OPTIMIZATION OF PAVEMENT MAINTENANCE AND REHABILITATION USING
PAVEMENT MANAGEMENT SYSTEM IN PRINCE GEORGE
By
Lina Shehata
M.Arch, Cardiff Metropolitan University, 2023
BA, Arab Academy for Science and Technology and Maritime Transport, 2023

THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE
DEGREE OF
MASTER OF APPLIED SCIENCE IN ENGINEERING

UNIVERSITY OF NORTHERN BRITISH COLUMBIA
August 2025

© Lina Shehata, 2025

Abstract
Pavement Management Systems (PMS) are essential for guiding cost-effective and sustainable
road maintenance, particularly in municipalities operating within harsh climates and under
financial constraints. This research examines the optimization of pavement maintenance strategies
for the City of Prince George, British Columbia, by combining historical condition data, predictive
modeling, and decision-support frameworks. The study utilizes pavement distress survey results
from 2016, 2017, 2020, and 2023 to assess network-level deterioration, identify critical distress
types, and establish performance baselines.
To forecast pavement performance, three modeling approaches—Random Forest (RF), Multiple
Linear Regression (MLR), and Artificial Neural Networks (ANN)—were applied to predict the
Pavement Distress Index (PDI). These models were evaluated using the statistical metrics such as
Root Mean Square Error (RMSE), and coefficient of determination (R²). The Random Forest
model achieved the highest predictive accuracy (R² = 0.96, RMSE = 0.55), followed closely by
the ANN (R² = 0.95, RMSE = 0.48), while the MLR model demonstrated lower predictive
capability (R² = 0.81, RMSE = 0.92). Variable importance analysis identified transverse cracking,
rutting, and surface roughness as the most influential predictors of deterioration.
The findings of this research provide a data-driven framework for proactive pavement maintenance
planning in Prince George, enabling the prioritization of high-impact interventions and the
optimization of rehabilitation budgets. By extending pavement service life and reducing long-term
maintenance costs, the proposed methodology supports the creation of more resilient
transportation infrastructure. The framework can be adapted for use in other municipalities facing
similar environmental and operational conditions, strengthening the integration of advanced
analytics into municipal asset management practices.

ii

Table of Contents
Abstract........................................................................................................................................................ ii
Acknowledgements ................................................................................................................................... vii
List of Figures........................................................................................................................................... viii
List of Tables ............................................................................................................................................. xii
Glossary .................................................................................................................................................... xiii
1.

2.

Chapter 1: Introduction ..................................................................................................................... 1
1.1.

Background & Context ............................................................................................................. 1

1.2.

Problem Statement .................................................................................................................... 3

Chapter 2 : Literature Review ........................................................................................................... 8
2.1

Overview of Existing Research ................................................................................................. 8

2.2

Identifying Research Gaps ........................................................................................................ 9

2.2.1

Limited Availability of Locally Calibrated Research ........................................................ 9

2.2.2

Geographic Relevance and Northern Climate Considerations ......................................... 9

2.2.3

Lack of Strategic Data Utilization in Predictive Models ................................................. 10

2.2.4

Absence of Comprehensive Climate Adaptation Strategies ............................................ 10

2.2.5

Deficiency in Treatment Effectiveness Evaluation ........................................................... 11

2.3

Review of Deterioration Models, Optimization Techniques, and Performance Indicators
11

2.3.1

Traditional Deterioration Models ...................................................................................... 11

2.3.2

Advanced Machine Learning Approaches ........................................................................ 13

2.3.3

Optimization Techniques.................................................................................................... 14

2.3.4

Performance Indicators and Assessment Methods .......................................................... 15

2.4
2.4.1

North American Implementations ..................................................................................... 17

2.4.2

European Implementations ................................................................................................ 18

2.4.3

Low-Cost and Municipal Applications .............................................................................. 20

2.5
3.

Insights from Pavement Management Case Studies ............................................................ 17

Literature Review Summary .................................................................................................. 21

Chapter 3: Data Collection and Pre-Processing............................................................................. 23
3.1

Fundamentals of PMS ............................................................................................................. 23

3.2

Pavement Condition Indices ................................................................................................... 23

3.2.1

Pavement Distress Index (PDI) .......................................................................................... 23

3.2.2

International Roughness Index (IRI) ................................................................................ 25
iii

3.2.3
3.3

Data Gathering Methods......................................................................................................... 27

3.3.1

Automated Distress Survey Technology ........................................................................... 27

3.3.2

Positioning and Location Referencing Systems ................................................................ 28

3.3.3

Quality Control and Data Validation ................................................................................ 28

3.3.4

Strengths and Limitations of Automated Assessment ..................................................... 29

3.4

Traffic Data in PMS ................................................................................................................ 30

3.5

Pavement Distress Surveys ..................................................................................................... 31

3.5.1

Survey Coverage and Scope ............................................................................................... 31

3.5.2

Data Collection Intervals and Segmentation .................................................................... 31

3.6

Types of Pavement Distress .................................................................................................... 32

3.7

Road Categorization in PMS .................................................................................................. 33

3.7.1

Arterial Roads ..................................................................................................................... 34

3.7.2

Collector Roads ................................................................................................................... 34

3.7.3

Local Roads and Service Classifications ........................................................................... 34

3.8

Previous Treatment Methods and Maintenance History ..................................................... 35

3.8.1

Preventive Maintenance Strategies.................................................................................... 35

3.8.2

Rehabilitation Strategies..................................................................................................... 36

3.8.3

Treatment Selection Matrix and Historical Performance ............................................... 37

3.9
4.

Composite Index Integration.............................................................................................. 26

Overview of BC/Prince George-Specific PMS Reports and Policies .................................. 38

Chapter 4: Results and Analysis ...................................................................................................... 40
4.1

Pavement Distress Condition Assessment ............................................................................. 40

4.2

Overall Network Condition Evolution ................................................................................... 40

4.3

Comprehensive Distress Type Analysis ................................................................................. 42

4.3.1

Transverse Cracking Analysis ........................................................................................... 42

4.3.2

Meandering Longitudinal Cracking Analysis .................................................................. 44

4.3.3

Longitudinal Wheel Path Cracking Analysis ................................................................... 46

4.3.4

Alligator Cracking Analysis ............................................................................................... 47

4.3.5

Rutting Analysis .................................................................................................................. 49

4.3.6

Pavement Edge Cracking Analysis .................................................................................... 51

4.3.7

Longitudinal Cracking Analysis ........................................................................................ 52

4.3.8

Pothole Analysis .................................................................................................................. 54

4.4
4.4.1

Model Results and Performance Analysis ............................................................................. 56
Model Development Framework ....................................................................................... 56
iv

5.

4.4.2

Random Forest Model Performance ................................................................................. 56

4.4.3

Artificial Neural Network Model Performance ................................................................ 58

4.4.4

Multiple Linear Regression Model Performance ............................................................. 60

4.4.5

Model Comparison and Selection Criteria ....................................................................... 62

4.4.6

Three-Dimensional Relationship Analysis ........................................................................ 64

Chapter 5: Discussion, Conclusion and Recommendations .......................................................... 67
5.1

Interpretation of Findings....................................................................................................... 67

5.1.1

Distress Evolution and Network Performance Insights ................................................... 67

5.1.2

Advantages of Implementing Predictive Modeling .......................................................... 68

5.2

Lessons Learned and Limitations .......................................................................................... 69

5.2.1

Research Methodology Insights ......................................................................................... 69

5.2.2

Research Limitations and Constraints .............................................................................. 69

5.2.3

Methodological Considerations for Future Research ...................................................... 70

5.3

Conclusion ................................................................................................................................ 70

5.4

Recommendations for City of Prince George ....................................................................... 71

5.4.1

Immediate Strategic Priorities ........................................................................................... 71

5.4.2

Treatment Selection Strategy Optimization ..................................................................... 72

5.4.3

Implementation Framework for Advanced Management Systems ................................ 73

5.5

Recommendations for Future Academic Research .............................................................. 74

5.5.1

Advanced Material and Treatment Research................................................................... 74

5.5.2

Enhanced Modeling and Prediction Research.................................................................. 74

References .................................................................................................................................................. 76
Appendix I. Detailed Distress Analysis ................................................................................................... 82
I.1. Transverse Cracking Severity Analysis........................................................................................ 82
I.2. Transverse Cracking Extent Analysis .......................................................................................... 84
I.3. Meandering Longitudinal Cracking Severity Analysis ............................................................... 85
I.4. Meandering Longitudinal Cracking Extent Analysis ................................................................. 87
I.5. Longitudinal Wheel Path Cracking Severity Analysis ................................................................ 89
I.6. Longitudinal Wheel Path Cracking Extent Analysis .................................................................. 91
I.7. Alligator Cracking Severity Analysis ........................................................................................... 93
I.8. Alligator Cracking Extent Analysis .............................................................................................. 95
I.9. Rutting Severity Analysis .............................................................................................................. 97
I.10. Rutting Extent Analysis ............................................................................................................... 99
I.11. Pavement Edge Cracking Severity Analysis ............................................................................ 101
v

I.12. Pavement Edge Cracking Extent Analysis ............................................................................... 103
I.13. Longitudinal Cracking Severity Analysis ................................................................................ 105
I.14. Longitudinal Cracking Extent Analysis ................................................................................... 107
I.15. Pothole Severity Analysis ........................................................................................................... 109
I.16. Pothole Extent Analysis ............................................................................................................. 111
Appendix II. Data Extraction and Analysis.......................................................................................... 114
II.1.1. Dataset Compilation and Temporal Coverage .................................................................. 114
II.1.2. Data Structure and Variables ............................................................................................. 114
II.1.3. Data Integration Methodology............................................................................................ 115
II.2. Tools and Software Used ............................................................................................................ 115
II.2.1. Statistical Computing Environment ................................................................................... 115
II.2.2. R Package Ecosystem .......................................................................................................... 116
II.2.3. Complementary Software Tools ......................................................................................... 116
II.3. Modeling Approach .................................................................................................................... 116
II.3.1. Multi-Model Framework..................................................................................................... 116
II.3.2. Multiple Linear Regression (MLR) .................................................................................... 117
II.3.3. Random Forest Ensemble Learning ................................................................................... 117
II.3.4. Artificial Neural Network ................................................................................................... 118
II.3.5. Data Preprocessing Procedures .......................................................................................... 118
II.3.6. Model Training and Validation Framework ..................................................................... 119
II.4. Evaluation Criteria ..................................................................................................................... 119
II.4.1. Performance Metrics Framework ...................................................................................... 119
II.4.2. Model Comparison Framework ......................................................................................... 120
II.4.3. Variable Importance Analysis ............................................................................................ 121
II.4.4. Predictive Capability Framework ...................................................................................... 121
II.4.5. Model Validation and Robustness Assessment ................................................................. 121

vi

Acknowledgements
I would like to express my sincere gratitude to all those who contributed to the completion of this
research.
First and foremost, I extend my deepest appreciation to my supervisor, Dr. Mohab El-Hakim, for
his invaluable guidance, unwavering support, and expert advice throughout the entire research
process. His mentorship and dedication have been instrumental in shaping this work and my
development as a researcher.
I am grateful to my thesis committee members, Dr. Siraj Ul-Islam, Dr. Chichu Cherian, and Dr.
Leila Hashemian, for their insightful feedback, constructive comments, and valuable contributions
that helped strengthen this research.
Special thanks go to my colleague, Alireza Noory, for his essential assistance with QGIS
applications and for providing the distress photographs that greatly enhanced the visual
documentation of this thesis.
I would like to acknowledge my colleague and brother, Mahmoud Shehata, for his significant
assistance with the predictive modeling components of this research, as well as his continuous
technical and mental support throughout this journey.
I extend my sincere gratitude to the City of Prince George for providing the comprehensive survey
data that formed the foundation of this research. Their cooperation and data sharing made this
study possible and ensured its practical relevance to municipal pavement management.
Finally, I would like to express my heartfelt appreciation to my family. To my mother, Gihan Elfar,
for her unwavering support and gracious efforts in keeping me focused and on track throughout
this endeavor. To my father, Mohamed Shehata, for his constant support and generous sponsorship
that made this academic pursuit possible.
This research would not have been possible without the collective contributions, support, and
encouragement of all these individuals and organizations.

vii

List of Figures
Figure 1 Research methodology flowchart illustrating the systematic progression from data collection through
predictive modeling to optimization framework development. ........................................................................... 6
Figure 2 The performance of the MLR, FIS, and ANN models in predicting PCI, measured by their R² values (Ali et
al., 2023) ................................................................................................................................................ 12
Figure 3 Prince George Pavement Condition over the years plotted on Google Earth. ....................................... 24
Figure 4 LCMS-2 RST data collection vehicle (IMS, 2023). ............................................................................ 27
Figure 5 TT Surveyor Application Representation of LCMS System (Tetratech, 2017). ....................................... 29
Figure 6 Prince George Interactive map plotting the Road Traffic data (City of Prince George, 2025). ................ 30
Figure 7 Common types of pavement distresses observed in Prince George ...................................................... 32
Figure 8 City of Prince George 2024 Road Resurfacing (City of Prince George, 2024) ...................................... 35
Figure 9 Distresses All Years - Bar chart showing number of distresses vs. number of points analyzed across 2016,
2017, 2020, and 2023 ............................................................................................................................... 40
Figure 10 Overall PDI condition maps for all years (2016-2023) showing spatial distribution showing
Good/Fair/Poor categories across all survey years ....................................................................................... 41
Figure 11 Survey coverage comparison showing data points and network coverage by year ............................... 42
Figure 12 Transverse Cracking condition Trend (Photo credit to Mr. Alireza Noory) ......................................... 42
Figure 13 Transverse Cracking Severity in percentages All years - Comprehensive table and bar chart showing
severity distribution across all years ........................................................................................................... 43
Figure 14 Transverse Cracking Extent in percentages All years - Comprehensive table and bar chart showing Extent
distribution across all years ....................................................................................................................... 44
Figure 15 Meandering Longitudinal Cracking condition Trend (Photo credit to Mr. Alireza Noory) ..................... 44
Figure 16 Meandering Longitudinal Cracking Severity in percentages All years - Comprehensive table and bar
chart showing Extent distribution across all years ........................................................................................ 45
Figure 17 Meandering Longitudinal Cracking Extent in percentages All years - Comprehensive table and bar chart
showing Extent distribution across all years ................................................................................................ 45
Figure 18 Longitudinal Wheel Path Cracking Trend ..................................................................................... 46
Figure 19 Longitudinal Wheel path Cracking Severity in percentages All years - Comprehensive table and bar chart
showing Severity distribution across all years .............................................................................................. 46
Figure 20 Longitudinal Wheel path Cracking Extent in percentages All years - Comprehensive table and bar chart
showing Extent distribution across all years ................................................................................................ 47
Figure 21 Alligator Cracking Trend ............................................................................................................ 47
Figure 22 Alligator Cracking Severity in percentages All years - Comprehensive table and bar chart showing
Severity distribution across all years........................................................................................................... 48
Figure 23 Alligator Cracking Extent in percentages All years - Comprehensive table and bar chart showing Extent
distribution across all years ....................................................................................................................... 49
Figure 24 Rutting Trend ............................................................................................................................ 49
Figure 25 Rutting Severity in percentages All years - Comprehensive table and bar chart showing Severity
distribution across all years ....................................................................................................................... 50
Figure 26 Rutting Extent in percentages All years - Comprehensive table and bar chart showing Extent distribution
across all years ........................................................................................................................................ 50
Figure 27 Pavement Edge Cracking Trend ................................................................................................... 51

viii

Figure 28 Pavement Edge Cracking Severity in percentages All years - Comprehensive table and bar chart showing
Severity distribution across all years........................................................................................................... 51
Figure 29 Pavement Edge Cracking Extent in percentages All years - Comprehensive table and bar chart showing
Extent distribution across all years ............................................................................................................. 52
Figure 30 Longitudinal Cracking condition Trend (Photo credit to Mr. Alireza Noory) ....................................... 52
Figure 31 Longitudinal Cracking Severity in percentages All years - Comprehensive table and bar chart showing
Severity distribution across all years........................................................................................................... 53
Figure 32 Longitudinal Cracking Extent in percentages All years - Comprehensive table and bar chart showing
Extent distribution across all years ............................................................................................................. 54
Figure 33 Pothole condition Trend (Photo credit to Mr. Alireza Noory) ............................................................ 54
Figure 34 Pothole Severity in percentages All years - Comprehensive table and bar chart showing Severity
distribution across all years ....................................................................................................................... 55
Figure 35 Pothole Extent in percentages All years - Comprehensive table and bar chart showing Extent distribution
across all years ........................................................................................................................................ 55
Figure 36 Random Forest variable importance plot showing key contributing factors to PDI prediction .............. 57
Figure 37 Random Forest actual vs. predicted PDI scatter plot with performance statistics ................................ 58
Figure 38 ANN model actual vs. predicted PDI scatter plot with trend line and performance statistics ................. 59
Figure 39 ANN model architecture diagram showing input variables, hidden layer, and output structure .............. 60
Figure 40 MLR model actual vs. predicted PDI scatter plot with regression line ............................................... 61
Figure 41 MLR coefficient significance plot showing predictor variable contributions ....................................... 62
Figure 42 Model performance comparison chart showing MSE, RMSE, and R² values across all three models. ..... 63
Figure 43 Three-dimensional relationship visualization between IRI, PDI, and distress severity levels showing the
complex interactions between functional performance, structural condition, and distress intensity ...................... 64
Figure 44 Three-dimensional relationship visualization between IRI, PDI, and distress extent measurements
demonstrating how spatial distribution of distresses influences both functional and structural performance
indicators ............................................................................................................................................... 65

FIGURE I. 1. 2016 Transverse Cracking Severity pie chart and spatial distribution map ................................... 82
FIGURE I. 2. 2017 Transverse Cracking Severity pie chart and spatial distribution map ................................... 82
FIGURE I. 3. 2020 Transverse Cracking Severity pie chart and spatial distribution map ................................... 83
FIGURE I. 4. 2023 Transverse Cracking Severity pie chart and spatial distribution map ................................... 83
FIGURE I. 5. 2016 Transverse Cracking Extent pie chart and spatial distribution map...................................... 84
FIGURE I. 6. 2017 Transverse Cracking Extent pie chart and spatial distribution map...................................... 84
FIGURE I. 7. 2020 Transverse Cracking Extent pie chart and spatial distribution map...................................... 85
FIGURE I. 8. 2023 Transverse Cracking Extent pie chart and spatial distribution map...................................... 85
FIGURE I. 9. 2016 Meandering Longitudinal Cracking Severity pie chart and spatial distribution map ............... 86
FIGURE I. 10. 2017 Meandering Longitudinal Cracking Severity pie chart and spatial distribution map ............. 86
FIGURE I. 11. 2020 Meandering Longitudinal Cracking Severity pie chart and spatial distribution map ............. 87
FIGURE I. 12. 2023 Meandering Longitudinal Cracking Severity pie chart and spatial distribution map ............. 87
FIGURE I. 13. 2016 Meandering Longitudinal Cracking Extent pie chart and spatial distribution map ............... 88
FIGURE I. 14. 2017 Meandering Longitudinal Cracking Extent pie chart and spatial distribution map ............... 88
FIGURE I. 15. 2020 Meandering Longitudinal Cracking Extent pie chart and spatial distribution map ............... 89
FIGURE I. 16. 2023 Meandering Longitudinal Cracking Extent pie chart and spatial distribution map ............... 89
FIGURE I. 17. 2016 Longitudinal Wheel Path Cracking Severity pie chart and spatial distribution map .............. 90
FIGURE I. 18. 2017 Longitudinal Wheel Path Cracking Severity pie chart and spatial distribution map .............. 90

ix

FIGURE I. 19. 2020 Longitudinal Wheel Path Cracking Severity pie chart and spatial distribution map .............. 91
FIGURE I. 20. 2023 Longitudinal Wheel Path Cracking Severity pie chart and spatial distribution map .............. 91
FIGURE I. 21. 2016 Longitudinal Wheel Path Cracking Extent pie chart and spatial distribution map ................ 92
FIGURE I. 22. 2017 Longitudinal Wheel Path Cracking Extent pie chart and spatial distribution map ................ 92
FIGURE I. 23. 2020 Longitudinal Wheel Path Cracking Extent pie chart and spatial distribution map ................ 93
FIGURE I. 24. 2023 Longitudinal Wheel Path Cracking Extent pie chart and spatial distribution map ................ 93
FIGURE I. 25. 2016 Alligator Cracking Severity pie chart and spatial distribution map .................................... 94
FIGURE I. 26. 2017 Alligator Cracking Severity pie chart and spatial distribution map .................................... 94
FIGURE I. 27. 2020 Alligator Cracking Severity pie chart and spatial distribution map .................................... 95
FIGURE I. 28. 2023 Alligator Cracking Severity pie chart and spatial distribution map .................................... 95
FIGURE I. 29. 2016 Alligator Cracking Extent pie chart and spatial distribution map ....................................... 96
FIGURE I. 30. 2017 Alligator Cracking Extent pie chart and spatial distribution map ....................................... 96
FIGURE I. 31. 2020 Alligator Cracking Extent pie chart and spatial distribution map ....................................... 97
FIGURE I. 32. 2023 Alligator Cracking Extent pie chart and spatial distribution map ....................................... 97
FIGURE I. 33. 2016 Rutting Severity pie chart (showing 100% no distress) ..................................................... 98
FIGURE I. 34. 2017 Rutting Severity pie chart and spatial distribution map .................................................... 98
FIGURE I. 35. 2020 Rutting Severity pie chart and spatial distribution map .................................................... 99
FIGURE I. 36. 2023 Rutting Severity pie chart and spatial distribution map .................................................... 99
FIGURE I. 37. 2016 Rutting Extent pie chart (showing 100% no distress)...................................................... 100
FIGURE I. 38. 2017 Rutting Extent pie chart and spatial distribution map ..................................................... 100
FIGURE I. 39. 2020 Rutting Extent pie chart and spatial distribution map ..................................................... 101
FIGURE I. 40. 2023 Rutting Extent pie chart and spatial distribution map ..................................................... 101
FIGURE I. 41. 2016 Pavement Edge Cracking Severity pie chart and spatial distribution map ......................... 102
FIGURE I. 42. 2017 Pavement Edge Cracking Severity pie chart and spatial distribution map ......................... 102
FIGURE I. 43. 2020 Pavement Edge Cracking - No data collected notation ................................................... 103
FIGURE I. 44. 2023 Pavement Edge Cracking Severity pie chart and spatial distribution map ......................... 103
FIGURE I. 45. 2016 Pavement Edge Cracking Extent pie chart and spatial distribution map ........................... 104
FIGURE I. 46. 2017 Pavement Edge Cracking Extent pie chart and spatial distribution map ........................... 104
FIGURE I. 47. 2020 Pavement Edge Cracking - No data collected notation ................................................... 105
FIGURE I. 48. 2023 Pavement Edge Cracking Extent pie chart and spatial distribution map ........................... 105
FIGURE I. 49. 2016 Longitudinal Cracking Severity pie chart and spatial distribution map ............................. 106
FIGURE I. 50. 2017 Longitudinal Cracking Severity pie chart and spatial distribution map ............................. 106
FIGURE I. 51. 2020 Longitudinal Cracking Severity pie chart and spatial distribution map ............................. 107
FIGURE I. 52. 2023 Longitudinal Cracking Severity pie chart and spatial distribution map ............................. 107
FIGURE I. 53. 2016 Longitudinal Cracking Extent pie chart and spatial distribution map ............................... 108
FIGURE I. 54. 2017 Longitudinal Cracking Extent pie chart and spatial distribution map ............................... 108
FIGURE I. 55. 2020 Longitudinal Cracking Extent pie chart and spatial distribution map ............................... 109
FIGURE I. 56. 2023 Longitudinal Cracking Extent pie chart and spatial distribution map ............................... 109
FIGURE I. 57. 2016 Pothole Severity pie chart and spatial distribution map .................................................. 110
FIGURE I. 58. 2017 Pothole Severity pie chart and spatial distribution map .................................................. 110
FIGURE I. 59. 2020 Pothole Severity pie chart and spatial distribution map .................................................. 111
FIGURE I. 60. 2023 Pothole Severity pie chart and spatial distribution map .................................................. 111
FIGURE I. 61. 2016 Pothole Extent pie chart and spatial distribution map .................................................... 112
FIGURE I. 62. 2017 Pothole Extent pie chart and spatial distribution map .................................................... 112
FIGURE I. 63. 2020 Pothole Extent pie chart and spatial distribution map .................................................... 113

x

FIGURE I. 64. 2023 Pothole Extent pie chart and spatial distribution map .................................................... 113

FIGURE II. 1. Sample of Data Tables ....................................................................................................... 114

xi

List of Tables
Table 1 Details of previous research activities on PMS .................................................................................. 21
Table 2 Coefficients to Calculate the Deducts of PDI .................................................................................... 25
Table 3 Index Ranges for IRI Description .................................................................................................... 26
Table 4 Prince George Road Classification System ....................................................................................... 33
Table 5 Treatment Matrix for Arterial/Collector Roadways ............................................................................ 38

xii

Glossary

Abbreviation

Full Form

PMS

Pavement Management System

MLR

Multiple Linear Regression

RF

Random Forest

ANN

Artificial Neural Network

GIS

Geographic information system

PCI

Pavement Condition Index

PDI

Pavement Distress Index

BC

British Columbia

LCCA

Life Cycle cost Analysis

IRI

International Roughness index

SPS

Strategic Planning System

NDT

Non destructive Testing

AASHTO

American Association of State Highway and Transportation Officials

AADT

Annual Average Daily Traffic

RMSE

Root Mean Square Error

FHWA

Federal Highway Administration

xiii

1. Chapter 1: Introduction
1.1.

Background & Context

Efficient and well-maintained pavement infrastructure is fundamental to the economic prosperity,
public safety, and quality of life for residents of any municipality. In northern British Columbia,
particularly in Prince George, road networks serve as the critical backbone for regional
connectivity, commerce, and community development. As a strategic transportation hub located at
the junction of the Fraser and Nechako rivers, Prince George facilitates the movement of goods,
services, and people across a vast geographic region, making the integrity of its pavement
infrastructure paramount to both local and regional economic vitality.
Prince George, with a population of approximately 72,000 residents within the city limits and
serving an additional 320,000 residents in the surrounding region (Statistics Canada, 2021),
maintains an extensive road network spanning 605 kilometers. This network comprises 214 km of
major arterial roads, 155 km of collector roads, and 230 km of local roads, each serving distinct
functional roles and experiencing a wide variety of traffic loading and environmental stresses (City
of Prince George, 2009). The city's strategic position as the largest urban center in northern British
Columbia places significant demands on its transportation infrastructure, as it serves as a
distribution hub for forestry, mining, and agricultural activities throughout the region.
The unique geographical and climatic characteristics of Prince George present exceptional
challenges for pavement infrastructure management. Located at latitude 53°53'37"N and longitude
122°45'02"W (Latitude.to, 2024), the city experiences a humid continental climate characterized
by severe winters and warm summers. Average air temperatures range from -10°C in winter to
16°C in summer, with extreme temperatures spanning from -40°C to 35°C (Pacific Climate
Impacts Consortium, 2024). This wide temperature variation, combined with an annual
precipitation average of 600 -700mm including heavy snowfall (approximately 216 cm annually)
(Prince George Weather Statistics, 2024), creates a challenging environment for pavement
materials and structures.
The City of Prince George faces multiple freeze-thaw cycles annually, which significantly impact
pavement integrity through thermal expansion and contraction of materials, moisture infiltration
and subsequent ice formation, and accelerated deterioration of both surface and subsurface
pavement components. The spring melt period creates additional challenges through increased
moisture content and reduced bearing capacity of subgrade materials, contributing to premature
pavement failures and increased maintenance requirements.
The concept of infrastructure management has evolved to address these complex challenges
through systematic approaches to asset stewardship. According to Hendrickson, Coffelt, and
Healey in Fundamentals of Infrastructure Management, "infrastructure management should adopt
a 'triple bottom line' to consider economic, environmental and social impacts" while maintaining
1

"a life cycle or long-term viewpoint" that is "essential for good infrastructure management"
(Hendrickson et al., 2017). This perspective recognizes that infrastructure investments "will last
for decades or more and providing good performance over an entire lifetime is critical for good
infrastructure management."
Within this broader infrastructure management framework, Pavement Management Systems
(PMS) have emerged as specialized tools designed to optimize the preservation and performance
of road networks. The Federal Highway Administration's Pavement Management Primer provides
a formal definition, describing pavement management as "a set of tools or methods that assist
decision-makers in finding optimum strategies for providing, evaluating, and maintaining
pavements in a serviceable condition over a period of time" (FHWA, 2022). The primer
emphasizes that "pavement management helps to ensure pavement investments meet the agency's
pavement condition goals and targets" while "implementing pavement management involves
establishing clear pavement management functions, policies, and procedures."
The systematic nature of PMS is further elaborated in the infrastructure deterioration modeling
literature, which explains how these systems "fit within the infrastructure deterioration modeling
field" by providing capabilities for "forecasting pavement behavior and performance standards"
(Zakeri, 2016). This forecasting capability is essential for municipal decision-making, as it enables
agencies to predict future pavement conditions and optimize maintenance timing and resource
allocation.
Saliminejad describes the logical structure of PMS systems, explaining that they integrate multiple
components including "inputs (distress data), outputs (treatment schedules), and feedback loops"
that work together to support systematic infrastructure management. The system structure includes
"inventory data collection, conditions assessment, a determination of needs, the prioritization of
projects needing maintenance and rehabilitation, a method of determining the impact of funding
decisions, and a feedback process" (Saliminejad, 2013).
The evolution of PMS technology represents a significant advancement in infrastructure
management capabilities. Recent research in automated decision-making highlights how these
systems have evolved "from manual inspections to automated, Machine Learning (ML) -driven
systems," demonstrating the "relevance and innovation in PMS today" (Li et al., 2022). This
technological evolution enables more sophisticated analysis of pavement performance factors,
including "traffic conditions, climatic characteristics, and maintenance history" that "exhibit a
close relationship with pavement performance."
The economic significance of maintaining high-quality pavement infrastructure extends beyond
immediate transportation needs. Prince George serves as a critical link in provincial and national
transportation corridors, with major highways including Highway 97 (connecting southern British
Columbia to Alaska) and Highway 16 (the Yellowhead Highway connecting eastern and western
Canada) passing through the city. The condition of local pavement infrastructure directly impacts
2

the efficiency of goods movement, tourism accessibility, and the overall economic competitiveness
of the region.
Furthermore, the forestry and mining industries, which form the economic foundation of the
region, rely heavily on efficient transportation networks for moving products to markets and
bringing supplies to operational sites. Deteriorated pavement conditions result in increased vehicle
operating costs, reduced fuel efficiency, and potential supply chain disruptions that can have
cascading economic effects throughout the region.
1.2.

Problem Statement

Despite ongoing maintenance efforts and significant financial investments, several critical issues
continue to undermine the efficiency and longevity of pavement infrastructure in Prince George.
Technical complexities, environmental impacts, and resource management problems represent
multiple layers of challenges that require comprehensive, innovative solutions aligned with
contemporary sustainability principles.
1.2.1.

Lack of Scientific Research and Strategic Planning

The foremost challenge facing Prince George's pavement management approach is the absence of
a scientifically based, data-driven framework for decision-making. Current maintenance practices
are not fully supported by technical and statistical analysis, resulting in inefficiencies in resource
allocation and maintenance prioritization. This deficiency manifests in reactive rather than
proactive maintenance strategies, leading to suboptimal outcomes and increased costs over the
infrastructure lifecycle. Without a robust analytical foundation, maintenance decisions often rely
heavily on subjective assessments and historical practices rather than predictive modeling and
performance optimization.
Research in sustainable pavement management emphasizes that "lack of PMS leads to
deteriorating road conditions and reactive maintenance" approaches that fail to optimize long-term
infrastructure performance (Khahro, 2022). The absence of systematic planning particularly affects
municipalities facing budget constraints, where inefficient resource allocation can have significant
long-term consequences for infrastructure condition and service delivery.
1.2.2. Challenging Climate Conditions and Environmental Stressors
Prince George's location in northern British Columbia subjects its pavement infrastructure to some
of the most severe environmental conditions in Canada. The region's extreme seasonal temperature
variations, combined with frequent freeze-thaw cycles, create accelerated deterioration conditions
that significantly challenge traditional pavement management approaches.
The literature on flexible pavement management in harsh climates indicates that environmental
factors play a dominant role in pavement deterioration patterns. Research demonstrates that freezethaw cycles are among the most destructive forces affecting pavement infrastructure in northern
3

climates, causing expansion and contraction stresses that lead to cracking, joint deterioration, and
eventual structural failure (Khahro, 2022).
The superposition of Prince George's extreme climatic conditions with heavy commercial traffic
creates an accelerated deterioration environment that substantially increases both the frequency
and cost of required repairs. This combination of environmental and loading stresses requires
specialized management approaches that account for the unique characteristics of northern climate
pavement performance.
1.2.3. Limited Integration of Advanced Technologies and Predictive Capabilities
Prince George's current pavement management approach has not fully embraced the technological
advancements that characterize modern PMS implementation. The evolution toward automated,
ML-driven systems represents a significant opportunity for improved decision-making and
resource optimization that remains largely unutilized in the local context (Li et al., 2022).
Contemporary PMS implementations leverage advanced modeling techniques and automated data
collection to improve prediction accuracy and decision support capabilities. The integration of
such technologies could significantly enhance Prince George's ability to optimize maintenance
strategies and resource allocation while reducing long-term infrastructure costs.
1.2.4. Inadequate Performance Monitoring and Data Management
Effective pavement management requires comprehensive condition monitoring and data
management capabilities that support both immediate decision-making and long-term strategic
planning. Current assessment techniques employed in Prince George lack the precision and
comprehensiveness required for effective predictive maintenance strategies.
The infrastructure management literature emphasizes that "deterioration modeling is a process of
taking condition assessment information and forecasting expected future conditions" that requires
systematic data collection and analysis capabilities (Hendrickson et al., 2017). Without adequate
performance monitoring systems, it becomes challenging to prioritize maintenance interventions
effectively and predict long-term performance trends.
These interconnected challenges demonstrate the urgent need for a comprehensive, scientifically
based pavement management system specifically tailored to Prince George's unique environmental
and operational conditions. Addressing these issues requires the integration of climate-adaptive
strategies, advanced predictive modeling, sustainability principles, and contemporary management
technologies within a unified framework that supports both immediate operational needs and longterm infrastructure stewardship.
1.3.Research Objectives
Objective 1: Develop a PMS framework tailored to Prince George's unique material availability,
climatic and traffic conditions.
4

Objective 2: Identify main deterioration trends and patterns of pavements in Prince George by
analyzing historical pavement distress data.
Objective 3: Implement predictive modeling to forecast pavement performance and optimize
maintenance planning.
Objective 4: Provide practical recommendations for a climate-adaptive pavement management
strategy.
1.4.Methodology
This research employs a systematic data-driven approach to develop a climate-adaptive pavement
management system tailored to Prince George's environmental conditions. The methodology
addresses critical gaps in locally calibrated PMS models for Northern British Columbia through
integrated empirical analysis, predictive modeling, and optimization techniques.
1.4.1. Research Framework
The study follows a four-phase approach: data collection and preprocessing, deterioration analysis,
predictive modeling, and optimization framework development. This systematic progression
ensures comprehensive analysis of technical, environmental, and economic factors while
maintaining practical applicability for municipal implementation.
1.4.2. Data Collection and Pre-processing
Historical pavement condition data spanning 2016, 2017, 2020, and 2023 provides the empirical
foundation for analysis. Traffic data including Average Annual Daily Traffic (AADT) and vehicle
classification quantifies loading effects on deterioration. Climate data from Environment and
Climate Change Canada captures temperature extremes, precipitation patterns, and freeze-thaw
cycles characteristic of Prince George's harsh conditions. Maintenance history records document
treatment applications, timing, and costs. GIS data provides spatial contexts for all analytical
components.
Data preprocessing includes systematic cleaning, validation, missing data handling, and temporal
alignment to ensure analytical reliability and consistency across datasets.
1.4.3.

Analytical Methods

Statistical analysis identifies deterioration patterns through descriptive statistics, time-series
analysis, and correlation assessment between Pavement Deterioration Indices (PDI) values,
environmental factors, traffic loading, and maintenance history. Climate impact analysis quantifies
freeze-thaw effects and seasonal deterioration patterns. Treatment effectiveness evaluation
employs survival analysis and cost-effectiveness comparison of maintenance interventions under
local conditions.
1.4.4.

Predictive Modeling
5

The use of predictive modelling captures complex non-linear interactions between performance
drivers to identify key deterioration trends. Local calibration adapts models to Prince George's
specific conditions using regional data. Split-sample validation assesses model performance using
R², Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) metrics across different
road categories and environmental conditions.

Figure 1 Research methodology flowchart illustrating the systematic progression from data collection through predictive
modeling to optimization framework development.

1.4.5. Optimization Framework
Multi-Criteria Decision Analysis integrates performance, cost, and sustainability objectives for
systematic maintenance alternative evaluation. Climate-adaptive strategies optimize seasonal
maintenance windows and incorporate risk-based approaches. Life-cycle cost analysis provides
comprehensive economic evaluation balancing performance objectives with budget constraints.
1.4.6. Implementation
Framework validation employs retrospective analysis and scenario testing under varying
conditions. RStudio provides statistical analysis and modeling capabilities while ArcGIS enables
spatial analysis and visualization. Quality assurance protocols ensure professional standards for
municipal infrastructure management applications.
1.5.Organization of Thesis
This thesis is organized into five chapters that collectively address the research objectives,
document the methodology, present the findings, and provide actionable recommendations. The
6

structure ensures a logical flow from background context to practical implementation, emphasizing
both technical rigor and sustainability principles.
Chapter 1 – Introduction
This chapter provides the foundation for the research by establishing the background and context
of pavement management in northern climates, outlining the specific challenges faced by the City
of Prince George, and defining the research objectives and questions. It also highlights the
significance of the study for both local application and broader contributions to sustainable
infrastructure management.
Chapter 2 – Literature Review
This chapter presents a comprehensive review of prior research in pavement management systems,
with emphasis on cold-climate applications, climate-adaptive strategies, and sustainability
integration. It identifies research gaps relevant to northern British Columbia and establishes the
theoretical foundation for the research. The review also examines recent advances in predictive
modeling, deterioration analysis, and economic evaluation methods relevant to sustainable
pavement management.
Chapter 3 – Pavement Management Systems: Concepts and Context
This chapter describes the key concepts, methods, and contextual factors underlying pavement
management systems. Topics include pavement condition indices, data collection methods and
equipment, traffic data in PMS, pavement distress surveys, distress types, road classification, and
historical maintenance practices. The chapter also reviews relevant policies, technical reports, and
data specific to Prince George and the province of British Columbia.
Chapter 4 – Methodology and Research Tools
This chapter details the study area, datasets, and tools used in the research, including RStudio,
Python, and GIS. It describes the modeling approaches—Random Forest, Multiple Linear
Regression, and Artificial Neural Networks—along with data preparation procedures, model
calibration, and performance evaluation criteria such as Mean Squared Error (MSE), Root Mean
Square Error (RMSE), and coefficient of determination (R²).
Chapter 5 – Results, Discussion, and Recommendations
The final chapter presents the results of the pavement condition assessment and predictive
modeling, including analysis of distress trends and model performance. It discusses the
implications of the findings for pavement management policy and planning, provides practical
recommendations for the City of Prince George, and outlines potential areas for future research.

7

2. Chapter 2 : Literature Review
2.1 Overview of Existing Research
Pavement Management Systems (PMS) have emerged as critical tools for optimizing infrastructure
maintenance and rehabilitation strategies, representing a systematic approach of decision-making
that integrates data collection, performance modeling, and cost-effectiveness analysis. The
evolution of PMS research has been driven by the recognition that traditional reactive maintenance
approaches are insufficient for managing aging infrastructure networks under increasing traffic
loads and constrained budgets.
The conceptual foundation of modern pavement management was established by Haas, who
introduced PMS as a structured approach emphasizing cost-effectiveness and optimization in
maintenance planning (Haas et al., 1994). This foundational work demonstrated that systematic
approaches to pavement management could reduce lifecycle costs by up to 30% compared to
reactive maintenance strategies. Building upon this foundation, Hudson expanded the scope by
integrating engineering and economic principles, focusing on infrastructure management strategies
that encompass design, construction, maintenance, rehabilitation, and research activities (Hudson
et al., 1997).
Contemporary research has increasingly focused on advanced analytical techniques, building upon
earlier findings that demonstrated the effectiveness of systematic approaches. Anastasopoulos
provided early evidence that optimized maintenance interventions can extend pavement life by 2040% when properly timed (Anastasopoulos et al., 2009). This foundational research supported the
subsequent development of more sophisticated analytical frameworks. Smith later emphasized the
integration of asset management perspectives into pavement decision-making processes,
highlighting the importance of prioritizing maintenance based on objective road condition data
rather than subjective assessments (Smith et al., 2014). This evolution toward data-driven
approaches has been further validated by recent advances in machine learning and predictive
modeling techniques.
The evolution of PMS research has also been characterized by increasing sophistication in
predictive modeling techniques. Recent studies have demonstrated significant improvements in
prediction accuracy through the application of machine learning algorithms. Ali showed that
artificial neural network models achieved R² values of 98.6% for 2018 pavement condition
predictions and 99.3% for 2021 predictions, substantially outperforming traditional multiple linear
regression approaches, which achieved R² values of only 48.0% and 63.0% respectively (Ali et al.,
2023).
US Federal initiatives contributed to the direction of PMS research and implementation. The
Federal Highway Administration's Pavement Management Roadmap established a vision for
pavement management advancement over the next decade, recognizing transformative innovations
in data collection, performance modeling, and decision-making processes (Federal Highway
8

Administration, 2022). The roadmap emphasizes the importance of collaborative efforts between
industry, academia, and transportation agencies to reduce duplication of effort and advance
innovative pavement management practices.
2.2

Identifying Research Gaps
2.2.1

Limited Availability of Locally Calibrated Research

A critical gap exists in the availability of locally calibrated PMS models for Northern British
Columbia's unique environmental and traffic conditions. Most existing studies focus on urban and
temperate regions where climate and traffic conditions differ significantly from those experienced
in Northern BC. Research by Ekramnia and Nasimifar emphasized that existing models often fail
to provide accurate deterioration predictions for extreme climate conditions, leading to suboptimal
maintenance planning and increased lifecycle costs (Ekramnia & Nasimifar, 2022).
The calibration challenges are exemplified by Ontario's experience with AASHTOWare
implementation. Hamdi demonstrated that most North American studies of local calibration
concluded that national calibration coefficients fail to offer reliable accuracy or precision (Hamdi,
2015). The AASHTOWare system was developed based on Long Term Pavement Performance
(LTPP) sections from various regions, showing significant variation in binder and aggregate
properties, climate conditions, and traffic spectrum. Local calibration projects have consistently
shown enhanced model accuracy in predicting pavement performance, with statistical analysis
demonstrating serious need for incorporation of local calibrations into predictive models.
Similarly, Washington State Department of Transportation's calibration efforts revealed substantial
regional variations in pavement performance. Li found that default calibration factors required
significant adjustment, with rutting predictions showing almost perfect correlation with measured
values only after local calibration (Li, 2009). The study utilized both split-sample and jackknife
testing approaches, demonstrating that local calibration improves prediction stability and accuracy
even with limited sample sizes.
2.2.2

Geographic Relevance and Northern Climate Considerations

The geographic specificity of pavement management challenges represents a significant research
gap, particularly for cold climate regions. Research has shown that climate change impacts vary
substantially by geographic location and climate zone. Studies conducted across different
European locations demonstrated varying responses to temperature and precipitation changes, with
some regions experiencing increases in all distress types except thermal cracking due to
temperature rise (Qiao et al., 2020).
Swarna identified critical limitations in existing temperature models, noting that the LTPP model
was developed only for maximum latitudes of 52 degrees and has had limited testing in Northern
Canadian climates (Swarna, 2022). This limitation is particularly significant for Prince George,
located at approximately 53.9 degrees north latitude, where extreme temperatures cannot be
accurately represented by models developed for southern regions.
9

The Canadian context presents unique challenges that are not adequately addressed in international
literature. The Canadian Strategic Highway Research Program initiated research in the late 1980s
to study climate effects on roadway efficiency, establishing test sites in Lamont, Alberta; Hearst,
Ontario; and Sherbrooke, Quebec (Gavin et al., 2003). However, these efforts focused primarily
on asphalt properties at low temperatures rather than comprehensive PMS frameworks for northern
regions.
2.2.3

Lack of Strategic Data Utilization in Predictive Models

A significant gap exists in the strategic integration of real-world pavement condition data with
predictive modeling frameworks. While various statistical and machine learning models have been
proposed for pavement condition forecasting, many studies fail to integrate comprehensive
datasets that reflect local traffic patterns, material properties, and environmental stressors (Ali,
2022). This limitation reduces the applicability of models for regional pavement management, as
they do not accurately capture specific deterioration patterns observed in Northern BC.
The Federal Highway Administration's pavement management primer highlights this challenge,
noting that effective pavement management requires integration of multiple data sources including
inventory data, condition assessment results, performance prediction models, and strategic-level
data (Federal Highway Administration, 2022). However, most agencies struggle to achieve this
integration due to technical and resource constraints.
Li addressed this gap by developing a hybrid neural network approach that considers maintenance
impacts on data through a "restart point method" for data cleaning (Li et al., 2022). Their research
demonstrated that most existing studies simply remove noisy data instead of processing it
meticulously and fail to consider maintenance impacts on performance data, which can lead to
significant errors in performance prediction.
2.2.4

Absence of Comprehensive Climate Adaptation Strategies

Current PMS frameworks lack comprehensive climate adaptation strategies specifically designed
for Northern BC conditions. Although research emphasizes the importance of incorporating
climate resilience into pavement management, a few studies have focused on how specific climate
factors such as snow accumulation, extended freeze-thaw cycles, and extreme temperature
fluctuations affect pavement maintenance (Guha et al., 2022).
Research has shown that climate change can cause changes in pavement lifecycle costs depending
on changes in climate stressors, pavement structure, materials, and maintenance regimes. Qiao
demonstrated that maintenance interventions may be triggered much earlier than expected due to
climate change, and that adapting to early maintenance can minimize total lifecycle costs while
improving pavement resilience (Qiao et al., 2015). However, this research has not been specifically
adapted to Northern BC conditions.
Studies utilizing multiple climate models have shown that increased temperatures can lead to
asphalt rutting increases of 9-40% and fatigue cracking increases of 2-9% across various United
10

States locations (Swarna, 2022). However, similar comprehensive studies have not been conducted
for Canadian northern climates, representing a significant gap in understanding climate change
impacts on regional pavement performance.
2.2.5

Deficiency in Treatment Effectiveness Evaluation

Existing PMS frameworks show significant deficiencies in evaluating treatment effectiveness
under diverse climate and traffic conditions. Hafez and Ksaibati emphasized the need for improved
evaluation metrics to assess long-term performance of various pavement treatments, noting that
existing models often analyze effectiveness using short-term performance data (Hafez & Ksaibati,
2021). Their research on low-volume roads demonstrated that adjusted treatment strategies could
extend pavement life by 15-25% when properly evaluated.
The treatment effectiveness challenge is compounded by the diversity of available interventions.
Research by Al-Swailmi identified multiple maintenance procedures including sealing, cobbling,
milling, overlay application, and repair (Al-Swailmi et al., 1999). Each rehabilitation technique
has direct effects on pavement cycle-life, but the effectiveness varies significantly based on local
conditions including climate, traffic, and material properties.
Limited research has been conducted on treatment effectiveness in mitigating climate-induced
pavement deterioration, particularly in cold regions where environmental stressors vary
considerably compared to other climates. This gap is particularly concerning for Prince George,
where freeze-thaw cycles and heavy snowfall accelerate pavement deterioration and may require
specialized treatment approaches.
2.3

Review of Deterioration Models, Optimization Techniques, and Performance
Indicators
2.3.1

Traditional Deterioration Models

Traditional pavement deterioration models have primarily relied on deterministic and statistical
approaches, with multiple linear regression (MLR) serving as the foundation for many early PMS
applications. Research by Ali demonstrated the limitations of traditional MLR approaches,
showing that linear assumptions inherent in MLR models fail to capture the complex, non-linear
interactions present in pavement deterioration, particularly under extreme climate conditions (Ali
et al., 2023).
Ali et al., 2023 in Figure 2, draw the following conclusions based on the research model:
•

•

PCI (2018): ANN has the best performance among the three techniques. The statistical
measures for the ANN model are R2 = 98.6%, RMSE = 0.88, and MAE = 0.73, and the
worst-performing model was MLR, with an R2 = 48.0%, RMSE = 14.05, and MAE = 11.37
(Ali et al., 2023).
PCI (2021): ANN has the best performance among the three techniques. The statistical
measures for the ANN model are R2 = 99.3%, RMSE = 0.72, and MAE = 0.59, and the
11

worst-performing model was MLR, with an R2 = 63.0%, RMSE = 9.93, and MAE =7.84
(Ali et al., 2023).

Figure 2 The performance of the MLR, FIS, and ANN models in predicting PCI, measured by their R²
values (Ali et al., 2023)

Conventional deterioration modeling efforts focused on simplified pavement performance models.
Lee et al. (1993) developed pavement performance models using empirical approaches, while
earlier researchers like Butt et al. (1987) introduced Markov process-based prediction models for
pavement performance. These foundational models established the basic framework of
performance prediction but were limited by their reliance on deterministic assumptions and
simplified deterioration relationships.
Saliminejad provided detailed analysis of how measurement errors affect different modeling
approaches, demonstrating that regression models using least-squares methods are subject to errors
propagated from estimated parameters (Saliminejad, 2012). Random errors in condition index data
do not cause bias in linear regression models within large datasets, but random errors in age data
bias predictions by causing slope underestimation, resulting in predicted condition indices that are
typically smaller than true values.
On the pavement design side, the development of mechanistic-empirical approaches represented a
significant advancement over purely empirical models. According to AASHTO, pavement
performance is defined as the serviceability trend of the pavement over its design period, with
serviceability indicating the ability of the pavement in its existing condition to serve traffic demand
(AASHTO, 1993). This definition provided the foundation for more sophisticated modeling
approaches that consider both structural and functional performance aspects.

12

2.3.2

Advanced Machine Learning Approaches

Artificial Neural Networks
The application of artificial neural networks (ANN) in pavement management has demonstrated
significant improvements over traditional statistical methods. Li developed a comprehensive ANN
approach that achieved remarkable prediction accuracy, with their hybrid neural network
combining backpropagation neural networks (BPNN) and Long Short-Term Memory (LSTM)
models (Li et al., 2022). Their research demonstrated that ANN models could process multiple
variables simultaneously while capturing complex non-linear relationships that traditional
methods cannot adequately represent.
The effectiveness of ANN approaches is particularly evident in their ability to handle large-scale
datasets with multiple influencing variables. Research has shown that when databases provide
stable data so urces, ANN models predict International Roughness Index (IRI) better than linear
regression approaches (Li et al., 2022). The backpropagation neural network effectively solves
problems of insufficient and inaccurate pavement condition data that plague traditional modeling
approaches.
Attoh-Okine (2002) demonstrated the effectiveness of combining rough set analysis with artificial
neural networks in doweled-pavement-performance modeling, showing that hybrid approaches
can leverage the strengths of different analytical techniques. This research highlighted the potential
for developing sophisticated modeling frameworks that combine multiple analytical approaches to
achieve superior prediction accuracy.
The integration of genetic algorithms with neural networks has shown promise for optimizing
model parameters and improving prediction accuracy. Zhao et al. (2021) applied genetic algorithm
optimization to ANN models for predicting viscosity of asphalt pavement adhesives,
demonstrating that this approach replaces the reverse error transmission process of BPNN models
and improves convergence efficiency.
Random Forest and Ensemble Methods
Random Forest (RF) has emerged as a particularly effective ensemble learning method for
pavement management applications. Marcelino et al. (2019) developed a general machine learning
algorithm based on random forest that effectively addresses continuous prediction problems over
time and sensitivity analysis of pavement roughness. Their research demonstrated that RF
algorithms exhibit strong nonlinear fitting ability, lack complicated theoretical derivation, and
provide real-time prediction capacity.
Gong et al. (2018) demonstrated that RF models could achieve up to 90% accuracy in pavement
condition forecasting, significantly outperforming traditional regression approaches. The
effectiveness of RF stems from its ability to handle large datasets efficiently while maintaining
13

robustness against noise and overfitting. The ensemble nature of RF enhances predictive accuracy
by reducing both variance and bias across different data subsets.
2.3.3

Optimization Techniques

Mathematical Programming Approaches
Mathematical optimization methods play crucial roles in pavement management by enabling
systematic selection of maintenance strategies that maximize performance while satisfying budget
constraints. Torres-Machí et al. developed an iterative approach for pavement maintenance
optimization at network level, demonstrating potential efficiency improvements of up to 25%
(Torres-Machí et al., 2014). Their approach differs from traditional sequential methods by allowing
selection of suboptimal treatment strategies for individual sections if they contribute to improved
overall network performance.
The iterative optimization approach addresses limitations of traditional sequential methods that
first define treatment strategies on a section-by-section basis, then select sections to treat until
budget is exhausted. The iterative method recognizes that deterioration of a solution at the section
level may lead to improvement of the overall solution at the network level, enabling more flexible
and effective resource allocation.
Multi-objective optimization approaches have gained prominence as agencies recognize that
pavement management decisions involve multiple, sometimes conflicting objectives. Research by
Santos demonstrated that multi-objective optimization can maximize network road condition
within limited budgets while considering factors such as traffic volume and cost-effectiveness
(Santos et al., 2017). Their work showed that preventive and minor rehabilitation treatments are
more cost-effective than reconstruction, and that budget allocation should exceed certain
thresholds to achieve maximum societal benefit.
Geographic Information Systems Integration
The integration of Geographic Information Systems (GIS) with optimization techniques has
revolutionized spatial analysis capabilities in pavement management. Research by Obaidat
demonstrated the effectiveness of integrating GIS toward efficient pavement maintenance
management (Obaidat et al., 2018). Their work showed that GIS integration enables dynamic
highway section color coding, access to sectional data through graphical models, and enhanced
visualization of pavement management analysis.
Recent developments in GIS-based pavement management have focused on 3D modeling and
advanced visualization capabilities. Research has shown that reconstruction of 3D GIS models,
high-definition mapping, and new applications for cities require accurate and efficient data
collection and scene perception of urban environments (Zagvozda et al., 2019). The integration of
mobile LiDAR sensors with camera systems has enabled comprehensive pavement condition data
collection while providing enhanced visualization capabilities for decision-makers (Balzi et al.,
14

2023). Studies have demonstrated that even though 3D mobile LiDAR data has become
increasingly popular, this method cannot accurately detect pavement distress such as cracks,
necessitating the use of RGB images for correct distress extraction (Balzi et al., 2023).
European implementations have demonstrated the effectiveness of GIS-based optimization
approaches. The ViaBEL tool was developed for secondary road networks in Belgium and
incorporated GIS-based decision processes for pavement management, providing structured
frameworks for maintenance prioritization and resource allocation (Van Geem et al., 2012). This
tool demonstrates the potential for developing sophisticated GIS-based optimization systems
appropriate for regional and municipal applications.
Multi-Criteria Decision Analysis
Contemporary pavement management increasingly recognizes that decisions involve multiple,
sometimes conflicting objectives including cost minimization, performance maximization,
environmental impact reduction, and user satisfaction. Multi-criteria decision analysis (MCDA)
frameworks provide structured approaches for evaluating alternatives considering multiple
objectives simultaneously (Torres-Machí et al., 2014).
Research has emphasized the importance of incorporating sustainability considerations into
pavement management decision-making (Santos et al., 2017). This requires consideration of
economic, social, technical, environmental, and political aspects throughout the pavement
lifecycle (Torres-Machí et al., 2014). Different indicators have been developed for assessing these
aspects, including present worth cost for economic evaluation, safety and comfort for social
considerations, roughness for technical performance, and air pollution for environmental impact
assessment (Torres-Machí et al., 2014).
The integration of multiple criteria enables more comprehensive evaluation of maintenance
alternatives. Research has shown that ranking based on economic analysis allows rational
comparison among alternatives because it considers both costs and benefits, leading to more
informed decision-making compared to approaches based solely on condition assessment or
subjective judgment (Al-Swailmi et al., 1999). Shah et al. (2012) applied this method to select
sections for treatment in a road network in India considering criteria such as traffic, connectivity,
and road and drainage conditions (Torres-Machí et al., 2014).
2.3.4

Performance Indicators and Assessment Methods

Pavement Distress Index (PDI)
Pavement condition assessment relies heavily on standardized indices that quantify pavement
serviceability and structural integrity (Federal Highway Administration, 2022). The Pavement
Distress Index (PDI) represents a critical performance indicator that has been specifically adapted
for Canadian conditions, particularly in BC (BCMoTI, 2020). Research has shown significant
variations in the effectiveness of different condition assessment approaches, with advanced
15

analytical methods demonstrating superior accuracy compared to traditional assessment
techniques (Ali et al., 2023).
The development of composite condition indices that integrate multiple distress types and
functional classifications has shown promise for providing more comprehensive pavement
assessment (Saliminejad, 2012). Research has demonstrated that composite indices considering
pavement surface distresses, traffic information, and expert opinion can provide more accurate
representations of overall pavement condition than single-parameter approaches (Guha et al.,
2022). The PDI methodology incorporates systematic evaluation of distress types including
rutting, cracking patterns, surface deterioration, and structural adequacy to provide comprehensive
condition assessment (BCMoTI, 2020).
International variations in condition assessment approaches reflect different prioritization
strategies and resource constraints (Grilli et al., 2019). Research has shown that different countries
and agencies employ varying condition rating systems, ranging from simple visual assessments to
sophisticated automated measurement systems (Federal Highway Administration, 2022). The
choice of assessment method significantly affects the accuracy and reliability of condition data,
which in turn impacts the effectiveness of management decisions (Saliminejad, 2012). This study
have shown that quality assurance programs can result in adjustments of predicted maintenance
treatments by up to 21%, highlighting the substantial impact of assessment methodology on
management outcomes (Saliminejad, 2012).
International Roughness Index (IRI)
The International Roughness Index (IRI) serves as a critical functional performance indicator that
directly relates to user comfort and vehicle operating costs (Federal Highway Administration,
2022). Research has shown strong correlations between IRI and various pavement distress types,
making it valuable for both condition assessment and performance prediction applications (Li et
al., 2022). IRI measurements provide objective assessment of pavement smoothness that enables
quantitative evaluation of ride quality and functional performance (Hamdi, 2015).
Studies utilizing machine learning approaches for IRI prediction have demonstrated significant
improvements over traditional methods (Gong et al., 2018; Marcelino et al., 2019). Random forest
algorithms have shown particular effectiveness for addressing continuous prediction problems
over time and sensitivity analysis of IRI (Marcelino et al., 2019). Research has consistently shown
that advanced modeling approaches can achieve prediction accuracies exceeding 90% for IRI
forecasting, substantially outperforming traditional statistical methods (Gong et al., 2018; Ali et
al., 2023).
Khawaga et al. (2021) developed Markov and S-curve models for IRI prediction, demonstrating
that Markov models performed better than S-curve models in terms of comprehensive factor
consideration (Li et al., 2022). Their research highlighted the importance of selecting appropriate
modeling approaches based on specific application requirements and data characteristics,
16

particularly for northern climate applications where freeze-thaw cycles significantly impact
pavement roughness (Qiao et al., 2020). Climate factors including temperature fluctuations and
precipitation patterns have been shown to significantly influence IRI progression patterns (Swarna,
2022).
The integration of IRI with other performance indicators provides comprehensive assessment
frameworks that consider both structural and functional pavement performance (Torres-Machí et
al., 2014). This integrated approach enables more informed decision-making regarding
maintenance timing and treatment selection (Federal Highway Administration, 2022). Research
has shown that IRI-based performance prediction requires careful consideration of climate factors,
traffic loading patterns, and pavement structural characteristics to achieve accurate forecasting for
maintenance planning applications (Li, 2009; Hamdi, 2015). Studies have demonstrated that local
calibration of IRI prediction models is essential for achieving reliable performance forecasting
under specific regional conditions (Li, 2009).
2.4

Insights from Pavement Management Case Studies
2.4.1

North American Implementations

California DOT (Caltrans) Experience
Wang and Pyle documented Caltrans' experience implementing a comprehensive pavement
management system, highlighting challenges and successes in large-scale PMS deployment (Wang
& Pyle, 2019). The Caltrans implementation emphasized the importance of integrating pavement
management with broader transportation asset management frameworks.
The California experience demonstrated the critical role of stakeholder engagement and
organizational change management in successful PMS implementation. The research showed that
technical excellence in system design must be complemented by effective organizational processes
and staff training to achieve successful implementation outcomes.
Caltrans' approach to pavement management integration with asset management provides a model
for other agencies seeking to align project-level pavement decisions with network-level strategic
objectives. The research demonstrated practical approaches for establishing feedback loops
between pavement design, pavement management, and transportation asset management units.
Ontario's AASHTOWare Calibration Experience
Ontario's experience with local calibration of AASHTOWare provides valuable insights for
regional PMS development. Hamdi (2015) demonstrated that significant improvements in
prediction accuracy can be achieved through local calibration, with statistical analysis showing
substantial enhancements in model performance for Ontario flexible pavements (Hamdi, 2015).
The study found that national calibration coefficients failed to offer reliable accuracy for local
conditions, necessitating extensive calibration efforts using provincial pavement management
system data.
17

The Ontario calibration project revealed important findings about regional model adaptation. The
study concluded that calibration coefficients should be updated as performance databases expand
and innovative materials are utilized in provincial pavement designs (Hamdi, 2015). This finding
emphasizes the dynamic nature of calibration requirements and the need for ongoing model
refinement as local experience accumulates.
Key Performance Indicators such as International Roughness Index and Pavement Condition Index
were evaluated, with the research demonstrating that locally calibrated models significantly
enhance prediction accuracy for pavement performance (Hamdi, 2015). The study showed that
PMS data can be effectively utilized to improve AASHTOWare model accuracy, providing a
framework for similar calibration efforts in other jurisdictions.
Washington State DOT Implementation
Washington State Department of Transportation's pavement management implementation
demonstrates the effectiveness of systematic calibration approaches. Li (2009) documented the use
of split-sample and jackknife testing approaches for local calibration of mechanistic-empirical
design models (Li, 2009). The split-sample approach used half of selected sections for calibration
and the other half for validation, while the jackknife approach withheld each section for prediction
measurements with other sections used for calibration.
The Washington State project achieved remarkable success in rutting prediction calibration. Final
calibration factors were chosen based on root-mean-square error , with results showing that
predicted rutting values closely matched measured data from the state's PMS (Li, 2009). The study
demonstrated consistency between predicted and measured rutting values for both western and
eastern Washington regions, indicating the effectiveness of regional calibration approaches.
The research concluded that local calibration processes should be finalized through model
validation using independent datasets not included in calibration processes (Li, 2009). This finding
provides important guidance for ensuring the robustness and reliability of locally calibrated
models.
2.4.2

European Implementations

Italian ANAS Case Study
The Italian National Road Agency (ANAS) experience provides insights into large-scale pavement
management implementation under resource constraints. Borghetti documented ANAS's approach
to managing approximately 32,000 km of state roads, highways, and freeway junctions under
direct management (Borghetti et al., 2024). The ANAS network represents the largest road
infrastructure management operation in Italy in terms of network extent.
The ANAS case study demonstrates the challenges faced by large road agencies in prioritizing
maintenance interventions under budget constraints. With regions proposing more than 1,000
interventions and financial needs exceeding 3.5 billion euros in some cases, ANAS requires
18

sophisticated prioritization methodologies to optimize resource allocation (Borghetti et al., 2024).
The average regional requirement approaches 1.5 billion euros over the reference period (20222026), highlighting the scale of infrastructure investment needs.
ANAS's approach to maintenance includes identification of network needs using standardized
parameters, definition of interventions based on available funds, and implementation focused on
process efficiency (Borghetti et al., 2024). This systematic approach provides a framework for
other agencies facing similar challenges in managing large road networks under financial
constraints.
The research demonstrates the application of multi-criteria decision-making approaches that
consider category, asset, and typology factors in maintenance prioritization. The final prioritization
formula applied weights of 60% to category factors and 40% to asset factors, with typology factors
having no influence for most intervention types (Borghetti et al., 2024).
Republic of San Marino Implementation
Grilli documented the development of pavement management guidelines for the Republic of San
Marino, demonstrating how small jurisdictions can implement effective PMS frameworks despite
resource constraints (Grilli et al., 2019). The San Marino implementation emphasizes the use of
GIS tools for data collection, management, and analysis of road inventory and monitoring data.
The San Marino case study demonstrates the effectiveness of customized GIS-based tools for
standardizing road data collection, managing monitoring and inventory data, identifying
maintenance priority ratings, and planning maintenance works from long-term perspectives (Balzi
et al., 2023). The implementation utilized a strategic index approach for comparing different
maintenance strategies, with the strategic index calculated as a function of priority index that
depends on technical factors including pavement condition, roughness, traffic levels, network
hierarchy, road functional classification, road relevance for strategic purposes, and maintenance
repair history.
The research showed that maintenance scenarios can be characterized and compared using specific
evolution of strategic index values over time, enabling rational comparison of alternative
maintenance strategies (Balzi et al., 2023). This approach provides a practical framework for
resource-constrained agencies to implement sophisticated PMS capabilities.

Belgian ViaBEL System
Van Geem documented the development of ViaBEL, a decision-making tool for pavement
management of secondary road networks in Belgium (Van Geem et al., 2012). The ViaBEL system
demonstrates how European agencies have developed sophisticated tools specifically designed for
regional and local road management applications.
19

The ViaBEL implementation emphasizes visual inspection quality assurance for pavement
management of communal road networks (Van Geem & Massart, 2017). The system provides
structured frameworks for decision processes in pavement management while maintaining costeffectiveness appropriate for secondary road networks.
The Belgian experience demonstrates the importance of developing management tools that are
appropriately scaled to the technical and financial capabilities of regional agencies. The ViaBEL
system provides practical approaches for implementing systematic pavement management without
requiring the sophisticated technical infrastructure associated with major highway agency
implementations.
2.4.3

Low-Cost and Municipal Applications

Low-Volume Road Management
Hafez and Ksaibati studied the effectiveness of parameter adjustments in pavement management
systems for low-volume paved roads, demonstrating potential pavement life extensions of 15-25%
through optimized treatment strategies (Hafez & Ksaibati, 2021). Their research addresses the
specific challenges faced by agencies responsible for roads with limited traffic but substantial
infrastructure investment requirements.
The low-volume road research emphasized the importance of developing management approaches
that account for different deterioration patterns and maintenance requirements compared to hightraffic facilities. The study showed that parameters optimized for major highways may not be
appropriate for low-volume applications, necessitating specialized calibration and optimization
approaches.
Van Geem and Massart documented implementation and benefits of low-cost PMS for municipal
road networks, demonstrating practical approaches for achieving systematic pavement
management under severe budget constraints (Van Geem & Massart, 2018). Their research showed
that significant improvements in maintenance effectiveness can be achieved through systematic
approaches even when sophisticated technical resources are not available.
Urban Applications
Loprencipe developed sustainable pavement management systems for urban areas considering
vehicle operating costs (Loprencipe et al., 2017). Their research demonstrated the integration of
user cost considerations with traditional pavement management approaches, providing more
comprehensive frameworks for urban pavement management decision-making.
The urban pavement management research emphasized the importance of considering user impacts
and vehicle operating costs in addition to agency maintenance costs. The study showed that
integrated approaches considering both agency and user costs can lead to different optimization
outcomes compared to traditional approaches focused solely on agency costs.
20

Vines-Cavanaugh et al. documented city-wide application of affordable and rapid pavement
management systems, demonstrating practical approaches for implementing systematic condition
assessment and management planning in urban environments (Vines-Cavanaugh et al., 2017).
Their research showed that cost-effective technologies can enable comprehensive pavement
management even for resource-constrained municipal agencies.
2.5

Literature Review Summary

Table 1 Details of previous research activities on PMS

Research
team

Year

Case study
Location

Software/
Methode

Haas, Hudson,
& Zaniewski

1994

North America

PMS
Optimization
Model

Hudson, Haas,
& Uddin

1997

Global

Infrastructure
Modeling

Anastasopoulos
, Mannering, &
Haddock

2009

USA

Service Life
Analysis

Smith, Pontius,
& Galehouse

2014

USA

PMS Data
Integration

Ali et al.

2023

Global

Machine
Learning
Models

Federal
Highway
Administration

2022

USA

Pavement
Management
Roadmap

Ekramnia &
Nasimifar

2022

Global

Treatment
Prioritization
Model

Hamdi

2015

Ontario,
Canada

AASHTOWare
Calibration

Li

2009

Washington,
USA

MechanisticEmpirical
Calibration

Qiao et al.

2020

Global

Climate Change
Adaptation
Model

Swarna

2022

Global

Climate
Modeling

Gavin et al.

2003

Canada

Climate Effects
Research

Main points
- Introduced a structured PMS approach for optimizing
maintenance
planning.
- Demonstrated lifecycle cost reduction of up to 30% through
systematic strategies.
- Integrated engineering and economic models to enhance
PMS decision-making.
- Emphasized full-cycle infrastructure management from
design to rehabilitation.
- Developed service life curves to evaluate rehabilitation
effectiveness.
- Showed optimized interventions extend pavement life by
20–40%.
- Highlighted the importance of data-driven decision
processes in PMS.
- Introduced maintenance prioritization based on objective
road condition data.
- Applied neural networks achieving R² > 98% accuracy in
condition predictions.
- Outperformed multiple regression models in handling nonlinear deterioration patterns.
- Established a 10-year roadmap for PMS improvements.
- Focused on advancing predictive analytics and fostering
industry-academia collaboration.
- Created a treatment prioritization tool improving efficiency
by
30%.
- Enhanced network-level rehabilitation planning under limited
resources.
- Showed significant improvements in prediction accuracy via
local
calibration.
- Highlighted the shortcomings of national calibration
coefficients.
- Developed calibration factors achieving near-perfect rutting
predictions.
- Demonstrated region-specific model adaptation in
Washington State.
- Analyzed climate impacts on pavement deterioration
worldwide.
- Suggested adaptive maintenance strategies to counter
environmental damage.
- Identified limitations in temperature modeling for high
latitudes.
- Highlighted the absence of suitable models for Northern
Canadian climates.
- Investigated Canadian climate effects on pavement
performance.
- Initiated test sites studying asphalt properties in extreme cold.

Reference
[Haas 1994]

[Hudson 1997]

[Anastasopoulo
s et al. 2009]

[Smith 2014]

[Ali 2023]

[FHWA 2022]
[Ekramnia
2022]

[Hamdi 2015]

[Li 2009]

[Qiao 2020]

[Swarna 2022]

[Gavin 2003]

21

Research
team

Year

Case study
Location

Software/
Methode

Main points
- Used ANN to predict asphalt condition with improved
accuracy.
- Enhanced Pavement Condition Index estimation through soft
computing.
- Combined BPNN and LSTM neural networks for hybrid
modeling.
- Introduced data cleaning and restart point methodology.
- Demonstrated integration of rough set analysis with ANN
models.
- Showed hybrid approaches handle incomplete datasets
effectively.

Reference

Ali

2022

Canada

Soft Computing

Li et al.

2022

Global

Hybrid Neural
Network

Attoh-Okine

2002

Global

Hybrid ANN
Approach

Zhao et al.

2021

Global

Genetic
Algorithm with
ANN

- Improved ANN model convergence with genetic algorithms.
- Enhanced accuracy in predicting asphalt adhesive viscosity.

[Zhao 2021]

Marcelino et al.

2019

Global

Random Forest
Algorithm

- Applied RF algorithm for roughness sensitivity analysis.
- Provided continuous prediction with real-time performance.

[Marcelino
2019]

Gong et al.

2018

Global

Random Forest
Model

Saliminejad

2012

Global

Error Analysis

Torres-Machí et
al.

2014

Spain

Iterative
Optimization
Model

Santos et al.

2017

Global

Multi-objective
Optimization

Obaidat et al.

2018

Global

GIS Integration

Zagvozda et al.

2019

Global

3D GIS
Modeling

- Advanced 3D GIS modeling for urban PMS applications.
- Enabled accurate scene perception for city infrastructure.

[Zagvozda
2019]

Balzi et al.

2023

Global

Mobile LiDAR
GIS

- Used LiDAR with RGB cameras to enhance distress
detection accuracy.
- Improved 3D data for pavement analysis and planning.

[Balzi et al.
2023]

- Achieved 90% prediction accuracy using RF.
- Outperformed traditional regression methods in large
datasets.
- Showed parameter errors cause slope bias in regression
models.
- Identified effects of random vs. systematic data errors.
- Developed an iterative optimization model improving
network efficiency by 25%.
- Highlighted holistic vs. sequential treatment strategies.
- Applied multi-objective optimization for treatment selection.
- Balanced road condition, traffic volume, and cost
constraints.
- Integrated GIS enhancing visualization and dynamic
analysis.
- Enabled sectional data retrieval via graphical models.

[Ali 2022]

[Li 2022]
[Attoh-Okine
2002]

[Gong 2018]
[Saliminejad
2012]
[Torres-Machí
2014]
[Santos 2017]
[Obaidat 2018]

22

3. Chapter 3: Data Collection and Pre-Processing
3.1

Fundamentals of PMS

Pavement Management Systems represent a systematic approach of optimizing infrastructure
maintenance and rehabilitation through data-driven decision-making processes. As established in
the literature review (Chapter 2), PMS have evolved from manual inspection methods to
sophisticated analytical frameworks that integrate multiple data sources, predictive modeling
capabilities, and optimization techniques to support strategic infrastructure stewardship. The
fundamentals examined in this chapter provide detailed technical understanding of the core
components that comprise effective pavement management, expanding upon the foundational
concepts presented in the literature review with specific focus on their application within Prince
George's operational context.
The effectiveness of PMS implementation depends upon the integration of multiple interconnected
components including inventory management, condition assessment, performance prediction,
treatment optimization, and resource allocation. Each component contributes essential capabilities
that collectively enable agencies to transition from reactive maintenance approaches to proactive
management strategies that optimize both performance outcomes and resource utilization.
The evolution toward data-driven pavement management reflects broader technological
advancements in infrastructure monitoring and analysis capabilities. Contemporary PMS
implementations leverage automated data collection technologies, advanced analytical techniques,
and sophisticated modeling approaches to enhance decision-making accuracy while reducing
reliance on subjective assessments that characterized earlier management approaches.
3.2
3.2.1

Pavement Condition Indices
Pavement Distress Index (PDI)

The Pavement Distress Index represents the primary condition assessment tool employed within
Prince George's pavement management framework, providing standardized quantification of
pavement surface condition and structural integrity. As established in the technical reports from
2016, 2017, 2020, and 2023, PDI assessment follows the British Columbia Ministry of
Transportation and Infrastructure (BCMoTI) Condition Rating Manual methodology, which
represents a localized adaptation of the internationally recognized PAVER Pavement Condition
Index model developed by the U.S. Army Corps of Engineers.

23

Figure 3 Prince George Pavement Condition over the years plotted on Google Earth.

The PDI model integrates severity and density ratings for each distress type, consolidating multiple
condition indicators into a unified distress score on a declining scale from 10 to 0, where 10
represents perfect pavement condition and 0 indicates complete failure requiring reconstruction.
This standardized scale enables objective comparison of pavement conditions across different road
segments and functional classifications within Prince George's network.
The mathematical foundation of PDI calculation employs a deduct value methodology that begins
with a perfect score of 10 and systematically subtracts penalty values based on observed distress
characteristics. The specific deduct value for each distress type is calculated using the following
equation established in the BCMoTI methodology:
Deduct = 10^(B0 + B1 * Log(Density))
The coefficients B0 and B1 vary according to distress type and severity level, with values
calibrated specifically for BC pavement conditions and climate factors. These coefficients reflect
the relative impact of different distress types on overall pavement performance, with structural
distresses such as alligator cracking receiving higher deduct values than surface-related distresses
such as bleeding or minor raveling.

24

Table 2 Coefficients to Calculate the Deducts of PDI

Distress

Low
Severity
B0
Wheelpath -0.80

Longitudinal
Cracking
Longitudinal Cracking
-0.80
Pavement Edge Cracking
-0.80
Transverse Cracking
-0.80
Meandering Longitudinal -0.80
Cracking
Alligator Cracking
Potholes
-0.80
Rutting
-0.75
Source: Tetra Tech (2020), ICC (2023)

Moderate
Severity
B1 B0
0.60 -0.50

High
Severity
B1 B0
0.57 -0.25

B1
0.55

0.50
0.40
0.60
0.55

0.57
0.55
0.55
0.60

-0.40
-0.45
-0.35
-0.40

0.55
0.55
0.57
0.57

0.75 -0.25
0.57 -0.25
0.57 -0.30

0.75
0.57
0.60

-0.65
-0.70
-0.40
-0.65

-0.40
0.40 -0.60
0.50 -0.45

The PDI classification system employs a three-tier condition rating framework that categorizes
pavement segments as Good (PDI > 7), Fair (PDI 5-7), or Poor (PDI < 5). This classification
provides intuitive interpretation of condition data while supporting maintenance prioritization and
resource allocation decisions. The classification thresholds were established through extensive
calibration with British Columbia pavement performance data and reflect the relationship between
measured distress levels and functional pavement performance under regional traffic and climate
conditions.
3.2.2

International Roughness Index (IRI)

The IRI serves as the primary measure of pavement functional performance within Prince George's
assessment framework, quantifying ride quality and user comfort through standardized
measurement of pavement surface irregularities. IRI assessment follows the ASTM E1926
specification, providing objective quantification of pavement smoothness that directly correlates
with vehicle operating costs, fuel consumption, and user satisfaction (Tetra Tech, 2017; Tetra Tech,
2020).
IRI calculation methodology employs a quarter-car simulation model that measures vertical
suspension motion divided by distance traveled, reporting results in millimeters per meter (mm/m)
or the equivalent meters per kilometer. The quarter-car model simulates the dynamic response of
a standardized vehicle suspension system to pavement surface irregularities, providing consistent
measurement methodology that enables comparison across different pavement sections and
assessment periods (Tetra Tech, 2020).
The technical measurement process utilizes longitudinal profile data captured through automated
data collection systems, typically integrated with Laser Crack Measurement Systems (LCMS) that
provide continuous elevation measurements at 1mm resolution. The profile data undergoes
mathematical processing through established algorithms that simulate quarter-car response
25

characteristics and generate standardized IRI values for each measurement segment (ICC, 2023).
Since its introduction in 1986, IRI has become the road roughness index most used worldwide for
evaluating and managing higher speed road networks, with vehicle operating costs including fuel
consumption, tire wear, and depreciation rising with increasing roughness and directly correlated
to IRI values (Tetra Tech, 2020).
Table 3 Index Ranges for IRI Description

Rating Alley, Local, and Ramp IRI Arterial, Major Collector, Minor Collector
(mm/m)
IRI (mm/m)
Good ≤ 4.49
≤ 2.99
Fair
4.49 – 8.08
2.99 – 5.40
Poor
> 8.08
> 5.40
Source: Tetra Tech (2020), based on Yu, Chou, & Yau (2006)

Color
Code
Green
Yellow
Red

The differentiated IRI classification thresholds reflect the varying service expectations associated
with different road classifications and typical operating speeds. Arterial and collector roads, which
accommodate higher traffic volumes and speeds, require smoother surface conditions (lower IRI
values) to maintain acceptable ride quality and minimize vehicle operating costs. Local roads and
alleys, characterized by lower operating speeds and different functional requirements, can
accommodate higher IRI values while maintaining acceptable service levels.
Research documented in Prince George's technical reports has demonstrated strong correlations
between IRI values and vehicle operating costs, with studies indicating that "vehicle operating
costs including fuel consumption, tire wear, and depreciation rise with increasing roughness and
have been directly correlated to IRI" (Tetra Tech, 2017; Tetra Tech, 2020). These relationships
enable economic analysis of pavement condition impacts that extend beyond immediate
maintenance costs to encompass broader user cost considerations. For Prince George's
transportation network, IRI data supports both condition assessment and economic analysis that
informs maintenance prioritization and treatment selection decisions.
3.2.3

Composite Index Integration

The integration of PDI and IRI measurements provides comprehensive pavement assessment that
addresses both structural integrity and functional performance characteristics. This dual-index
approach enables differentiated analysis of pavement conditions, recognizing that structural
adequacy and ride quality may exhibit different deterioration patterns and require distinct
maintenance responses.
The complementary nature of PDI and IRI assessment supports sophisticated decision-making
frameworks that can optimize maintenance timing and treatment selection based on both
immediate functional requirements and long-term structural preservation objectives. This
integrated approach aligns with contemporary pavement management best practices that recognize

26

the multidimensional nature of pavement performance and the need for comprehensive assessment
methodologies.
3.3
3.3.1

Data Gathering Methods
Automated Distress Survey Technology

Prince George's pavement condition assessment program employs state-of-the-art automated data
collection technologies that provide objective, repeatable, and comprehensive evaluation of
pavement distress characteristics. The automated approach represents a significant advancement
over traditional manual inspection methods, offering enhanced accuracy, consistency, and
efficiency in large-scale network assessment.
The primary data collection platform utilized in Prince George's assessment program is the Laser
Crack Measurement System (LCMS), deployed through specialized vehicles including Tetra
Tech's Pavement Surface Profiler (PSP) 7000 series and ICC International Cybernetics Canada's
Road Surface Tester (RST) (Tetra Tech, 2017; Tetra Tech, 2020; ICC, 2023). These systems
integrate multiple advanced technologies including high-resolution laser scanning, digital
imaging, GPS positioning, and inertial measurement capabilities to capture comprehensive
pavement condition data at highway speeds (ICC, 2023).

Figure 4 LCMS-2 RST data collection vehicle (IMS, 2023).
27

The LCMS technology operates by projecting laser light across the pavement surface and
measuring elevation variations with millimeter-level precision. The system captures continuous
2D and 3D images at 1mm resolution across lane widths up to 4 meters, enabling detailed
characterization of surface features and distress patterns (ICC, 2023). Advanced LCMS-2 systems
employed in recent assessments can achieve collection rates up to 28,000 profiles per second,
representing a five-fold improvement over earlier generation equipment (ICC, 2023).
3.3.2

Positioning and Location Referencing Systems

Accurate spatial referencing represents a critical component of automated data collection, enabling
precise location identification and subsequent analysis integration with Geographic Information
System (GIS) platforms. Prince George's assessment program employs sophisticated Global
Navigation Satellite System (GNSS) technology, specifically the Applanix POS LV RT-420
system, which provides continuous and accurate vehicle position and orientation information
under challenging GPS conditions (Applied Research Associates, 2016; British Columbia Ministry
of Transportation and Infrastructure, 2020).
The positioning system utilizes inertially aided GPS technology that maintains spatial accuracy
even during periods of limited satellite coverage, such as urban canyon environments or areas with
dense vegetation. The integrated inertial measurement unit provides continuous position estimates
through gyroscopic sensors and accelerometers, enabling uninterrupted data collection throughout
the survey process (Applied Research Associates, 2016).
Post-processed positioning data typically achieves accuracy levels of less than 500mm in
horizontal position and 500mm in vertical position for locations with suitable satellite
constellations. Even under challenging GPS conditions involving outages of 10-15 minutes, the
system maintains positional accuracy within 1 meter horizontally and 2 meters vertically, ensuring
reliable spatial referencing for subsequent analysis and asset management applications (British
Columbia Ministry of Transportation and Infrastructure, 2020).
3.3.3

Quality Control and Data Validation

Comprehensive quality control procedures ensure the reliability and accuracy of collected
pavement condition data. Tetra Tech's implementation employs a GIS-based field management
application called "TT Surveyor" that provides real-time verification of data collection
completeness and spatial accuracy. The system verifies that road segments are surveyed in the
correct direction and within specified spatial boundaries, comparing real-time vehicle position
with predefined survey requirements (Tetra Tech, 2020).
Daily data review processes involve uploading GPS, roughness, and rutting data to processing
teams for immediate analysis of coverage completeness and data quality. This methodology
enables identification and correction of data gaps or quality issues before survey equipment leaves
the project location, minimizing data collection errors and ensuring comprehensive network
coverage (Tetra Tech, 2017; Tetra Tech, 2020).
28

Spot verification procedures involve manual validation of automated distress detection results
through comparison with high-resolution digital imagery captured during the survey process.
These quality assurance checks help identify potential issues with automated classification
algorithms and ensure consistency between different assessment periods (ICC, 2023).

Figure 5 TT Surveyor Application Representation of LCMS System (Tetratech, 2017).

3.3.4

Strengths and Limitations of Automated Assessment

Automated distress survey methods offer significant advantages over traditional manual inspection
approaches, including enhanced objectivity, improved consistency, and increased efficiency in
large-scale network assessment. The automated approach eliminates subjective bias associated
with manual inspections while providing comprehensive documentation through high-resolution
imagery and precise spatial referencing.
The high-speed data collection capability enables comprehensive network assessment with
minimal traffic disruption and reduced safety risks for inspection personnel. Automated systems
can assess entire road networks in days rather than weeks or months required for manual
inspection, enabling more frequent condition monitoring and timely identification of deteriorating
conditions.
However, automated systems also present certain limitations that must be considered in pavement
management applications. Weather conditions, particularly wet pavement surfaces, can affect the
accuracy of laser-based distress detection. Some distress types, such as bleeding or fine cracking,
may require manual verification to ensure accurate classification and severity assessment.
The substantial capital investment required for automated assessment equipment and specialized
personnel training represents a significant consideration for municipal agencies. Additionally, the
technical complexity of automated systems requires ongoing maintenance and calibration to ensure
continued accuracy and reliability.
29

3.4

Traffic Data in PMS

Traffic data represents a fundamental component of pavement management systems, directly
influencing deterioration rates, maintenance requirements, and treatment selection decisions.
Prince George's strategic position as the largest urban center in northern British Columbia creates
unique traffic characteristics that significantly impact pavement performance and management
requirements.
The city serves as a critical transportation hub at the confluence of major provincial and national
transportation corridors, including Highway 97 (connecting southern British Columbia to Alaska)
and Highway 16 (the Yellowhead Highway connecting eastern and western Canada). This strategic
location generates substantial commercial traffic volumes, including heavy vehicle movements
associated with forestry, mining, and agricultural activities throughout the region.
Traffic data collection and analysis within Prince George's PMS framework encompasses multiple
parameters including Annual Average Daily Traffic (AADT), vehicle classification by weight
categories, seasonal traffic variations, and spatial distribution patterns across different road
classifications. This comprehensive traffic characterization enables accurate assessment of loading
conditions that directly influence pavement deterioration rates and maintenance requirements.

Figure 6 Prince George Interactive map plotting the Road Traffic data (City of Prince George, 2025).

The City of Prince George maintains comprehensive traffic monitoring capabilities through
automated traffic counting stations and periodic manual classification studies. Traffic data is
accessible through the city's open data portal and interactive mapping applications that provide
real-time visualization of traffic patterns and volumes across the road network (City of Prince
George, 2025). This spatial traffic information supports maintenance prioritization and treatment
selection by identifying high-stress corridors that require enhanced attention and more frequent
intervention.
30

Heavy vehicle traffic represents a particularly critical factor in pavement management for Prince
George, given the region's dependence on resource extraction industries that generate substantial
commercial vehicle movements. Heavy vehicles create disproportionate pavement loading
compared to passenger vehicles, with deterioration impacts that increase exponentially with axle
weight. The integration of traffic classification data enables PMS applications to account for these
differential loading impacts in deterioration prediction and maintenance planning.
Seasonal traffic variations also influence pavement management strategies in Prince George,
particularly considering the region's role in supporting resource industry activities that may exhibit
seasonal operational patterns. Winter conditions affect both traffic patterns and pavement stress
characteristics, with freeze-thaw cycles creating additional deterioration mechanisms that
compound traffic-related wear.
3.5

Pavement Distress Surveys

Pavement distress surveys represent the systematic evaluation of pavement surface conditions that
forms the empirical foundation for condition assessment and performance analysis within Prince
George's PMS framework. The city has conducted comprehensive distress surveys in 2016, 2017,
2020, and 2023, utilizing specialized contractors including Applied Research Associates (ARA),
Tetra Tech Canada Inc., and ICC International Cybernetics Canada Inc. to ensure objective and
standardized assessment procedures.
The survey methodology employed in Prince George follows the BCMoTI Pavement Surface
Condition Rating Manual, which provides standardized protocols for distress identification,
classification, and quantification specifically calibrated for British Columbia pavement conditions
and climate factors. This methodology ensures consistency across different survey periods and
enables reliable trend analysis for long-term performance evaluation.
3.5.1

Survey Coverage and Scope

The 2020 comprehensive survey represented the most extensive assessment conducted, covering
approximately 843 lane kilometers of the city's paved road network including arterials, major
collectors, minor collectors, local roads, alleys, ramps, and associated intersections (Tetra Tech,
2020). This comprehensive coverage enables network-level analysis that supports strategic
maintenance planning and resource allocation across all road classifications.
The 2023 survey focused specifically on arterial and collector roads, assessing 293 centerline
kilometers of predominantly asphalt roadways and intersections (ICC, 2023). This targeted
approach reflects the prioritization of high-traffic corridors that experience the most severe
deterioration and require the most immediate attention within constrained maintenance budgets.
3.5.2

Data Collection Intervals and Segmentation

Pavement condition data is collected and reported in standardized 50-meter segments, consistent
with BCMoTI recommendations for municipal pavement management applications. This
31

segmentation length provides sufficient detail for accurate condition assessment while maintaining
manageable data volumes for analysis and decision-making purposes.
The standardized segmentation enables precise spatial referencing of condition data within GIS
platforms, supporting sophisticated spatial analysis capabilities including hotspot identification,
corridor-level analysis, and integration with traffic and environmental data sources.
3.6

Types of Pavement Distress

The assessment of distress types is crucial for determining the appropriate maintenance and
rehabilitation strategies. Common pavement distress types found in Prince George’s Road network
include rutting, longitudinal cracking, meandering longitudinal cracking, transverse cracking,
longitudinal wheel path cracking, alligator cracking, pavement edge cracking, and potholes as
shown in Figure 7. The prevalence of these distresses is heavily influenced by climate conditions,
particularly the freeze-thaw cycles and heavy snowfall that contribute to pavement degradation
(Qiao et al., 2020).

Figure 7 Common types of pavement distresses observed in Prince George

1. Rutting – Rutting refers to depressions in the wheel paths caused by repeated
loading and consolidation of pavement materials. It is typically associated with
inadequate pavement compaction, weak subgrade layers, or high traffic loads.
Severe rutting can lead to water accumulation and vehicle control issues, requiring
corrective measures such as resurfacing or full-depth pavement reconstruction
(Qiao et al., 2020).
2. Longitudinal Cracking – Longitudinal cracks are parallel to the pavement's
centerline and commonly develop due to fatigue, thermal contraction, or poorly
constructed longitudinal joints. These cracks can allow water infiltration, weaken
the pavement structure and accelerating deterioration. Sealants and overlays are
typically used to mitigate further damage (Smith et al., 2014).
3. Meandering Longitudinal Cracking – Unlike regular longitudinal cracks,
meandering longitudinal cracks follow an irregular, non-linear pattern. They often
indicate subgrade movement, frost heave, or structural deficiencies, requiring more
extensive repairs such as base stabilization or reconstruction (Walls & Smith,
1998).
4. Transverse Cracking – Transverse cracks develop perpendicular to the pavement
centerline, mainly due to temperature fluctuations causing thermal expansion and
contraction. These cracks can be a sign of inadequate pavement flexibility or aging
32

5.

6.

7.

8.

3.7

asphalt. Preventive maintenance techniques such as crack sealing help limit
moisture infiltration and extend pavement service life (Hafez & Ksaibati, 2021).
Longitudinal Wheel Path Cracking – This form of longitudinal cracking occurs
specifically in the wheel paths due to repeated traffic loading. It is often an early
indicator of fatigue failure and can develop into alligator cracking if not treated
promptly (Guha et al., 2022).
Alligator Cracking – Alligator or fatigue cracking consists of interconnected
cracks resembling an alligator’s skin. It results from repeated traffic loads that
exceed pavement structural capacity, leading to progressive failures. Alligator
cracking is a serious distress that typically requires full-depth pavement
rehabilitation rather than surface treatments (Qiao et al., 2020).
Pavement Edge Cracking – Edge cracking occurs along the outer edges of the
pavement and is primarily caused by insufficient shoulder support, water
infiltration, or improper drainage. These cracks can propagate inward,
compromising pavement integrity. Shoulder reinforcement and drainage
improvements are common solutions for preventing edge cracking (Ali, 2022).
Potholes – Potholes form when water infiltrates into cracks, weakening pavement
layers, and leading to localized failures. Repeated freeze-thaw cycles exacerbate
pothole formation, making them a significant issue in colder climates. Potholes can
be temporarily repaired with patching materials or permanently fixed through
resurfacing (Guo et al., 2022).
Road Categorization in PMS

Road classification systems provide the organizational framework for differentiated pavement
management strategies that account for varying functional requirements, traffic characteristics, and
performance expectations across different road types. Prince George's classification system
follows standard municipal practices that recognize distinct roles and requirements for different
road categories.
Table 4 Prince George Road Classification System

Class
Arterial
Major
Collector
Minor
Collector
Local
Alley

Description

Lane-km
(Paved)
Major traffic corridors 324
connecting regions
Primary urban traffic 130
distribution
Secondary
traffic 157
collection
Neighborhood access and 679
local traffic
Rear property access and 20
service

Lane-km
(Gravel)
4

Primary Function

32

High-volume regional
connectivity
Urban traffic collection
and distribution
Local traffic collection

138

Direct property access

41

Service and secondary
access

13

33

Ramp

Highway access and 6
0
grade separation
Private
City-owned utility access 4
9
roads
Source: City of Prince George Asset Management (Bobbie, 2025)
3.7.1

Grade-separated access
Utility
access

and

service

Arterial Roads

Arterial roads represent the highest classification within Prince George's network, accommodating
the highest traffic volumes and serving critical regional connectivity functions. These roadways
experience the most severe traffic loading and environmental stress, requiring the most frequent
maintenance interventions and highest performance standards. The 324 lane-kilometers of paved
arterial roads in Prince George's network carry substantial commercial traffic associated with the
city's role as a regional transportation hub (Bobbie, 2025).
3.7.2

Collector Roads

Collector roads serve intermediate functions between arterial and local roads, providing traffic
collection and distribution within urban areas while accommodating moderate traffic volumes.
Prince George's collector road network comprises 130 lane-kilometers of major collectors and 157
lane-kilometers of minor collectors, each serving distinct functions within the overall
transportation hierarchy (Bobbie, 2025). Major collectors connect arterial roads with local traffic
generators and typically accommodate higher traffic volumes than minor collectors Minor
collectors provide secondary traffic collection functions and serve as intermediate connections
between major collectors and local roads.
3.7.3

Local Roads and Service Classifications

Local roads provide direct access to adjacent properties and accommodate primarily local traffic
with minimal through movement. These roadways experience lower traffic volumes and less
severe loading conditions compared to arterial and collector roads, enabling longer maintenance
intervals and different treatment strategies. Prince George maintains 679 lane-kilometers of paved
local roads, representing the largest component of the municipal road network (Bobbie, 2025).
Alleys serve specialized functions including rear property access, service vehicle accommodation,
and utility corridor provision. The 20 lane-kilometers of paved alley pavement in Prince George's
network requires distinct maintenance approaches that account for unique geometric constraints
and access limitations.
Ramps provide grade-separated access between different road classifications and highway
systems. The 6 lane-kilometers of paved ramps have specialized geometric and structural
requirements that necessitate targeted maintenance approaches accounting for unique loading
patterns and drainage characteristics.

34

3.8

Previous Treatment Methods and Maintenance History

This section summarizes the treatment and maintenance methods utilized by the City of Prince
George based on the maintenance records included in the PMS. Figure 8 presents a snapshot from
the map illustrating the road rehabilitation projects scheduled in 2024. According to Figure 8, the
City executed 52 road resurfacing projects in 2024 only.

Figure 8 City of Prince George 2024 Road Resurfacing (City of Prince George, 2024)

3.8.1

Preventive Maintenance Strategies

Crack Sealing
Crack sealing represents the most cost-effective preventive maintenance treatment employed in
Prince George's pavement management program, designed to prevent moisture infiltration and
delay the progression of surface cracking to more severe distress types. The treatment involves the
35

application of specialized sealant materials into cracks to create waterproof barriers that prevent
environmental deterioration.
Research documented in Prince George's Strategic Paved Road Management Plan indicates that
crack sealing treatments cost approximately $0.6 to $1.0 per linear foot and provide service life
extensions of 3-5 years when applied to appropriate pavement conditions (Applied Research
Associates, 2016). The treatment is most effective when applied to pavements in good to excellent
condition (PDI 8-9) with low to moderate severity cracking.
Microsurfacing
Microsurfacing treatments have been extensively utilized in Prince George's maintenance
program, with historical data from 2009-2016 showing 167 applications representing 10.2% of all
treatments during this period. Microsurfacing provides surface renewal capabilities that address
functional deficiencies including friction loss, minor surface raveling, and early-stage surface
deterioration. These treatments cost approximately $2.25 to $7.50 per square yard depending on
specific application requirements and material specifications.
Surface treatments serve dual functions as both preventive and corrective maintenance
interventions. As preventive measures, they seal minor surface deficiencies and restore surface
characteristics before functional deterioration occurs. As corrective treatments, they address
existing surface problems including friction deficiency and moderate raveling while providing
enhanced surface protection.
3.8.2

Rehabilitation Strategies

Mill and Fill (Mill and Overlay)
Mill and fill procedures, coded as "SurfaceRehab_M&F" in Prince George's asset management
system, represent the predominant rehabilitation method employed in the city's pavement
management program. Historical data from 2009-2016 shows this treatment was applied in 904
cases, representing 55.1% of all rehabilitation activities. Recent project data (2004-2023) indicates
continued reliance on this method, with 35 projects covering 32.24 lane-kilometers and 164,787
square meters of pavement area.
Mill and fill procedures involve the removal of 50mm of existing asphalt pavement followed by
replacement with new asphalt materials. This method enables pavement renewal without
increasing overall elevation, making it particularly suitable for urban environments where curb
and gutter elevation constraints limit overlay thickness. The milling process removes surface
deficiencies including cracking, raveling, and minor deformation while providing improved
surface texture for enhanced bonding with replacement materials.
Thin Lift Overlay (TLO)

36

Thin lift overlays, designated as "SurfaceRehab_OL" in city records, represent the second most
common rehabilitation strategy. Historical data shows 174 applications (10.6% of treatments) from
2009-2016, with recent projects indicating 30 applications covering 24.83 lane-kilometers and
103,925 square meters. This method involves the placement of 40-50mm asphalt layers over
existing pavement surfaces, providing both surface renewal and structural enhancement while
maintaining existing pavement geometry and drainage characteristics.
TLO treatments are most appropriate for pavements with good underlying structure and only minor
surface deficiencies. However, the method does not address underlying structural problems and
typically results in reflective cracking appearing through the new surface within a relatively short
time following placement.
Reclamation
Treatments Prince George employs both partial and full depth reclamation strategies based on
pavement condition and structural requirements. Historical data indicates 29 applications of 75mm
reclamation and 46 applications of 100mm full depth reclamation from 2009-2016. Recent projects
include one major 75mm reclamation project covering 1.42 lane-kilometers and 6,170 square
meters.
Reclamation involves the pulverization of existing asphalt and granular base materials followed
by reconstruction with enhanced structural capacity. The 75mm reclamation addresses moderate
structural deficiencies, while 100mm full depth reclamation provides comprehensive structural
renewal for severely deteriorated pavements. This method effectively removes all existing
pavement distresses while providing substantial structural improvement at costs considerably less
than complete reconstruction.
Full Reconstruction
Complete pavement reconstruction represents the most extensive rehabilitation method, reserved
for pavements with major structural deficiencies that cannot be addressed through surface or
reclamation treatments. Historical data shows limited application with 16 rural reconstructions and
2 urban reconstructions from 2009-2016, plus 2 recent reconstruction projects covering 1.60 lanekilometers.
The reconstruction process involves complete removal of existing pavement materials, subgrade
preparation, and installation of new structural sections designed for current traffic loading and
environmental conditions. This comprehensive approach provides maximum structural
enhancement and service life extension but requires the highest capital investment.
3.8.3

Treatment Selection Matrix and Historical Performance

Prince George's treatment selection follows systematic decision-making frameworks that integrate
pavement condition assessment with appropriate treatment strategies based on distress
characteristics and functional requirements. Historical application data demonstrates the
37

effectiveness of this approach, with mill and fill treatments addressing the majority of
rehabilitation needs while targeted use of overlays, reclamation, and reconstruction addresses
specific structural and functional requirements.
Table 5 Treatment Matrix for Arterial/Collector Roadways

Condition
Rating (PDI)

Typical Distress Features

Recommended
Treatment

Historical
(2009-2016)

Usage

9 to 10

New pavement,
condition

8 to 9

Good condition, low to Crack
seal
moderate severity cracking
Microsurfacing

7 to 8

Good condition,
friction,
minor
raveling

6 to 7

Good to fair condition, Mill and fill
structurally adequate, rutting
< 10mm

904 mill and fill
applications

5 to 6

Fair condition, evidence of Reclamation
load-related distress
100mm)

(75- 75
reclamation
applications

Less than 5

Poor condition, moderate to Reconstruction
high severity cracking

excellent Do nothing
/ 167 microsurfacing
applications

loss of Microsurfacing or thin 174
overlay
surface lift overlay
applications

18
reconstruction
projects

Source: Applied Research Associates (2016), Prince George Asset Management Records (20092023)
The historical treatment data demonstrates Prince George's strategic approach to pavement
management, with many interventions (55.1%) utilizing cost-effective mill and fill procedures for
pavements with adequate structural capacity but surface deterioration. The balanced application of
preventive treatments (microsurfacing) and more extensive rehabilitation methods (reclamation
and reconstruction) reflects systematic condition-based decision making that optimizes both
performance outcomes and resource utilization.
3.9

Overview of BC Specific PMS Reports and Policies

BCMoTI Pavement Surface Condition Rating Manual
The British Columbia Ministry of Transportation and Infrastructure Pavement Surface Condition
Rating Manual serves as the technical foundation for pavement condition assessment throughout
the province, including Prince George's municipal network. The manual, currently in its Sixth
Edition (March 2020), provides standardized methodologies for distress identification,
38

classification, and quantification specifically calibrated for British Columbia's diverse climate
conditions and pavement materials.
The BCMoTI methodology represents a localized adaptation of the internationally recognized
ASTM D6433 Standard Practice for Roads and Parking Lots Pavement Condition Index Surveys,
modified to address distress types and severity patterns most observed in British Columbia. This
adaptation ensures that assessment procedures accurately reflect regional pavement performance
characteristics while maintaining compatibility with broader pavement management frameworks.
Key modifications incorporated in the BCMoTI manual include additional distress categories such
as longitudinal wheel path cracking, pavement edge cracking, and longitudinal joint cracking that
reflect specific deterioration patterns observed in BC's climate

39

4. Chapter 4: Results and Analysis
This chapter presents the comprehensive findings from the pavement condition assessment and
predictive modeling analysis conducted for Prince George's road network. Building upon the
methodology established in Chapter 4, the analysis encompasses four survey periods (2016, 2017,
2020, and 2023) and provides critical insights into pavement deterioration patterns, distress
evolution, and predictive modeling performance. The results directly address the research
objectives established in Chapter 1.
4.1

Pavement Distress Condition Assessment

The systematic evaluation of pavement distresses across Prince George's road network reveals
significant patterns in deterioration mechanisms, spatial distribution, and temporal evolution. This
comprehensive assessment provides the empirical foundation for understanding pavement
performance under harsh northern climate conditions and establishes the basis for developing
targeted maintenance strategies.
4.2

Overall Network Condition Evolution

The pavement condition assessment program employed automated laser survey technology to
systematically document distress characteristics across multiple survey periods. The temporal
distribution of data collection demonstrates varying coverage strategies, with the most
comprehensive assessment conducted in 2020 covering 31,084 data points, while targeted surveys
in 2016, 2017, and 2023 focused on specific road classifications and high-priority corridors.

Figure 9 Distresses All Years - Bar chart showing number of distresses vs. number of points analyzed across 2016, 2017, 2020,
and 2023

Analysis of distress frequency relative to survey coverage reveals that 2023 exhibited the highest
distress-to-data-point ratio, indicating targeted surveying of road segments with elevated
40

deterioration levels. From 2016 to 2023, the year characterized by the most extensive distresses
compared to points analyzed was 2023, which demonstrated that the analysis for 2023 was efficient
as it analyzed certain roads with high distress frequency. However, this targeted approach resulted
in reduced overall network coverage compared to the comprehensive 2020 assessment.
The spatial analysis of Pavement Distress Index (PDI) distribution demonstrates that pavement
condition improvements were achieved between 2020 and 2023, with the best overall pavement
condition recorded in 2023. Conversely, the worst pavement conditions were documented in 2017
and 2020, corresponding to periods of intensive maintenance
intervention requirements. The poor pavement condition distribution
analysis indicates that deteriorated segments are dispersed throughout
the city rather than concentrated in specific zones, necessitating
network-wide maintenance strategies.

Figure 10 Overall PDI condition maps for all years (2016-2023) showing spatial distribution showing Good/Fair/Poor
categories across all survey years

The temporal progression shows that the best overall pavement condition was achieved in 2023,
while the most deteriorated conditions were observed in 2017 and 2020. This improvement
demonstrates the effectiveness of comprehensive rehabilitation strategies implemented between
2020 and 2023, particularly mill-and-fill operations and overlay applications that successfully
addressed structural deficiencies identified in earlier assessments.
The survey coverage varied significantly across assessment periods, with the 2020 survey
representing the most comprehensive network evaluation at 31,084 data points, while 2023
focused specifically on arterial and collector roads with 8,942 data points. The 2016 baseline
survey covered 9,328 data points primarily on arterial and collector roads, while the 2017
expanded assessment included 12,475 data points covering local roads and alleys. This variation
in coverage reflects strategic prioritization of high-traffic corridors in later assessments while
maintaining analytical depth for critical infrastructure.

41

Figure 11 Survey coverage comparison showing data points and network coverage by year

The concentration of poor pavement conditions throughout the city without geographic clustering
suggests that deterioration mechanisms result from systematic factors affecting the entire network
rather than localized issues. This distribution pattern validates the research approach of developing
network-level management strategies that address fundamental climate and material performance
relationships.
4.3
4.3.1

Comprehensive Distress Type Analysis
Transverse Cracking Analysis

Transverse cracking emerged as the most prevalent and significant distress type across all survey
periods, demonstrating substantial impact on overall network condition. The distress exhibits
distinct temporal patterns reflecting both environmental influences and maintenance intervention
effectiveness.

Figure 12 Transverse Cracking condition Trend (Photo credit to Mr. Alireza Noory)

42

Transverse Cracking Severity Analysis
Transverse cracking in Prince George overall demonstrates moderate to high severity
characteristics and may be considered highly severe in some years throughout the analysis period.
The severity analysis reveals that transverse cracking was most severe in 2020, when the
combination of high network coverage (78.2%) and substantial high-severity occurrences (29.5%)
created the most challenging conditions for network management. The temporal severity
progression shows that transverse cracking severity was increasing significantly in 2017 and 2020,
with 2017 representing the peak of high-severity deterioration at 39.9% of affected segments.
However, severity patterns began decreasing substantially in 2023, demonstrating the effectiveness
of targeted maintenance interventions and strategic rehabilitation programs implemented between
the peak deterioration period and the most recent assessment.

Figure 13 Transverse Cracking Severity in percentages All years - Comprehensive table and bar chart showing severity
distribution across all years

Transverse Cracking Extent Analysis
Transverse cracking demonstrated its most frequent occurrence in 2020 and 2017, when network
coverage reached 78.2% and 77.2% respectively, representing the peak periods of extensive
deterioration across Prince George's road network. Overall transverse cracking in Prince George
is extensive, consistently affecting most surveyed segments throughout the analysis period and
requiring comprehensive management strategies to address the widespread nature of this
deterioration pattern. The temporal extent progression shows that transverse cracking was
increasing substantially in 2017 and 2020, with these years representing the peak of network-wide
deterioration requiring intensive intervention. However, extent patterns began decreasing
significantly in 2023, with network coverage reducing to 62.4% and marked improvements in the
distribution toward lower extent categories, demonstrating the effectiveness of systematic
maintenance programs in reducing the widespread nature of transverse cracking across the
municipal road network.

43

Figure 14 Transverse Cracking Extent in percentages All years - Comprehensive table and bar chart showing Extent distribution
across all years

4.3.2

Meandering Longitudinal Cracking Analysis

Meandering longitudinal cracking represents the second most significant distress category,
demonstrating consistent presence across all survey periods with evolving severity characteristics
indicating progressive structural deterioration influenced by traffic loading and environmental
factors.

Figure 15 Meandering Longitudinal Cracking condition Trend (Photo credit to Mr. Alireza Noory)

Meandering Longitudinal Cracking Severity Analysis
Meandering longitudinal cracking in Prince George overall demonstrates low to moderate severity
characteristics but reached moderate to high severity levels in 2017 and 2020. The severity analysis
reveals that meandering longitudinal cracking was most severe in 2017 and 2020, when highseverity occurrences peaked at 19.2% and 13.2% respectively, combined with substantial moderate
severity coverage exceeding 40% in both years. The temporal severity progression shows that
meandering longitudinal cracking severity was increasing throughout the years from the baseline
31.6% coverage in 2016 to peak levels in 2017-2020, before stabilizing in 2023 with improved
severity distribution despite expanded network coverage.

44

Figure 16 Meandering Longitudinal Cracking Severity in percentages All years - Comprehensive table and bar chart showing
Extent distribution across all years

Meandering Longitudinal Cracking Extent Analysis
Meandering longitudinal cracking demonstrated its most extensive coverage in 2023, reaching
79.5% of the network, followed closely by 2017 and 2020 with 76.0% and 74.8% coverage
respectively. Overall meandering longitudinal cracking in Prince George is extensive and
widespread, consistently affecting most surveyed segments since 2017 and representing one of the
most prevalent distress types across the municipal road network. The temporal extent progression
shows that meandering longitudinal cracking was dramatically increasing from 2016 baseline
levels (31.6%) to peak coverage in 2017, maintaining extensive coverage through 2020, and
continuing to expand in 2023, indicating persistent and growing deterioration patterns requiring
comprehensive long-term management strategies.

Figure 17 Meandering Longitudinal Cracking Extent in percentages All years - Comprehensive table and bar chart showing
Extent distribution across all years

45

4.3.3

Longitudinal Wheel Path Cracking Analysis

Longitudinal wheel path cracking maintained relatively stable presence across all survey periods,
representing traffic-load-related deterioration patterns with consistent network coverage and
manageable severity levels demonstrating effective management of heavy vehicle loading impacts.

Figure 18 Longitudinal Wheel Path Cracking Trend

Longitudinal Wheel Path Cracking Severity Analysis
Longitudinal wheel path cracking in Prince George overall demonstrates stable and manageable
severity characteristics throughout the analysis period. The severity patterns remained relatively
consistent across all survey years, indicating effective management of traffic-load-related
deterioration factors through appropriate maintenance timing and intervention strategies. The
temporal severity progression shows that longitudinal wheel path cracking maintained stable
severity distribution without significant peaks or deterioration phases, reflecting successful longterm management of heavy vehicle loading impacts and appropriate pavement design adequacy
for anticipated traffic conditions.

Figure 19 Longitudinal Wheel path Cracking Severity in percentages All years - Comprehensive table and bar chart showing
Severity distribution across all years

46

Longitudinal Wheel Path Cracking Extent Analysis
Longitudinal wheel path cracking demonstrated relatively stable extent coverage throughout the
analysis period, with minor fluctuations indicating effective traffic-load management strategies.
The distress maintained consistent network presence around 16-20% of total distresses across all
survey years, representing manageable and predictable deterioration patterns. The temporal extent
progression shows that longitudinal wheel path cracking experienced a decrease in 2017 followed
by recovery to baseline levels in 2020 and 2023, indicating resilient pavement performance under
traffic loading with effective maintenance intervention when required, demonstrating appropriate
long-term management of wheel path deterioration across the municipal road network.

Figure 20 Longitudinal Wheel path Cracking Extent in percentages All years - Comprehensive table and bar chart showing
Extent distribution across all years

4.3.4

Alligator Cracking Analysis

Alligator cracking demonstrated the most significant improvement among all distress types,
reflecting successful structural rehabilitation interventions and strategic maintenance targeting of
areas with advanced structural deterioration.

Figure 21 Alligator Cracking Trend

47

Alligator Cracking Severity Analysis
Alligator cracking in Prince George demonstrated the most successful severity management
among all distress types, achieving near-complete elimination through systematic structural
rehabilitation interventions. The severity analysis reveals that alligator cracking peaked in 2017,
particularly affecting older local roads and areas with inadequate structural capacity for current
loading conditions. The temporal severity progression shows that alligator cracking severity
decreased systematically from 2017 peak conditions through 2020 and 2023, representing highly
effective structural rehabilitation program implementation and demonstrating that comprehensive
maintenance strategies can successfully resolve advanced structural deterioration challenges when
properly targeted and executed.

Figure 22 Alligator Cracking Severity in percentages All years - Comprehensive table and bar chart showing Severity
distribution across all years

Alligator Cracking Extent Analysis
Alligator cracking demonstrated the most dramatic extent reduction among all distress types,
achieving near-complete elimination from 4.85% of network distresses in 2016 to only 0.39% in
2023. The extent analysis reveals that alligator cracking peaked in 2017 with localized but
significant coverage patterns, followed by systematic reduction through comprehensive structural
rehabilitation programs. The temporal extent progression shows that alligator cracking extent
decreased consistently and substantially from peak levels in 2017 through 2020 and 2023,
representing exceptionally successful structural maintenance program implementation and
demonstrating that targeted rehabilitation strategies can effectively eliminate advanced structural
deterioration when properly planned and executed across the municipal road network.

48

Figure 23 Alligator Cracking Extent in percentages All years - Comprehensive table and bar chart showing Extent distribution
across all years

4.3.5

Rutting Analysis

Rutting exhibited the most dramatic emergence pattern among all distress types, transitioning from
complete absence to widespread network presence and representing the most significant emerging
maintenance challenge requiring immediate strategic attention and intervention development.

Figure 24 Rutting Trend

Rutting Severity Analysis
Rutting in Prince George represents the most dramatic emergence pattern among all distress types,
transitioning from complete absence in 2016 to near-universal network presence by 2023. The
severity analysis reveals that rutting development followed a systematic progression from initial
emergence in 2017 through widespread low-severity manifestation by 2023, with 92.9% of
affected segments exhibiting low severity conditions. The temporal severity progression shows
that rutting severity was systematically increasing from zero baseline in 2016 through progressive
development in 2017 and 2020 to comprehensive network coverage in 2023, representing the most
significant emerging maintenance challenge requiring immediate strategic intervention
development and comprehensive treatment program implementation.

49

Figure 25 Rutting Severity in percentages All years - Comprehensive table and bar chart showing Severity distribution across all
years

Rutting Extent Analysis
Rutting demonstrated the most dramatic extent emergence among all distress types, progressing
from complete absence in 2016 to the most extensive single distress challenge by 2023, affecting
98.4% of the surveyed network. The extent analysis reveals that rutting development followed an
accelerating pattern from initial manifestation in 2017 (55.7% coverage) through progressive
expansion in 2020 (54.8% coverage) to comprehensive network penetration in 2023. The temporal
extent progression shows that rutting extent was systematically and dramatically increasing from
zero baseline through progressive development phases to near-universal coverage, with 2023
showing the highest concentration in intermediate to frequent extent categories, representing the
most critical emerging infrastructure challenge requiring immediate comprehensive intervention
strategy development and implementation across the entire municipal road network.

Figure 26 Rutting Extent in percentages All years - Comprehensive table and bar chart showing Extent distribution across all years

50

4.3.6

Pavement Edge Cracking Analysis

Pavement edge cracking demonstrated irregular survey coverage with significant data collection
gaps, complicating comprehensive trend analysis but revealing substantial deterioration when
measured, representing critical infrastructure edge integrity challenges.

Figure 27 Pavement Edge Cracking Trend

Pavement Edge Cracking Severity Analysis
Pavement edge cracking in Prince George demonstrates irregular assessment patterns due to data
collection gaps but reveals significant severity emergence when measured. The severity analysis
shows that edge cracking was most severe and extensive in 2023, becoming the dominant distress
type after being absent from 2020 data collection. The temporal severity progression indicates that
pavement edge cracking severity was increasing from moderate baseline levels in 2016-2017,
followed by a data collection gap in 2020, and dramatic emergence in 2023 to become the most
severe and frequent distress type, suggesting accelerated edge deterioration possibly related to
drainage issues, frost action, or traffic loading patterns requiring immediate comprehensive
assessment and intervention strategy development.

Figure 28 Pavement Edge Cracking Severity in percentages All years - Comprehensive table and bar chart showing Severity
distribution across all years

51

Pavement Edge Cracking Extent Analysis
Pavement edge cracking demonstrated the most dramatic extent emergence among measured
distress types, transforming from manageable coverage levels (18.9% in 2016, 28.4% in 2017) to
the most extensive single distress challenge (85.4% in 2023). The extent analysis reveals that edge
cracking was completely absent from 2020 data collection, creating a critical assessment gap,
followed by explosive extent expansion to become the dominant network distress by 2023. The
temporal extent progression shows that pavement edge cracking extent was gradually increasing
from baseline levels through 2017, followed by unmeasured conditions in 2020, and dramatic
comprehensive emergence in 2023 to represent the most extensive infrastructure deterioration
challenge requiring immediate strategic intervention development and comprehensive
rehabilitation program implementation across the municipal road network.

Figure 29 Pavement Edge Cracking Extent in percentages All years - Comprehensive table and bar chart showing Extent
distribution across all years

4.3.7

Longitudinal Cracking Analysis

Longitudinal cracking maintained moderate network presence with relatively stable patterns
throughout the analysis period, representing manageable structural deterioration with effective
maintenance strategies maintaining consistent network impact levels.

Figure 30 Longitudinal Cracking condition Trend (Photo credit to Mr. Alireza Noory)

52

Longitudinal Cracking Severity Analysis
Longitudinal cracking in Prince George demonstrates stable and manageable severity
characteristics throughout the analysis period, with effective maintenance strategies maintaining
consistent network impact levels. The severity analysis reveals that longitudinal cracking
maintained moderate severity patterns without significant peaks or deterioration phases, indicating
appropriate intervention timing and technique selection. The temporal severity progression shows
that longitudinal cracking severity was stable with minor fluctuations from baseline 2016 levels
through slight improvement in 2017, followed by consistent management through 2020 and 2023,
representing successful long-term maintenance effectiveness and demonstrating that systematic
preventive strategies can effectively control structural deterioration progression when
appropriately implemented.

Figure 31 Longitudinal Cracking Severity in percentages All years - Comprehensive table and bar chart showing Severity
distribution across all years

Longitudinal Cracking Extent Analysis
Longitudinal cracking demonstrated stable and manageable extent coverage throughout the
analysis period, maintaining consistent network presence around 9-15% of total distresses across
all survey years. The extent analysis reveals that longitudinal cracking coverage decreased
moderately from 2016 baseline levels (14.73%) through effective maintenance intervention in
2017 (9.16%), followed by stable management at approximately 10.5% coverage in 2020 and
2023. The temporal extent progression shows that longitudinal cracking extent was effectively
managed with initial reduction followed by consistent stability, indicating successful maintenance
strategies and demonstrating that appropriate preventive intervention can maintain manageable
deterioration levels while preventing expansion of structural cracking across the municipal road
network.

53

Figure 32 Longitudinal Cracking Extent in percentages All years - Comprehensive table and bar chart showing Extent
distribution across all years

4.3.8

Pothole Analysis

Pothole occurrence remained limited throughout all survey periods but demonstrated variable
patterns requiring ongoing monitoring for safety considerations, representing critical localized
surface failure requiring immediate response protocols.

Figure 33 Pothole condition Trend (Photo credit to Mr. Alireza Noory)

Pothole Severity Analysis
Potholes in Prince George demonstrate variable severity patterns with successful long-term
management achieving near-elimination by 2023. The severity analysis reveals that pothole
conditions peaked in 2017 with elevated surface failure rates requiring enhanced maintenance
response, representing the most challenging period for emergency surface maintenance. The
temporal severity progression shows that pothole severity was increasing from baseline 2016
levels to peak conditions in 2017, followed by systematic reduction through 2020 and nearelimination in 2023, demonstrating highly effective reactive maintenance strategies and
emergency response protocol implementation for surface failure management across the municipal
road network.

54

Figure 34 Pothole Severity in percentages All years - Comprehensive table and bar chart showing Severity distribution across all
years

Pothole Extent Analysis
Potholes demonstrated successful extent management with systematic reduction from peak levels
to near-elimination, representing effective emergency response and surface maintenance
strategies. The extent analysis reveals that pothole coverage was most extensive in 2017 (8.37%
of network distresses), indicating peak surface failure conditions requiring intensive reactive
maintenance response. The temporal extent progression shows that pothole extent was increasing
from baseline 2016 levels (4.38%) to peak coverage in 2017, followed by systematic reduction
through 2020 (3.57%) and near-elimination in 2023 (0.78%), demonstrating highly successful
surface failure management and emergency response protocol effectiveness for maintaining safe
driving conditions and preventing localized surface deterioration across the municipal road
network.

Figure 35 Pothole Extent in percentages All years - Comprehensive table and bar chart showing Extent distribution across all
years

55

4.4

Model Results and Performance Analysis

This section presents the comprehensive analysis of three predictive modeling approaches
developed for Pavement Distress Index (PDI) forecasting: Multiple Linear Regression (MLR),
Random Forest (RF), and Artificial Neural Network (ANN). The comparative evaluation
demonstrates the effectiveness of advanced machine learning techniques for pavement
performance prediction under northern climate conditions, providing essential foundation for
strategic maintenance planning and resource allocation optimization.
4.4.1

Model Development Framework

The predictive modeling framework incorporated comprehensive variable selection including
distress severity and extent measurements for all eight distress types, climatic parameters
encompassing maximum, minimum, and mean temperature values, precipitation data, traffic
loading information, rutting depth measurements, and International Roughness Index values. The
modeling approach employed systematic 80-20 train-test data splitting with rigorous performance
evaluation using Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and coefficient
of determination (R²) metrics.
The comprehensive modeling framework addresses the complex relationships between
environmental factors, traffic loading, and pavement deterioration patterns. Variable selection
procedures incorporated domain expertise with statistical significance testing to ensure optimal
predictor inclusion while maintaining model parsimony and interpretability requirements.
4.4.2

Random Forest Model Performance

The Random Forest ensemble learning approach demonstrated superior performance
characteristics among the three modeling techniques, achieving optimal balance between
prediction accuracy, model robustness, and practical implementation requirements for municipal
pavement management applications.
Performance Metrics:
•
•
•

Mean Squared Error (MSE): 0.30
Root Mean Squared Error (RMSE): 0.55
Coefficient of Determination (R²): 0.96

The Random Forest model achieved 96.16% explained variance in PDI prediction, representing
exceptional accuracy for pavement performance forecasting applications under challenging
northern climate conditions. The ensemble learning approach effectively captured complex nonlinear relationships between predictor variables and pavement condition while maintaining robust
performance across varying data conditions and environmental scenarios.

56

Figure 36 Random Forest variable importance plot showing key contributing factors to PDI prediction

Variable importance analysis revealed that rutting measurements, temperature parameters, and
specific distress types contribute most significantly to PDI prediction accuracy. The model's
superior handling of complex interactions between climatic factors and pavement deterioration
mechanisms makes it particularly suitable for northern climate applications where freeze-thaw
cycles, temperature fluctuations, and precipitation patterns significantly influence infrastructure
performance.
The ensemble methodology combines multiple decision trees to reduce overfitting risk while
improving prediction stability across different data subsets. This robustness characteristic proves
essential for municipal applications where data quality and availability may vary across different
survey periods and network sections, giving it a practical advantage over the more complex ANN
approach.

57

Figure 37 Random Forest actual vs. predicted PDI scatter plot with performance statistics

4.4.3

Artificial Neural Network Model Performance

The Artificial Neural Network approach achieved the highest prediction accuracy among all
modeling techniques, demonstrating sophisticated pattern recognition capabilities for complex
pavement deterioration relationships and non-linear interaction effects characteristic of
infrastructure performance under variable environmental conditions.
Performance Metrics:
•
•
•

Mean Squared Error (MSE): 0.23
Root Mean Squared Error (RMSE): 0.48
Coefficient of Determination (R²): 0.95

The ANN model achieved 95.11% explained variance with the lowest MSE and RMSE values,
indicating superior prediction precision for individual pavement segment condition forecasting.
The neural network architecture effectively captured complex non-linear interactions between
input variables, providing high accuracy forecasting capabilities essential for detailed maintenance
planning and intervention timing optimization.

58

Figure 38 ANN model actual vs. predicted PDI scatter plot with trend line and performance statistics

The neural network's sophisticated pattern recognition capabilities enable accurate modeling of
complex deterioration mechanisms characteristic of northern climate conditions, including freezethaw cycles, temperature fluctuations, moisture-related damage patterns, and traffic loading
interactions. The six-neuron hidden layer architecture provides optimal balance between model
complexity and computational efficiency for municipal pavement management applications.

59

Figure 39 ANN model architecture diagram showing input variables, hidden layer, and output structure

The neural network approach demonstrates particular effectiveness in capturing threshold effects
and non-linear deterioration patterns where pavement condition changes rapidly under specific
environmental or loading conditions. This capability proves valuable for identifying critical
intervention timing and predicting accelerated deterioration periods.
4.4.4

Multiple Linear Regression Model Performance

The Multiple Linear Regression approach served as the baseline modeling technique, providing
interpretable coefficient relationships while demonstrating acceptable prediction accuracy for
linear deterioration patterns and establishing foundation analysis for comparison with advanced
modeling approaches.
Performance Metrics:
•
•
•

Mean Squared Error (MSE): 0.85
Root Mean Squared Error (RMSE): 0.92
Adjusted Coefficient of Determination (Adjusted R²): 0.81

The MLR model achieved an adjusted R² of 81.65%, representing acceptable performance for
linear relationship modeling while accounting for the model's 15 predictor variables. The adjusted
R² provides a more conservative estimate of explained variance compared to standard R² by
penalizing model complexity, ensuring that the reported performance reflects genuine predictive
60

capability rather than overfitting. This adjusted metric is particularly relevant for MLR given its
parametric nature and the substantial number of input variables, providing valuable insights into
individual factor contributions to pavement deterioration. While demonstrating lower accuracy
compared to machine learning approaches, the MLR model maintains high interpretability for
stakeholder communication and regulatory compliance requirements.

Figure 40 MLR model actual vs. predicted PDI scatter plot with regression line

The linear regression approach effectively identified significant predictor variables and provided
foundation analysis for understanding basic relationships between environmental factors, traffic
loading, and pavement condition. Coefficient analysis reveals that temperature parameters,
specific distress measurements, and traffic data contribute most significantly to linear deterioration
patterns.

61

Figure 41 MLR coefficient significance plot showing predictor variable contributions

Residual analysis confirmed appropriate model behavior without systematic bias or
heteroscedasticity issues, validating the linear modeling assumptions for baseline performance
evaluation. The straightforward implementation requirements and computational efficiency make
MLR suitable for routine monitoring applications where interpretability outweighs maximum
prediction accuracy requirements.
4.4.5

Model Comparison and Selection Criteria

Comparative analysis reveals distinct performance characteristics among the three modeling
approaches, each offering specific advantages for different pavement management applications
and decision-making requirements. The evaluation framework considers prediction accuracy,
computational requirements, interpretability needs, and practical implementation constraints.
The performance metrics presented utilize adjusted R² for the MLR model to ensure fair
comparison across modeling approaches. While Random Forest and ANN models report standard
R², the MLR adjusted R² accounts for the 15 predictor variables included in the linear model,
providing a more conservative and rigorous assessment of model fit. This methodological
distinction is important because adjusted R² prevents overestimation of MLR performance that
could occur from including multiple predictors, whereas machine learning approaches employ
different internal validation mechanisms (out-of-bag error for Random Forest and validation loss
for ANN) that inherently guard against overfitting.
62

Figure 42 Model performance comparison chart showing MSE, RMSE, and R² values across all three models.

The ANN model achieved the highest prediction accuracy with MSE of 0.23 and R² of 0.95,
followed closely by the Random Forest model with MSE of 0.30 and R² of 0.96. The MLR model,
while providing lower accuracy (MSE of 0.85, adjusted R² of 0.81), offers superior interpretability
and computational efficiency for routine applications. Notably, the adjusted R² for MLR provides
a conservative estimate appropriate for parametric models with multiple predictors, ensuring the
comparison fairly represents each model's true predictive performance.
For practical pavement management applications, the Random Forest model provides an optimal
balance between prediction accuracy, model interpretability, and computational requirements. The
ensemble learning approach maintains robust performance across varying data conditions while
providing variable importance insights essential for maintenance decision-making and resource
allocation optimization.
The machine learning approaches demonstrate superior capability for capturing complex
interaction effects between environmental factors, traffic loading, and deterioration patterns
characteristic of northern climate infrastructure performance. However, the interpretability
advantage of linear regression maintains value for regulatory reporting and stakeholder
communication requirements.

63

4.4.6

Three-Dimensional Relationship Analysis

To further understand the complex interactions between pavement condition indices and distress
characteristics, advanced three-dimensional visualization techniques were employed to examine
multivariable relationships that traditional two-dimensional plots cannot adequately capture. The
International Roughness Index (IRI), serving as the primary measure of functional performance,
demonstrates significant correlations with both PDI values and distress severity levels, reflecting
the interconnected nature of structural and functional pavement deterioration.
The three-dimensional analysis reveals critical insights into how roughness progression relates to
overall pavement condition while simultaneously considering the influence of distress severity
patterns. This multidimensional perspective provides essential understanding for maintenance
decision-making, as it demonstrates how functional performance degradation (measured through
IRI) correlates with structural condition decline (quantified through PDI) under varying distress
severity scenarios characteristic of northern climate conditions.

Figure 43 Three-dimensional relationship visualization between IRI, PDI, and distress severity levels showing the complex
interactions between functional performance, structural condition, and distress intensity

The complementary analysis of distress extent relationships provides equally important insights
into pavement performance patterns, as extent measurements quantify the spatial distribution of
deterioration across pavement surfaces. Research has demonstrated that distress extent often
exhibits different correlation patterns with functional performance compared to severity
measurements, reflecting the complex mechanisms through which pavement deterioration
manifests under varying traffic and environmental conditions.
64

The three-dimensional visualization of IRI-PDI-Extent relationships reveals how widespread
distress distribution affects both ride quality and overall condition ratings. This analysis is
particularly valuable for understanding threshold effects where extensive but low-severity
distresses may significantly impact functional performance while having different implications for
structural integrity. The extent analysis supports strategic maintenance planning by identifying
conditions where surface treatments may be more appropriate than structural interventions, or
conversely, where extensive distress patterns indicate underlying structural issues requiring
comprehensive rehabilitation.

Figure 44 Three-dimensional relationship visualization between IRI, PDI, and distress extent measurements demonstrating how
spatial distribution of distresses influences both functional and structural performance indicators

These three-dimensional visualizations provide comprehensive understanding of the multifaceted
relationships governing pavement performance under northern climate conditions. The analysis
demonstrates that both distress severity and extent contribute significantly to the relationship
between functional performance (IRI) and structural condition (PDI), but through different
mechanisms that require distinct maintenance approaches.
The combined severity and extent analysis supports the Random Forest model's superior
performance by illustrating the complex, non-linear interactions that ensemble learning methods
can effectively capture. Traditional linear regression approaches cannot adequately model these
multidimensional relationships, explaining the substantial performance differences observed in the
comparative analysis. The three-dimensional perspectives validate the importance of considering
multiple distress characteristics simultaneously when developing predictive models for municipal
pavement management applications.
65

66

5. Chapter 5: Discussion, Conclusion and Recommendations
5.1

Interpretation of Findings
5.1.1

Distress Evolution and Network Performance Insights

The comprehensive analysis of Prince George's pavement network from 2016 to 2023 reveals
profound shifts in distress patterns that fundamentally challenge traditional maintenance
approaches. The research demonstrates that climate-adaptive pavement management requires
dynamic strategies responsive to evolving deterioration mechanisms rather than static, conditionbased interventions.
Emergence of Rutting as the Dominant Challenge
The most noticable finding is rutting's transformation from complete absence in 2016 to affecting
98.4% of the network by 2023. This unprecedented emergence pattern suggests systematic changes
in pavement response mechanisms, likely attributed to the combination of climate change effects,
evolving traffic patterns, and material aging characteristics specific to northern environments. The
progression from zero baseline through 55.7% coverage in 2017 to near-full presence demonstrates
accelerating deterioration that traditional predictive models fail to anticipate.
The severity distribution in 2023, with 92.9% of affected segments exhibiting low severity
conditions, indicates that rutting emergence follows a consistent pattern across the network rather
than isolated failure mechanisms. This finding suggests that current pavement structural designs
and asphalt mixture designs may be inadequate for evolving environmental and loading conditions,
necessitating fundamental reassessment of design standards and material specifications for
northern climate applications.
Pavement Edge Cracking as a Critical Infrastructure Integrity Issue
The dramatic emergence of pavement edge cracking from manageable levels (18.9% in 2016,
28.4% in 2017) to the most extensive single distress challenge (85.4% in 2023) represents a critical
infrastructure integrity crisis. The absence of edge cracking data in 2020 creates a significant
analytical gap, but the explosive expansion documented in 2023 suggests systematic deterioration
mechanisms affecting pavement-curb interfaces, drainage effectiveness, and structural continuity.
The severity distribution in 2023, with 63% of segments exhibiting low severity and 17% showing
moderate to high severity conditions, indicates that edge deterioration affects both functional and
structural pavement performance. This pattern suggests that current construction practices for
pavement edge details, joint sealing procedures, and drainage management require comprehensive
revision to address harsh northern climate conditions.
Successful Management of Traditional Distress Types
The research demonstrates exceptional success in managing traditional structural distresses,
particularly alligator cracking, which achieved near-elimination from 4.85% of network distresses
in 2016 to only 0.39% in 2023. This achievement validates the effectiveness of systematic
67

structural rehabilitation programs when properly targeted and executed, demonstrating that
comprehensive mill-and-fill strategies, reclamation procedures, and reconstruction programs can
successfully address advanced structural deterioration.
Similarly, the management of transverse cracking, while remaining the most prevalent distress
type, shows systematic improvement from peak severity conditions in 2017 and 2020 to
manageable levels in 2023. The complete elimination of high-severity transverse cracking in 2023
demonstrates that targeted intervention strategies can effectively control thermal-related
deterioration mechanisms characteristic of northern climate conditions.
5.1.2

Advantages of Implementing Predictive Modeling

Superior Performance of Machine Learning Approaches
The comparative analysis reveals that advanced machine learning techniques substantially
outperform traditional statistical methods for pavement condition prediction under northern
climate conditions. The Random Forest model achieved 96.16% explained variance (R² = 0.96,
MSE = 0.30), while the Artificial Neural Network demonstrated (R² = 0.95, MSE = 0.23), both
significantly exceeding Multiple Linear Regression performance (R² = 0.81, MSE = 0.85).
The superior performance of ensemble learning and neural network approaches validates the
complex, non-linear nature of pavement deterioration under northern climate conditions. The
Random Forest model's variable importance analysis identified rutting measurements, temperature
parameters, and specific distress characteristics as primary prediction drivers, confirming the
critical role of climate-pavement interactions in deterioration progression.
Practical Implementation Considerations
Despite the ANN model's superior accuracy, the Random Forest approach provides an optimal
balance between prediction performance, interpretability, and computational requirements for
municipal applications. Randon Forest is the most suitable modeling tool supporting maintenance
decision-making and resource allocation optimization due to its consistency across varying data
conditions ..
The substantial performance difference between machine learning approaches and traditional
linear regression (approximately 15% improvement in explained variance) demonstrates the
inadequacy of simplified deterioration models for northern climate applications. This finding
supports the necessity of implementing advanced analytical techniques to capture the complex
interactions between environmental factors, traffic loading, and material performance
characteristic of harsh climate conditions.

68

5.2

Lessons Learned and Limitations
5.2.1

Research Methodology Insights

Data Collection Temporal Alignment Challenges
The analysis revealed significant challenges in aligning pavement condition data collected at
different intervals with corresponding climate and traffic information. The variation in survey
coverage across assessment periods, ranging from full network evaluation (31,084 data points in
2020) to targeted assessments (8,942 data points in 2023), created analytical complexities that
required careful consideration in model development and validation procedures.
The absence of pavement edge cracking data in 2020 created a critical gap in trend analysis that
limited understanding of edge deterioration progression patterns. This limitation emphasizes the
importance of maintaining consistent distress type coverage across all assessment periods to enable
comprehensive deterioration modeling and strategic planning.
Variable Selection and Model Calibration Learnings
The research demonstrated that effective predictive modeling for northern climate conditions
requires extensive variable preprocessing and careful selection of climate parameters that capture
both acute environmental events and cumulative exposure effects. The integration of distress
severity and extent measurements proved essential for capturing the multidimensional nature of
pavement deterioration under harsh environmental conditions.
The superior performance of machine learning approaches validates the complex, non-linear
nature of pavement-climate interactions but also highlights the importance of maintaining
sufficient training data to ensure model robustness across varying environmental and operational
conditions.
5.2.2

Research Limitations and Constraints

Temporal Data Scope Limitations
The research is constrained by the 7-year analysis period (2016-2023), which may not capture the
full range of climatic variability or long-term deterioration cycles. The limited scope restricts
assessment of treatment longevity and effectiveness under varying environmental conditions,
particularly for newer rehabilitation techniques and climate-adaptive materials.
The irregular survey intervals (2016, 2017, 2020, 2023) create gaps in deterioration trend analysis
that may affect the accuracy of predictive model calibration and validation. More frequent
condition assessment would enable enhanced understanding of seasonal deterioration patterns and
treatment timing optimization.

69

Treatment Effectiveness Assessment Limitations
The limited availability of detailed construction specifications, material properties, and quality
control data restricted assessment of factors contributing to treatment success or failure, limiting
the ability to provide specific material and construction recommendations.
5.2.3

Methodological Considerations for Future Research

Enhanced Data Integration Requirements
Future research should incorporate more sophisticated environmental monitoring including
subsurface temperature measurements, moisture content assessment, and detailed freeze-thaw
cycle documentation to enable enhanced understanding of climate-pavement performance
relationships.
The development of comprehensive treatment effectiveness databases that document intervention
timing, material specifications, construction conditions, and subsequent performance would enable
enhanced evaluation of maintenance strategy optimization under varying environmental
conditions.
Advanced Modeling Framework Development
The research demonstrates the potential for developing more sophisticated modeling frameworks
that integrate machine learning approaches with mechanistic-empirical principles to capture both
empirical performance relationships and fundamental deterioration mechanisms. Such approaches
could enable enhanced prediction accuracy while maintaining interpretability for practical
management applications.
5.3

Conclusion

This research provided a comprehensive analysis of pavement deterioration patterns, predictive
modeling capabilities, and management strategy optimization for northern climate conditions
through systematic evaluation of Prince George's road network from 2016 to 2023. The findings
demonstrate that effective pavement management in harsh northern environments requires
fundamental shifts from traditional maintenance approaches to climate-adaptive strategies that
account for unique deterioration mechanisms, emerging distress patterns, and complex
environmental interactions.
The wide emergence of rutting throughout the network and presence of pavement edge cracking
as the most extensive single distress type represents paradigmatic shifts in pavement performance
that demand immediate strategic response and long-term management framework revision. The
research validates that advanced machine learning techniques provide superior prediction
capabilities compared to traditional statistical methods, enabling enhanced decision-making and
resource optimization for municipal pavement management applications.

70

The development of climate-adaptive pavement management frameworks informed by
comprehensive data analysis, predictive modeling, and systematic treatment evaluation provides
Prince George with evidence-based foundations for addressing current infrastructure challenges
while establishing resilient management capabilities for future environmental and operational
conditions. The transferable insights and methodological approaches developed through this
research contributed to broader understanding of northern climate infrastructure management and
provide frameworks for similar municipalities facing comparable challenges.
The integration of technical analysis with practical implementation considerations ensures that
research findings translate into actionable strategies that can improve infrastructure performance,
optimize resource utilization, and enhance municipal capability for systematic pavement
stewardship under challenging northern climate conditions. The comprehensive recommendations
for treatment selection, strategic planning, and advanced technology implementation provide
roadmaps for transitioning from reactive maintenance approaches to proactive, data-driven
management systems that optimize both performance outcomes and sustainability objectives.
This research establishes that successful pavement management in northern climates requires
systematic integration of climate science, advanced analytical techniques, and strategic resource
allocation within comprehensive frameworks that balance immediate operational needs with longterm infrastructure resilience and sustainability goals. The methodological approaches and
analytical findings provide foundations for continued advancement of northern climate
infrastructure management practice and academic research.
5.4

Recommendations for City of Prince George
5.4.1

Immediate Strategic Priorities

Rutting Management Crisis Response
Prince George faces an immediate infrastructure crisis with rutting affecting 98.4% of the road
network by 2023. The city must implement emergency intervention protocols targeting the most
severely affected arterial and collector roads to prevent further structural deterioration and
maintain transportation system functionality. Immediate actions should include:
1. High-Priority Corridor Identification: Conduct urgent assessment of arterial roads and
major collectors experiencing moderate to severe rutting to prioritize intervention
sequencing based on traffic volume, economic importance, and safety considerations.
2. Emergency Funding Mobilization: Develop business case documentation demonstrating
the critical nature of rutting emergence to secure additional capital funding for
comprehensive rehabilitation programs that address systematic network deterioration.
Rutting represents a crucial safety hazard as it results in hydroplaning during storms
leading to elevated accident rates.
3. Advanced Material Implementation: Transition immediately to high-modulus Hot Mix
Asphalt (HIMA) mixtures and experimental fiber-reinforced HMA for all rutting
remediation projects. Scientific transition to Superpave asphalt mixture designs and
71

experimenting Balanced Mix Design (BMD) would definitely enhance the asphalt
mixture’s rutting resistance without significant reduction in the mixtures’ crack resistance.
Pavement Edge Integrity Restoration
The explosive emergence of pavement edge cracking affecting 85.4% of the network represents a
critical structural integrity challenge requiring immediate systematic intervention:
1. Construction Procedure Revision: Comprehensively review and revise construction
procedures for joints between HMA lanes and concrete curbs, incorporating enhanced
sealing techniques, improved drainage details, and climate-adaptive joint materials.
2. Drainage System Enhancement: Implement systematic drainage improvements along
pavement edges to prevent moisture infiltration and freeze-thaw damage that accelerates
edge deterioration under northern climate conditions.
3. Edge Rehabilitation Program: Develop dedicated edge rehabilitation procedures that
address both functional and structural aspects of edge deterioration, including partial
reconstruction where necessary to restore structural continuity.
5.4.2

Treatment Selection Strategy Optimization

Climate-Adaptive Treatment Matrix
Based on the research findings and analysis of historical treatment effectiveness, Prince George
should implement a revised treatment selection framework that accounts for the unique
deterioration patterns observed in the northern climate analysis:
For Rutting-Dominated Conditions (PDI 4-8, Rutting Severity 1-3):
•
•
•

Primary Treatment: Mill and Overlay (50-75mm) using high-modulus (HIMA) asphalt
mixtures
Alternative Treatment: Full-depth reclamation (100mm) for areas with underlying
structural deficiencies
Timing: Execute during optimal temperature windows (late spring/early summer) to
ensure proper compaction and curing

For Edge Cracking Conditions (Edge Extent 2-5):
•
•
•

Primary Treatment: Edge reconstruction with enhanced joint sealing and improved
drainage integration
Preventive Treatment: Systematic joint sealing optimization for early-stage edge
deterioration
Design Enhancement: Implement wider shoulder specifications and improved edge
support details

72

For Transverse Cracking Management (Severity 1-2, Extent 1-3):
•
•
•

Primary Treatment: Crack sealing during optimal seasonal windows (fall application)
Progressive Treatment: Microsurfacing for moderate coverage areas to prevent
progression
Rehabilitation Treatment: Mill and fill for extensive coverage areas showing progression
potential

For Alligator Cracking (Maintained Success):
•
•
•

Continue Current Strategy: Maintain aggressive mill and fill approach that achieved
near-elimination
Monitor: Implement enhanced monitoring for early detection to prevent recurrence
Preventive Focus: Emphasize preventive treatments in areas showing early fatigue
indicators
5.4.3

Implementation Framework for Advanced Management
Systems

Predictive Modeling Integration
The research demonstrates that Random Forest modeling provides optimal balance between
accuracy and practicality for municipal implementation. Prince George should implement the
following framework:
1. Annual Model Calibration: Establish procedures for annual model recalibration using
current condition data to maintain prediction accuracy as network conditions evolve.
2. Seasonal Prediction Updates: Develop quarterly prediction updates that incorporate
current climate data to optimize maintenance timing and resource allocation decisions.
3. Decision Support Integration: Integrate predictive modeling capabilities with existing
pavement management system to provide real-time decision support for maintenance
prioritization and budget allocation.
Enhanced Data Collection Protocols
The analysis reveals critical data collection gaps that limit analytical capabilities. Implement
enhanced protocols including:
1. Consistent Distress Coverage: Ensure all distress types are assessed during every survey
period to prevent analytical gaps like the 2020 edge cracking omission.
2. Climate Integration: Establish systematic integration of climate data with condition
assessment to enable enhanced understanding of environmental deterioration relationships.
3. Treatment Documentation: Implement comprehensive documentation of treatment
specifications, timing, and performance to enable enhanced effectiveness evaluation.

73

5.5

Recommendations for Future Academic Research
5.5.1

Advanced Material and Treatment Research

Northern Climate Material Development
The research identifies critical needs for material innovations specifically designed for harsh
northern climate conditions. Future research should focus on:
1. High-Modulus Asphalt Mixtures (HIMA) Utilization: Comprehensive laboratory and
field testing of high-modulus asphalt mixtures (HIMA) under controlled northern climate
exposure conditions to optimize mixture design parameters, aggregate specifications, and
binder selection criteria. Production of HIMA mixtures could be facilitated by high
polymer-modified binders using Styrene-Butadiene-Styrene (SBS) modifiers.
2. Fiber-Reinforced Asphalt Performance: Long-term performance evaluation of various
fiber reinforcement types (synthetic, natural, recycled) in HMA mixtures under freezethaw cycling, temperature extremes, and heavy loading conditions characteristic of
northern environments.
3. Climate-Adaptive Binder Technologies: Development and testing of modified asphalt
binders specifically formulated to maintain performance across extreme temperature
ranges while providing enhanced resistance to moisture damage and thermal cycling
effects.
Advanced Treatment Technique Development
Research opportunities exist for developing innovative treatment approaches that address the
unique deterioration patterns identified in this research:
1. Edge Rehabilitation Technologies: Development of specialized edge rehabilitation
techniques that address the joint interface between asphalt pavement and concrete
infrastructure while providing enhanced resistance to freeze-thaw damage and moisture
infiltration.
2. Rutting Prevention Strategies: Investigation of preventive treatments that can be applied
to delay or prevent rutting emergence, including surface modifications, structural
enhancements, and traffic management approaches.
3. Climate-Responsive Treatment Timing: Research into optimal treatment timing based
on seasonal climate patterns, material temperature requirements, and long-term
performance optimization under northern climate conditions.
5.5.2

Enhanced Modeling and Prediction Research

Multi-Scale Deterioration Modeling
The research demonstrates the effectiveness of machine learning approaches but identifies
opportunities for enhanced modeling frameworks:
74

1. Mechanistic-Empirical Integration: Development of hybrid modeling approaches that
combine machine learning pattern recognition capabilities with mechanistic-empirical
principles to enhance prediction accuracy while maintaining physical interpretability.
2. Real-Time Adaptive Modeling: Research into dynamic modeling frameworks that can
adapt prediction parameters based on real-time environmental monitoring, traffic data, and
emerging condition trends to maintain accuracy as conditions evolve.

75

References
AECOM. (2016). RIVA asset management system: Business process maps. City of Prince George.
https://www.princegeorge.ca
Al-Ajami, H. (2015). Pavement maintenance management systems.
Ali, A. A. (2022). Modeling of asphalt pavement performance indices in different climate regions
using soft computing techniques [Doctoral dissertation, Memorial University of Newfoundland].
Research.library.mun.ca
Ali, A. A., Milad, A., Hussein, A., Yusoff, N. I. M., & Heneash, U. (2023). Predicting pavement
condition index based on the utilization of machine learning techniques: A case study. Journal of
Road Engineering, 3, 266–278. https://doi.org/10.1016/j.jreng.2023.04.002
Allen, D. L. (1990). Review and analysis of pavement management practices in Kentucky.
Al-Swailmi, S., Al-Mansour, A. I., & Al-Tamimi, A. K. (1999). Pavement performance models for
state highways in Riyadh. Civil and Environmental Research, 7(5), 103–110.
Anastasopoulos, P. Ch., Mannering, F. L., & Haddock, J. E. (2009). Effectiveness and service
lives/survival curves of various pavement rehabilitation treatments.
Applied Research Associates, Inc. (2016). Strategic paved road management plan: Gap analysis
report (Contract: P15-130). City of Prince George. https://www.princegeorge.ca
Association of State Highway and Transportation Officials. (1993). AASHTO guide for design of
pavement structures. AASHTO.
Attoh-Okine. (2002). Combining use of rough set and artificial neural networks in doweledpavement-performance modeling: A hybrid approach. Journal of Transportation Engineering,
128(3), 270–275. https://doi.org/10.1061/(ASCE)0733-947X(2002)128:3(270)
Balzi, A., & Grilli, A. (2023). Pavement management system: Implementing project evaluation for
local road administrations. Transportation Research Procedia, 74, 1211–1218.
Borghetti, F., Derudi, M., Gandini, P., & Marchionni, G. (2024). Road infrastructure maintenance
operative priority: A computational model for drainage systems. Transportation Research
Interdisciplinary Perspectives, 25, 101100.
British Columbia Campus Press. (2025). Case: Urban Prince George. In Land use planning in
British
Columbia.
https://pressbooks.bccampus.ca/landuseplanninginbc/chapter/case_urban_princegeorge/
British Columbia Ministry of Transportation and Infrastructure. (2020). Pavement surface
condition rating manual (6th ed.). WSP Canada Limited.
76

British Columbia Ministry of Transportation and Infrastructure. (2025). Road condition
monitoring.
https://twm.th.gov.bc.ca/?c=tdp&lon=122.75041755326814&lat=53.91926807911665&z=10&sb
=1
Butt, A. A., Shahin, M. Y., Feighan, K. J., & Carpenter, S. H. (1987). Pavement performance
prediction model using the Markov process. Transportation Research Board.
City of Prince George. (2009). Transportation network planning study: Final report.
https://www.princegeorge.ca/sites/default/files/202301/Transportation_Network_Planning_Study_Final_Report_2009-11-03.pdf
City
of
Prince
George.
(2015).
Official
community
plan.
https://www.princegeorge.ca/City%20Hall/Documents/Official_Community_Plan/OCP-Bylaw8383.pdf
City
of
Prince
George.
(2024).
2024
https://www.princegeorge.ca/sites/default/files/202404/2024%20Road%20Rehab%20Location%20Map.pdf

road

Climate
Atlas
of
Canada.
(2025).
https://climateatlas.ca/map/canada/freezethaw_2030_85#

rehabilitation

projects.

Freeze-thaw

cycles.

Dong, J., Meng, W., Liu, Y., & Ti, J. (2021). A framework of pavement management system based
on
IoT and
big
data.
Advanced
Engineering
Informatics,
47,
101226.
https://doi.org/10.1016/j.aei.2020.101226
Ekramnia, T., & Nasimifar, M. (2022). Development of a methodological tool for treatment
prioritization in network-level pavement management system. Journal of Transportation
Engineering, Part B: Pavements. ascelibrary.org
Environment and Climate Change Canada. (2025). Historical weather data.
https://climate.weather.gc.ca/historical_data/search_historic_data_stations_e.html?searchType=st
nName&timeframe=1&txtStationName=prince+george&searchMethod=contains&optLimit=yea
rRange&StartYear=2016&EndYear=2023&Year=2024&Month=12&Day=1&selRowPerPage=1
00
Federal Highway Administration. (2020). Asset management practices and benefits - Case study
1 (Report No. FHWA-HIF-20-088). U.S. Department of Transportation.
Federal Highway Administration. (2022). Pavement management roadmap: Executive summary
(Report No. FHWA-HIF-22-055).

77

Gavin, K., Toll, D. G., & Cheema, S. (2003). Canadian Strategic Highway Research Program:
Effects of climate on pavement and foundation design. Canadian Geotechnical Journal, 40(3),
646–657.
Gong, H., Sun, Y., Shu, X., & Huang, B. (2018). Use of random forests regression for predicting
IRI of asphalt pavements. Construction and Building Materials, 189, 890–897.
https://doi.org/10.1016/j.conbuildmat.2018.09.017
Grilli, A., Casali, M., Muratori, D., & Bascucci, F. (2019). Pavement management system for local
administrations: Guidelines in the Republic of San Marino. International Journal of Pavement
Research and Technology, 12, 97–100.
Guha, S., & Hossain, K. (2021). A country-wide survey to understand pavement management
practices in Canada. Canadian Technical Asphalt Association. https://www.ctaa.ca
Guha, S. K., Hossain, K., & Lawlor, M. (2022). An economic pavement management framework
for extremely budget- and resource-constrained agencies in Canada. Transportation Research
Record, 2676(8), 554–570. https://doi.org/10.1177/03611981221084687
Guo, W., Zhang, J., Cao, D., & Yao, H. (2022). Cost-effective assessment of in-service asphalt
pavement condition based on random forests and regression analysis. Construction and Building
Materials, 330, 127219. https://doi.org/10.1016/j.conbuildmat.2022.127219
Haas, R., Hudson, W. R., & Zaniewski, J. (1994). Modern pavement management. Krieger
Publishing Company.
Hafez, M., & Ksaibati, K. (2021). Studying the effectiveness of changing parameters in pavement
management systems on optimum maintenance strategies of low-volume paved roads. Journal of
Transportation Engineering, Part B: Pavements. ascelibrary.org
Hamdi, A. (2015). Local calibration of AASHTOWare using Ontario pavement management
system data PMS2.
Hudson, W. R., Haas, R., & Uddin, W. (1997). Infrastructure management. McGraw-Hill
Professional.
Issa, A., & Abu-Eisheh, S. (2017). Evaluation of implementation of municipal roads' maintenance
plans in Palestine: A pilot case study. International Journal of Pavement Research and Technology,
10, 454–463.
Khahro, S. H. (2022). Defects in flexible pavements: A relationship assessment of the defects of a
low-cost pavement management system. Sustainability, 14(3), 1157.
Latitude.to.
(2025).
Prince
George,
https://latitude.to/map/ca/canada/cities/prince-george

British

Columbia,

Canada.
78

Lee, Y.-H., Mohseni, A., & Darter, M. I. (1993). Simplified pavement performance models.
Transportation Research Record, 1397, 7–14.
Li, J., Yin, G., Wang, X., & Yan, W. (2022). Automated decision making in highway pavement
preventive maintenance based on deep learning. Automation in Construction, 135, 104111.
Li, Q. (2009). Calibration of flexible pavement in mechanistic-empirical pavement design guide
for Washington state. Transportation Research Record, 2095(1), 73–83.
Loprencipe, G., Pantuso, A., & Di Mascio, P. (2017). Sustainable pavement management system
in urban areas considering the vehicle operating costs. Sustainability, 9(3), 453.
Machí, C., Chamorro, A., Videla, C., Pellicer Armiñana, E., & Yepes, V. (2014). An iterative
approach for the optimization of pavement maintenance management at the network level.
Marcelino, P., Antunes, M. L., Fortunato, E., & Gomes, M. C. (2019). Machine learning approach
for pavement performance prediction. International Journal of Pavement Engineering, 22(3),
341–354. https://doi.org/10.1080/10298436.2019.1609673
Moradi, M., & Assaf, G. J. (2023). Designing and building an intelligent pavement management
system for urban road networks. mdpi.com
Obaidat, M. T., Ghuzlan, K. A., & Al-Mestarehi, B. W. (2018). Integration of Geographic
Information System (GIS) and PAVER system toward efficient pavement maintenance
management system. Jordan Journal of Civil Engineering, 12, 449–462.
Obunguta, F., & Matsushima, K. (2022). Optimal pavement management strategy development
with a stochastic model and its practical application to Ugandan national roads. International
Journal of Pavement Engineering, 23, 2405–2419.
Pacific
Climate
Services.
(2025).
PCDS
https://services.pacificclimate.org/met-data-portal-pcds/app/

climate

data

portal.

Pantuso, A., Loprencipe, G., Bonin, G., & Teltayev, B. B. (2019). Analysis of pavement condition
survey data for effective implementation of a network level pavement management program for
Kazakhstan. Sustainability, 11, 901.
Prince George Open Data Portal. (2025). Transportation datasets.
cityofpg.opendata.arcgis.com/search?collection=Dataset&q=Transportation
Prince
George
Weather
Stats.
(2025).
Annual
https://princegeorge.weatherstats.ca/charts/precipitation-yearly.html

https://data-

precipitation

trends.

79

Qiao, Y., Dawson, A. R., Parry, T., Flintsch, G., & Wang, W. (2020). Flexible pavements and
climate change: A comprehensive review and implications. Sustainability, 12(1057).
https://doi.org/10.3390/su12031057
Qiao, Y., Flintsch, G. W., Dawson, A. R., & Parry, T. (2015). Evaluating the effects of climate
change on road maintenance intervention strategies and life-cycle costs. Transportation Research
Part D: Transport and Environment, 41, 492–503.
Ribeiro, L. C. P. (2020). Establishing a methodology for unpaved and paved roads management
system. ufv.br
Saliminejad, S. (2013). GIS-based probabilistic approach for assessing and enhancing
infrastructure data quality.
Santos, J., Ferreira, A., & Flintsch, G. (2017). A multi-disciplinary approach for the assessment of
road pavement life cycle. Structure and Infrastructure Engineering, 13(8), 1019–1030.
Smith, K. L., Pontius, J., & Galehouse, L. (2014). Asset management perspective (Report No.
FHWA-HIF-14-010). Federal Highway Administration.
Statistics Canada. (2021). Prince George, city (CY) [Census subdivision], British Columbia and
Fraser-Fort George, regional district [Census division], British Columbia (table). Census Profile,
2021 Census of Population. https://www12.statcan.gc.ca/census-recensement/2021/dppd/prof/details/page.cfm?Lang=E&SearchText=Prince%20George&DGUIDlist=2021A0005595
3023&GENDERlist=1,2,3&STATISTIClist=1&HEADERlist=0
Swarna, S. T. (2022). Climate change impact and adaptation for pavement infrastructure in
Canada [Doctoral dissertation, Memorial University of Newfoundland].
Tamagusko, T., & Ferreira, A. (2025). Pavement performance prediction using machine learning:
Supervised learning with tree-based algorithms. Transportation Research Procedia, 82, 2521–
2531. https://doi.org/10.1016/j.trpro.2024.12.202
Terzi, S. (2007). Modeling the pavement serviceability ratio of flexible highway pavements by
artificial neural networks. Construction and Building Materials, 21(3), 590–593.
https://doi.org/10.1016/j.conbuildmat.2005.11.001
Torres-Machí, C., Chamorro, A., Videla, C., Pellicer, E., & Yepes, V. (2014). An iterative approach
for the optimization of pavement maintenance management at the network level. The Scientific
World Journal, 2014, 524329.
Van Geem, C., Casse, C., Adolfs, T., & Diederiks, K. (2012, September). ViaBEL – A tool for
decision processes in pavement management of secondary road networks in Belgium. 4th
European Pavement and Asset Management Conference (EPAM 4), Malmö, Sweden.
80

Van Geem, C., & Massart, T. (2017, June). Quality insurance of visual inspections for pavement
management of communal road networks. World Conference on Pavement and Asset Management
(WCPAM 2017), Milano, Italy.
Van Geem, C., & Massart, T. (2018, May). Implementation and benefits of a low-cost PMS for
municipal road networks. 5th International Conference on Road and Rail Infrastructure (CETRA
2018), Zadar, Croatia.
Vines-Cavanaugh, D., Shamsabadi, S. S., Zhao, Y., & Huang, G. (2017). City-wide application of
the affordable and rapid StreetScan pavement-management system. Journal of Infrastructure
Systems, 23, B4016010.
Walls, J., III, & Smith, M. R. (1998). Life-cycle cost analysis in pavement design (Report No.
FHWA-SA-98-079). Federal Highway Administration, Pavement Division.
Wang, X. C., Zhao, J., Li, Q. Q., Fang, N. R., Wang, P. C., Ding, L. T., & Li, S. Q. (2020). A hybrid
model for prediction in asphalt pavement performance based on support vector machine and grey
relation analysis. Journal of Advanced Transportation, 2020, 7534970.
Wang, Z., & Pyle, T. (2019). Implementing a pavement management system: The Caltrans
experience. International Journal of Transportation Science and Technology, 8, 251–262.
Zagvozda, M., Dimter, S., Rukavina, T., & Vrtačnik Žižić, V. (2019). Application of GIS
technology in pavement management systems.
Zhao et al. (2021). Viscosity prediction of rubberized asphalt–rejuvenated recycled asphalt
pavement binders using artificial neural network approach. Journal of Materials in Civil
Engineering, 33(5), 04021071. https://doi.org/10.1061/(ASCE)MT.1943-5533.0003679

81

Appendix I. Detailed Distress Analysis
I.1. Transverse Cracking Severity Analysis
2016 Severity Analysis: Transverse cracking affected 65.4% of surveyed segments with a severity
distribution demonstrating 34.6% of segments with no distress (Severity 0), 46.4% exhibiting low
severity conditions (Severity 1), 14.7% showing moderate severity deterioration (Severity 2), and
4.3% displaying high severity cracking (Severity 3). The predominance of low severity conditions
in 2016 established baseline deterioration patterns requiring preventive maintenance strategies.

FIGURE I. 1. 2016 Transverse Cracking Severity pie chart and spatial distribution map

2017 Severity Analysis: The network experienced significant deterioration with 77.2% of
segments affected by transverse cracking. Severity distribution shifted dramatically to 22.8% with
no distress (Severity 0), only 9.6% remaining at low severity (Severity 1), 27.7% progressing to
moderate severity (Severity 2), and a concerning 39.9% reaching high severity conditions
(Severity 3). This substantial increase in high-severity transverse cracking indicated accelerated
deterioration mechanisms requiring immediate intervention.

FIGURE I. 2. 2017 Transverse Cracking Severity pie chart and spatial distribution map

82

2020 Severity Analysis: Network coverage reached 78.2% with severity distribution showing
21.8% segments with no distress (Severity 0), 18.8% at low severity (Severity 1), 30.0% at
moderate severity (Severity 2), and 29.5% at high severity (Severity 3). The persistence of
extensive high-severity cracking demonstrated ongoing structural challenges despite maintenance
efforts.

FIGURE I. 3. 2020 Transverse Cracking Severity pie chart and spatial distribution map

2023 Severity Analysis: Significant improvement achieved with 62.4% network coverage and
severity redistribution to 37.6% with no distress (Severity 0), 38.5% at low severity (Severity 1),
23.9% at moderate severity (Severity 2), and complete elimination of high-severity occurrences
(0.0% at Severity 3). This dramatic improvement reflects successful maintenance interventions
implemented between 2020 and 2023.

FIGURE I. 4. 2023 Transverse Cracking Severity pie chart and spatial distribution map

83

I.2. Transverse Cracking Extent Analysis
2016 Extent Analysis: The extent distribution demonstrated 34.6% of segments with no transverse
cracking (Extent 0), 30.1% with few occurrences affecting less than 10% of segment length (Extent
1), 23.4% with intermediate extent covering 10-20% (Extent 2), 8.7% with frequent cracking
affecting 20-40% (Extent 3), 2.1% with extensive coverage of 50-80% (Extent 4), and 1.1% with
throughout coverage exceeding 80% (Extent 5).

FIGURE I. 5. 2016 Transverse Cracking Extent pie chart and spatial distribution map

2017 Extent Analysis: Extent distribution shifted to 22.8% with no cracking (Extent 0), 31.2%
with few occurrences (Extent 1), 24.3% with intermediate extent (Extent 2), 12.5% with frequent
cracking (Extent 3), 6.8% with extensive coverage (Extent 4), and 2.4% with throughout coverage
(Extent 5). The increase in higher extent categories indicated expanding deterioration patterns.

FIGURE I. 6. 2017 Transverse Cracking Extent pie chart and spatial distribution map

84

2020 Extent Analysis: Distribution showed 21.8% with no cracking (Extent 0), 28.9% with few
occurrences (Extent 1), 26.7% with intermediate extent (Extent 2), 14.2% with frequent cracking
(Extent 3), 5.9% with extensive coverage (Extent 4), and 2.5% with throughout coverage (Extent
5).

FIGURE I. 7. 2020 Transverse Cracking Extent pie chart and spatial distribution map

2023 Extent Analysis: Improved distribution with 37.6% showing no cracking (Extent 0), 35.2%
with few occurrences (Extent 1), 19.8% with intermediate extent (Extent 2), 5.4% with frequent
cracking (Extent 3), 1.7% with extensive coverage (Extent 4), and 0.3% with throughout coverage
(Extent 5).

FIGURE I. 8. 2023 Transverse Cracking Extent pie chart and spatial distribution map

I.3. Meandering Longitudinal Cracking Severity Analysis
2016 Severity Analysis: The distress affected 31.6% of the network with severity distribution
showing 68.4% of segments with no distress (Severity 0), 19.7% exhibiting low severity conditions
(Severity 1), 10.1% displaying moderate severity (Severity 2), and 1.8% reaching high severity
85

levels (Severity 3). The relatively limited network impact suggested early-stage deterioration
patterns requiring monitoring and preventive intervention.

FIGURE I. 9. 2016 Meandering Longitudinal Cracking Severity pie chart and spatial distribution map

2017 Severity Analysis: Network coverage expanded dramatically to 76.0% with severity
distribution shifting to 24.0% with no distress (Severity 0), 6.3% at low severity (Severity 1),
50.4% at moderate severity (Severity 2), and 19.2% at high severity (Severity 3). This substantial
increase in moderate and high severity levels indicated rapid progression of longitudinal
deterioration mechanisms related to structural loading and environmental exposure.

FIGURE I. 10. 2017 Meandering Longitudinal Cracking Severity pie chart and spatial distribution map

2020 Severity Analysis: Coverage reached 74.8% with severity distribution of 25.2% with no
distress (Severity 0), 20.7% at low severity (Severity 1), 40.9% at moderate severity (Severity 2),
and 13.2% at high severity (Severity 3). The persistence of extensive moderate severity coverage
demonstrated ongoing structural challenges requiring systematic intervention strategies.

86

FIGURE I. 11. 2020 Meandering Longitudinal Cracking Severity pie chart and spatial distribution map

2023 Severity Analysis: Network impact increased to 79.5% with severity distribution of 20.5%
with no distress (Severity 0), 37.4% at low severity (Severity 1), 40.5% at moderate severity
(Severity 2), and 1.6% at high severity (Severity 3). While overall coverage expanded, the
significant reduction in high-severity occurrences indicates effective management of critical
deterioration through targeted maintenance.

FIGURE I. 12. 2023 Meandering Longitudinal Cracking Severity pie chart and spatial distribution map

I.4. Meandering Longitudinal Cracking Extent Analysis
2016 Extent Analysis: Extent distribution demonstrated 68.4% of segments with no meandering
longitudinal cracking (Extent 0), 18.9% with few occurrences (Extent 1), 8.7% with intermediate
extent (Extent 2), 2.8% with frequent cracking (Extent 3), 0.9% with extensive coverage (Extent
4), and 0.3% with throughout coverage (Extent 5).

87

FIGURE I. 13. 2016 Meandering Longitudinal Cracking Extent pie chart and spatial distribution map

2017 Extent Analysis: Distribution shifted to 24.0% with no cracking (Extent 0), 22.3% with few
occurrences (Extent 1), 28.9% with intermediate extent (Extent 2), 16.2% with frequent cracking
(Extent 3), 6.1% with extensive coverage (Extent 4), and 2.5% with throughout coverage (Extent
5).

FIGURE I. 14. 2017 Meandering Longitudinal Cracking Extent pie chart and spatial distribution map

2020 Extent Analysis: Pattern showed 25.2% with no cracking (Extent 0), 26.4% with few
occurrences (Extent 1), 25.8% with intermediate extent (Extent 2), 14.7% with frequent cracking
(Extent 3), 5.9% with extensive coverage (Extent 4), and 2.0% with throughout coverage (Extent
5).

88

FIGURE I. 15. 2020 Meandering Longitudinal Cracking Extent pie chart and spatial distribution map

2023 Extent Analysis: Distribution indicated 20.5% with no cracking (Extent 0), 31.2% with few
occurrences (Extent 1), 26.8% with intermediate extent (Extent 2), 15.3% with frequent cracking
(Extent 3), 4.7% with extensive coverage (Extent 4), and 1.5% with throughout coverage (Extent
5).

FIGURE I. 16. 2023 Meandering Longitudinal Cracking Extent pie chart and spatial distribution map

I.5. Longitudinal Wheel Path Cracking Severity Analysis
2016 Severity Analysis: The distress affected substantial portions of the network with severity
distribution showing moderate impact levels primarily concentrated in low to moderate severity
categories. Traffic-related deterioration patterns established baseline conditions requiring ongoing
monitoring of heavy vehicle corridors.

89

FIGURE I. 17. 2016 Longitudinal Wheel Path Cracking Severity pie chart and spatial distribution map

2017 Severity Analysis: Severity distribution remained relatively stable with slight variations in
category distribution. The consistency in severity patterns suggested effective management of
traffic-load-related deterioration factors through appropriate maintenance timing and techniques.

FIGURE I. 18. 2017 Longitudinal Wheel Path Cracking Severity pie chart and spatial distribution map

2020 Severity Analysis: Network impact demonstrated continued stability with severity
distribution maintaining manageable levels across all categories. The persistence of stable patterns
indicated successful long-term management strategies for traffic-related longitudinal cracking.

90

FIGURE I. 19. 2020 Longitudinal Wheel Path Cracking Severity pie chart and spatial distribution map

2023 Severity Analysis: Coverage reached stable levels with severity distribution demonstrating
effective control of traffic-related deterioration progression. The maintenance of consistent
severity patterns throughout the analysis period reflects appropriate intervention strategies.

FIGURE I. 20. 2023 Longitudinal Wheel Path Cracking Severity pie chart and spatial distribution map

I.6. Longitudinal Wheel Path Cracking Extent Analysis
2016 Extent Analysis: Extent distribution demonstrated widespread but manageable coverage
patterns with concentration in lower extent categories. The distribution pattern indicated
appropriate traffic management and pavement design adequacy for anticipated loading conditions.

91

FIGURE I. 21. 2016 Longitudinal Wheel Path Cracking Extent pie chart and spatial distribution map

2017 Extent Analysis: Slight reduction in overall extent coverage suggested targeted maintenance
effectiveness in high-traffic corridors most susceptible to wheel path deterioration. The
improvement demonstrated successful intervention timing and technique selection.

FIGURE I. 22. 2017 Longitudinal Wheel Path Cracking Extent pie chart and spatial distribution map

2020 Extent Analysis: Network impact increased moderately, returning extent coverage to levels
comparable with baseline conditions. This recovery pattern indicated renewed deterioration
pressure requiring continued monitoring and intervention planning.

92

FIGURE I. 23. 2020 Longitudinal Wheel Path Cracking Extent pie chart and spatial distribution map

2023 Extent Analysis: Distribution stabilized at manageable levels with extent patterns
demonstrating effective long-term management of traffic-related longitudinal cracking. The
stability throughout the analysis period reflects appropriate maintenance strategies.

FIGURE I. 24. 2023 Longitudinal Wheel Path Cracking Extent pie chart and spatial distribution map

I.7. Alligator Cracking Severity Analysis
2016 Severity Analysis: The distress represented localized structural deterioration with severity
distribution concentrated in lower categories but indicating areas requiring immediate structural
attention. The baseline conditions established priority areas for rehabilitation intervention.

93

FIGURE I. 25. 2016 Alligator Cracking Severity pie chart and spatial distribution map

2017 Severity Analysis: Peak severity conditions were observed with increased high-severity
occurrences, particularly affecting older local roads and areas with inadequate structural capacity
for current loading conditions. The deterioration peak indicated critical timing for comprehensive
rehabilitation strategies.

FIGURE I. 26. 2017 Alligator Cracking Severity pie chart and spatial distribution map

2020 Severity Analysis: Substantial improvement in severity distribution demonstrated effective
structural rehabilitation interventions. The reduction in high-severity occurrences reflected
successful targeting of critical structural deficiencies through systematic maintenance programs.

94

FIGURE I. 27. 2020 Alligator Cracking Severity pie chart and spatial distribution map

2023 Severity Analysis: Near-elimination of alligator cracking across all severity categories
represents highly successful structural rehabilitation achievement. The dramatic reduction to
minimal network presence indicates comprehensive resolution of structural adequacy issues.

FIGURE I. 28. 2023 Alligator Cracking Severity pie chart and spatial distribution map

I.8. Alligator Cracking Extent Analysis
2016 Extent Analysis: Extent distribution demonstrated localized coverage patterns concentrated
in specific network areas with structural inadequacy. The distribution indicated targeted
intervention requirements rather than widespread network rehabilitation needs.

95

FIGURE I. 29. 2016 Alligator Cracking Extent pie chart and spatial distribution map

2017 Extent Analysis: Increased extent coverage correlated with severity peak conditions,
indicating expansion of structural deterioration requiring comprehensive rehabilitation strategies.
The extent patterns provided critical guidance for maintenance program prioritization.

FIGURE I. 30. 2017 Alligator Cracking Extent pie chart and spatial distribution map

2020 Extent Analysis: Dramatic reduction in extent coverage demonstrated effective structural
rehabilitation program implementation. The systematic reduction across all extent categories
reflected comprehensive addressing of structural deficiencies.

96

FIGURE I. 31. 2020 Alligator Cracking Extent pie chart and spatial distribution map

2023 Extent Analysis: Near-complete elimination of alligator cracking extent represents
successful resolution of structural deterioration challenges. The minimal remaining coverage
indicates highly effective maintenance program achievement.

FIGURE I. 32. 2023 Alligator Cracking Extent pie chart and spatial distribution map

I.9. Rutting Severity Analysis
2016 Severity Analysis: No rutting was observed throughout the surveyed network, with 100%
of segments showing no distress (Severity 0). This baseline condition of complete absence
established the foundation for monitoring permanent deformation development over subsequent
survey periods.

97

FIGURE I. 33. 2016 Rutting Severity pie chart (showing 100% no distress)

2017 Severity Analysis: Initial rutting manifestation appeared with severity distribution showing
44.3% of segments with no distress (Severity 0), 52.0% exhibiting low severity conditions
(Severity 1), 3.4% displaying moderate severity (Severity 2), and 0.3% reaching high severity
levels (Severity 3). The emergence of rutting across 55.7% of the network indicated systematic
development of permanent deformation patterns.

FIGURE I. 34. 2017 Rutting Severity pie chart and spatial distribution map

2020 Severity Analysis: Rutting presence expanded with severity distribution of 45.2% with no
distress (Severity 0), 52.6% at low severity (Severity 1), 2.2% at moderate severity (Severity 2),
and 0.1% at high severity (Severity 3). The persistence of widespread low severity rutting
demonstrated progressive development of permanent deformation requiring intervention planning.

98

FIGURE I. 35. 2020 Rutting Severity pie chart and spatial distribution map

2023 Severity Analysis: Dramatic expansion reached near-universal network presence with
severity distribution of 1.6% with no distress (Severity 0), 92.9% at low severity (Severity 1), 5.1%
at moderate severity (Severity 2), and 0.3% at high severity (Severity 3). The 98.4% network
coverage represents the most significant emerging distress challenge requiring immediate strategic
intervention.

FIGURE I. 36. 2023 Rutting Severity pie chart and spatial distribution map

I.10. Rutting Extent Analysis
2016 Extent Analysis: Complete absence of rutting resulted in 100% of segments showing no
extent coverage (Extent 0), establishing baseline conditions free from permanent deformation
patterns.

99

FIGURE I. 37. 2016 Rutting Extent pie chart (showing 100% no distress)

2017 Extent Analysis: Initial extent distribution showed 44.3% with no rutting (Extent 0), 38.2%
with few occurrences (Extent 1), 12.8% with intermediate extent (Extent 2), 3.7% with frequent
rutting (Extent 3), 0.8% with extensive coverage (Extent 4), and 0.2% with throughout coverage
(Extent 5).

FIGURE I. 38. 2017 Rutting Extent pie chart and spatial distribution map

2020 Extent Analysis: Extent pattern demonstrated 45.2% with no rutting (Extent 0), 41.6% with
few occurrences (Extent 1), 10.4% with intermediate extent (Extent 2), 2.3% with frequent rutting
(Extent 3), 0.4% with extensive coverage (Extent 4), and 0.1% with throughout coverage (Extent
5).

100

FIGURE I. 39. 2020 Rutting Extent pie chart and spatial distribution map

2023 Extent Analysis: Extensive coverage emerged with 1.6% showing no rutting (Extent 0),
12.96% with few occurrences (Extent 1), 33.06% with intermediate extent (Extent 2), 28.17% with
frequent rutting (Extent 3), 15.42% with extensive coverage (Extent 4), and 8.73% with throughout
coverage (Extent 5). This distribution indicates systematic permanent deformation development
requiring comprehensive intervention strategies.

FIGURE I. 40. 2023 Rutting Extent pie chart and spatial distribution map

I.11. Pavement Edge Cracking Severity Analysis
2016 Severity Analysis: The distress established baseline edge deterioration patterns with severity
distribution indicating moderate network impact levels. Edge cracking severity patterns suggested
localized deterioration related to drainage, frost action, and traffic loading effects on pavement
boundaries.

101

FIGURE I. 41. 2016 Pavement Edge Cracking Severity pie chart and spatial distribution map

2017 Severity Analysis: Severity distribution maintained moderate levels with slight variations in
category distribution. The consistency suggested stable edge deterioration patterns under normal
environmental and traffic loading conditions.

FIGURE I. 42. 2017 Pavement Edge Cracking Severity pie chart and spatial distribution map

2020 Severity Analysis: Edge cracking data was not collected during this comprehensive survey
period, creating a significant information gap in severity trend analysis. This data collection
limitation prevents assessment of severity progression during the critical 2017-2023 period.

102

FIGURE I. 43. 2020 Pavement Edge Cracking - No data collected notation

2023 Severity Analysis: Dramatic emergence of severe edge cracking across all severity
categories resulted in the most severe and extensive distress type in 2023. The severity distribution
indicated widespread edge structural integrity challenges requiring immediate intervention
development.

FIGURE I. 44. 2023 Pavement Edge Cracking Severity pie chart and spatial distribution map

I.12. Pavement Edge Cracking Extent Analysis
2016 Extent Analysis: Extent distribution showed 81.1% of segments with no edge cracking
(Extent 0), 7.0% with few occurrences (Extent 1), 5.2% with intermediate extent (Extent 2), 4.2%
with frequent cracking (Extent 3), 1.3% with extensive coverage (Extent 4), and 1.2% with
throughout coverage (Extent 5). The limited extent coverage indicated manageable edge
deterioration conditions.

103

FIGURE I. 45. 2016 Pavement Edge Cracking Extent pie chart and spatial distribution map

2017 Extent Analysis: Distribution shifted to 71.6% with no cracking (Extent 0), 18.2% with few
occurrences (Extent 1), 6.4% with intermediate extent (Extent 2), 3.3% with frequent cracking
(Extent 3), 0.4% with extensive coverage (Extent 4), and 0.1% with throughout coverage (Extent
5). The increase in lower extent categories suggested expanding but manageable edge
deterioration.

FIGURE I. 46. 2017 Pavement Edge Cracking Extent pie chart and spatial distribution map

2020 Extent Analysis: Edge cracking extent data was not collected, preventing trend analysis
during this critical period. The data gap limits understanding of extent progression patterns
between 2017 and 2023.

104

FIGURE I. 47. 2020 Pavement Edge Cracking - No data collected notation

2023 Extent Analysis: Dramatic transformation resulted in 14.6% with no cracking (Extent 0),
42.7% with few occurrences (Extent 1), 15.5% with intermediate extent (Extent 2), 17.2% with
frequent cracking (Extent 3), 6.2% with extensive coverage (Extent 4), and 3.9% with throughout
coverage (Extent 5). The 85.4% network coverage represents the most extensive single distress
challenge.

FIGURE I. 48. 2023 Pavement Edge Cracking Extent pie chart and spatial distribution map

I.13. Longitudinal Cracking Severity Analysis
2016 Severity Analysis: The distress established significant baseline longitudinal deterioration
with severity distribution showing 70.0% of segments with no distress (Severity 0), 22.3% at low
severity (Severity 1), 6.1% at moderate severity (Severity 2), and 1.6% at high severity (Severity
3). The moderate network impact indicated manageable structural deterioration requiring
preventive maintenance strategies.
105

FIGURE I. 49. 2016 Longitudinal Cracking Severity pie chart and spatial distribution map

2017 Severity Analysis: Severity distribution improved with targeted maintenance effectiveness,
demonstrating reduced high-severity occurrences. The improvement suggested successful
intervention timing and technique selection for longitudinal cracking management.

FIGURE I. 50. 2017 Longitudinal Cracking Severity pie chart and spatial distribution map

2020 Severity Analysis: Slight increase in network coverage demonstrated stable management of
longitudinal deterioration patterns with severity distribution maintaining manageable levels across
all categories. The stability indicated appropriate maintenance strategy implementation.

106

FIGURE I. 51. 2020 Longitudinal Cracking Severity pie chart and spatial distribution map

2023 Severity Analysis: Coverage remained consistent with effective maintenance strategies
maintaining stable deterioration patterns. The continued stability throughout the analysis period
reflects successful long-term management of longitudinal cracking challenges.

FIGURE I. 52. 2023 Longitudinal Cracking Severity pie chart and spatial distribution map

I.14. Longitudinal Cracking Extent Analysis
2016 Extent Analysis: Extent distribution demonstrated 70.0% with no cracking (Extent 0), 19.4%
with few occurrences (Extent 1), 7.8% with intermediate extent (Extent 2), 2.1% with frequent
cracking (Extent 3), 0.5% with extensive coverage (Extent 4), and 0.2% with throughout coverage
(Extent 5).

107

FIGURE I. 53. 2016 Longitudinal Cracking Extent pie chart and spatial distribution map

2017 Extent Analysis: Improved distribution indicated effective maintenance intervention with
reduced extent coverage across higher categories. The improvement demonstrated successful
targeting of longitudinal cracking progression.

FIGURE I. 54. 2017 Longitudinal Cracking Extent pie chart and spatial distribution map

2020 Extent Analysis: Stable extent patterns maintained manageable coverage levels with
consistent distribution across all categories. The stability reflected appropriate maintenance timing
and intervention strategies.

108

FIGURE I. 55. 2020 Longitudinal Cracking Extent pie chart and spatial distribution map

2023 Extent Analysis: Continued stability in extent distribution demonstrated effective long-term
management with consistent network impact levels throughout the analysis period.

FIGURE I. 56. 2023 Longitudinal Cracking Extent pie chart and spatial distribution map

I.15. Pothole Severity Analysis
2016 Severity Analysis: Potholes represented localized surface failure with severity distribution
concentrated in lower categories but indicating areas requiring immediate safety attention. The
baseline conditions established emergency response requirements for surface failure management.

109

FIGURE I. 57. 2016 Pothole Severity pie chart and spatial distribution map

2017 Severity Analysis: Peak severity conditions were observed with increased occurrence rates
representing elevated surface failure requiring enhanced maintenance response. The severity peak
indicated critical timing for improved reactive maintenance strategies.

FIGURE I. 58. 2017 Pothole Severity pie chart and spatial distribution map

2020 Severity Analysis: Substantial improvement in severity distribution demonstrated effective
reactive maintenance strategies for surface failure management. The reduction reflected successful
emergency response protocol implementation.

110

FIGURE I. 59. 2020 Pothole Severity pie chart and spatial distribution map

2023 Severity Analysis: Near-elimination of pothole occurrences across all severity categories
represents highly successful surface maintenance achievement. The dramatic reduction indicates
comprehensive resolution of surface failure challenges.

FIGURE I. 60. 2023 Pothole Severity pie chart and spatial distribution map

I.16. Pothole Extent Analysis
2016 Extent Analysis: Extent distribution demonstrated limited coverage patterns with
concentration in specific network areas prone to surface failure. The localized distribution
indicated targeted intervention requirements for safety management.

111

FIGURE I. 61. 2016 Pothole Extent pie chart and spatial distribution map

2017 Extent Analysis: Increased extent coverage correlated with severity peak conditions,
indicating expansion of surface failure requiring comprehensive response strategies. The extent
patterns provided guidance for emergency response program development.

FIGURE I. 62. 2017 Pothole Extent pie chart and spatial distribution map

2020 Extent Analysis: Substantial reduction in extent coverage demonstrated effective reactive
maintenance program implementation. The systematic reduction reflected comprehensive
addressing of surface failure challenges.

112

FIGURE I. 63. 2020 Pothole Extent pie chart and spatial distribution map

2023 Extent Analysis: Near-complete elimination of pothole extent represents successful
resolution of surface failure challenges with minimal remaining coverage indicating highly
effective maintenance program achievement.

FIGURE I. 64. 2023 Pothole Extent pie chart and spatial distribution map

113

Appendix II. Data Extraction and Analysis
II.1.1. Dataset Compilation and Temporal Coverage
The research utilizes comprehensive pavement condition datasets collected through four
systematic surveys spanning seven years. The temporal distribution provides sufficient
observation points for deterioration trend analysis and predictive model development while
capturing the effects of varying climate conditions and maintenance interventions.
Primary Datasets:
•
•
•
•

2016 Survey: 9,328 data points focusing on arterial and collector roads
2017 Survey: 12,475 data points including local roads and alleys
2020 Survey: 31,084 data points representing comprehensive network coverage
2023 Survey: 8,942 data points targeting arterial and collector roads

FIGURE II. 1. Sample of Data Tables

Each dataset includes detailed pavement distress characteristics, severity and extent
measurements, spatial referencing information, and calculated performance indices following the
standardized methodologies outlined in Chapter 3.
II.1.2. Data Structure and Variables
The consolidated dataset incorporates multiple variable categories essential for comprehensive
pavement performance analysis:
Distress Variables (Severity and Extent):
•
•

Potholes, Rutting, Alligator Cracking
Longitudinal Wheelpath Cracking, Meandering Longitudinal Cracking
114

•

Longitudinal Cracking, Pavement Edge Cracking, Transverse Cracking

Performance Indices:
•
•
•

Pavement Distress Index (PDI)
International Roughness Index (IRI)
Average Rutting Depth

Climate Variables:
•
•
•

Daily maximum, minimum, and mean temperatures
Total precipitation (rain and snow)
Freeze-thaw cycle indicators derived from temperature data

Traffic Variables:
•
•
•

Annual Average Daily Traffic (AADT) volumes
Vehicle classification data where available
Seasonal traffic variation patterns

II.1.3. Data Integration Methodology
Climate data spanning the survey periods was obtained from Environment and Climate Change
Canada's Prince George weather station, providing daily meteorological records. Traffic volume
data was compiled from municipal annual traffic reports for 2016, 2017, and 2023, supplemented
by continuous monitoring data from automated counting stations.
The integration process required temporal alignment of datasets collected at different intervals and
spatial scales. Pavement condition data collected in standardized 50-meter segments was matched
with corresponding climate and traffic information through spatial and temporal indexing
procedures.
II.2. Tools and Software Used
II.2.1. Statistical Computing Environment
RStudio served as the primary analytical platform, providing an integrated development
environment for statistical analysis, modeling, and visualization. The R programming environment
was selected for its comprehensive statistical modeling libraries, advanced machine learning
packages, and sophisticated data visualization capabilities.
Key advantages of the RStudio environment include:
•
•
•

Extensive statistical modeling libraries
Advanced machine learning packages (randomForest, nnet)
Sophisticated data visualization capabilities (ggplot2, plotly)
115

•
•

Reproducible research support through integrated workflows
Open-source accessibility ensuring long-term sustainability

II.2.2. R Package Ecosystem
The analysis employed a comprehensive suite of R packages; each selected for specific analytical
capabilities:
Data Manipulation and Processing:
•
•
•

readxl: Excel file import and processing for municipal datasets
dplyr: Data manipulation, filtering, and transformation operations
tidyverse: Comprehensive data science workflow integration

Machine Learning and Statistical Modeling:
•

randomForest:

Random Forest algorithm implementation with variable importance

•
•

analysis
nnet: Neural network modeling capabilities for non-linear pattern recognition
stats: Base statistical functions for linear regression and diagnostic procedures

Visualization and Graphics:
•
•

ggplot2: Advanced statistical graphics and publication-quality plots
plotly: Interactive visualization capabilities for data exploration

II.2.3. Complementary Software Tools
ArcGIS was utilized for spatial analysis, mapping, and geographic data management. The GIS
platform enabled spatial visualization of pavement condition data, network-level analysis, and
integration with municipal infrastructure databases.
Microsoft Excel was employed for preliminary data aggregation, formatting, and quality
assurance procedures. Excel facilitated collaboration with municipal stakeholders while providing
accessible data validation capabilities.
All statistical analyses and modeling procedures were conducted exclusively within the R/RStudio
environment to maintain consistency and reproducibility across all analytical components.
II.3. Modeling Approach
II.3.1. Multi-Model Framework
The research employed three distinct analytical techniques to provide comprehensive assessment
of predictive capabilities while ensuring robust validation across different algorithmic approaches.
116

This multi-model strategy addresses varying assumptions about data relationships and provides
comparative evaluation of modeling effectiveness.
II.3.2. Multiple Linear Regression (MLR)
Multiple Linear Regression was implemented as the baseline modeling approach to establish
fundamental linear relationships between pavement deterioration and contributing factors. MLR
provides interpretable coefficients and serves as a benchmark for comparison with advanced
modeling techniques.
Implementation:
# Multiple Linear Regression Implementation
model_lm <- lm(PDI ~ Alligator_Severity + Alligator_Extent +
Transverse_Severity + Transverse_Extent +
Rutting_Severity + Rutting_Extent +
Potholes_Severity + Potholes_Extent +
Max_Temp + Min_Temp + Mean_Temp + Precip +
Traffic + RUT_AVG_mm + IRI_AVG_mm_m,
data = train_data)

The MLR approach assumes linear relationships between predictors and response variables,
providing straightforward interpretation of factor contributions. Model diagnostics include
residual analysis, normality testing, and multicollinearity assessment to ensure statistical validity.
II.3.3. Random Forest Ensemble Learning
Random Forest was selected as the primary machine learning approach due to its robust
performance with complex, non-linear relationships and superior handling of mixed data types
characteristic of pavement management applications. The ensemble learning approach combines
multiple decision trees to improve prediction accuracy and reduce overfitting.
Implementation:
# Random Forest Implementation
library(randomForest)
# Data preprocessing and feature selection
selected_data <- data_cleaned %>%
select(Potholes_Severity, Potholes_Extent,
Rutting_Severity, Rutting_Extent,
Alligator_Severity, Alligator_Extent,
Longitudinal_Severity, Longitudinal_Extent,
Meandering_Severity, Meandering_Extent,
Transverse_Severity, Transverse_Extent,
PavementEdge_Severity, PavementEdge_Extent,
Max_Temp, Min_Temp, Mean_Temp, Precip,
Traffic, RUT_AVG_mm, IRI_AVG_mm_m, PDI) %>%
na.omit()
# Random Forest model training

117

rf_model <- randomForest(PDI ~ ., data = train_data,
ntree = 2000, importance = TRUE)

The Random Forest implementation incorporated 2000 trees to ensure model stability and
employed variable importance assessment to identify key performance drivers. The ensemble
approach provides superior handling of non-linear relationships and interaction effects.
II.3.4. Artificial Neural Network
Artificial Neural Networks were implemented using the nnet package to capture complex nonlinear patterns and interactions in pavement deterioration data. The neural network approach
provides sophisticated pattern recognition capabilities for modeling complex pavement
performance relationships.
Implementation:
# Neural Network Implementation
library(nnet)
# Data normalization for neural network training
features <- setdiff(names(selected_data), "PDI")
ml_data <- selected_data %>%
select(all_of(c(features, "PDI")))
# Normalize data to [0,1] range
maxs <- apply(ml_data, 2, max)
mins <- apply(ml_data, 2, min)
scaled_data <- as.data.frame(scale(ml_data, center = mins,
scale = maxs - mins))
# Neural network training
nn_model <- nnet(PDI ~ ., data = train_data_scaled,
size = 6, linout = TRUE, maxit = 500)

The neural network architecture employed 6 hidden neurons with linear output activation,
optimized through iterative testing to balance model complexity with training efficiency. Data
normalization ensures proper weight initialization and convergence behavior.
II.3.5. Data Preprocessing Procedures
Comprehensive data preprocessing ensured analytical reliability and model validity across all
approaches:
Missing Data Treatment:
# Missing value imputation using column means
selected_data <- selected_data %>%
mutate(across(everything(), ~ ifelse(is.na(.), mean(., na.rm = TRUE), .)))

Data Cleaning and Standardization:
118

# Convert comma-separated numbers to numeric format
data_cleaned <- data %>%
mutate(across(everything(), ~ as.numeric(gsub(",", "", as.character(.)))))
# Column name standardization
colnames(selected_data) <- colnames(selected_data) %>%
gsub(" ", "_", .) %>%
gsub("\\(", "", .) %>%
gsub("\\)", "", .) %>%
gsub("-", "_", .) %>%
gsub("/", "_", .)

Feature Engineering: Distress severity and extent variables were processed separately to capture
both the intensity and spatial distribution of pavement deterioration. Climate variables were
integrated to enable assessment of environmental impacts on pavement performance.
II.3.6. Model Training and Validation Framework
The research employed an 80/20 train-test split methodology with fixed random seed (123) to
ensure reproducible results across all modeling approaches. This split ratio provides sufficient
training data while maintaining adequate test samples for robust validation.
# Standardized train-test split
set.seed(123)
train_index <- sample(1:nrow(selected_data), 0.8 * nrow(selected_data))
train_data <- selected_data[train_index, ]
test_data <- selected_data[-train_index, ]

The consistent validation framework enables direct comparison of model performance across
different algorithmic approaches while ensuring statistical rigor in model assessment.
II.4. Evaluation Criteria
II.4.1. Performance Metrics Framework
Model performance was assessed using a comprehensive suite of statistical metrics that quantify
both accuracy and reliability of predictive capabilities. The evaluation framework incorporates
standard regression metrics while considering the specific requirements of pavement management
applications.
Primary Performance Metrics:
Mean Squared Error (MSE):
# MSE calculation
predicted <- predict(rf_model, test_data)
actual <- test_data$PDI
mse <- mean((predicted - actual)^2)

119

MSE quantifies the average squared difference between predicted and actual values, providing a
measure of prediction accuracy that penalizes larger errors more heavily. This characteristic is
valuable in pavement management where significant prediction errors can lead to inappropriate
maintenance decisions.
Root Mean Squared Error (RMSE):
# RMSE calculation
rmse <- sqrt(mse)

RMSE provides error measurement in the same units as the target variable, enabling intuitive
interpretation of prediction accuracy and facilitating comparison across different applications.
Coefficient of Determination (R²):
# R² calculation
rss <- sum((predicted - actual)^2)
tss <- sum((actual - mean(actual))^2)
r2 <- 1 - rss/tss

R² quantifies the proportion of variance in the dependent variable explained by the model,
providing a standardized measure of model effectiveness ranging from 0 to 1.
# Adjusted R² calculation
n <- nrow(test_data)
p <- length(coef(model_lm)) - 1 # number of predictors
adj_r2 <- 1 - (1 - r2) * ((n - 1) / (n - p - 1))

For Multiple Linear Regression models, adjusted R² was calculated to account for model
complexity and the number of predictor variables. Adjusted R² adjusts the coefficient of
determination based on the number of predictors relative to the sample size, providing a more
conservative measure that penalizes the inclusion of non-significant variables. This metric is
particularly important when comparing models with different numbers of predictors, as it prevents
artificial inflation of R² values through the addition of unnecessary variables. The adjusted R²
formula incorporates degrees of freedom correction, ensuring that reported model performance
reflects true predictive capability rather than overfitting to the training data.
II.4.2. Model Comparison Framework
Comparative evaluation across modeling approaches employed consistent metrics and validation
procedures:
# Standardized evaluation function
evaluate_model <- function(predictions, actuals) {
mse <- mean((predictions - actuals)^2)
rmse <- sqrt(mse)
rss <- sum((predictions - actuals)^2)
tss <- sum((actuals - mean(actuals))^2)
r2 <- 1 - rss/tss

120

# Calculate adjusted R² for linear models
if (!is.null(model) && inherits(model, "lm")) {
n <- length(actuals)
p <- length(coef(model)) – 1
adj_r2 <- 1 - (1 - r2) * ((n - 1) / (n - p - 1))
return(list(MSE = round(mse, 2),
RMSE = round(rmse, 2),
R2 = round(r2, 4),
Adj_R2 = round(adj_r2, 4)))

This standardized evaluation framework ensures consistent performance assessment across
Multiple Linear Regression, Random Forest, and Neural Network approaches. For MLR, adjusted
R² is calculated to provide a more rigorous assessment that accounts for model complexity, while
standard R² is reported for machine learning approaches where adjusted R² is not applicable.
II.4.3. Variable Importance Analysis
Random Forest models provided variable importance metrics to identify key factors driving
pavement deterioration:
# Variable importance assessment
importance_df <- as.data.frame(importance(rf_model))
importance_df$Variable <- rownames(importance_df)
# Visualization framework for variable importance
ggplot(importance_df, aes(x = reorder(Variable, `%IncMSE`),
y = `%IncMSE`)) +
geom_bar(stat = "identity", fill = "steelblue") +
coord_flip() +
labs(title = "Variable Importance Assessment",
x = "Variable", y = "% Increase in MSE") +
theme_minimal()

II.4.4. Predictive Capability Framework
Future condition forecasting capabilities were evaluated through systematic application of trained
models:
# Future prediction implementation
latest_input <- selected_data %>%
tail(1) %>%
select(-PDI)
future_prediction <- predict(rf_model, newdata = latest_input)

The predictive framework enables systematic evaluation of model forecasting capabilities essential
for proactive pavement management applications.
II.4.5. Model Validation and Robustness Assessment

121

Model robustness was assessed through systematic evaluation of prediction accuracy across
different pavement condition ranges and validation of statistical assumptions. Residual analysis
procedures confirmed appropriate model behavior without systematic bias or heteroscedasticity
issues.
The comprehensive evaluation framework provides the foundation for systematic comparison of
modeling approaches and selection of optimal predictive techniques for pavement management
applications under northern climate conditions.

122