

160

The comparison with Song et al. (2006), although the most complete in the
literature, is a very strict test for our model as we are attempting to predict hourly
average concentrations rather than daily average concentrations as in the Song et
al. study.

Typically, as the averaging period increases the performance of a

dispersion model will improve (Schutte et al. 2005). In general, our performance
measures improve when daily average concentrations are considered.

The

exception to this is still the correlation coefficient.
Our performance measures compare more favourably with some other
studies. Chang et al. (2003) reported a CALPUFF FAC2 of 52% and MG of 1.069.
Barna and Gimson (2002) saw higher FGE values ranging from 0.69 to 0.99, lower
index of agreement values ranging from 0.67 to 0.87, and correlation coefficients
ranging from 0.49 to 0.77 during their evaluation of hourly predicted PM
concentrations during PM episodes in New Zealand. It appears that CALPUFF in
Golden is performing better during our episodes than those modeled by Barna and
Gimson (2002), at least at the CPR and TOWN stations. Assimakopoulos (2005)
reported index of agreement values ranging from 0.08 to 0.73 and correlation
coefficients ranging from 0.4 to 0.61 depending on the characteristics of the
modeling environment. CALPUFF in Golden compares well with this study.
Since air quality health problems are most severe during peak PM
concentration hours, comparison of

NRMSE

during the top

10

observed

concentrations for each episode was performed. CALPUFF tends to show larger
errors when only the top 10 observed concentrations are considered indicating
poorer performance at characterizing higher levels of PM.

161

The suite of performance measures used here indicates that the model is not
performing at an optimal level at all stations. However, based on comparison with
some similar studies the model performance is adequate at the CPR and TOWN
station. The GOLF station's poor performance in predicting PM concentrations was
not expected as the station performed well in the CALMET evaluation (Chapter 3). It
is evident in both episodes that there are periods of time when the model predictions
at the GOLF station are extremely low for a period of several hours or even days
(Figures 6-6 and 6-11).

An analysis of the wind direction during these times

indicates that both the observed and predicted winds are from the northwest.
Northwest winds would disperse pollutants released from the point and area sources
near town down valley to the Southeast. None of the pollutants sources would be
carried towards the GOLF receptor.
During the hours where modelled PM concentrations at the GOLF stations
are extremely low, there were PM concentrations observed at the station.

This

would suggest that although most of the sources modelled were in the Town of
Golden and south of the GOLF station, there exists an unknown source or
background level of PM that affects the GOLF station even when winds are from the
northwest.

Holmes and Morawska (2006) noted that CALPUFF typically shows

reasonable agreement with pollutant concentrations with discrepancies usually
accounted for by failure to model an unknown source. Based on our study, it would
be reasonable to assume that the GOLF station's poor model performance is due to
a failure to identify all sources of PM in the modeling domain. A PM emission source
to the north of Golden could have been left out of the model. Or, the area sources

162

already included in the model could have been characterized improperly. The area
sources were only modeled as releasing from the more populated areas of the Town
and this left the GOLF station as the only station outside the area source
boundaries.

The low levels at the GOLF stations could also be attributable to not

including a background PM level due to natural sources (i.e forest fires and
windblown fugitive dust) in the model.
An additional analysis of the modelled PM concentrations versus wind
direction was performed (Figure 6-16, 6-17) and compared to the observed PM
concentrations versus wind direction presented previously in Figures 4-7 and 4-8.
The analysis lends further support to the hypothesis that there is a missing source
unaccounted for to the north-northeast of the GOLF station, with a lower occurrence
of elevated levels of PM concentrations modelled with north or northwest winds,
when compared with the observations (Figure 6-16c, Figure 6-17c). In contrast, the
observed and modelled wind directions that result in the highest PM concentrations
compare well at the other two stations.
Analysis of the PM modeling results is presented in the following chapter.
The results of the model performance evaluation at the three stations must be taken
into account when interpreting the PM results, especially given the GOLF station's
poor performance.

163

(a)
• PM10 |

• PM10

• PM2 s l

• PM? S

Wind direction (degrees)

Wind direction (degree#)

• PMtO
• PMiO 1

• PM2.5

• PM2.5

Wind direction (d. »>...)

Figure 6-16

wind arte** (!»*•«)

Hourly PM concentrations and wind direction for Episode 1 at (a) Town station;
(b) CPR station; (c) Golf station. Graphs to the left shows the measurements
observed during each Episode 1 and the graph on the right shows the model
predicted concentrations by wind direction.

164

• PM10
• PMtO

• PM2 5

PM2 S

13$

180

225

Wind direction (d«gr««a)

Wind Direction

• PM1Q
• PM2.5
• PMlO
• PM2 5

Wind Direction

• PM10
• PM2.5

Wind direction (d*gr«««)

Figure 6-17

Hourly PIUI concentrations and wind direction for Episode 2 at (a) Town station;
(b) CPR station; (c) Golf station. Graphs to the left shows the measurements
observed during each Episode 2 and the graph on the right shows the model
predicted concentrations by wind direction.

165

7

PARTICULATE MATTER ANALYSIS AND SOURCE APPORTIONMENT
The CALPUFF model was evaluated and the model found adequate at

estimating PM concentrations at the CPR and TOWN stations. PM concentrations
are not accurately estimated at the GOLF station. Following model evaluation, an
analysis of the overall PM concentration predictions was performed, along with an
analysis of the sources contributing to the two episodes under investigation.
A goal of this study was to determine the meteorological conditions that
prevail during the episodes of PM under investigation.

Most importantly the

conditions during the hours with the highest PM concentrations can indicate specific
meteorological parameters that lead to spikes in PM that may affect community
health in Golden.
Source apportionment allows for analysis of what sources contribute to the
high PM concentrations at each station. Diurnal variation in emissions and hourly
meteorological conditions will determine the percent contribution of each source at
the three monitors. This will allow for the investigation of any spatial variability in
emission impacts.

7.1 PARTICULATE MATTER AND METEOROLOGICAL ANALYSIS
A comparison of basic meteorological parameters routinely recorded at
weather stations with particulate matter levels during both episodes has been
performed (Chapter 4). The analysis determined that high PM levels occur most
often at low wind speeds. High PM levels occurred during both northwesterly and
southeasterly winds along the valley alignment. During both episodes, the highest
PM concentrations occur in the late afternoon and into the evening (Figure 4-1 and

166

4-2). This was hypothesized to occur because of a combination of meteorological
and emissions related processes. Downslope flows during this part of the day could
bring emissions toward the valley floor, where PM monitors in Golden are located.
In addition, the along-valley flow is typically slowing (Figure 7-1) and reversing at this
time.
The mixing height typically lowers during this time as well (Figure 7-2).
Coinciding with the shift in wind and collapse of the mixing height, emission
generating activities are increased at this time of the day. During the late afternoon
and early evening, traffic increases as residents drive home from work or school and
space heating emissions increase as residents burn wood, oil or gas to heat their
homes upon arrival.
Meteorological parameters (wind speed, mixing height, temperature, and
Monin-Obukhov Length) estimated by CALMET were compared to the predicted
particulate matter levels during both episodes. The comparison can show the
influence of the predicted parameters on particulate matter concentrations. Time
series of predicted meteorological parameters were compared to time series of PM
levels to investigate correlation (Table 7-1). High correlation values, either negative
or positive, would indicate that there is a possible effect of the parameter on
predicted PM concentrations. A multiple regression analysis was also performed
(Table 7-2).
Negative correlations between wind speed and PM levels, as well as mixing
height and PM levels, are moderate for both episodes, especially at the CPR and

167

TOWN stations. This is expected with lower wind speeds or mixing height hindering
dispersion of particulate matter as it is emitted.

(a)

Wind Speed

20 .2

PM10
« 0.6

v*

0

2

4

6

8

10 12

14

16

18

20

PM2.5

22

Hour of Day

(b)

r70
- 60
E
50 "5>
3
c
40 0

•a
s
- 30 c

*
c

^ 0.6

—•—Wind Speed
PM10
a

PM2.5

20 8

s

10

CL

0
8

10 12

14

16

18

20

22

Hour of Day

Figure 7-1

Modeled wind speed and observed particulate matter concentrations at the
TOWN station for (a) Episode 1; (b) Episode 2. The wind speed and PM
averages are from all days in each episode.

168

(a)

25 6>
- Mixing Height
-PM10
-PM2.5

8

10

12

14 16

18

20 22

Hour of Day

(b)

40 .2

o> 500

Mixing Height
PM10
PM2.5

* 300
S
200

20 <3

8

10 12 14 16 18 20 22
Hour of Day

Figure 7-2

Modeled mixing height and observed particulate matter concentrations at the
TOWN station for (a) Episode 1; (b) Episode 2, The mixing height and PM
averages are from all days in each episode.

169

Temperature is also moderately, negatively correlated with observed values
meaning that lower temperatures correspond to higher PM concentrations. This
lends support to the hypothesis that at lower temperatures emissions from sources
such as space heating would increase.

It should also be recognized that other

factors, such as mixing height and wind speed, tend to co-vary with temperature, so
a negative correlation with temperature may also be related to dispersion. However,
under this hypothesis stronger correlation coefficients should be calculated for PM2.5,
the main pollutant emitted by space heating. This occurs at the TOWN station but
not at the CPR station. Emissions from space heating were varied diurnally, but
were not adjusted to reflect temperature fluctuations. However, by coincidence, the
diurnal fluctuations calculated from the woodstove survey would follow typical daily
temperature patterns.
Multiple regression analysis was used to investigate the significance of wind
speed, mixing height and temperature on PM concentrations. Table 7-2 summarizes
the individual p-values (representing the significance of each parameter in predicting
the corresponding PM concentration) as well as the multiple regression r-squared
(representing the amount of variability in PM accounted for by the variables wind
speed, temperature and mixing height) and p-values of the F variable (representing
the significance of the multiple parameters included in the regression; values <0.05
indicate that the multiple regression results are significant).
Episode 1: Wind speed is a significant variable only at the TOWN station.
Mixing height is significant in determining PM concentrations in all scenarios except
PM10 at the TOWN station. Temperature is a significant factor at both the CPR and

170

TOWN stations. Overall, the three variables tested account for 30-40% of the PM
concentration variability
Table 7-1

Correlation coefficients between predicted meteorological parameters and
predicted PM levels during both episodes. Wspd = wind speed, Mix hgt =
mixing height, Temp = temperature, Mon-obu = Monin-Obukhov length.
CPR

TOWN

GOLF

PM25 | PM10

PM .5 I PM10

PM25

-0.48
-0.34
-0.56
0.02

-0.46
-0.38
-0.51
0.01

-0.54
-0.42
-0.23
-0.05

-0.56
-0.48
-0.35
0.05

-0.14
-0.29
-0.08
0.09

-0.11
-0.27
-0.11
0.15

-0.51
-0.52
-0.40
0.62

-0.50
-0.51
-0.38
0.65

-0.56
-0.40
-0.23
0.26

-0.58
-0.42
-0.25
0.25

-0.22
-0.13
0.04
0.13

-0.22
-0.16
-0.01
0.22

Episode 1

I PM10

Wspd
Mix hgt
Temp
Mon-obu

2

Episode 2
Wspd
Mix hgt
Temp
Mon-obu

Table 7-2

Multiple regression results for wind speed (WS), temperature (T) and mixing
height (Mix Hgt) as compared to predicted PM10 and PM2.5 levels during each
episode.
E1 PM10
CPR
TOWN
GOLF

WS

Individual p-value
Mix Hgt

Multiple Regression
R-squared
Significance F
0.409
0.000
0.325
0.000

0.000

0.001

0.016
0.687

0.146
0.002

0.068

0.002

0.000

0.000
0.000

0.012
0.026

0.000
0.000

0.864

0.425

0.003

0.325
0.406
0.059

E2 PM10
CPR
TOWN
GOLF

0.000
0.000
0.000

0.694
0.131

0.000
0.000

0.006

0.019

0.325
0.399
0.063

0.000
0.000
0.000

E2 PM2.5
CPR
TOWN
GOLF

0.000
0.000
0.000

0.501
0.073
0.074

0.000
0.000

0.317
0.436
0.057

0.000
0.000
0.000

E1 PM2.5
CPR
TOWN
GOLF

0.111
0.000
0.641

0.068

0.012

0.005

171

Episode 2: Wind speed and mixing height are significant variables at all
stations. Temperature is not a significant factor except at the GOLF station for PMi0.
Once again, the variables account for 30-40% of the PM variation.
During both episodes it appears that high PM concentrations are largely
affected by low wind speeds and low mixing height. This would also be supported by
the time series plots presented in Chapter 6 (Figure 6-6Figure 6-6d and Figure 611d). Temperature can also be significant, however it is likely an indicator of the
"built-in" diurnal variation of the area sources modeled.

Low wind speeds and

mixing height are crucial to pollutant dispersion, especially during the night.
CALMET uses the vertical temperature profile above the height of the previous
hour's mixing height from the HOBO/upper air data as well as the calculated surface
heat fluxes, and calculates the daytime convective mixing height. It then calculates
the mechanical mixing height based on Venkatram (1980) and the surface wind
speed and surface roughness. The daytime mixing height is the maximum of the
calculated convective or mechanical mixing height.

In the absence of surface

convective mixing, surface wind speeds and surface roughness are used to estimate
only the mechanical mixing height at night, resulting in lower mixing height
predictions.

In this study, the daytime mixing heights seem reasonable, but the

night-time mixing heights are extremely low, especially during episode 2, which may
contribute to the over-prediction of PM concentrations.
Variability
(meteorological)

in

PM

concentrations

conditions

and

is

emission

caused
processes.

by

differing
PM

dispersal

variability

not

172

characterized by meteorological factors is likely due to changes in emissions not
captured by the level of detail in the emission scenarios.

7.2

SOURCE APPORTIONMENT
Source apportionment determines the types and amounts of PM that come

from specific emission sources. As demonstrated in Golden by the Evans and Jeong
(2007), source apportionment analysis can be accomplished through receptor
modeling. Receptor modelling is typically expensive as it involves laboratory time to
chemically analyze air filter samples. In this section, CALPUFF results are used to
apportion the sources contributing to the PM concentrations recorded at all three
monitoring station.
The CALPUFF setup described and evaluated in Chapter 6 was used to
model individual source contributions. Sources were combined to produce an overall
PM10 and PM2 5 concentration. By modeling individual sources, the contribution of
each source to the total can be estimated. The GOLF station is analyzed in this
section, however the analysis should be taken in the context of the poor model
performance at this location.

It should be noted that there is likely an unknown

source (background or otherwise) that was not included in the model that affects PM
levels at the GOLF station.

Inclusion of this unknown source would significantly

reduce the impacts of other sources at the GOLF station.

7.2.1 Episode 1 Source Apportionment
Source contributions over the entire first episode (Figure 7-3) show that space
heating, road dust , LP operations and construction operations account for the
majority of PM10 at all three monitors. Space heating emissions impact the CPR and

173

TOWN stations the most, while LP emissions contribute the most to the GOLF
station.

100%
• RDUST

80%

• CONS
• BURN

60%

• TROAD
• SPCHT

40%

• LP
• RAIL

20% --

• HWAY

0%

CPR

Figure 7-3

TOWN

GOLF

PM10 source contribution to the entire first episode. RDUST = road dust; CONS
= construction; BURN = burning; TROAD = town road traffic; SPCHT = space
heating; LP = Lousiana Pacific Wood Products Ltd.; Rail = railway; HWAY =
highway

During the top 25 hourly concentrations, the impact of space heating is higher
at the CPR and TOWN stations, while the GOLF station record over 75% of its
ambient concentration from emissions at LP (Figure 7-4).

174

100%

• RDIIST

80%

• CONS
• BURN

60%

• TROAD
•SPCHT
• LP

40%

• RAIL

20%

• HWAY

0%

CPR

Figure 7-4

TOWN

GOLF

PM10 source contribution to the top 25 hourly PM concentrations during the
first episode

Source contributions to PM2.5 concentrations over the entire first episode
show that space heating emissions are the largest constituent at all three stations
(Figure 7-5). When only the top 25 concentrations are considered the impact of the
space heating emissions is further heightened (Figure 7-6).

100%
80%
60%

I THI

• RDUST
• CONS
• BURN
• TROAD
• SPCHT

40%

• LP
• RAIL

20%

• HWAY

0%

CPR
Figure 7-5

TOWN

GOLF

PM2.$ source contribution to the entire first episode

175

100%
D RDUST

80%

• CONS
• BURN

60%

• TROAD
•SPCHT

40%

•LP
• RAIL

20%

• HWAY

CPR
Figure 7-6

TOWN

GOLF

PM2.5 source contribution top 25 hourly PM concentrations during the first
episode

7.2.2 Episode 2 Source Apportionment
Source contributions over the entire second episode (Figure 7-7) show that
road dust, LP operations, space heating and construction operations account for the
majority of PM10 at all three monitors. Road dust and LP emissions impact the CPR
and TOWN stations the most, while LP emissions contribute the most to the GOLF
station.

When the top 25 hourly concentrations are isolated (Figure 7-8), the

emissions from road dust contribute a larger proportion to the overall PM
concentrations.

176

H RDUST
• CONS
• BURN
• TROAD
•SPCHT
•LP
• RAIL
• HWAY

CPR
Figure 7-7

TOWN

GOLF

PM10 source contributions to the entire second episode.

100%

• RDUST
• CONS
• BURN
• TROAD
• SPCHT
• LP
• RAIL
• HWAY

CPR
Figure 7-8

TOWN

GOLF

PM10 source contribution to the top 25 hourly PM concentrations during the
second episode.

Source contributions to PM2.5 over the entire second episode (Figure 7-9)
showed that space heating emissions impact all three monitors the most. Other
area sources made up the majority of the remaining emissions. The contribution of
space heating emissions

was also emphasized when the top 25 hourly

concentrations are isolated (Figure 7-10).

177

• RDUST
• CONS
• BURN
• TROAD
•SPCHT
• LP
• RAIL
• HWAY

CPR
Figure 7-9

TOWN

GOLF

PM25 source contributions to the entire second episode.
100%

O RDUST
• CONS
• BURN
• TROAD
O SPCHT
• LP
• RAIL
• HWAY

CPR
Figure 7-10

TOWN

GOLF

PM2.S source contribution to top 25 hourly PM concentrations during the
second episode.

The source apportionment showed similar source contributions to both the
CPR and TOWN stations, with quite different emissions contributing to the PM
concentrations at the GOLF station. Overall, space heating emissions contributed
the most to both episodes at the CPR and TOWN stations. LP emissions were the
main contributor at the GOLF station.

178

7.3 INDIVIDUAL SOURCE RESULTS
This section describes each individual modeled source's contribution to the
predicted PM levels at the CPR, TOWN and GOLF receptors.

7.3.1 LP Emissions
The contributions from LP operations were 13%, 11% and 69% for PMi0
during episode 1 at CPR, TOWN and GOLF respectively. LP operations contributed
2%, 2% and 15% of PM2.5 during episode one at CPR, TOWN and GOLF
respectively. During episode two, LP operations contributed 23%, 22% and 72% to
PM10 at CPR, TOWN and GOLF respectively. LP operations contributed 4%, 3% and
24% of PM2.5 during episode two at CPR, TOWN and GOLF respectively (Table 7-

3).
Table 7-3

LP emission source contributions to predicted PM levels at the CPR,
TOWN and GOLF receptors. For comparative purposes the total PM
loading for all sources during the episode (ng/m3) is also included.
Percent Contribution to PM

Episode
1
2

Pollutant
PM10
PM2.5
PM10

PM2.5

CPR
13
2
23
4

TOWN
11
2
22
3

GOLF
69
15
72
24

Total PM Loading All
Sources (pg/m3)
CPR
GOLF
TOWN
2,318
4,772
589
6,128
10,763
401
1,676
3,414
389
3,098
5,311
107

7.3.2 Highway Emissions
The contributions from highway emissions were 2%, 0.5% and 3% for PM10
during episode 1 at CPR, TOWN and GOLF respectively. Highway emissions
contributed 1%, 0.3% and 2% of PM25 during episode one at CPR, TOWN and
GOLF respectively.

During episode two, highway emissions contributed 4%, 1%

and 1% to PM10 at CPR, TOWN and GOLF respectively. Highway emissions

179

contributed 2%, 0.5% and 2% of PM25 during episode two at CPR, TOWN and
GOLF respectively (Table 7-4).

Table 7-4

Highway emission source contributions to predicted PM levels at the
CPR, TOWN and GOLF receptors. For comparative purposes the total
PM loading for all sources during the episode (iig/m3) is also included.
Percent Contribution to PM
Pollutant
LC

Episode

O

1

PM .5
PM10
PM .5
2

2

2

CPR
2
1
4
2

TOWN
0.5
0.3
1
0.5

GOLF
3
2
1
2

Total PM Loading All
Sources (ng/m3)
CPR
TOWN
GOLF
2,318
589
4,772
6,128
10,763
401
1,676
3,414
389
3,098
5,311
107

7.3.3 Railway Emissions
To investigate the effects of the rail operations in the Golden airshed this
analysis combines the PM outputs for both railway line and yard emissions. The
contributions from railway emissions were 3%, 1% and 2% for PM10 during episode
1 at CPR, TOWN and GOLF respectively. Railway emissions contributed 1%, 0.4%
and 1% of PM2.5 during episode one at CPR, TOWN and GOLF respectively. During
episode two, railway emissions contributed 5%, 1% and 12% to PM10 at CPR,
TOWN and GOLF respectively. Railway emissions contributed 3%, 1% and 14% of
PM2.5 during episode two at CPR, TOWN and GOLF respectively (Table 7-5).

Table 7-5

Railway emission source contributions to predicted PM levels at the
CPR, TOWN and GOLF receptors. For comparative purposes the total
PM loading for all sources during the episode (|jg/m3) is also included.
Percent Contribution to PM

Episode
1

Pollutant
PM10
PM .5
PM10
PM .5
2

2

2

CPR
3
1
5
3

TOWN
1
0.4
1
1

GOLF
2
1
12
14

Total PM Loading All
Sources (ng/m3)
CPR
TOWN
GOLF
2,318
589
4,772
6,128
10,763
401
389
1,676
3,414
3,098
5,311
107

180

7.3.4 Space Heating Emissions
The contributions from space heating were 37%, 42% and 14% for PM10
during episode 1 at CPR, TOWN and GOLF respectively. Space heating contributed
65%, 66% and 58% of PM25 during episode one at CPR, TOWN and GOLF
respectively. During episode two, space heating contributed 16%, 17% and 4% to
PM10 at CPR, TOWN and GOLF respectively. Space heating contributed 36%, 36%
and 25% of PM25 during episode two at CPR, TOWN and GOLF respectively (Table
7-6).
Table 7-6

Space heating emission source contributions to predicted PM levels at
the CPR, TOWN and GOLF receptors. For comparative purposes the
total PM loading for all sources during the episode (Md/m3) is also
included.
Percent Contribution to PM

Episode
1

Pollutant
PM10
PM .5
PM10
PM .5
2

2

2

CPR
37
65
16
36

TOWN
42
66
17
36

GOLF
14
58
4
25

Total PM Loading All
Sources (pg/m3)
CPR
TOWN
GOLF
2,318
4,772
589
6,128
10,763
401
3,414
389
1,676
3,098
5,311
107

7.3.5 Road Dust
The contributions from road dust were 23%, 23% and 6% for PM10 during
episode 1 at CPR, TOWN and GOLF respectively. Road dust contributed 8%, 8%
and 5% of PM2 5 during episode one at CPR, TOWN and GOLF respectively. During
episode two, road dust contributed 23%, 27% and 4% to PM10 at CPR, TOWN and
GOLF respectively. Road dust contributed 11%, 13% and 6% of PM25 during
episode two at CPR, TOWN and GOLF respectively (Table 7-7).

181

Table 7-7

Road dust emission source contributions to predicted PM levels at the
CPR, TOWN and GOLF receptors. For comparative purposes the total
PM loading for all sources during the episode (Mg/m3) is also included.
Percent Contribution to PM

Episode
1

Pollutant
PM10

PM25
PM10

2

PM25

CPR
23
8
23
11

TOWN
23
8
27
13

GOLF
6
5
4
6

Total PM Loading All
Sources (Mg/m3)
CPR
TOWN
GOLF
2,318
4,772
589
10,763
6,128
401
1,676
3,414
389
3,098
5,311
107

7.3.6 Local Traffic Emissions
The contributions from local traffic emissions were 5%, 6% and 2% for PMio
during episode 1 at CPR, TOWN and GOLF respectively. Local traffic emissions
contributed 7%, 7% and 6% of PM2.5 during episode one at CPR, TOWN and GOLF
respectively. During episode two, local traffic emissions contributed 8%, 9% and 2%
to PMio at CPR, TOWN and GOLF respectively. Local traffic emissions contributed
14%, 15% and 9% of PM2.5 during episode two at CPR, TOWN and GOLF
respectively (Table 7-8).
Table 7-8
Local traffic emission source contributions to predicted PM levels at the
CPR, TOWN and GOLF receptors. For comparative purposes the total PM loading for
all sources during the episode (Mg/m3) is also included.
Percent Contribution to PM
Episode
1
2

Pollutant
PM10

PM25
PM10
PM .5
2

CPR
5
7
8
14

TOWN
6
7
9
15

GOLF
2
6
2
9

Total PM Loading All
Sources (Mg/m3)
CPR
TOWN
GOLF
2,318
4,772
589
6,128
10,763
401
1,676
3,414
389
3,098
5,311
107

7.3.7 Construction Emissions
The contributions from construction operations were 11%, 12% and 4% for
PM10 during episode 1 at CPR, TOWN and GOLF respectively. Construction

182

operations contributed 10%, 10% and 7% of PM2.s during episode one at CPR,
TOWN and GOLF respectively.

During episode two, construction operations

contributed 16%, 17% and 4% to PM10 at CPR, TOWN and GOLF respectively.
Construction operations contributed 18%, 19% and 12% of PM2.5 during episode two
at CPR, TOWN and GOLF respectively. The construction emissions were based on
prorated annual estimates. It was not known what actual construction operations
were being conducted during the episodes (Table 7-9).
Table 7-9

Construction emission source contributions to predicted PM levels at
the CPR, TOWN and GOLF receptors. For comparative purposes the
total PM loading for all sources during the episode (pg/m3) is also
included.
Percent Contribution to PM

Episode
1

Pollutant
PM10
PM .5
PM10
PM2.5
2

2

CPR
11
10
16
18

TOWN
12
10
17
19

GOLF
4
7
4
12

Total PM Loading All
Sources (pg/m3)
CPR
TOWN
GOLF
2,318
589
4,772
6,128
10,763
401
3,414
1,676
389
3,098
5,311
107

7.3.8 Burning Emissions
The contributions from burning were 4%, 5% and 1% for PM10 during episode
1 at CPR, TOWN and GOLF respectively. Burning contributed 7%, 7% and 5% of
PM2.5 during episode one at CPR, TOWN and GOLF respectively. During episode

two, burning contributed 6%, 7% and 1% to PM10 at CPR, TOWN and GOLF
respectively. Burning contributed 13%, 13% and 8% of PM2.5 during episode two at
CPR, TOWN and GOLF respectively (Table 7-10).

183

Table 7-10

Burning emission source contributions to predicted PM levels at the
CPR, TOWN and GOLF receptors. For comparative purposes the total
PM loading for all sources during the episode (pg/m3) is also included.
Percent Contribution to PM

Episode
1
2

Pollutant

CPR

TOWN

GOLF

PM10

4
7
6
13

5
7
7
13

1
5
1
8

PM2.s
PM10
PM .5
2

7.4

Total PM Loading All
Sources (ug/m3)
CPR
TOWN
GOLF
2,318
6,128
1,676
3,098

4,772
10,763
3,414
5,311

589
401
389
107

COMPARISON WITH RECEPTOR MODELING
Figure 7-9 provides a comparison of the modeled CALPUFF source

apportionment of PM2.5 with the receptor modeling of PM2.5 conducted by BCMOE
(Evans and Jeong 2007) during the speciation project in Golden. PMF analysis data
were obtained for the specific days that receptor modeling data were collected
during each episode. Only 4 filters were collected during each episode. Due to the
small sample size, a seasonal comparison is also provided which compares the
modeled data with the larger sample size used to provide the seasonal PM25
breakdown (Figure 6-5). Since the final emission rates in CALPUFF were adjusted
using the results of the receptor modelling it is noted that the comparison here is
somewhat circular. However, the comparison can be used to illustrate whether the
adjusted emission rates were mirrored in the results at the receptors in CALPUFF.
To review, the solution from the receptor modeling in Golden revealed seven
factors contributing to PM.

The factors identified were: Road Salt, Sulphate,

Residential Wood Burning, Wood Processing, Crustal Material, Traffic and
Residential Winter Heating.

It was assumed that the winter heating and the

residential wood burning factors would be an appropriate comparison with modeled
space heating emissions, road dust emissions would be compared to the crustal

184

factor, the wood processing factor would be compared directly with LP emissions
and traffic emissions would be compared with traffic emissions as suggested by
Evans and Jeong (2007). Filters for receptor modeling were collected at the TOWN
station, therefore comparative modeled data was from the TOWN receptor.
During episode 1, the CALPUFF modeled space heating emissions were a
smaller contributor to PM levels in comparison to the residential wood burning factor
presented in the receptor modeling, but are still the most significant factor. The
other sources showed reasonable agreement with the receptor modeling analysis.
The modeled data showed the best agreement with the seasonal receptor data.
During episode 2, the seasonal receptor modeling and the CALPUFF modeling
source apportionment did not show as good agreement. Specifically the receptor
modelling indicated a higher contribution from LP and a lower contribution from
space heating than was modeled with Calpuff. When compared with the receptor
modeling from the episode, the LP contribution was similar, but the space heating
contribution was still lower.

185

(a)
c
o

•3

3

ja

5

• Space heating

c

• Road Dust
• LP
• Traffic

®
0.
s
Q.

Receptor
(Seasonal)

Receptor
(episode)

CALPUFF

(b)

• Space heating
• Road dust
• LP
• Traffic

Receptor
(Seasonal)

Figure 7-11

Receptor
(episode)

CALPUFF

Source apportionment comparison between seasonal receptor modeling,
receptor modeling specific to the episode and CALPUFF predicted percentages
of total estimated PM2.5 is provided for (a) episode 1, and (b) episode 2.

186

7.5

DISCUSSION
CALPUFF was used to model eight sources identified as contributors to PM in

the El. CALPUFF results show that, as hypothesized, space heating was the top
contributor to both PM10 and PM2.5 during episode 1. During episode 2, LP and road
dust were the top contributors to PM10 and space heating is the top contributor to
PM2.5- Road dust was hypothesized to be a major contributor to this "spring-time"
episode and its contribution to the PM10 portion supports this hypothesis.
CALMET predicted a typical daily pattern in the boundary layer. Stability
increases as the sun goes down and the earth's surface is cooled by emitting
infrared radiation. This causes the convective mixing layer to shrink and the cooling
continues into the night and morning. Typically, the poor dispersal in the evening is
predicted by CALMET to remain through night-time hours and into the morning. This
leads to extended periods of high PM as pollutants from all sources remain "trapped"
creating a build-up of PM. CALMET predicts neutral or stable atmospheric conditions
at these times. PM concentrations rise as both vertical and horizontal dispersion are
hindered by the increasing stability and decreasing wind speeds. As the sun rises in
the morning, heating of the surface causes the mixing layer to grow. An increase in
horizontal wind speeds is accompanied by better vertical dispersion as the
atmosphere becomes more unstable. CALMET predicts a fairly repetitive cycle
throughout the two episodes. Thus, high PM concentrations typically occur during
the morning and evening hours. It is unclear to what extent the upper air profiles
input into CALMET (Figure 5-3) contribute to the build-up of pollutants, but the
persistence of temperature inversions throughout the episodes would hinder the

187

growth of the mixing layer and the dispersion of the PM. A more accurate vertical
temperature and wind profile is essential for accurate modeling within the Columbia
River Valley.
The source apportionment of the 25 highest PM concentrations for each
episode was used to investigate which sources were contributing to the high hours
of PM. These hours are of importance because of the health related consequences
of extreme peaks in PM (Section 2-1). Overall, these hours contribute a large portion
of the loading during each episode. Isolation of the highest PM concentrations in
episode 1 at the CPR and TOWN monitors saw a jump in contributions of space
heating emissions for both PMi0 and PM2.5. There are several coinciding factors that
contribute to the prevalence of space heating during these times. A spike in the
diurnal variation of space heating emissions occurs at the same time that CALMET
predicts lower mixing heights, slowing of wind speeds and at times a reversal of flow
(see for example, Figure 6-6, Table 6-12). CALMET predicts these poor dispersal
conditions during the early morning hours (low mixing heights, low wind speeds) and
evening hours (low mixing heights, low wind speeds and at times flow reversal).
Typically, the highest concentrations occur during these hours as well (Figure 4-1
and 4-2). This supports the predictions made in Chapter 4 that a slowing of the wind
speed combined with a low mixing height and possible reversal of the along-valley
flow may contribute to higher PM concentrations. It is also supported by the
correlation and multiple regression analysis (Section 7-1). These confirmed that
both wind speed and mixing height were significant factors in the prediction of PM
concentrations.

188

The same factors are allowing for road dust to contribute more to PM10 and
space heating more to PM2.5 during episode 2. Spikes in the diurnal variation of road
dust emissions corresponds to the timing of the same meteorological factors
described during episode 1 making them more prevalent during high hours of PM.
Road dust emits mostly in the coarse fraction of PM10 and thus does not have as
great of an effect on the PM2.5 subset. Space heating emissions contribute more to
the higher PM2.5 concentrations because of similar factors as those described for
episode 1.
The largest industrial emitter in Golden is LP. As such, residents are
interested in its effects on PM concentrations. As expected, the modeling shows
that LP emissions affect all stations in Golden. The effects are largely in the PM10
range. Over the past 10 years LP has performed several upgrades designed to
reduce PM25 emissions. The major improvements affecting air emissions from the
mill and powerhouse from the LP site include: installing a baghouse on the plywood
sander reducing emissions of sander dust significantly; installing a dry electrostatic
precipitator on the powerhouse boiler, and; removing two old 1960-era dryers and
replacing them with a highly efficient dryer coupled with a new wet electrostatic
precipitator. In an effort to control PM10, over the past 7 years has had a more
consistent road dust control program for mill haul roads using dust suppressant and
a consistent watering program during dry summer days.
A pattern exists in the apportionment of predicted LP emissions to each
monitor. Predicted LP emissions account for a higher percentage of the PM at the
GOLF station as compared to the other two stations. At first it may seem LP's

189

location to the north of the Town causes increased predicted concentrations at the
GOLF station. This pattern, however, is deceiving because the apportionment is
displayed on a percentage basis. The overall amount of PM loading from LP
modeled at all three stations is similar for episode 1. During episode 2, the total LP
emissions at the CPR and TOWN stations are considerably more than those at the
GOLF station. LP emissions are accounting for a higher percentage of the modeled
PM at the GOLF station because the unknown source(s) or background PM level
likely impacting this location is not modelled.

Conversely, the CPR and TOWN

stations, represented as discrete receptors within the area sources defined
boundaries, are impacted much more from area sources. This clarification shows
that despite LP's lower percent contribution at the CPR and TOWN stations the
potential exposure to residents of Golden is higher at these locations compared to
the GOLF station during modeling of episode 2 and similar at all three stations
during modeling of episode 1.
The GOLF station results must be taken in context with the poor predictive
performance at this station as described in the CALPUFF model performance
evaluation (Chapter 6).

Although LP emissions are a high percentage of the

emissions recorded at this station, it must be considered that the predicted PM
values are much too low. It is likely that there are sources not modeled or not
modeled properly that contribute to the PM observed at the GOLF station. Other
CALPUFF studies have shown that when CALPUFF is not in reasonable agreement
with pollutant concentrations it tends to be as a result of an unknown source
(Holmes and Morawska 2006). The modeling scenario places the GOLF station as

190

the only monitor outside the boundaries of the area sources. The area sources in
the model do not disperse to the GOLF monitor as readily as the elevated stacks
producing the LP emissions.
LP emissions are more readily dispersed because of the effective release
height of the stacks. When the winds are from the north, the peaks in PM have more
contribution from the LP emissions. Northerly winds bring the LP emission towards
the residents in the Town area and further south. Southerly winds bring the LP
emissions away from the TOWN and towards the GOLF monitor. High PM events at
the CPR and TOWN station during southerly wind events are entirely dominated by
the area sources modeled.
There is an important limitation to this modeling analysis. Only two short
episodes were modeled and different meteorological conditions during other parts of
the year will influence emission dispersion. For example, predominantly stagnant
conditions or northerly winds could cause more of an influence of LP emissions on
the populated areas of Golden.
Line source emissions from highway and railway emission do not have much
effect on PM concentrations. Examining outputs of CALPUFF line source dispersion
shows that modeled emissions do not disperse far from the roadway or rail line
segment. Unless a monitor is located very near to the roadway, the impacts are
likely not to be great; the effects are on a micro-scale. Conversely, residents living
along highways, busy roadways or rail lines may have quite a different
apportionment of PM sources than those living further away.

191

Figure 7-11 shows the modeled sources and receptor modeling contributions
to PM loading during the episode. Episode 1 CALPUFF results show reasonable
agreement with the receptor modeling. The wintertime PM2.5 emissions from the
receptor modeling predicted space heating as the major contributor.

CALPUFF

modeled a lower contribution, but predicts that space heating is the most significant
contributor of PM2.5 at 62.5%. The other modeled emission sources show good
agreement with the receptor modeling.

The CALPUFF modeling supports the

seasonal receptor modeling analysis for episode 1, however this is expected given
that the emission rates were adjusted based on the receptor modeling. The limited
number of receptor modeling samples taken during the episodes agreed reasonably
well with the modeled results, but more samples would be needed to attempt any
further analysis.
Episode 2 CALPUFF results show less agreement with the receptor modeling
(Figure 7-11). Springtime receptor modeling was used for the comparison with the
February episode. The wintertime receptor modeling did not compare well with
Episode 2 CALPUFF results either. For episode 2 the CALPUFF modeling does not
support the results of the receptor modeling. Given that CALPUFF emission rates
were adjusted based on the receptor modeling results it is unexpected that the
results of the CALPUFF and receptor modeling do not agree. This suggests that the
CALMET meteorology modeled or the CALPUFF source characterization incorrectly
captured the emissions from source to receptor.
It was observed in initial model runs that CALPUFF modeling using the
existing emissions

inventory significantly over-predicted PM concentrations.

192

Sarigiannis et al. (2004) showed that using inaccurate or out-of-date emissions
inventories can cause significant error in dispersion modeling.

The results of

CALPUFF were improved using results from the recent receptor modeling analysis
(Evans and Jeong 2007). This highlights the importance of having accurate, detailed
emissions information for any modeling project. Certainly it would provide for a more
accurate picture of dispersal in Golden if some of the area sources in this study were
defined as individual point sources or more detailed emission information regarding
the area emissions were known.
The source apportionment from CALPUFF modeling has provided an
estimate of the sources contributing to PM during two episodes. Although model
performance was not ideal, based on the available emission information and
meteorological network CALPUFF predictions generally agree with the results of the
receptor modeling indicating that the sources of concern are space heating, road
dust and, at times, LP emissions.

193

8

MODELING SUMMARY

Application of the CALPUFF modeling system allowed for investigation of two
episodes of high PM concentrations in Golden. The study set out to answer three
questions:
•

What are the dispersal mechanisms and emission processes that contribute
to episodes of PM in Golden?

•

Can these conditions be modeled in CALPUFF with the existing emissions

information to accurately represent PM concentrations during these episodes?
•

Will CALPUFF predicted PM source apportionment compare well with
receptor modeling conducted by BCMOE (Evans and Jeong 2007)?
Examining

the

dispersion

and

emissions

characteristics

within

and

surrounding the Town of Golden prior to modeling identified that high PM
concentrations during the episodes typically occurred during hours with low wind
speeds. In addition, high PM levels were usually initiated during the late afternoon
or early evening, with high concentrations persisting sometimes all night. It is likely
at this time the along-valley wind flows are slowing and reversing, and, combined
with escalated emissions from increased traffic and space heating contributed to
higher PM levels.
Testing of the CALMET model revealed poor comparative performance of the
meteorological model when surface stations were omitted in the "leave one out"
method.

However, CALMET showed reasonable agreement with observed

conditions when run with a full suite of surface stations. Furthermore, CALMET was

194

able to capture along-valley wind speed slowing and reversal in direction during late
afternoon and early evening.
The CALPUFF dispersion model was not able to capture the hourly PM
concentrations during the episodes with great accuracy. CALPUFF has had varying
degrees of success in modeling pollutant concentrations accurately in complex
terrain. The accuracy of this modeling exercise was comparable to other CALPUFF
studies of similar application.

CALPUFF's ability to characterize dispersion in

complex terrain depends upon input of meteorological parameters from surface
stations and upper air stations. Ideally, these stations would be located within or
very close to the modeling domain. The upper air station closest to Golden was
more than 400 kilometers away; although site specific temperature and SODAR data
were used as well.

The upper air profiles may have compromised CALPUFF's

ability to capture the PM concentrations accurately.
Wind speed and mixing height are key factors in the dispersion of pollutants
in Golden. In the absence of a local upper air station, data from Prince George,
combined with local upper air wind data from the SODAR and HOBO temperature
transect, were used as a surrogate upper air input file for CALMET. Mixing height is
determined based on the vertical temperature profile, surface heat fluxes, horizontal
winds, and surface roughness parameters. The upper air data, although partially
local, likely does not provide the most accurate picture of the complete vertical
temperature profile. CALPUFF over and under predictions are largely attributed to
times when the mixing height is extremely low or high, respectively, for consecutive
hours. Low mixing height during the night is attributable to the low wind speeds

195

observed during the episodes.

The unrealistically low mixing heights predicted

during this study can be adjusted by applying an appropriate minimum mixing height.
The daytime mixing heights could be better predicted by better defining the vertical
temperature profile from the upper air data. Providing more accurate upper air data
is a necessary component to any future modeling studies in Golden. Ideally, twicedaily rawisonde upper air data would be collected in Golden to allow for the local
vertical temperature and wind profile to be established.
Inaccurate emission rate estimates, broad definition of area sources within
the modeling domain and missing PM sources or background PM levels limited
CALPUFF modeling accuracy.

The emission rates originally tabulated from the

emissions inventory and later revised from the receptor modeling did not depict the
actual emission rates. Point source emissions are easily characterized because of
legislated emissions reporting. Therefore, the accuracy of the LP emissions are
likely the highest. However, area source emission estimates are hampered not only
by the accuracy of the emission rates, but the definition of the emitting area.
Additional information on the spatial variability of area emissions would improve
modeling efforts. At the GOLF station, it is likely that a missing PM source or a
failure to apply a background level of PM to the modeling, resulted in large periods
of under-prediction. Future studies should identify any missing PM source (likely to
the North of town) or apply an appropriate background PM level. A background PM
level could be captured by establishing an additional monitoring station in an pristine
environment near Golden, well away from anthropogenic sources.

196

The source apportionment of the CALPUFF results identified the major
contributors to degraded air quality levels during the two episodes under
investigation as space heating, road dust and, intermittently, LP operations. These
results compared favourably with receptor modeling conducted during the same
period in Golden.
CALPUFF modeling was commissioned in Golden, along with a number of
other studies, to provide insight on the sources and relative contributions of PM in
Golden. Our study set further goals to accurately model hourly predictions during
episodes of high PM. CALPUFF identified PM source contributions in Golden. In
addition, the prediction of hourly PM concentrations was comparable to other similar
studies despite the identified limitations of meteorological and emissions data.

197

9

REFERENCES

Abel, T, et al. 2006. An Air Emissions Inventory for Golden, British Columbia.
British Columbia Ministry of Environment.
http://wlapwww.gov.bc.ca/kor/epd/air_emissions_golden/air_emiss_golden
Abdul-Wahab S.A. 2004. Evaluation of the Industrial Source Complex short term
model: dispersion over terrain. Journal of the Air and Waste Management
Association 54: 396-408.
Allwine, J.K et al. Peer Review of the CALMET / CALPUFF Modeling System. The
Kevric Company Inc. August 28, 1998.
http://www.epa.gov/scram001/7thconf/calpuff/calpeer.pdf
Arena, N., Georgieva E., Ratto, C.F. 1997. Diagnostic wind field modeling for
complex terrain: new development and testing. Physica Medica 8 (Supplement
1): 373-374.
Arya, S.P. 1999. Air pollution meteorology and dispersion. Oxford University
Press, New York.
Australian Environment Protection Authority, 2000. AUSPLUME Gaussian-plume
Dispersion Model Technical User Manual. Victoria, Australia.
Assimakopoulos, V. 2005. Evaluation of CALPUFF modeling system performance:
an application over the Greater Athens Area, Greece. International Journal of
Environment and Pollution 24: 22:32.
Barna, M.G. and N.R. Grimson. 2002. Dispersion modeling of a wintertime
particulate pollution episode in Christchurch, New Zealand. Atmospheric
Environment 36: 3531-3544.
Bellasio, R. 2005. Algorithms to account for topographic shading effects and surface
temperature dependence on terrain elevation in diagnostic meteorological
models. Boundary-Layer Meteorology 114: 595-614.
British Columbia Ministry of Environment. 2008. Guidelines for Air Quality
Dispersion Modelling in British Columbia.
http://www.env.gov.bc.ca/epd/bcairquality/reports/pdfs/air_disp_model_08.pdf
British Columbia Ministry of Environment. 2002. Fine Particulates: What They Are
and How They Affect Us. Water, Air and Climate Branch. Victoria, BC.
http://www.env.gov.bc.ca/air/particulates/fpwtaaht.html

198

British Columbia. Provincial Health Officer. (2003). Report on the Health of British
Columbians. Provincial Health Officer's Annual Report 2002.The health and wellbeing of people in British Columbia. Victoria, B.C.: Ministry of Health Planning.
Brechler, J. 2000. Model assessment of air pollution in Prague. Environmental
Monitoring and Assessment 65: 269-276.
Burkholder, B. 2005. PM in Golden. British Columbia Ministry of Environment.
Golden BC.
Caton, R.B. and D.V. Bates. 2002. Updating BC Provincial Air Quality Objectives An Options Discussion Paper. British Columbia Ministry of Environment.
http://www.env.qov.bc.ca/air/airqualitv/pdfs/aqo paper.pdf
Chang, J.C. and S.R. Hanna. 2004. Air quality model performance evaluation.
Meteorology and Atmospheric Physics 87: 167-196.
Chang, J.C. et al. 2003. Evaluations of CALPUFF, HPAC, and VLSTRACK with two
mesoscale field datasets. Journal of Applied Meteorology 42: 453-466.
Cimorelli, A.J. et al. 2005. AERMOD: A dispersion model for industrial source
applications. Part I: General model formulation and boundary layer
characterization Journal of Applied Meteorology 44 (5): 682-693.
Cox, R.M., Sontowski, J. and C. Dougherty. 2005. An Evaluation of three
diagnostic wind models (CALMET, MCSCIPUF, and SWIFT) with wind data from
the Dipole Pride 25 experiments. Meteorological Applications 12:4: 329-341.
Curry, J.A. and P.J. Webster. 1999. Thermodynamics of atmospheres and oceans.
Academic Press, San Diego.
Dabberdt, Walter F. and Erik Miller. 2000. Uncertainty, ensembles and air quality
dispersion modeling: applications and challenges. Atmospheric Environment 34:
4667 - 4673.
Dockery, W.D., Cunningham, J., Damokosh, A.I., Neas, L.M., Spengler, J.D.,
Koutrakis, P., Ware, J.H., Raizenne, M., Speizer, F.E., 1996. Health effects of
acid aerosols on North American children: respiratory symptoms. Environmental
Health Perspectives 104: 500.
Dosio, A. et al. 2001. Assessing the meteorological conditions of a deep Italian
Alpine valley system by means of a measuring campaign and simulations with
two models during a summer smog episode. Atmospheric Environment 35:
5441-5454.

199

Enstrom, J.E. 2005. Fine particulate air pollution and total mortality among elderly
Californians, 1973-2002. Inhalation Toxicology 17 (14): 803-816.
Elbir, Tolga. 2003. Comparison of model predictions with the data of an urban air
quality monitoring network in Izmir, Turkey. Atmospheric Environments7:
2149-2157.
EPA. 2003. AERMOD: Latest Features and Evaluation Results. US Environmental
Protection Agency, http://www.epa.gov/scram001/7thconf/aermod/eval.pdf.
EPA. Draft User's Guide for the Industrial Source Complex (ISCST3) Dispersion
Models. EPA-454/ B-95-003a. US Environmental Protection Agency: North
Carolina.
Evans, G.J., and Cheol-Heon Jeong. 2007. Data Analysis and Source
Apportionment of PM2.5 in Golden, BC using Positive Matrix Factorization (PMF).
Southern Ontario Centre for Atmospheric Aerosol Research, University of
Toronto.
Gera, B.S. et al. 1996. Sodar in Dispersion Modeling. Journal of Applied
Meteorology 35: 1632-1636.
Ghose, MK, et al. 2005. Assessment of the status of urban air pollution and its
impact on human health in the city of Kolkata. Environmental Monitoring
Assessmenf\ 08 (1-3): 151-167.
Irwin, J.S. et al. A comparison of CALPUFF air quality simulation results with
monitoring data for Krakow Poland. 7th Int. Conf. on Harmonisation within
Atmospheric Dispersion Modeling for Regulatory Purposes. Belgirate, Italy. May
2001. http://rtmod.irc.it/~thunis/harmo7/P217.pdf. September 25, 2005.
Isakov, V. et al. 2004. Near-field dispersion modeling for regulatory applications.
Journal of Air and Waste Management Association 54: 473-482.
Jansen, K.L. et al. 2005. Associations between health effects and PM and black
carbon in subjects with respiratory disease. Environmental Health Perspectives
113 (12).
Kaur, S. 2005. Pedestrian exposure to air pollution along a major road in Central
London, UK. Atmospheric Environment 39 (38): 7307-7320.
Levy, Jl et al. 2002. Using CALPUFF to evaluate the impacts of power plant
emissions in Illinois: model sensitivity and implications. Atmospheric
Environment 36: 1063-1075.

200

McEwen, B.A. 2003. Applicatioin of high resolution mesoscale model fields with the
CALPUFF dispersion modeling system in Prince George, B.C. Thesis (M.Sc.)
University of Northern Britsh Columbia, Faculty of Natural Resources and
Environmental Studies.
Moraes, Osvaldo L.L. et al. 2005. Surface layer parameters over a complex terrain.
Atmospheric Environment 39: 3103-3112.
Ministry of Water, Land and Air Protection (MWLAP). 2001. Air Contaminant
Emissions (ACE) Model GIS User Manual and Report.
National Institute of Water and Atmospheric Research, NIWA. 2004. Good Practice
Guide for Atmospheric Dispersion Modeling. Ministry for the Environment, New
Zealand, www.mfe.govt.nz
Oettl, D. eta al. 2001. Evaluation of a Gaussian and a Lagrangian model against a
roadside data set, with emphasis on low wind speed conditions. Atmospheric
Environment 35: 2123-2132.
Oke, T.R. Boundary layer climates. Methuen, New York. 1987.
Olesen, H.R. 2005. User's Guide to the Model Validation Kit. Initiative on
Harmonisation within Dispersion Modelling for Regulatory Purposes.
www.harmo.org/kit
Ostermann, K., and Alex Schutte. 2004. CALMET Modeling for the Williams Lake
Airshed. Prepared for the Ministry of Environment, Cariboo Region.
http://www.env.gov.bc.ca/epd/bcairquality/reports/pdfs/CALMET_wlmslake.pdf
Ostro, B., 1993. The association of air pollution and mortality: examining the case for
inference. Archives of Environmental Health 48: 336.
Ovadnevaite, J. et al. 2006. 2002 summer fires in Lithuania: Impacts on the Vilnius
city air quality and the inhabitants health. Science of the Total Environment 356:
11-21.

Pasquill F. 1961. The estimation of the dispersion of windborne material.
Meteorological Magazine 90: 33-49.
Ratto, CF et al. (1994) Mass consistent models for wind fields over complex terrain:
the state of the art. Environmental Software 9: 247-268.
Robe, F.R. and J.S. Scire. 1998. Combining mesoscale prognostic and diagnostic
wind models: A practical approach for air quality applications in complex terrain.
10th Joint Conference on the Application of Air Pollution Meteorology. January
11-16,1998. Phoenix, Arizona.

201

Ross, D.G. et al. (1988) Diagnostic wind field modeling for complex terrain: model
development and testing. Journal of Applied Meteorology 7: 785-796.
Sakiyama, S. et al. 2005. Guidelines for air quality dispersion modeling in British
Columbia. British Columbia Ministry of Environment, Air Protection Section.
Victoria, BC.
http://wlapwww.gov.bc.ca/air/airquality/pdfs/airdispmodelguidejuly%2005.pdf
Schmitz, R. 2005. Modeling of air pollution dispersion in Santiago de Chile.
Atmospheric Environment 39: 2035-2047
Schnelle, K.B. and P.R. Dey. 2000. Atmospheric Dispersion Modeling Compliance
Guide. McGraw-Hill, New York.
Scire, J. and J. Godfrey. 2005. Earth Tech Inc. CALPUFF Training Course. Course
Notes. Denver, Colorado, 14-16 December 2005.
Scire, J.S. and F.R. Robe. 1997. Fine-scale application of the CALMET
meteorological model to a complex terrain site. Air and Waste Management
Association's 9Cfh Annual Meeting and Exhibitioin. June 8-13, 1997. Toronto,
Ontario.
Scire, J.S. et al. 2000. A User's Guide for the CALPUFF Dispersion Model. Earth
Tech, Inc.
Scire, J.S. et al. 2000b. A User's Guide for the CALMET Meteorological Model.
Earth Tech, Inc.
Scire, J.S. et al. 1999. CALPUFF Modeling System. Earth Tech, Inc.
Scire, J.S. et al. 1989. CALGRID: A Mesoscale Photochemical Grid Model Volume
II: User's Guide. California Air Resources Board. Sacremento, California.
http://www.src.com/calgrid/files/CALGRID_Users_Guide.pdf
Song, Y., Zhang, M., Cai, X. 2006. PM10 modeling of Beijing in the winter.
Atmospheric Environment 40: 4126-4136.
Triantafyllou, AG and PA Kassomenos. 2002. Aspects of atmospheric flow and
dispersion of air pollutants in a mountainous basin. The Science of the Total
Environment 297: 85-103.
US EPA, US Environmental Protection Agency, Office of Air Quality Planning and
Standards, 1996. Review of National Ambient Air Quality Standards for PM:
Policy Assessment of Scientific and Technical Information. US EPA, Washington,
DC. Report No. EPA-452/R-96-013.

202

Vedal, S. 1997. Critical review—ambient particles and health: lines that divide,
Journal of the Air and Waste Management Association 47 (5): 551-581.
Vedal, S. 1995. Health Effects of Inhalable Particles: Implications for British
Columbia. Air Resources Branch, British Columbia Ministry of Environment.
Victoria B.C.
Venkatram, A. 1988. Topics in Applied Dispersion Modeling in Lectures on Air
Pollution Modeling, eds. Venkatram, A. and J.C. Wyngaard . American
Meteorological Society, Boston.
Watson, J.G. 2002. Visibility: Science and regulation, of the Air and Waste
Management Association 52 (6): 628-713.
Whiteman, D.C. 2000. Mountain Meteorology: Fundamentals and Applications.
Oxford University Press, New York.
WHO, World Health Organization, 1994. Update and Revision of the Air Quality
Guidelines for Europe. WHO Regional Office for Europe, WHO, Copenhagen.
Report no. EUR/ ICP/EHAZ 94-05/PB0, pp. 14.
Win Lee , S et al. 2004. Important aspects in source PM2 5 emissions measurement
and characterization from stationary combustion systems. Fuel Processing
Technology 85 (6): 687-699.
Wyngaard, J.C. 1988. Structure of the PBL in Lectures on Air Pollution Modeling.
eds.
Venkatram, A. and J.C. Wyngaard . American Meteorological Society,
Boston.
Yamartino, R.J., Scire, J., Carmichael, G., Chang, Y., 1992. The CALGRID
mesoscale photochemical grid model: model formulation. Atmospheric Environment
26A (8): 1493-1512.

203