The Impacts of Climate and Climate Change on Aviation in the Canadian North

Andrew Chi Wai Leung

A thesis submitted in conformity with the requirements for the degree of Doctor of Philosophy Department of Physical and Environmental Sciences University of Toronto

Andrew Chi Wai Leung

Doctor of Philosophy

Department of Physical and Environmental Science University of Toronto

2019 Abstract

Aviation is inherently linked to meteorology as severe weather is often responsible for flight delays, cancellations and sometimes accidents. Climate change is expected to change the Arctic environment and the warming rate in this region is greater than most locations on Earth. With a changing climate, the risks of flying will also be changing. In Canada, many Arctic communities in Hudson Bay, Nunavik in northern Quebec and western Labrador rely heavily on aviation to transport passengers, mail and groceries because they lack road networks or railway to access larger settlements and shipping is limited to brief periods in summer. Using historical hourly and daily climate data, this thesis examines four topics related to flying: 1) wind pattern changes

(1971 to 2010) at seven locations around Hudson Bay, northern Quebec and western Labrador; 2) fog and visibility trends at 16 Hudson Bay communities (1953-2014); 3) historic long-term soil temperature trends at 5 to 150 cm depths and future projections under three greenhouse gas concentration trajectories at Kuujjuaq, Quebec; 4) appearance and climate conditions for frostquakes. The results of these topics are: 1) an increase in hourly average and daily maximum wind speed around Hudson Bay region and declining trends in western Labrador, plus prevailing wind direction changed at two communities; 2) fog and ice fog frequencies declined but reduced

ii and low visibility trends varied spatially within the Hudson Bay region; 3) soil warming at approximately 1oC per decade from 1967 to 1995 and future soil temperature will be above 0oC under all projected trajectories at Kuujjuaq; 4) identified that water-saturated soil, minimal snow cover and rapid temperature drop to below freezing causes frostquakes and that observations were somewhat dependent on the density of the observational network. Passengers travelling to and from Hudson Bay will benefit from the results of this research to better understand the risks associated with flying to these communities. Pilots, airport operators and airlines will improve their awareness of this issue and increase their understandings of the risks caused by climate change in this area. Improved safety will be achieved by anticipating, adapting to and mitigating these changes.

iii

Acknowledgements

I would like to thank my supervisor, Professor William Gough, for allowing me to combine my pursuit for research and my passion in aviation. I would like to express my gratitude for his unwavering support and guidance in my research.

I would also like to give my thanks to the Ph.D committee members: Professor George

Arhonditsis, Professor Yuhong He and Dr. Tanzina Mohsin. Their input and suggestions helped with the development of this thesis. In addition, I would like to recognize Prof. Ken Butler for his assistance with programming and statistics. I am grateful for Dr. Angela Masson who took the time to provide expert knowledge in the aviation field and comments from a pilot’s perspective. I would like to recognize my colleagues in the Archive and Data Services in

Environment and Climate Change Data for assisting with the acquisition and quality control of various climate data which form the backbone of this thesis. Also, I would like to acknowledge all the individuals who offered help and assistance at some point during my doctoral research.

Finally, I would like to thank my parents for supporting my graduate journey along the way.

Statement of Co-authorship

At the time of the thesis submission, Chapters 2 to 4 are research manuscripts that are currently in revision for publication in peer-reviewed journals. Chapter 5 was published as a peer-reviewed book chapter in Citizen Empowered Mapping by Springer Nature in 2017. For each chapter, they serve as standalone research projects. In each chapter, I was the lead author responsible for identifying the research objective, collecting data, conducting statistical analysis and writing the manuscript. In Chapters 2 to 5, Prof. William Gough supervised the research and provided helpful suggestions to explain key research findings. Prof. Ken Butler provided assistance with data processing, statistical methods and interpretation of the results in Chapters 2 and 3. Prof. Tanzina Mohsin provided feedback and suggestions for the manuscript in Chapter 2 and assisted with the statistical downscale modelling in Chapter 4. Mr. Yehong Shi produced the

GIS maps and identified the clusters of observations in Chapter 5.

Table of Contents

Acknowledgements ...... iv

Statement of Co-authorship ...... v

Table of Contents ...... vi

List of Tables ...... xi

List of Figures ...... xiii

Chapter 1: Introduction ...... 1

1.1 Importance of Aviation Climatology Research ...... 1

1.2 Wind Study ...... 5

1.3 Fog and Low Visibility ...... 7

1.4 Climate Change Impact Assessment on Soil Temperature ...... 9

1.5 Frostquakes...... 11

1.6 Research Objectives ...... 12

1.7 References ...... 12

Chapter 2: Wind Analysis ...... 17

2.1 Abstract ...... 17

2.2 Introduction ...... 18

2.3 Methods ...... 21

2.3.1 Data Collection ...... 21

2.3.2 Wind Speed ...... 23

2.3.2.1 Calm Wind ...... 25

2.3.3 Wind Direction...... 25

2.3.4 Aviation Perspective ...... 26

2.3.5 Uncertainties and Assumptions...... 26

2.4 Results ...... 28

2.4.1 Average Daily Wind Speed ...... 28

2.4.2 Maximum Daily Wind Speed ...... 32

2.4.3 Time of Daily Maximum Wind Speed...... 37

2.4.4 Calm Wind ...... 40

2.4.5 Wind Direction...... 44

2.4.6 Aviation Perspective ...... 54

2.5 Discussion ...... 57

2.5.1 Average Wind Speed ...... 57

2.5.2 Maximum Wind Speed ...... 62

2.5.3 Calm Wind ...... 66

2.5.4 Wind Direction...... 67

2.5.5 Aviation Perspective ...... 70

2.6 Conclusion ...... 74

2.7 Acknowledgement ...... 76

2.8 References ...... 76

Chapter 3: Fog and Low Visibility ...... 83

3.1 Abstract ...... 83

3.2 Introduction ...... 84

3.3 Methods ...... 87

3.3.1 Weather Conditions ...... 87

3.3.2 Data Collection ...... 89

3.4 Results ...... 93

3.4.1 Fog ...... 93

3.4.2 Ice Fog ...... 99

3.4.3 Reduced and Low Visibilities ...... 105

vii

3.5 Discussion ...... 118

3.5.1 Fog and Ice Fog ...... 118

3.5.1.1 Uncertainties in Fog and Ice Fog Observation ...... 124

3.5.2 Visibility ...... 127

3.5.2.1 Uncertainties in Visibility Records...... 130

3.5.3 Weather Monitoring ...... 131

3.5.4 Flight Safety ...... 132

3.6 Conclusion ...... 133

3.7 Acknowledgements ...... 134

3.8 References ...... 135

Chapter 4: Soil Temperature ...... 139

4.1 Abstract ...... 139

4.2 Introduction ...... 140

4.3 Methods ...... 142

4.3.1 Site Characteristics...... 142

4.3.2 Data Collection ...... 145

4.3.3 Historical Data Analysis ...... 145

4.3.4 Statistical Downscaling ...... 146

4.3.5 Climate Projections ...... 148

4.3.5.1 Assumptions ...... 150

4.4 Results ...... 151

4.4.1 Historical Analysis ...... 151

4.4.2 Statistical Downscaling ...... 153

4.4.3 Projections...... 162

4.4.3.1 Annual ...... 162

viii

4.4.3.2 Winter ...... 166

4.5 Discussion ...... 169

4.5.1 Historical Trends and Model Fitting ...... 169

4.5.2 Projections...... 172

4.5.3 Uncertainties ...... 176

4.5.3.1 Observation Uncertainties ...... 176

4.5.3.2 Modelling and Projections Uncertainties ...... 178

4.5.4 Implications...... 181

4.5.4.1 Implications to the Environment ...... 181

4.5.4.2 Implications to Airports ...... 182

4.5.5 Long-term Monitoring Network ...... 183

4.6 Conclusion ...... 184

4.7 Acknowledgements ...... 186

4.8 References ...... 186

Chapter 5: Frostquakes ...... 192

5.1 Abstract ...... 192

5.2 Introduction ...... 193

5.3 Methods ...... 196

5.3.1 Climate Data Analysis ...... 197

5.3.2 Social Media ...... 200

5.4 Results ...... 202

5.4.1 Climate Data Analysis ...... 202

5.4.1.1 January 18, 2000 ...... 202

5.4.1.2 Winter of 2013–14 ...... 202

5.4.2 Social Media ...... 205

5.4.3 Coincidence of Frostquake Reporting and Weather Conditions ...... 210

5.5 Discussion ...... 212

5.5.1 Weather Conditions ...... 212

5.5.2 Social Media ...... 213

5.5.3 Frostquake Clusters ...... 217

5.5.4 Aviation Impacts ...... 219

5.6 Conclusion ...... 220

5.7 Acknowledgements ...... 221

5.8 References ...... 221

Chapter 6: Conclusion ...... 224

List of Tables

Table 2.1. Student’s T-test for trends in monthly averaged wind speed between 1971 to 1990 and 1991 to 2010...... 29

Table 2.2. Mann-Kendall test with Theil-Sen slope estimator for change in average daily wind speed time series in monthly and seasonal scales...... 32

Table 2.3. Student’s T-test for trends in maximum daily wind speed (≥ 28 km/h) between 1971 to 1990 and 1991 to 2010...... 33

Table 2.4. Trends for days with maximum daily wind speed ≥ 28 km/h over time...... 36

Table 2.5. Frequency of calm winds at each location...... 44

Table 2.6. Mann-Kendall test on wind direction...... 46

Table 2.7. Aircraft movements based on weekday scheduled passenger flights as of July 1, 2017...... 55

Table 2.8. Types of aircraft flown to serve the communities in the wind study...... 56

Table 2.9. Prevailing wind direction and runway configuration...... 56

Table 3.1. The length of study in the selected communities and their regional groupings...... 91

Table 3.2. Trends in fog hours per year with autocorrelation present...... 98

Table 3.3. Trends in fog hours per year with significance level adjusted for autocorrelation...... 98

Table 3.4. Trends in ice fog hours per year with autocorrelation present...... 104

Table 3.5. Trends in ice fog hours per year with significance level adjusted for autocorrelation...... 104

Table 3.6. Historical percentage of the reduced and low visibilities during study airports’ operational hours...... 114

Table 3.7. Trends for change in number of reduced and low visibility hours with significance level before and after adjusted for autocorrelation...... 115

Table 3.8. Top three causes and their proportions for reduced and low visibilities at study airports...... 116

Table 4.1. Historic average soil temperature and the rate of change over time for soil temperature at depths from 5 to 150 cm from 1967 to 1995...... 151

Table 4.2. Pearson r correlation between air temperature at 2 m and soil temperatures...... 152

Table 4.3. Pearson r correlation between the four selected variables and observed soil temperature...... 155

Table 4.4. Modelling Efficiency calculation for the CanESM2 model output...... 162

Table 5.1. Breakdown of date ranges with frostquake and the number of reports within each range...... 210

Table 5.2. Daily maximum, minimum and the difference between the maximum and minimum temperatures for the Toronto City weather station...... 211

xii

List of Figures

Figure 2.1. Weather stations selected to analyze wind conditions in the Hudson Bay region...... 23

Figure 2.2. Hour of the day when daily maximum wind speed was recorded from 1971 to 2010...... 38

Figure 2.3. Wind rose for Baker Lake from 1971-1990 and 1991-2010...... 41

Figure 2.4. Wind rose for Churchill from 1971-1990 and 1991-2010...... 41

Figure 2.5. Wind rose for Inukjuak from 1971-1990 and 1991-2010...... 42

Figure 2.6. Wind rose for Kuujjuarapik from 1971-1990 and 1991-2010...... 42

Figure 2.7. Wind rose for Nitchequon from 1971-1990 and 1991-2010...... 43

Figure 2.8. Wind rose for Schefferville from 1971-1990 and 1991-2010...... 43

Figure 2.9. Wind rose for Wabush from 1971-1990 and 1991-2010...... 44

Figure 2.10. LOWESS curve for Nitchequon showing the perceived trend and actual trend in monthly wind directional change...... 47

Figure 2.11. LOWESS curves for wind direction at Baker Lake, Churchill, Inukjuak, Kuujjuarapik, Schefferville and Wabush...... 48

Figure 2.12. The percentage change in wind speed, in five-knot increments, and direction at the study sites...... 52

Figure 3.1. Location of airport sites used in fog and visibility study...... 91

Figure 3.2. Number of fog hours per year at the study sites...... 94

Figure 3.3. Number of ice fog hours per year at the study sites...... 100

Figure 3.4: Time series trend for reduced visibility at the study sites...... 105

Figure 3.5. Time series trend for low visibility at the study sites...... 110

Figure 3.6. Illustration of the human observer’s present weather conditions reporting interface...... 127

Figure 4.1. Site location for soil temperature study at Kuujjuaq, Quebec...... 143

xiii

Figure 4.2. Photographs taken during the installation of earth thermistors at Kuujjuaq on October 14, 1966...... 144

Figure 4.3. Comparison between the observed monthly 5 cm soil temperature and the predicted values from the CanESM2 and NCEP models for mean and variance...... 156

Figure 4.4. Comparison between the observed monthly 10 cm soil temperature and the predicted values from the CanESM2 model for mean and variance...... 157

Figure 4.5. Comparison between the observed monthly 20 cm soil temperature and the predicted values from the CanESM2 model for mean and variance...... 158

Figure 4.6. Comparison between the observed monthly 50 cm soil temperature and the predicted values from the CanESM2 model for mean and variance...... 159

Figure 4.7. Comparison between the observed monthly 100 cm soil temperature and the predicted values from the CanESM2 model for mean and variance...... 160

Figure 4.8. Comparison between the observed monthly 150 cm soil temperature and the predicted values from the CanESM2 model for mean and variance...... 161

Figure 4.9. Annual soil temperature projections from 1997 to 2086 for depths at 5, 10, 20, 50, 100 and 150 cm...... 163

Figure 4.10. Winter soil temperature projections from 1997 to 2086 for depths at 5, 10, 20, 50, 100 and 150 cm...... 166

Figure 5.1. Number of non-mobile page views on cryoseism article on English Wikipedia..... 194

Figure 5.2. Weather stations in Canada and the US chosen for temperature and snow depth analysis...... 198

Figure 5.3. Temperature at weather stations in Canada and the US during the winter of 2013– 2014...... 204

Figure 5.4. Snow depth at weather stations in Canada and the US...... 205

Figure 5.5. Plots of reported frostquake locations on January 2/3, January 6/7 and January 20–22 of 2014...... 206

Figure 5.6. Density report map for frostquakes...... 208

xiv

Figure 5.7. The local time in Canada and the US at which the public reported to have heard a frostquake...... 209

Chapter 1: Introduction

1.1 Importance of Aviation Climatology Research

Aviation began on December 17, 1903, with the Wright brothers successfully launching their aircraft, albeit for only 12 seconds. Today, the longest flight in the world is from Auckland to Doha, lasting more than 18 hours (Rosen, 2018). Air transport proves to be an effective way of moving a large number of passengers, mail and freight across continents and oceans in a timely fashion. While aspects of aviation have had some negative impacts on the environment, this is balanced by its essential function to humankind (Upham et al., 2012).

The negative environmental impacts of aviation include noise, air quality and groundwater contamination from the antifreeze used to de-ice aircraft and hydrocarbons in fuel and maintenance facilities. Yet, arguably, the greatest and longest lasting environmental impact is its contribution to climate change due to aircraft emitting greenhouse gases into the atmosphere. Unlike land and sea-based transportation, aviation directly emits these greenhouse gases into the stratosphere. These gases and their interactions within the stratosphere have been studied extensively to determine the change in radiative forcing (e.g. Lee et al., 2009), yet there remains a large degree of uncertainty for some factors such as water vapour, aerosol and induced cloudiness (Dessens et al., 2014; IPCC, 1999). In contrast, there is far less research on the impacts of climate change on aviation itself. This is the main focus of my doctoral work. In the

United Nation’s Intergovernmental Panel on Climate Change (IPCC) Fifth Assessment Report

(AR5), there was an extensive review of how aviation-induced radiative forcing would affect the climate (IPCC, 2013). However, only a subsection with two paragraphs was devoted to how climate change affects aviation (IPCC, 2014). While there are some recent studies that were published after the IPCC’s AR5 publication cutoff date and thus not included in the AR5 report,

1 this paucity of focus demonstrates this is an area ripe for research. Thus, additional research on climate change impacts is necessary to improve the understanding of how a changing climate will affect air travel in the Canadian North. Since aircraft use the atmosphere regularly, any type of weather has the potential to impact aviation (Sasse & Hauf, 2003). Passengers travelling in northern Canada will benefit from the results of this research through anticipated improved safety arising from the awareness of increased risk and ameliorative responses to that risk.

Aviation in the Canadian north presents unique challenges. Some of these include vast distance between communities, low population density and harsh climates. However, some pilots such as Max Ward who later founded Wardair (“Wardair Canada Ltd.”, 1970), thrive in these demanding environment to meet the customer needs (Klassen, 2004). Aviation in the north also faces a higher risk than in the south. According to Gultepe et al. (2015), more than a quarter of all aviation accidents were weather-related. They also found that in northern Canada, aviation accidents and the resulting fatalities occurred at a much higher rate than the national average.

Similar patterns were also observed in the United States. Weather accounted for 23% of all accidents (Kulesa, 2002) and in Alaska, the accident rate was more than double the national average (Quirk, 2013). As climate is an extension of daily weather, a changing climate will likely be manifested as altered weather. Since Arctic temperatures are warming faster than the global average, their weather may experience a more rapid change. Some of these changes will affect flying directly while others will affect the infrastructure that supports aviation, such as terminals and runways. Statistics Canada (2009) calculated that the aviation passenger trips per capita in the three territorial capitals were some of the highest in the country in 2006 as passengers need to transfer at these airports to connect to other flights. Aviation passenger trips per capita at

Whitehorse airport was comparable to Vancouver and Toronto, while Iqaluit and Yellowknife

2 had the highest trips per capita in the entire country. Flights between each settlement in the

Canadian Arctic were considered to be short and lasted less than an hour. However, passengers often had multiple stopovers in the same plane to reach their final destination as the aircraft flew to different communities along the way to pick up and drop off cargo and passengers. Given that the aviation industry is expected to grow at a pace of 5-6% per year over the next 15 years

(Schäfer & Waitz, 2014), more passengers may be exposed to this risk. Moreover, Koetse and

Rietveld (2009) described climate change implications on wind, fog and visibility to be highly uncertain and worthy of more investigation.

Most of the airports in the Hudson Bay region have short runways because low passenger and cargo traffic do not economically justify using jet aircraft. While jets are heavier than turboprops and require a longer runway, all aircraft need to generate sufficient lift from its wings to take off. Less dense air from warmer surface temperature will generate less lift for the aircraft.

The solution to this issue is to either extend the runway length or carry less load in the aircraft. A study has identified that four US airports will experience 50-200% more days with weight restriction on Boeing 737-800, a popular model of passenger jet aircraft worldwide, because of extreme heat by 2070 (Coffel & Horton, 2015). While most of the airports in Hudson Bay region are serviced by turboprop aircraft, variants of this model are used by WestJet, Air Inuit,

Canadian North and First Air in Canada’s three territories (Yukon, Northwest Territories and

Nunavut). Similar, older models of Boeing 737 aircraft are flown into Churchill, Rankin Inlet and La Grande Riviere airports in the Hudson Bay region, making this US study highly relevant to the Canadian context. Even though extending the runway is a solution to counter the weight restriction issue (U.S. Transportation Research Board, 2008), constructing a longer runway is a major fiscal challenge at northern remote airports due to high construction cost shared by a small

3 number of airport users. Transporting building materials often require year-long planning to take advantage of lower shipping cost by sea-lift in summer (Kloepfer, 2017). In some cases, runway extension may not be possible due to local topography, close proximity to the community and lack of flat land. Aside from limitation on freight capacity, hotter air also reduces the effectiveness of the control of the aircraft. Due to the fact that many remote northern airports are surrounded by mountains, warmer temperature will increase the difficulty of manoeuvering the aircraft due to lower air density and will elevate the risk of a crash in summer. Overall, warmer temperatures would lead to a reduction of snow depth (Dyer & Mote, 2006) and an increase in soil temperature (Qian et al., 2011). However, higher air temperature is unlikely to cause weight restriction issues on aircraft in the Hudson Bay region since the local climate is cold. Therefore, the direct impacts of warmer air temperature on aviation are not considered in this study.

Recent research has shown that contrail clouds have the ability to raise the nighttime minimum temperature and lower daytime maximum temperature (Bernhardt & Carleton, 2015), similar to the characteristics exhibited by urban heat island effect to some degree. Aside from temperature changes, weather patterns in the region will also be affected. Aviation emission of gases in the upper troposphere also led to fewer naturally formed clouds (Burkhardt & Kärcher,

2011). Soot emitted by aircraft was attached to cloud condensation nuclei to form contrail- induced cirrus clouds. In addition, at cruising altitude, winter transatlantic flights will face a significant increase in turbulence (Williams & Joshi, 2013), resulting in a higher chance of injury.

Air mass types were observed to be changing in the Hudson Bay and Foxe Basin region (Leung

& Gough, 2016). The type was changing from DP (dry polar) to MP (moist polar), which suggested that the frequency and the location of the fronts were altering and that the characteristics of the temperature and moisture content of the air parcels in the region had shifted.

This latter effect may directly influence the three aviation factors covered in this thesis: wind, fog and soil temperature.

1.2 Wind Study

While there was no Canadian statistics available specifically for wind-related aviation accidents, wind was responsible for close to half of all accidents in the United States (Jenama &

Kumar, 2013). Within this category, a third of accidents were attributed to strong crosswinds and almost another third was attributed to wind gust (Jenama & Kumar, 2013). Crosswind landings occur when the wind direction does not align with the runway orientation. It is particularly dangerous because the wind is trying to push the aircraft off the runway’s heading while the pilot needs to steer the aircraft into the wind so that the resulting force’s vector aligns with the runway.

Most of the Hudson Bay communities have a small population and smaller, turboprop aircraft usually services them. Only a few airports have large, jet service. Turboprop aircraft are more susceptible than jets to high crosswinds because turboprops have lower crosswind limits

(Shepherd, 2018). The largest type of aircraft regularly serving the Hudson Bay is Boeing 737, flown by Air Inuit, Canadian North and First Air airlines, and the type has a crosswind limit of

31 to 35 knots (57 to 65 km/h) (Brady, 2018). Therefore, high wind speeds combined with unfavourable wind directions can cause delays or cancelled flights in addition to presenting a safety risk. However, minimal or calm winds are also not ideal for aircraft to land and takeoff because lift is generated by wind flowing over the wings (Beckwith, 1985). Low wind flow reduces the amount of lift generated such that a longer runway is required during takeoff. Lift is also beneficial during landing as it allows the pilots to land the aircraft at a slower speed. In general, high wind speeds and crosswinds present a greater risk factor than low or calm winds.

Wan et al. (2010) used homogenized average wind speeds to detect trends from over 100 stations in Canada. Their study analyzed the long-term mean wind speeds at four Hudson Bay airports (Baker Lake, Churchill, Coral Harbour and Kuujjuarapik) and found that the speed was strongest in fall followed by winter. In fall, the long-term means at those airports were greater than 20 km/h. In winter, only the two stations in the western side of Hudson Bay (Baker Lake,

Churchill) had an average wind speed greater than 20 km/h. The average speed at stations on the eastern side of Hudson Bay was between 10 to 20 km/h. During spring and summer, the average wind speed at all four stations was between 10 to 20 km/h. Wan et al. (2010) also found that wind speeds at those stations were declining according to raw wind speeds. However, after homogenizing the wind speeds to account for station re-location, anemometer height adjustment and change in measuring instruments, the trends at Baker Lake, Churchill and Coral Harbour switched to positive and that the speeds were increasing significantly. There was no trend for

Kuujjuarapik’s wind speed.

Wind gusts in Canada are expected to increase according to climate change simulations

(Cheng et al., 2014). In Clyde River, Nunavut, individuals noted that the winds became stronger in summer and winter based on their traditional knowledge (Gearheard et al., 2010). But a lack of surface wind research in the Canadian Arctic prevented the validation of their views

(Gearheard et al., 2010). Higher wind speed and gust will intensify storm surges (Burbidge,

2017). The roof of Inuvik airport in Northwest Terrorities was damaged by high winds in 2012

(CBC, 2012a). Koetse and Rietveld (2009) suggested that sea level rise was virtually certain to occur in the US. They identified that in four eastern seaboard states, up to 3% of airport property and runways may be at risk of damage from storm surges in the future. Increase in intense precipitation events would likely occur while seasonal precipitation and flooding patterns are

6 likely to be altered. Qikiqtarjuaq airport in Nunavut and Fort Simpson airport in Northwest

Territories were threatened by flooding in 2012 and 2015 respectively (CBC, 2012b; CBC, 2015).

Also, an increase in cold-season storms would likely occur. These factors are expected to put the airport infrastructure at risk. At London’s Heathrow airport, climate models projected increased precipitation events which could negatively affect the airport operations if the tarmac and drainage were not adapted to more intense rainfall (Pejovic et al., 2009). As many Hudson Bay airport and the communities they serve are located near the shorelines, higher storm surges could cause additional damage to both the airports and the communities.

Chapter 2 examines the wind record in the Arctic region in detail and presents a trend analysis of wind speed, maximum daily wind speed and wind direction changes across seven stations around Hudson Bay region, northern Quebec and northern Labrador from 1971 to 2010.

1.3 Fog and Low Visibility

Proportion to other countries, Canada has a higher rate of aviation accidents caused by low visibility than in India and the United States (Jenama & Kumar, 2013; Gultepe et al., 2015).

One of the major factors for low visibility in the Hudson Bay region is fog. Fog makes it difficult for pilots to find the runway to land. While the plane is on the ground, fog blocks a pilot’s vision from maintaining safe clearance distance away from other aircraft and obstructions. With low wind speed and high humidity, these environmental conditions are unfavourable for flying because the fog becomes stagnant and persists. The lack of sea ice during summer in Hudson

Bay allows fog to advect inland through onshore winds (Hanesiak & Wang, 2005). A longer ice- free period in Hudson Bay (Kowal et al., 2017) potentially creates favourable conditions for more fog to form off the waters.

A study conducted by the Canadian military in the early 1950s found that ice fog and blowing snow were the two main causes of reduced visibility and that these factors were heavily influenced by local topography and wind patterns (Rae, 1954). That study found that the probability of fog formation decreased from 0oC to about -28.9oC, followed by an increase in probability to -34.4oC. At -40oC, Rae (1954)’s study described ice fog was almost certain to occur but often not severe enough to affect airport operations. The study also offered insight into the historical relationship between wind speed, blowing snow and visibility. Using Resolute,

Nunavut in 1947 to 1949 as an example, Rae (1954) reported that visibility was often exceeding

9 km if the wind speed was below 24.1 km/h because the blowing snow was limited to drifting snow on the ground. When wind speed increased past 24.1 km/h, blowing snow began to obstruct visibility but the severity for a given wind speed was variable. Once the wind speed exceeded 48.3 km/h, blowing snow would cause visibility to drop to less than 0.8 km. More recent research also found that 30 to 40% of blowing snow conditions occurred with less than 0.8 km of visibility (Hanesiak & Wang, 2005). This is identical to the modern day’s threshold of reduced visibility as defined by Transport Canada (Transport Canada, 2017). For that reason, visibility could be expected to be less than 0.8 km if blowing snow and wind speed exceeded 48 km/h.

Fog and low visibility not only hinders aviation but also other land and sea-based activities as well as tourism (Denstadli & Jacobsen, 2014). However, in general, fog is most observed at night and early morning (Gough & He, 2015) while most Hudson Bay airports are not operating at these hours and the observer would not be making records of fog occurrences.

Chapter 3 examines the frequency of encountering fog and low visibility through time series analysis at 16 communities around Hudson Bay region. This chapter analyzes whether fog

8 and low visibility conditions were improving (i.e. less frequent) in this area. It identifies the main causes for the reduction in visibility at these communities.

1.4 Climate Change Impact Assessment on Soil Temperature

There are not as many soil temperature measuring stations as weather observation stations in Canada (Qian et al., 2011). Between 1901 and 1995, the annual mean soil temperature across Canada was 2.5oC and the soil temperature at 20 cm depth increased by an average of

0.6oC (Zhang et al., 2005). Their study revealed that soil temperature did not warm at the same rate as air temperature. In Hudson Bay and northern Quebec regions, both the annual air and soil temperatures were cooling by up to 2oC between 1901 and 1995 and the cooling rates between air and soil temperatures were similar, with the exception of northwestern Hudson Bay area from

Churchill in the south to Baker Lake and beyond in the north (Zhang et al., 2005). In this particular area, the average annual air temperature changed by +0.5oC to +2.5oC while the average annual soil temperature changed by -1oC to +0.3oC. Soil temperature at other depths showed a general rising trend across Canada from 1958 to 2008 and the warming was associated with the rise of air temperature (Qian et al., 2011).

In northern Quebec, permafrost degradation led to differential subsidence and settlement on the tarmac and the access roads (Doré et al., 2016). For the access roads in Umiujaq, Quebec, differential subsidence as much as 85 cm was found (Fortier et al., 2011). Drainage systems were also disrupted (Boucher & Guimond, 2012), which caused even greater impacts when intense precipitation events occur. With deteriorating permafrost conditions in the Canadian

Arctic and Hudson Bay region (Tam et al., 2015), airport infrastructure built on top of permafrost are at risk of compromising the structural integrity due to differential settlement and will require

9 additional cost allocated to building inspection and maintenance (Nelson et al., 2002).

Intervention techniques such as installing heat drains to mitigate against permafrost thaw would require yet more expense (Boucher & Guimond, 2012).

Climate change impact assessment (CCIA) provides a framework for analyzing the potential impacts of climate change (Carter et al., 1994). CCIA utilizes General Circulation

Model (GCM) for climate projections based on different greenhouse gases concentration trajectories. GCM simulates the interactions between the Earth’s atmosphere, terrestrial system, oceans and cryosphere through physical processes such as advection, energy transfers and land surface processes. GCMs consist of multiple layers of atmospheres and oceans, divided into horizontal grids of 250 to 600 km wide depending on the model chosen (IPCC, 2013b). Due to

GCM’s large grid size, it is not ideal to use GCM to analyze change on a local scale nor to produce climate scenarios on a daily scale (Pielke & Wilby, 2012). Approaches to address this issue include using regional climate model (RCM) and downscaling. Downscaling is divided into dynamic downscaling and statistical downscaling. Dynamic downscaling uses RCMs that have a higher resolution of surface features such as terrain and soil (Castro et al., 2005). Statistical downscaling uses transfer functions between the mesoscale weather systems and the observed variable to determine which predictor variables had the strongest relationship with the predictant

(Pielke & Wilby, 2012). This ensures the relationship between predictor and predictant are reasonably linked while capturing weather patterns that are higher in resolution than dynamic downscaling. The atmospheric predictor variables include surface temperature, humidity, sea level pressure, surface wind, upper air flow and others. The predictor variables with the strongest relationships and share the most similar characteristics with the predictant variable were used on past climate for validation on accuracy and future climate for projections.

Chapter 4 uses statistical downscaling model approach and applies to observed, historical soil temperature at Kuujjuaq, Quebec to project future soil temperature conditions. The results of this chapter provide an insight into the soil temperature from 5 to 150 cm depths in the future and whether the soil at those depths will be exposed to temperatures above 0oC, conditions unfavourable for permafrost.

1.5 Frostquakes

Rapid drop in temperature while the soil is fully saturated with water can lead to a rare phenomenon known as cryoseism or frostquake (Leung et al., 2017). This occurs when the ground is saturated with water, has a minimal amount of snow cover and a large, sudden drop causes the water to freeze inside the soil. While the water expands into nearby pores during the freeze-up process, the pressure builds as the saturated ground does not have sufficient space for the ice to expand. Eventually, the ground would give way due to the pressure and a noticeable crack may form. Audible cracking sound may also be heard or small vibration can be felt inside nearby buildings. This was experienced by the general public shortly after the ice storm in

December 2013 and this rare phenomenon was covered by mainstream media shortly after people reported a cracking sound or felt the vibration when the pressure from the ground was released. While this specific event from 2013 mainly affected central Canada, the cracks could jeopardize the integrity of the runway or the foundation of other airport infrastructure.

Chapter 5 examines frostquakes in the winter of 2013 through social media monitoring to determine the atmospheric conditions that were favourable to cause frostquakes. It provides, for the first time, a large-scale and widespread tracking of frostquake due to severe weather. It also identified the affected areas by time, total reports and by population density. Finally, using the

11 weather forecast, this chapter offered forecast on specific dates in areas that were prone to frostquakes.

1.6 Research Objectives

In this thesis, the projects will examine various factors that affect aviation and related cold climate phenomena:

1) What is the historical wind patterns (wind speed and direction) at the community

airports in Hudson Bay and northern Quebec? Has the wind patterns changed over a

40-year period?

2) How often were the community airports affected by fog and restricted visibility

conditions?

3) What was the baseline soil temperature at a northern Quebec airport location? How

would the soil temperature change under climate change?

4) What were the environmental conditions that caused frostquakes? Which areas were

affected? Could frostquakes be predicted based on known weather conditions from

forecasts?

1.7 References

Bernhardt, J., & Carleton, A. M. (2015). The impacts of long-lived jet contrail ‘outbreaks’ on surface station diurnal temperature range. International Journal of Climatology, 35(15), 4529-4538. Boucher, M., & Guimond, A. (2012). Assessing the Vulnerability of Ministère des Transports du Québec Infrastructure in Nunavik in a Context of Thawing Permafrost and the Development of an Adaptation Strategy. In Guy Doré & Brian Morse (Eds.), Cold Regions Engineering. Quebec City: American Society of Civil Engineers.

Brady, C. (2018). The Boeing 737 Technical Guide (Version 78). Frodsham, Cheshire, United Kingdom: Tech Pilot Services Limited. Burbidge, R. (2017). Climate-proofing the airport of the future. Journal of Airport Management, 11(2), 114-128. Burkhardt, U., & Kärcher, B. (2011). Global radiative forcing from contrail cirrus. Nature Climate Change, 1, 54-58. Carter, T. R., Parry, M. L., Harasawa, H., & Nishioka, S., (1994). IPCC Technical Guideline for Assessing Climate Change Impacts and Adaptations with a Summary for Policy Markers and a Technical Summary. Department of Geography, University College, London, and the Centre for Global Environmental Research. National Institute for Environment studies, Japan. Castro, C. L., Pielke, R. A. Sr., & Leoncini, G. (2005). Dynamical downscaling: Assessment of value retained and added using the Regional Atmospheric Modeling System (RAMS). Journal of Geophysical Research: Atmospheres, 110, D05108. CBC. (2012a). Inuvik, N.W.T., airport roof damaged in blizzard. Retrieved from https://www.cbc.ca/news/canada/north/inuvik-n-w-t-airport-roof-damaged-in-blizzard- 1.1130330 CBC. (2012b). Qikiqtarjuaq airport repair costs double after more floods. Retrieved from https://www.cbc.ca/news/canada/north/qikiqtarjuaq-airport-repair-costs-double-after- more-floods-1.1197325 CBC. (2015). Fort Simpson wants N.W.T. to help fight riverbank erosion. Retrieved from https://www.cbc.ca/news/canada/north/fort-simpson-wants-n-w-t-to-help-fight-riverbank- erosion-1.3002516 Coffel, E., & Horton, R. (2015). Climate Change and the Impact of Extreme Temperatures on Aviation. Weather, Climate, and Society, 7(1), 94-102. Dessens, O., Köhler, M. O., Rogers, H. I., Jones, R. I., & Pyle, J. A. (2014). Aviation and climate change. Transport Policy, 34, 14-20. Dyer, J. L., & Mote, T. L. (2006). Spatial variability and trends in observed snow depth over North America. Geophysical Research Letters, 33(16), L16503. Fortier, R., LeBlanc, A., & Yu, W. (2011). Impacts of permafrost degradation on a road embankment at Umiujaq in Nunavik (Quebec), Canada. Canadian Geotechnical Journal, 48(5), 720-740. Gough, W. A., & He, D. (2015). Diurnal temperature asymmetries and fog at Churchill, Manitoba. Theoretical and Applied Climatology, 121(1-2), 113-119. Gough, W. A., & Leung, A. (2002). Nature and fate of Hudson Bay permafrost. Regional Environmental Change, 2(4), 177-184. Gultepe, I., Zhou, B., Milbrandt, J., Bott, A., Li, Y., Heymsfield, A., . . . Cermak, J. (2015). A review on ice fog measurements and modeling. Atmospheric Research, 151, 2-19.

IPCC. (1999). Special Report, Aviation and the Global Atmosphere. Cambridge, United Kingdom: Cambridge University Press. IPCC. (2013a). Climate Change 2013: The Physical Science Basis. Working Group I Contribution to the Fifth Assessment Report of the IPCC. Cambridge, United Kingdom: Cambridge University Press. IPCC. (2013b). What is a GCM? Retrieved from http://www.ipcc- data.org/guidelines/pages/gcm_guide.html IPCC. (2014). Climate Change 2014: Impacts, Adaptation, and Vulnerability. Working Group II Contribution to the Fifth Assessment Report of the IPCC Cambridge, United Kingdom: Cambridge University Press. Jenama, R. K., & Kumar, A. (2013). Bad weather and aircraft accidents – global vis-à-vis Indian scenario. Current Science, 104(3), 316-325. Klassen, H. C. (2004). Max Ward and Wardair. Journal of the West, 43(2), 54-62. Kloepfer, C. (2017). An airport public–private partnership in the Canadian Arctic. Journal of Airport Management, 11(2), 136-146. Koetse, M. J., and Rietveld, P. (2009). The impact of climate change and weather on transport: An overview of empirical findings. Transportation Research Part D, 14(3), 205-221. Kowal, S., Gough, W. A., & Butler, K. (2017). Temporal evolution of Hudson Bay Sea Ice (1971–2011). Theoretical and Applied Climatology, 127(3-4), 753-760. Kulesa, G. (2002). Weather and Aviation: How does Weather Affect the Safety and Operations of Airports and Aviation, and How Does FAA Work to Manage Weather-related Effects? Washington, DC: US Department of Transport Center for Climate Change and Environmental. Lee, D. S., Fahey, D. W., Forster, P. M., Newton, P. J., Wit, R. C., Lim, L. L., . . . Sausen, R. (2009). Aviation and global climate change in the 21st century. Atmospheric Environment, 43(22-23), 3520-3537. Leung, A. C. W., & Gough, W. A. (2016). Air Mass Distribution and the Heterogeneity of the Climate Change Signal in the Hudson Bay / Foxe Basin region, Canada. Theoretical and Applied Climatology, 125(3), 583-592. Leung, A. C. W., Gough, W. A., & Shi, Y. (2017). Identifying Frostquakes in Central Canada and Neighbouring Regions in the United States with Social Media. In M. Leitner & J. Jokar Arsanjani (Eds.), Citizen Empowered Mapping (pp. 201-222). Cham, Switzerland: Springer Nature. Nelson, F. E., Anisimov, O. A., & Shiklomanov, N. I. (2002). Climate Change and Hazard Zonation in the Circum-Arctic Permafrost Regions. Natural Hazards, 26, 203-225. Pejovic, T., Williams, V. A., Noland, R. B., & Toumi, R. (2009). Factors Affecting the Frequency and Severity of Airport Weather Delays and the Implications of Climate

Change for Future Delays. Transportation Research Record: Journal of the Transportation Research Board, 2139, 97-106. Pielke, R. A., & Wilby, R. L. (2012). Regional Climate Downscaling: What’s the Point? Eos, Transactions American Geophysical Union, 93(5), 52-53. Qian, B., Gregorich, E. G., Gameda, S., Hopkins, D., & Wang, X. L. (2011). Observed soil temperature trends associated with climate change in Canada. Journal of Geophysical Research: Atmospheres, 116, D02106. Quirk, W. A. (2013). Bush Pilot Way: Flying and Training in Alaska to Become the Best Pilot You Can Be. Anchorage, Alaska, United States of America: Publication Consultants. Rosen, E. (2018). The 10 Longest Flights in the World by Distance. Retrieved from https://www.forbes.com/sites/ericrosen/2018/03/26/the-10-longest-flights-in-the-world- by-distance Schäfer, A. W., & Waitz, I. A. (2014). Air transportation and the environment. Transport Policy, 34(1), 1-4. Shepherd, M. (2018). Should Airplanes Land In 40-60 Mph Nor'easter Winds? Experts Weigh In. Retrieved from https://www.forbes.com/sites/marshallshepherd/2018/03/02/should- airplanes-land-in-40-60-mph-noreaster-winds-experts-weigh-in Statistics Canada. (2009). Transportation in the North. Tam, A., Gough, W. A., & Xie, C. (2015). An Assessment of Potential Permafrost along a North-South Transect in Canada under Projected Climate Warming Scenarios from 2011 to 2100. The International Journal of Climate Change: Impacts and Responses, 6(2), 1-8. Transport Canada. (2017). Canadian Aviation Regulations 300 Series – Aerodromes and Airports. Gatineau, Quebec. Upham, P., Maughan, J., Raper, D., & Thomas, C. (2012). Towards Sustainable Aviation. Oxon, United Kingdom: Routledge. U.S. Transportation Research Board. (2008). Potential Impacts of Climate Change on U.S. Transportation. Washington, DC: National Research Council of the National Academies. Wan, H., Wang, X. L., & Swail, V. R. (2010). Homogenization and Trend Analysis of Canadian Near-Surface Wind Speeds. Journal of Climate, 23, 1209-1224. Wardair Canada Ltd. (1970, March 26). Flight International, 509. Williams, P. D., & Joshi, M. M. (2013). Intensification of winter transatlantic aviation turbulence in response to climate change. Nature Climate Change, 3, 644-648. Zhang, Y., Cheng, W., Smith, S. L., Riseborough, D. W., & Cihlar, J. (2005). Soil temperature in Canada during the twentieth century: Complex responses to atmospheric climate change. Journal of Geophysical Research, 110, D03112.

Characterizing Observed Surface Wind Speed and Direction Trends in the Hudson Bay Region from an Aviation Perspective

Andrew C.W. Leung1, William A. Gough1, Ken A. Butler2, Tanzina Mohsin1 1Department of Physical & Environmental Sciences, University of Toronto Scarborough 2Department of Computer & Mathematical Sciences, University of Toronto Scarborough 1265 Military Trail, Toronto, Ontario, Canada, M1C 1A4

Chapter 2: Wind Analysis

2.1 Abstract

Wind analysis is important for informing airport operation and safety. Many communities in the Hudson Bay and Labrador regions (Canada) are remote communities that rely heavily on aircraft for passenger and freight movement. Wind speed, maximum daily wind speed and wind direction from 1971 to 2010 were examined to determine how they influenced aviation in seven northern communities. Significant increases in wind speed and maximum daily wind speed were found for some of the months and seasons of the year for the Hudson Bay region and a strongly significant decrease in those variables for the Labrador communities. Average wind speeds at multiple locations are approaching the threshold where takeoff and landing would be restricted to one direction. There were significant changes to wind direction at two communities. An assessment was also made to determine whether the existing airport runways are built aligned to prevailing wind directions while also considering potential impacts of greater wind speed and shifting wind directions.

2.2 Introduction

Travel in northern Canadian communities is being impacted by climate change. In the

James Bay region, a warmer environment has resulted in a shorter operational period for ice roads in winter (Hori et al., 2017) and a longer window for re-supply of goods by ships in summer (Prowse et al., 2009). Many remote communities continue to rely on air travel for year- round transportation, delivering perishable goods and medical evacuation (medivac). However, aviation in the Canadian north faces greater risks than in the south. Gultepe et al. (2015) reported that 27% of all aviation accidents were weather-related. Aviation accidents were 25 times higher in the Canadian north and fatalities were 18 times more likely than the national average (Gultepe et al., 2015). Similar disparities were also observed in the United States where weather accounted for 23% of all accidents (Kulesa, 2002); and in Alaska, the accident rate was more than two times higher than the national average (Quirk, 2013). Flying in the north is inherently more challenging. Pilots are more frequently exposed to hazardous flying conditions like blizzards. Similar to automobiles, higher performance demands are required for aircraft engines in extreme cold. Many northern airports lack advanced navigation or landing aids for low visibility that are commonly found in southern airports. Northern airports are very often far apart from one another and thus it takes much longer to return to an originating airport or divert to a nearby airport in case of any in-flight emergencies. In Canada, if an accident occurs in the

Hudson Bay or the Arctic region, search and rescue experts are dispatched from Trenton, Ontario

(Canadian Armed Forces, 2015), typically requiring many hours to reach the crash site.

As climate is the statistical mean of daily weather, a changing climate will likely alter the weather. Since Arctic temperatures are warming faster than the global average, Arctic weather is likely to experience a more rapid change (Gearheard et al., 2010; Weatherhead et al., 2010). Air

18 mass types were observed to be changing in the Hudson Bay and Foxe Basin region (Leung &

Gough, 2016). The shift was from dry polar (DP) to moist polar (MP) indicating that the frequencies and locations of the fronts between air masses are changing and that the characteristics of the temperature and moisture content of the air masses in the region have also evolved. Some of these changes will affect flights enroute while others will affect the infrastructure that supports aviation, such as terminals and runways. More passengers will be exposed to delays, cancellations and flight safety risks as the global aviation industry grows at a pace of 5-6% per year over the next 15 years (Schäfer & Waitz, 2014). Yet, most of the climate- related research in aviation examines the impacts of aircraft and fuel use on the environment as a source of greenhouse gas emissions. IPCC AR5 covered how climate would be affected by various aviation-induced radiative forcing (IPCC, 2013) but not a lot was devoted to how climate change affects aviation (IPCC, 2014). Koetse and Rietveld (2009) mentioned that the uncertainty of climate change on wind was high. Coumou et al. (2015) reported that historical zonal winds in the mid-latitudes of Northern Hemisphere slowed down due to a smaller temperature gradient between the equator and the North Pole while Francis and Vavrus (2015) found that meridional wind increased due to Arctic warming. Under future climate model scenarios, there were studies that showed significant differences for surface wind conditions under different scenarios in different locations (Eichelberger et al., 2008). For example, while Ma et al. (2016) found that global average surface wind speed would diminish in the future, both Eichelberger et al. (2008) and Ma et al. (2016) concluded that North America wind speed would be largely unchanged. But both studies also showed that eastern Hudson Bay region would experience the greatest increase in wind speed in the entire North America. Hence, the future wind speed changes appear to be location-dependent.

Surface weather conditions appear to play a large role in aviation-related delays and accidents. Pejovic et al. (2009) found that an increase of 1 knot (1.852 km/h) in average wind speed would increase the probability of weather-related delays by 8% at London Heathrow airport. Since London Heathrow airport is severely congested, any small delays could quickly cascade into longer delays. Whereas, the traffic at northern Canadian airports is considerably less than at London Heathrow. Thus, the airlines are less likely to face cascading delays from backlogged traffic but not immune to crew’s maximum duty time or extended weather delays.

However, ramp space and processing capacity in northern Canadian airports are limited. This presents a challenge to airport operations when multiple aircraft were delayed due to unfavourable weather conditions and those aircraft arrive at the same time after the adverse weather was cleared.

Wind has been described by Budd and Ryley (2012) as having “the greatest potential to impact the safety of the air transport system.” Jenama and Kumar (2013) found that 48% of all weather-related aviation accidents in the US were caused by wind. According to Jenama and

Kumar (2013), the prevailing causes of wind-related accidents from 1994 to 2003 in the US were crosswind (34%), wind gust (29%), high wind speed (8%) and sudden wind shift (3%). Larger and heavier aircraft are less affected by crosswind (Shepherd, 2018). Since wind gusts in the

Canadian north are projected to increase by up to 300% in the highest wind gust category (90 km/h, or 48.6 knots) by 2100 (Cheng et al., 2014), the risk of more accidents resulting from wind gusts and crosswinds is substantial. However, the Canadian Arctic lacks historical data that serves as a baseline to assess regional climate change (Reidlinger & Berkes, 2001) as well as a lack of surface wind studies conducted in the region (Gearheard et al., 2010), highlighting an important research needs. Aside from aviation, the safety of other transportation methods such as

20 shipping (Strefford, 2002) and the duration of winter roads (Hori et al., 2017) are affected by wind. Socioeconomic impacts and opportunities such as subsistence hunting (Gearheard et al.,

2010), insurance payout for wind gust damage (Klawa & Ulbrich, 2003) and wind energy generation potential in northern Ontario (Arriaga et al., 2013) and northern Quebec (Hooshangi,

2014) may benefit from additional wind analysis studies.

This study explored the potential impacts of flying into the communities around the

Hudson Bay area by:

1) establishing a historical wind speed and direction patterns that would serve as a

baseline for further analysis,

2) analyzing the variations and changes in wind patterns to assess temporal trends, and

3) assessing the airports’ runway orientations to determine whether the prevailing wind

direction matched the runway’s orientation.

2.3 Methods

2.3.1 Data Collection

Environment and Climate Change Canada’s Climate Data Online (CDO) website contains surface weather records from airports in the Hudson Bay region since around 1953. This data provides hourly and daily measurements of temperature, precipitation, visibility, sky condition and wind patterns. These records are of high quality because of their stationarity and long record interval measurements, while also not being susceptible to urban heat island effects nor tall buildings. The measurements were made by trained professionals. Wind data from hourly observed weather records measured at 10 m above ground level at seven airport locations were obtained from CDO (Figure 2.1). The anemometers used were type 45B, U2A, and more recently,

21 digital 78D (Wan et al., 2010). All seven airports in this study operated 24-hours a day. These airports were chosen because they have relatively high passenger traffic and often serve as a local hub for cargo and passengers destined for nearby communities. Two stations (Baker Lake and Churchill) are located in western Hudson Bay. Baker Lake is located about 200 km inland of northwestern Hudson Bay while Churchill is located on the coastline. Two stations (Inukjuak and

Kuujjuarapik) are located on the eastern shores of Hudson Bay. Nitchequon is an abandoned airstrip and represented the wind conditions in central Quebec. The last two stations are

Schefferville and Wabush, which are located at the border of Quebec and Labrador. With the exception of Nitchequon, all other locations are predominantly Inuit communities. The data were downloaded through a data-mining program written in R program that is adapted for bulk download of contents in the CDO website in a fast, accurate and highly efficient manner (R-

CDO). Data obtained were from 1971 to 2010 except Nitchequon, which was from 1953 to 1985.

Nitchequon’s weather station ceased operation after November 1985. The wind data, in particular, wind direction, were visualized with WRPlot View software.

Figure 2.1: Environment and Climate Change Canada weather stations selected to analyze wind conditions in the Hudson Bay region.

2.3.2 Wind Speed

Wind speed measures the two-minute average of wind at 10 m above ground at the beginning of each hour. Seasonal time periods are as follows: winter (December, January,

February), spring (March, April, May), summer (June, July, August) and fall (September,

October, November). Daily average wind speed is calculated by averaging 24 hourly wind speeds each day. Daily maximum wind speed is the highest wind speed, if it exceeds 28 km/h, recorded from all hourly wind speed within a calendar day. This is similar to the daily maximum wind gust, which is the highest instantaneous wind speed recorded within a calendar day that exceeds 28 km/h, or 15 knots (Cheng et al., 2014). The maximum daily wind speed was analyzed to determine which hours of a day were most likely to encounter the strongest wind. This approach was used because CDO website reports the speed of maximum gust, an instantaneous peak value at any time of the day, and the direction of the maximum gust but does not include

23 the time of such occurrence. The Student’s t-test was used to compare the first half of the period

(1971 to 1990) with the second half (1991 to 2010), except Nitchequon which were 1953 to 1969 and compared with 1970 to 1985 to determine the significant trends, if any, for daily average and daily wind gust. Mann-Kendall test was used to assess the trends in number of days that met or exceeded the 28 km/h daily maximum wind speed threshold. Similar to other climatological trend analysis (Mohsin & Gough, 2010), a time series analysis was conducted with Mann-

Kendall statistical test (Mann, 1945; Kendall, 1975) on averaged daily wind speed on monthly and seasonal scales. Serial independent observations in a time series are required for the Mann-

Kendall test. If autocorrelation is present in the data, a modified Mann-Kendall test was performed to account for the influence of autocorrelation on significance testing by adjusting the variance of the dataset (Hamed & Rao, 1998). This is achieved by dividing the total number of observations by the number of “effective” observations to generate a correction factor. The correction factor is adaptable to both positive and negative autocorrelations. Positive autocorrelation will have a correction factor > 1 and negative autocorrelation will have a correction factor between 0 and 1. A smaller p-value is achieved if the variance is multiplied by the correction factor with a value between 0 and 1. Vice versa, positive autocorrelation will cause the variance to increase and results in a larger p-value. This increases the probability of the null hypothesis not being rejected because the significance level was not reached to reject the null hypothesis. Theil-Sen slope estimator was used to assess the magnitude of change (Theil,

1950; Sen, 1968).

2.3.2.1 Calm Wind

According to Environment Canada (2015), calm wind is defined as the average wind speed less than 2 knots (3.7 km/h) when the observation was taken. Cook (2015) stated that calm wind can be broken down into two categories, true calms and incidental calms. True calms mean that wind was not blowing and usually persisted in consecutive observations. Incidental calms differ from true calms in that wind was blowing prior to and just after the observation was taken but wind speed happened to be calm while the instrument measured wind speed at the last two minutes of each hour or while the wind direction was changing. To determine the breakdown of true calms and incidental calms, true calms were calculated by looking at how frequently wind speed was calm in one hour and the next hour.

2.3.3 Wind Direction

By Environment Canada’s definition, calm winds do not have any wind direction

(Environment Canada, 2015). Similar to the calculation of daily wind speed values, the daily wind direction was calculated by averaging the hourly wind directions with R’s circular package

(Agostinelli & Lund, 2011) and hours with calm winds were excluded from the daily wind direction calculation step. The monthly averaged wind direction was calculated based on the previous step using the daily averaged wind direction. Given that previous studies in the region reported a non-linear rate of change in air mass patterns and reduction in days with sea ice within this time period (Leung & Gough, 2016; Kowal et al., 2017), a locally weighted scatterplot weighting (LOWESS) was applied to reflect a non-linear trend in a time series plot (Cleveland,

1981). Since wind direction is a vector variable, some data points were shifted up by 360o to reveal the true change in direction while retaining the same direction without limiting to the

25 direction from 10o to 360o. Mann-Kendall test (Mann, 1945; Kendall 1975) was used to assess if the change in wind direction is significant.

2.3.4 Aviation Perspective

Combining the trends in wind speed and wind direction, the potential impacts of flying at the airports used in this study was examined since pilots have to continuously assess weather conditions from before departure, enroute and approaching the destination. Likewise, airport staff have to forecast different weather conditions, including wind speed and direction, which affect airport operations such as assigning the proper runway direction for use, limiting the traffic or closing the airport until the weather improves. Published flight schedules from a weekday in July 2017 were obtained from airline companies to obtain a snapshot of the scheduled arrival and departure time of aircraft as well as the type of aircraft flying to these communities.

2.3.5 Uncertainties and Assumptions

Anemometer upgrade over the years led to changes in the measuring method and accuracy. The oldest anemometer, Type 45B, required the human observer to estimate wind speed based on the number of flashes on the wind speed indicator lamp while the newer U2A anemometer had a dial to indicate the average wind speed (Wan et al., 2010). The newest anemometers, 78D, used a microcomputer to measure wind speed in 5-second intervals to calculate the mean. Sampling frequency also changed over time. Prior to 1996, hourly average wind speeds at aviation sites were taken in the last one minute of each hour (Wan et al., 2010).

Since 1996, this was changed to last two minutes of each hour. Non-aviation sites, not used in

26 this study, used the average hourly wind speed in the last ten minutes of each hour. This study assumed that the change in sampling frequency did not affect the wind speed values. According to Wan et al. (2010), changes in anemometer height also played a huge role in the measured wind speeds. In the past, anemometers placements were not standardized and they were often installed at the top of the terminal building or control tower, whose heights varied by locations.

Standardization took place at different time based on their maintenance schedule and anemometers were moved to 10 metres. As the maintenance schedule of these sites were not made publicly available, it was impossible to determine when the standardization at a particular site happened other than homogenizing the dataset to detect step changes.

Daily maximum wind speed was less than or equal to the daily gust speed as the former took a single hour with highest wind speed in the last one or two minutes of the hour while the latter measured the maximum wind speed in any ten-minute interval within the 24-hour (Wan et al., 2010). Therefore, the latter would almost always measured a higher value because it would capture any events happening outside of the last one or two minutes of each hour. But the latter cannot be attributed to a particular hour as Environment and Climate Change Canada did not publish the time of occurrence for the gust speed. Calm winds were equally susceptible to the sampling bias related to the hourly observations. It was possible to have calm wind in the last one or two minutes of each hour while having non-calm conditions in the rest of the hour. The anemometer also required a certain amount of force exerted by the wind to push the wind vane or the cup from calm conditions. Similarly, low winds conditions could fail to sufficiently push the wind vane such that the recorded wind direction was the direction before the wind died down. As a result, there is a bias towards low speed winds.

Instruments were assumed to be in good working conditions at all times. This was often not the case especially at the northern sites used in this study. Northern weather stations were frequently exposed to icing conditions in winter that froze the propeller of the anemometer, causing the anemometer to falsely report calm wind for multiple hours or days. An observer needs to be diligent to notice the discrepancy between the windy conditions outdoors and the reported calm conditions before attempting to remove the icing from the propeller. The bearings inside the anemometer also wear out over time and lead to inaccurate wind speed or direction being measured. Birds perching on top of the anemometer also led to inaccurate wind speed and direction. Power failure or unstable power current could lead to missing or erroneous data.

Sending spare parts or dispatching repair crews often took longer because these airports were harder to reach. These factors all contributed to uncertainties in the wind speed and wind direction values in the dataset.

2.4 Results

2.4.1 Average Daily Wind Speed

Using Student’s t-test, overall average wind speed appeared to increase at Baker Lake,

Churchill, Inukjuak, Kuujjuarapik and Nitchequon, although some of the locations did not show a significant increase in all seasons (Table 2.1). Baker Lake showed an increase in wind speed in winter, spring and fall but the only statistically significant change was a decrease in wind speed in summer. The two Labrador locations, Schefferville and Wabush, showed statistically significant declines in wind speed on all temporal scale (Table 2.1). The wind speed values from these locations vary, but all were close to 18.5 km/h (10 knots), the threshold value for aircraft to takeoff into headwinds. In the monthly trends, Baker Lake had the least number of months with

28 statistically significant trends. Baker Lake experienced an increase of 1.47 km/h in wind speed in

May but experienced a decline of 2.22 km/h in July. Churchill and Inukjuak each had three months of statistically significant changes (at p < 0.05) although all three months in Churchill had statistically significant increasing wind speed while Inukjuak had two months of statistically significant higher wind speed and one month of significantly lower wind speed. Similarly, at

Nitchequon, out of five statistically significant months, one showed lower wind speed and the rest showed higher wind speed. At Kuujjuarapik, there were five months that exhibited statistically significant increases in wind speed. At Schefferville and Wabush, declining wind speed was observed in the months with significant changes in wind speed. But the magnitude of decline at Wabush was less than that of Schefferville. Other months did not exhibit a statistically significant change.

Table 2.1: Student’s T-test for trends in monthly averaged wind speed between 1971 to 1990 and 1991 to 2010 (except Nitchequon, from 1953 to 1969 compared with 1970 to 1985). Bolded numbers indicate trends that are significant at p-value less than 0.001(***), 0.01 (**), 0.05 (*) and 0.10 (^). Location Baker Lake Churchill Period Mean of Mean of ∆ (km/h) Mean of Mean of ∆ (km/h) 1971-1990 1991-2010 1971-1990 1991-2010 wind speed wind speed wind speed wind speed (km/h) (km/h) (km/h) (km/h) January 22.04 22.43 +0.39 23.14 22.19 -0.94^ February 21.50 20.79 -0.72 21.73 21.80 +0.07 March 20.42 20.19 -0.23 20.43 20.80 +0.37 April 18.74 20.21 +1.47* 20.55 20.90 +0.36 May 16.89 16.14 -0.76 19.38 20.63 +1.24** June 17.23 15.02 -2.22*** 18.18 18.35 +0.17 July 17.49 17.69 +0.20 16.55 17.37 +0.82* August 19.02 19.27 +0.25 17.62 18.38 +0.76^ September 21.08 21.71 +0.63 20.49 22.19 +1.69** October 22.54 21.69 -0.85 22.75 23.42 +0.66 November 22.05 22.71 +0.66 22.97 22.54 -0.43 December 22.61 22.69 +0.08 21.48 21.56 +0.08 Winter 20.23 20.41 +0.18 22.13 21.86 -0.27 Spring 17.22 16.28 -0.94*** 20.13 20.78 +0.65*

Summer 20.89 20.91 +0.02 17.44 18.03 +0.59** Fall 20.22 20.05 -0.17 22.08 22.72 +0.64^ Annual 23.66 22.85 -0.81 20.43 20.84 +0.41** Location Inukjuak Kuujjuarapik Period Mean of Mean of ∆ (km/h) Mean of Mean of ∆ (km/h) 1971-1990 1991-2010 1971-1990 1991-2010 wind speed wind speed wind speed wind speed (km/h) (km/h) (km/h) (km/h) January 18.99 19.20 +0.21 17.11 17.17 +0.06 February 18.22 18.45 +0.23 15.25 15.70 +0.45 March 17.61 20.08 +2.47*** 14.91 15.68 +0.77^ April 20.57 20.99 +0.42 15.08 16.16 +1.08** May 20.02 20.55 +0.53 14.84 15.83 +0.99** June 19.19 19.83 +0.64 14.39 15.50 +1.11** July 18.76 19.12 +0.36 14.60 14.08 -0.52 August 19.55 20.08 +0.53 15.95 16.59 +0.64 September 19.80 22.14 +2.34*** 18.68 20.02 +1.34** October 21.90 21.82 -0.08 20.03 19.54 -0.49 November 23.68 22.31 -1.37* 21.46 21.10 -0.36 December 21.08 22.03 +0.95 19.32 20.38 +1.06* Winter 19.47 19.99 +0.52 17.29 17.82 +0.53* Spring 19.38 20.54 +1.16*** 14.93 15.88 +0.95*** Summer 19.17 19.68 +0.51* 14.98 15.38 +0.40^ Fall 21.78 22.09 +0.31 20.06 20.20 +0.14 Annual 19.94 20.60 +0.66*** 16.80 17.31 +0.51*** Location Nitchequon Schefferville Period Mean of Mean of ∆ (km/h) Mean of Mean of ∆ (km/h) 1971-1990 1991-2010 1971-1990 1991-2010 wind speed wind speed wind speed wind speed (km/h) (km/h) (km/h) (km/h) January 13.65 16.41 +2.76*** 16.60 15.59 -1.01* February 14.31 16.11 +1.80** 17.65 15.09 -2.56*** March 13.57 16.15 +2.58*** 17.69 16.09 -1.60*** April 14.82 15.76 +0.94^ 16.98 16.34 -0.64 May 14.14 14.42 +0.28 16.22 15.29 -0.93* June 14.36 14.44 +0.08 16.32 15.73 -0.59^ July 15.11 15.10 -0.01 15.55 13.91 -1.64*** August 16.37 15.30 -1.07* 15.80 14.83 -0.97** September 18.19 17.90 -0.29 16.93 16.59 -0.34 October 17.79 18.13 -0.34 18.23 16.57 -1.66*** November 15.89 17.29 +1.40** 17.88 16.48 -1.40** December 14.23 15.05 +0.82 15.94 15.19 -0.75 Winter 14.06 15.87 +1.81*** 16.70 15.32 -1.38*** Spring 14.17 15.42 +1.25*** 16.98 15.90 -1.08*** Summer 15.30 14.94 -0.36 15.89 14.82 -1.07***

Fall 17.30 17.76 +0.46 17.68 16.54 -1.14*** Annual 15.21 15.99 +0.78*** 16.80 15.65 -1.15*** Location Wabush Period Mean of Mean of ∆ (km/h) 1971-1990 1991-2010 wind speed wind speed (km/h) (km/h) January 14.18 13.88 -0.30 February 14.93 13.67 -1.26** March 16.11 14.51 -1.60*** April 15.39 14.80 -0.59 May 13.83 13.98 +0.15 June 14.26 14.22 -0.04 July 13.82 12.44 -1.38*** August 13.76 12.74 -1.02*** September 15.04 14.56 -0.48 October 16.20 15.05 -1.15*** November 15.44 14.41 -1.03** December 13.31 13.40 +0.09 Winter 14.11 13.66 -0.45^ Spring 15.10 14.42 -0.68** Summer 13.94 13.13 -0.81*** Fall 15.57 14.67 -0.90*** Annual 14.68 13.97 -0.71***

The averaged daily wind speed on monthly and seasonal scales is presented in Table 2.2.

Autocorrelation was present in most of these time series and the significance levels were adjusted by changing the variance to account for the effects of positive autocorrelation (Hamed and Rao, 1998). Baker Lake showed a significant decline in wind velocity in July and summer.

Churchill, Inukjuak and Kuujjuarapik showed an increase in wind speed in most months or seasons that saw statistically significant changes. Nitchequon saw statistically significant increases in wind speed from December to April and in winter and spring. Schefferville and

Wabush had statistically significant slower wind speeds.

Table 2.2: Mann-Kendall test with Theil-Sen slope estimator for change in average daily wind speed (∆km/h) time series in monthly and seasonal scales. Bolded numbers indicate trends that are significant at p-values less than 0.001(***), 0.01 (**), 0.05 (*) and 0.10 (^). Month Baker Lake Churchill Inukjuak Kuujjuarapik January -0.61 -0.42 -0.10 -0.22 February -0.62 -0.48 No change -0.13 March -0.33 +0.56* +1.50*** +0.33^ April -0.16 +0.56* +0.64* +0.46^ May +0.27 +0.60** +0.64*** +0.38^ June -0.42 No change +0.46* +0.39 July -1.08** +0.23 +0.32 -0.22 August No change +0.17 +0.33 +0.23 September -0.21 +0.30 +0.67* +0.15 October -0.30 +0.11 -0.32 -0.57* November -0.28 -0.14 -0.46^ -0.56* December +0.52 +0.06 +0.67*** +0.20 Winter -0.22 -0.28^ +0.20 -0.03 Spring -0.06 +0.58** +0.96*** +0.39* Summer -0.50* +0.13 +0.36** +0.12 Fall -0.25 +0.11 No change -0.31*** Nitchequon Schefferville Wabush January +1.86*** -0.65** -0.05 February +0.71^ -1.06*** -0.38 March +1.45*** -0.84*** -0.61*** April +0.75* -0.27 -0.12 May +0.27 -0.67** +0.07 June +0.33 -0.57** -0.08 July +0.16 -1.11*** -0.67*** August -0.74 -0.78*** -0.45* September -0.08 -0.50* -0.30* October +0.27 -1.18*** -0.52* November +0.80 -0.76** -0.37^ December +0.87* -0.44^ +0.15 Winter +1.17** -0.70*** -0.08 Spring +0.82** -0.61*** -0.22* Summer -0.05 -0.83*** -0.41*** Fall +0.29 -0.81*** -0.40**

2.4.2 Maximum Daily Wind Speed

Using the wind gust threshold of ≥ 28 km/h (Cheng et al., 2014), the maximum daily wind speed was found to have increased significantly around the Hudson Bay region (except at

Baker Lake) and decreased significantly in central Quebec and western Labrador (Table 2.3). On an annual scale, Baker Lake, Nitchequon, Schefferville and Wabush have experienced statistically significantly lower maximum daily wind speed while Churchill and Inukjuak experienced statistically significantly higher maximum daily wind speed. In Churchill, maximum daily wind speed increased statistically significantly at the 95% confidence level (p < 0.05) in three seasons and at the 90% confidence level (p < 0.10) in all four seasons. On the other hand,

Schefferville experienced statistically significant (p < 0.05) slower maximum daily wind speed in three of the seasons. Statistically significant (p < 0.001) lower maximum daily wind speed were found in all four seasons at Nitchequon and Wabush. In Baker Lake, Churchill, Inukjuak and Kuujjuarapik, statistically significant changes (p < 0.05) were observed mostly in the colder months from November to March. In summary, at Baker Lake, Nitchequon, Schefferville and

Wabush’s months with statistically significant changes (p < 0.05) in maximum daily wind speed were all found to be declining. However, the maximum daily wind speed at Nitchequon and

Wabush declined much more rapidly than at Baker Lake and Schefferville. Whereas, all the statistically significant (p < 0.05) monthly changes at Churchill and Inukjuak were increases in maximum daily winds.

Table 2.3: T-test for trends in maximum daily wind speed (≥ 28 km/h) between 1971 to 1990 and 1991 to 2010 (except Nitchequon, from 1953 to 1969 compared with 1970 to 1985). Bolded numbers indicate trends that are statistically significant at p-value less than 0.001(***), 0.01 (**), 0.05 (*) and 0.10 (^). Location Baker Lake Churchill Period Mean of Mean of ∆ (km/h) Mean of Mean of ∆ (km/h) 1971-1990 1991-2010 1971-1990 1991-2010 max daily max daily max daily max daily wind wind wind wind (km/h) (km/h) (km/h) (km/h) January 45.18 44.20 -0.98 39.16 39.65 +0.49 February 44.13 43.94 -0.19 37.41 39.81 +2.40*** March 42.89 41.16 -1.73* 38.64 38.05 -0.59

April 40.69 40.14 -0.55 36.94 38.14 +1.20^ May 38.17 39.02 +0.85 35.35 37.03 +1.68** June 35.75 35.54 -0.21 34.16 34.32 +0.16 July 36.09 35.69 -0.40 34.55 34.99 +0.44 August 36.75 36.20 -0.50 35.83 37.27 +1.44* September 37.37 37.30 -0.07 38.56 40.35 +1.79* October 39.57 39.39 -0.18 40.76 41.32 +0.56 November 43.81 41.61 -2.20** 40.36 40.44 +0.08 December 42.92 43.00 +0.07 39.59 40.11 +0.52 Winter 44.11 43.69 -0.42 38.71 39.86 +1.15** Spring 40.74 40.17 -0.58 36.95 37.75 +0.80* Summer 36.21 35.82 -0.39 34.86 35.53 +0.67* Fall 40.43 39.55 -0.88* 39.95 40.71 +0.76^ Annual 40.70 40.12 -0.58** 37.83 38.61 +0.78*** Location Inukjuak Kuujjuarapik Period Mean of Mean of ∆ (km/h) Mean of Mean of ∆ (km/h) 1971-1990 1991-2010 1971-1990 1991-2010 max daily max daily max daily max daily wind wind wind wind (km/h) (km/h) (km/h) (km/h) January 38.06 39.27 +1.21^ 36.29 35.21 -1.08^ February 38.05 38.35 +0.30 35.36 35.60 +0.24 March 37.58 39.77 +2.19** 34.95 35.45 +0.50 April 39.52 39.02 -0.50 35.82 34.99 -0.83 May 37.15 37.26 +0.11 33.96 34.37 +0.41 June 35.44 35.97 +0.53 34.42 34.82 +0.40 July 35.43 35.53 +0.10 34.86 34.17 -0.69 August 36.36 37.01 +0.65 35.84 35.09 -0.75 September 38.97 39.90 +0.93 36.59 37.95 +1.36* October 40.25 40.23 -0.02 37.71 37.75 +0.04 November 41.82 41.08 -0.74 38.08 39.56 +1.48* December 39.40 41.20 +1.80* 36.86 38.53 +1.67** Winter 38.53 39.73 +1.20** 36.27 36.74 +0.47 Spring 38.13 38.63 +0.50 34.92 34.96 +0.04 Summer 35.76 36.17 +0.41 35.10 34.73 -0.37 Fall 40.41 40.41 No 37.53 38.43 +0.90* change Annual 38.32 38.80 +0.48* 36.16 36.45 +0.29 Location Nitchequon Schefferville Period Mean of Mean of ∆ (km/h) Mean of Mean of ∆ (km/h) 1971-1990 1991-2010 1971-1990 1991-2010 max daily max daily max daily max daily wind wind wind wind (km/h) (km/h) (km/h) (km/h) January 37.74 36.30 -1.44 36.99 35.68 -1.31*

February 37.03 36.62 -0.41 37.09 35.72 -1.37^ March 35.27 35.57 +0.30 36.27 35.40 -0.87 April 38.19 34.44 -3.75*** 35.70 34.64 -1.06* May 34.66 34.15 -0.51 34.60 34.69 +0.09 June 35.99 34.22 -1.77* 33.86 34.36 +0.50 July 36.24 33.20 -3.04*** 33.80 33.38 -0.42 August 36.59 33.46 -3.13*** 33.85 32.90 -0.95* September 38.58 37.08 -1.50^ 34.96 34.77 -0.19 October 37.16 35.46 -1.70** 36.01 35.61 -0.40 November 38.10 35.61 -2.49** 37.26 35.76 -1.50** December 38.22 34.84 -3.38*** 36.38 35.30 -1.08^ Winter 37.75 35.95 -1.80*** 36.82 35.59 -1.23** Spring 36.00 34.77 -1.23*** 35.55 34.90 -0.65* Summer 36.30 33.58 -2.72*** 33.83 33.57 -0.26 Fall 37.94 36.06 -1.88*** 36.10 35.35 -0.75* Annual 37.05 35.14 -1.94*** 35.59 34.88 -0.71*** Location Wabush Period Mean of Mean of ∆ (km/h) 1971-1990 1991-2010 max daily max daily wind wind (km/h) (km/h) January 34.68 32.27 -2.41*** February 34.53 31.60 -2.93*** March 33.85 31.62 -2.23*** April 33.63 31.43 -2.20*** May 32.69 30.96 -1.73*** June 32.29 30.87 -1.42*** July 32.51 30.47 -2.04*** August 32.08 30.04 -2.04*** September 33.32 30.93 -2.39*** October 33.75 31.79 -1.96*** November 34.33 31.84 -2.49*** December 34.23 31.65 -2.58*** Winter 34.48 31.89 -2.59*** Spring 33.42 31.35 -2.07*** Summer 32.29 30.50 -1.79*** Fall 33.80 31.50 -2.30*** Annual 33.47 31.33 -2.14***

A time series analysis was conducted to determine the trend in the number of days when the maximum wind speed exceeded 28 km/h (Table 2.4). Baker Lake and Kuujjuarapik did not

35 show any statistically significant trends over time. Churchill and Nitchequon had statistically significantly more days with the maximum wind speed over 28 km/h while Schefferville had significantly less, annually. Churchill and Inukjuak both had a statistically significant increasing number of days with maximum wind speed over 28 km/h in the spring and fall. Nitchequon recorded a statistically significant increasing number of days with maximum wind speed greater than 28 km/h in all four seasons at the 90% confidence interval (p < 0.10) but only the spring showed statistical significance at the 95% confidence interval (p < 0.05). The annual number of daily maximum wind speed days, from 33.47 to 31.33 days, declined the most at Schefferville where all four seasons had a statistically significant lower number of days with maximum wind speed greater than 28 km/h over time (p < 0.05).

Table 2.4: Trends for days with maximum daily wind speed ≥ 28 km/h over time. Bolded numbers indicate trends that are statistically significant for the Mann-Kendall test at p-value less than 0.001(***), 0.01 (**), 0.05 (*) and 0.10 (^). Location Baker Lake Churchill Inukjuak Kuujjuarapik Period Number ∆ (days Number ∆ (days / Number ∆ (days Number ∆ (days of days / year) of days year) of days / year) of days / year) Winter 69.4 -0.03 55.3 No 52.4 +0.25 47.1 No change change Spring 62.8 +0.08 54.0 +0.38** 54.5 +0.53* 40.0 +0.21 Summer 49.4 -0.08 46.6 +0.29 53.8 +0.38 39.5 +0.04 Fall 61.9 +0.12 61.8 +0.19* 64.4 +0.19* 59.6 -0.11 * Annual 243.4 +0.02 217.6 +0.89** 225.1 +1.00 186.1 +0.10 Location Nitchequon Schefferville Wabush Winter 37.9 +0.45^ 42.9 -0.42*** 34.6 -0.29 Spring 37.9 +0.67* 45.0 -0.25* 40.2 -0.28 Summer 40.0 +0.40^ 42.5 -0.54*** 36.3 -0.57* Fall 48.3 +0.46^ 47.5 -0.42** 38.8 -0.31^ Annual 163.7 +2.00** 177.8 -1.54*** 149.9 -1.59^

2.4.3 Time of Daily Maximum Wind Speed

An analysis was performed to examine which hours the maximum daily wind speed were most likely to be observed (Figure 2.2). If there was a tie for maximum wind gust for more than one hour in a day, each of the hours that was tied was considered as the hour that maximum wind speed was reached. Maximum wind occurred mostly overnight from midnight to 2 am (all local time) and also during the afternoon from 1 pm to 5 pm. Morning (6 am to 10 am) and evening (7 pm to 10 pm) were 20% less likely to encounter daily maximum wind than in the afternoon and about 70% less likely than overnight.

(a) Baker Lake

(b) Churchill

(d) Kuujjuarapik

(e) Nitchequon

(f) Schefferville

(g) Wabush

Figure 2.2: Hour of the day when daily maximum wind speed was recorded from 1971-2010 at (a) Baker Lake, (b) Churchill, (c) Inukjuak, (d) Kuujjuarapik, (e) Nitchequon, (f) Schefferville and (g) Wabush.

2.4.4 Calm Wind

With the exception of Nitchequon and Baker Lake, all of the other locations had a much lower percentage of calm winds from 1991-2010 compared to 1971-1990 (Figures 2.3 to 2.9).

The greatest reductions in calm winds were observed at Churchill, Kuujjuarapik and

Schefferville, where a third of the winds were no longer classified as calm in 1991-2010. There was a great range of calm wind frequency from 1.23% at Churchill to 14.55% at Wabush (Table

2.5). Similarly, true calms were the lowest at Churchill and highest at Wabush. Churchill and

Inukjuak had the least number of hours of calm wind and they also had the least amount of true calm winds. Other locations had over 50% of calm winds classified as incidental.

Figure 2.3: Wind rose for Baker Lake from (a) 1971-1990 and (b) 1991-2010.

Figure 2.4: Wind rose for Churchill from (a) 1971-1990 and (b) 1991-2010.

Figure 2.5: Wind rose for Inukjuak from (a) 1971-1990 and (b) 1991-2010.

Figure 2.6: Wind rose for Kuujjuarapik from (a) 1971-1990 and (b) 1991-2010.

Figure 2.7: Wind rose for Nitchequon from (a) 1953-1969 and (b) 1970-1985.

Figure 2.8: Wind rose for Schefferville from (a) 1971-1990 and (b) 1991-2010.

Figure 2.9: Wind rose for Wabush from (a) 1971-1990 and (b) 1991-2010.

Table 2.5: Frequency of calm winds at each location. True calms and incidental calms are subsets of total calm and their percentages represent their portions of all calm observations. Location Total calm (%) True calms (%) Incidental calms (%) Baker Lake 9.44 61.5 38.5 Churchill 1.23 38.2 61.8 Inukjuak 4.01 49.5 50.5 Kuujjuarapik 5.27 52.1 47.9 Nitchequon 5.26 59.3 40.7 Schefferville 5.97 54.6 45.4 Wabush 14.55 67.7 32.3

2.4.5 Wind Direction

From the wind roses, westerly winds were frequently dominant at each location (Figures

2.3 to 2.9). Baker Lake and Churchill had dominant winds from the due west, northwest and due north (Figures 2.3 and 2.4). At Inukjuak, wind primarily came from due west and northeast

(Figure 2.5). Further south, Kuujjuarapik’s winds came most frequently from due north, southeast, due south and due west (Figure 2.6). Nitchequon’s dominant winds came from due south, southwest, due west and northwest (Figure 2.7). At Schefferville, winds primarily came

44 from northwest and south-southeast (Figure 2.8). At Wabush, winds mainly came from due west and due south (Figure 2.9). Aside from Nitchequon, each location experienced higher magnitude of wind speed in the prevailing wind directions during the 1991-2010 period than during the

1971-1990 period (Figures 2.3 to 2.6, 2.8 to 2.9). Nitchequon also showed a shift in predominant wind direction from southwest and northwest wind in 1971-1990 towards mainly westerly wind in 1991-2010 (Figure 2.7).

Using LOWESS curves and Mann-Kendall test, significant change in wind direction were found in some of the locations. Since wind direction is a vector, apparent trends and the direction may actually be artificial. To reduce the biased perception of trends associated with vector variable, 360o was added to shift the wind direction points from 10o to 140o. The 150o was chosen as a cut-off point because all of the locations showed that wind were not generally coming from this direction (Figures 2.3 to 2.9). As seen in Gearheard et al. (2010)’s wind patterns study at a community further north in Clyde River, Nunavut, they also used a similar cut-off point to present graphically due to the similarities in prevailing wind directions. Using

Nitchequon as an example, the monthly wind direction would appear to be changing from 100o in the 1950s to 250o in mid-1960s and stabilized afterwards (Figure 2.10a). After adjusting and accounting for the characteristics of a vector variable, the trend for wind direction was changing from about 350o in the 1950s to 270o in the mid-1960s, followed by a slower change in direction in the late1960s towards 250o by the 1980s (Figure 2.10b). The adjusted wind direction plots for the remaining locations are presented in Figure 11. Mann-Kendall test showed that Baker Lake and Nitchequon experienced statistically significant changes (p < 0.001) in wind direction while other locations did not (Table 2.6). Baker Lake’s wind direction significantly changed from 360o

(or 0o) to 340o.

Table 2.6: Mann-Kendall test on wind direction. Statistically significant values are bolded with p-value less than 0.001(***), 0.01 (**), 0.05 (*) and 0.10 (^). Location Mann-Kendall significance value Baker Lake 8.3 X 10-5*** Churchill 0.37 Inukjuak 0.26 Kuujjuarapik 0.57 Nitchequon 1.25 X 10-9*** Schefferville 0.87 Wabush 0.26

Figure 2.10: LOWESS curve for Nitchequon showing the (a) perceived trend and (b) actual trend in monthly wind directional change.

(a) Baker Lake

(b) Churchill

(d) Kuujjuarapik

(e) Schefferville

(f) Wabush

Figure 2.11: LOWESS curves for wind direction at (a) Baker Lake, (b) Churchill, (c) Inukjuak, (d) Kuujjuarapik, (e) Schefferville and (f) Wabush.

Percentage changes in wind speed and wind direction were quantified between the 1971-

1990 and 1991-2010 periods (1953-1969 vs. 1970-1985 for Nitchequon) from Figures 2.3 to 2.9 by examining both variables together. Donut charts for each location were created (Figure 2.12).

In the donut charts, wind speed in 5-knot increments were plotted with the direction from the inner circle towards the outer circle. Areas shaded in darker red indicate the greatest increase in percentage and areas shaded in darker blue indicate the greatest decrease in percentage.

Nitchequon data in the latter period was converted from 36 points (in 10s of degrees) into 16 points of the compass to allow for comparison because of data recording format in the earlier period.

At Baker Lake, there was a large percentage increase of winds between 5 and 20 knots coming from the south, southwest, west and northwest. A drop in <10-knot winds in the west and northwest and >5-knot winds were observed in northeast and east directions. At Churchill, winds under 5 knots were drastically increased in almost all directions, but in particular from 50o to 90o and 170o to 360o. Greatest percentage increase in winds was observed at ≥15 knots in all quadrants except in the southwest and northeast. At Inukjuak, the wind increased the greatest in the east, south and southwest directions while decreased the most in the <15-knot category in the northeast and southeast directions. At Kuujjuarapik, winds <10 knots were mostly increased in all directions except from 0o to 30o. Winds >25 knots appeared to be more frequently observed from 0o to 160o, except between 50o and 80o in which there was no change. At Nitchequon, winds became more frequently coming from northeast, southeast and west directions especially at ≥20 knots. At Schefferville, winds >15 knots were generally reduced by at least 30%. Winds from most directions which were <10 knots appeared to have increased. At Wabush, winds >20

51 knots were also reduced by about 30% or more. There was an increased percentage of winds between 5 and 15 knots from the south to northeast.

a) Baker Lake b) Churchill

c) Inukjuak d) Kuujjuarapik

52 e) Nitchequon f) Schefferville

g) Wabush

Legend +1 to +29% -1 to -29% +30 to +59% -30 to -59% ≥ +60% ≥ -60% No change

Figure 2.12: The percentage change in wind speed, in five-knot increments, and direction at (a) Baker Lake, (b) Churchill, (c) Inukjuak, (d) Kuujjuarapik, (e) Nitchequon, (f) Schefferville and (g) Wabush. Red represents an increase in percentage and blue represents a decrease in percentage of winds from that direction.

2.4.6 Aviation Perspective

Based on weekday scheduled departure and arrival times from the airlines serving these communities as of July 1, 2017 (Table 2.7). There were no scheduled flights after 10 pm or before 8 am. Wabush had the highest total of scheduled passenger flights in a day while Inukjuak had the highest proportion of flights within a single hour (3 departures and 2 arrivals between 2 to 3 pm). Nitchequon was not included since the airport was closed in 1985 and no historical flight schedules were available. The types of aircraft being used were also considered. Other than

Churchill, all of the other airports were served by turboprop aircraft (Table 2.8). Churchill was served by both turboprop and jet aircraft. Wabush was served by the greatest variety of turboprop aircraft. All of the regularly scheduled aircraft flying to these communities are classified as light or medium weight according to the International Civil Aviation Organization (ICAO) definition.

Table 2.7: Aircraft movements based on weekday scheduled passenger flights as of July 1, 2017. Numbers represent the amount of flights arriving or departing within that hour. Local Baker Lake Churchill Inukjuak Kuujjuarapik Schefferville Wabush Time Dep. Arr. Dep. Arr. Dep. Arr. Dep. Arr. Dep. Arr. Dep. Arr. 0:00 1:00 2:00 3:00 4:00 5:00 6:00 7:00 8:00 1 9:00 1 1 1 1 10:00 2 1 2 11:00 2 1 2 2 12:00 2 1 1 1 13:00 1 2 14:00 1 1 3 2 1 1 1 15:00 1 16:00 2 3 1 17:00 1 1 18:00 19:00 1 1 20:00 1 2 1 21:00 1 22:00 1 23:00

Table 2.8: Types of aircraft flown to serve the communities in the study. Location Scheduled aircraft types Jet Turboprop Baker Lake None ATR-42, ATR-72 Churchill Boeing 737-200, Fairchild ATR-42, ATR-72 Dornier 328JET Inukjuak None Beechcraft King Air, de Havilland Canada Dash 8-300 Kuujjuarapik None de Havilland Canada DHC-6 Twin Otter, de Havilland Canada Dash 8-100, de Havilland Canada Dash 8-300 Schefferville None Beechcraft King Air, de Havilland Canada Dash 8-300 Wabush None Beechcraft 1900D, Beechcraft King Air, British Aerospace Jetstream 32, de Havilland Canada Dash 8-100, de Havilland Canada Dash 8-300, Swearingen Metroliner

All of the runways in these communities are bi-directional and have no restrictions on using a particular direction due to terrain or other factors. Even though only Baker Lake and

Nitchequon experienced significant changes in wind direction over time, results showed that all but Baker Lake and Churchill airports’ runways were not aligned with the prevailing wind (Table

2.9).

Table 2.9: Prevailing wind direction and runway configuration. Location Runway direction (in o) Prevailing crosswind? Baker Lake 160 / 340 No Churchill 70 / 250, 150 / 330 No Inukjuak 70 / 250 Yes Kuujjuarapik 30 / 210 Yes Nitchequon 30 / 210 Yes Schefferville 180 / 360 Yes Wabush 180 / 360 Yes

2.5 Discussion

2.5.1 Average Wind Speed

While the actual change in daily average wind speed was minimal (less than 1 km/h of increase or decrease) on an annual scale, these values approached the threshold value of 18.5 km/h (10 knots). If the wind speed exceeds this threshold value, planes are required to take off and land into the wind. Failure to assign proper runway direction based on the wind speed will lead to an increased risk for pilots and passengers. Given that the values in Table 2.1 were daily averaged wind speed, a significant increase in wind speed at Churchill, Inukjuak, Kuujjuarapik and Nitchequon meant some hourly winds had crossed the threshold value in the 1991-2010 period. This observation also holds true on seasonal and monthly scales. The above-average wind speed appeared to be linked to above-average surface air temperature in the Canadian

Arctic maritime ecosystem (Steiner et al., 2015), which was consistent with the increase in wind speed at the sites around the Hudson Bay in this study. Schefferville, on the other hand, had significant decreases in average wind speed that would allow planes to take off and land in either direction of the runway more often. Wabush was least susceptible to the risks associated with changes in wind speed of the seven locations examined given that their daily average wind speeds were all below the 18.5 km/h threshold value.

Between the average wind speed results from Student’s t-test (Table 2.1) and Mann-

Kendall test (Table 2.2), they both presented similar findings on trends and significance levels although minor differences did exist. For example, at Baker Lake, both tests showed a significant decline in wind velocity in July and summer, yet t-test also identified a significant increase in wind velocity in May that Mann-Kendall test failed to identify. On the other hand, the Mann-

Kendall test sometimes found significant trends that t-test did not find. For example, Mann-

Kendal found a significant increase in wind velocity from March to May but t-test only found similar findings for May within those three months. These differences were minor because all of the trends that were significant in both tests were in agreement with each other in terms of direction, but the magnitudes were slightly different. If there was disagreement for significance between Mann-Kendall test and t-test, t-test significance should be used because Mann-Kendall test compares the slope in pairs between the data points to determine the direction of trends and

Mann-Kendall test has less statistical power than t-test. Although test for normality would often fail in large data and thus using Mann-Kendall test would be preferred over t-test, the wind speed dataset is sufficiently large and each hourly observation could be interpreted as randomly sampled in a way that is no different than the current hour’s temperature being somewhat similar to the past hour, the overall dataset could be considered to be approaching normal distribution as defined by central limit theorem. Central limit theorem states that the sample mean approaches the population mean when the sample size is sufficiently large, even if the dataset is not normally distributed (e.g. skewed or the presence of outliers). Student’s t-test gains more statistical power as the sample size grows. In fact, it has been suggested that t-test would be applicable regardless of the distribution if the sample size is sufficiently large (Lumley et al., 2002). Therefore, in any discrepancy related to significance levels between Mann-Kendall and Student’s t-test, the t-test is considered to be more accurate as t-test combined with central limit theorem bypasses the issue of normality.

Positive autocorrelation was widespread in the daily averaged wind data which suggested that the day-to-day variability of wind speed was rather low. The significance levels in Table 2.2, after taking positive autocorrelation into consideration, were conservative because the autocorrelation was calculated to lag 30 (days) for monthly data and lag 90 (days) for seasonal

58 data. Many data points, particularly those near the end of each month or season, were compared with subsequent data points in the same month or season in the next year. This approach included subsequent data points that were very unlikely to be autocorrelated. Any perceived autocorrelation between data at the end of one month/season and the data at the beginning of the same month/season in the following year are likely by chance and not because true autocorrelation exists between the two values. While there are other approaches, such as pre- whitening, to identify and remove autocorrelation, this process removes some of the trend in the dataset while the modified Mann-Kendall test (Hamed & Rao, 1998) does not.

In a similar study, Wan et al. (2010) used a different time frame to examine wind speed across Canada, five of the stations from this study (Baker Lake, Churchill, Kuujjuarapik,

Schefferville and Wabush) were also examined. Comparing the long-term seasonal average wind speed between the two studies, both achieved virtually identical results. On a few occasions, seasonal average wind speed in this study showed higher values that could be attributed to a difference in the months used to characterize the seasons between the two studies. Wan et al.

(2010) also examined the trends of wind speed and found a significant month-to-month decline in wind speed at Kuujjuarapik and Wabush and an increase at Schefferville. This finding was surprising given that Schefferville showed slower wind speed in all of the categories in this study, for wind speed, wind gust and the number of gust days. These differences could be explained by the time scale and method used to assess significant changes. Wan et al. (2010) used monthly mean time series data to examine homogenized wind speed from 1953 to 2006, whereas this study used Student t-tests and examined daily non-homogenized data from 1971 to 2010.

Nitchequon’s data would be more similar to their study (from 1953 to 1985), but their study did not include this location.

Spatial trend differences were observed between the Hudson Bay and Labrador regions.

Churchill, Inukjuak and Kuujjuarapik are located on the shores of Hudson Bay and each location showed a generally increasing trend in average wind speed. On the other hand, Schefferville and

Wabush are both in the Labrador region and both locations recorded a clear decline in average wind speed. Baker Lake, although located near to Hudson Bay, did not show the clear trends that other sites around the Hudson Bay exhibited. This may have to do with the fact that Baker Lake was much further north and inland than all of the sites. Nitchequon served as a representative location in central Quebec that wind speed in the area was increasing. However, the difference in the sampling period must be taken into consideration. Since Labrador locations had a declining trend in average wind speed and less calm wind, this suggested that the distribution of wind speed at Schefferville and Wabush appeared to move towards the mean and away from the extremely high wind speed and calm wind. On the other hand, the Hudson Bay locations had less calm wind and an increase in average wind speed which suggested that these sites would become more windy with a general upward shift in wind speed. It can be seen that wind speed exceeding

20 km/h (in dark blue, dark green and light blue colours) was more prevalent in 1991-2010

(Figures 2.3b, 2.4b and 2.5b) than in 1971-1990 (Figures 2.3a, 2.4a and 2.5a).

There could be multiple factors that could explain the variability in trends between different locations. For example, the presence of warm air mass has shifted northwards and became more frequent (Leung & Gough, 2016). A smaller temperature difference between the tropics and the poles caused weaker jet stream (Coumou et al., 2015), which led to greater meandering of the Rossby waves (Mann et al., 2017) and shifting the wind patterns. Others factors could be related to geography. For example, the latitude of the location determines the theoretical maximum duration of sunshine hours and the time of sunrise and sunset. Both played

60 a role in convection and creating atmospheric instability. Also, the differences and timing of delayed freeze-up and accelerated break-up of sea ice between western and eastern Hudson Bay

(Gagnon & Gough, 2005; Gagnon & Gough, 2006) could also lead to difference in wind speed trends due to onshore winds from the unfrozen portions of the bay (Hanesiak & Wang, 2005).

In a study that examined indigenous traditional knowledge towards wind at a northern

Canadian community, Clyde River in Nunavut, local Inuit observed stronger winds in summer and winter and that strong winds in winter tend to last longer (Gearheard et al., 2010). At Sachs

Harbour, Northwest Territories, indigenous people described winds in the region were changing in direction and velocity and that wind storms were becoming more intense than in the past

(Riedlinger & Berkes, 2001). This study’s data showed that four communities, Baker Lake,

Inukjuak, Kuujjuarapik and Nitchequon experienced stronger winds in winter, the latter two at statistically significant levels at p < 0.05 (Table 2.1).

Stronger winds offer an opportunity for many off-grid communities in remote and northern settlements to reduce their greenhouse gas emissions from electricity generation by switching from diesel fuel to renewable energy by wind (Arriaga et al., 2013). According to

Arriaga et al. (2014), Nunavut’s electrical generators were almost entirely running on diesel.

Increasing wind speed at Baker Lake could be beneficial to future wind energy projects at this location. Churchill would not benefit as much from higher wind speeds because the power transmission line for the town is connected to electrical grid to the south and powered by hydroelectric. Most of the other study communities also would not benefit from the increased potential of higher wind speeds. Nitchequon is an abandoned town and therefore has no energy demands. Schefferville and Wabush already use hydroelectric power (Nalcor, n.d.) while

Inukjuak has rejected a wind turbine project in favour of a hydroelectric project (Pollon, 2017).

Kuujjuarapik is proposing a new power plant with a biomass power generator and three wind turbines to replace the existing diesel generator (Natural Resources Canada, 2018). The new power plant will generate more electricity under greater wind conditions. However, excessively high wind speeds could force the turbines to be shut down for safety reasons. In addition, it was possible that the increase in wind speed were coming from a direction that the turbines were not facing. Also, the capacity and cost of battery storage could also limit the amount of energy generated by wind power to be stored for future use. Seasonality was also a factor. Ideally, highest winds in winter would be most beneficial to wind-generated electricity because demands from energy and heating were highest in winter. This was not the case for some communities

(Table 2.1). In any of these cases, the community might not fully benefit from the increased energy output through higher winds.

2.5.2 Maximum Wind Speed

Spatial differences between different geographic regions were found. In western Hudson

Bay, the daily maximum wind was 33% to 80% more prominent around midnight (12 am to 1 am) than during the day from 7 am to 5 pm (Figures 2.2a and 2.2b). However, in eastern Hudson Bay, the daily maximum wind in the afternoon was observed just as frequently as during the overnight period (Figures 2.2c and 2.2d). Further east at Nitchequon, the maximum wind was most often observed in the afternoon, especially at 2 pm, than at any other time (Figure 2.2e). At

Schefferville and Wabush (Figures 2.2f and 2.2g), the maximum wind was 1.5 to 2 times more likely to be observed during the afternoon period than the morning, evening and overnight periods. The diurnal pattern for maximum wind speed could be explained by sunrise which caused differential heating between the land and Hudson Bay. In the morning, the land heats up more quickly than the water because water has higher latent heat capacity. The result caused air

62 above the land to rise and forced the wind from the Bay to move towards inland. With a greater temperature gradient, the speed of onshore wind increases. The night time increase in wind speed could be due to offshore wind when the wind pattern was reversed. The largest differential heating between land and water occurred when the Hudson Bay was not frozen in sea ice (from around June to November). Since sunset takes place very late in northern latitudes in summer, this could explain the increase in maximum wind speed around midnight.

The three Hudson Bay locations, Churchill, Inukjuak and Kuujjuarapik experienced an increase in maximum daily wind speed, statistically significant on an annual scale (p < 0.05) at the first two locations (Table 2.3). Seasonal changes were very strong at Churchill, Nitchequon,

Schefferville and Wabush. Interestingly, Baker Lake and Nitchequon had lower maximum gust wind speed in 1981-2010 but faster average wind speed in the months that observed significant trends. At the same time, Nitchequon had more days where maximum wind speed greater than

28 km/h was observed (Table 2.4). This suggested that the variances of wind speed lowered as the overall wind speeds at this location were increasing while the frequencies of upper extreme gusts were reduced. This is contrast to a study by Bartlett et al. (2003) who found that as mean annual maximum wind speed increased, the standard deviation of this variable also increased.

Using five decades of historic wind gusts data, Cheng (2014) found that daily wind gusts at ≥ 50 km/h were significantly correlated with positive daily temperature anomalies and negative daily pressure anomalies in the Hudson Bay and northern Labrador regions. This suggested that localized convective wind storms could be more frequent, especially in summer, due to favourable conditions for their development. A shorter time-period of hourly wind gusts from 1994 to 2009 plus eight GCMs in IPCC (2007) Fourth Assessment Report (AR4)’s Special

Report on Emissions Scenarios (SRES) A2 and B1 scenarios generating future projections up to

2100 were analyzed by Cheng et al. (2014). Their findings are somewhat in disagreement with this study’s results. Between this study and their study, the stations in common were Baker Lake,

Churchill, Kuujjuarapik, Schefferville and Wabush. They found neutral to an increase in wind gust events in the stations around the Hudson Bay and northern Labrador regions across the four seasons. They also stated that wind gusts occurred least frequently in summer across a number of the stations in this study. However, this study reported that gust speeds were either decreasing or had no significant trends at each location except Churchill (Table 2.3). The number of gust days were also decreasing or not showing significant trends at each station other than Nitchequon

(Table 2.4). The results of this study suggested that the direction of observed wind gust changes was contrary to the GCM outputs because the AR4 GCMs projected an increase in wind gust while this study had observed a decrease in this attribute. However, Cheng et al. (2014) did not examine Inukjuak nor Nitchequon. In addition, this study grouped all of the Hudson Bay stations into one region and Schefferville and Wabush were grouped into the Labrador region. Whereas,

Cheng et al. (2014) assigned Schefferville into their Hudson Bay regional group while Wabush and the area represented by Nitchequon were grouped into northern Ontario and the central

Quebec which included stations much further west and south, such as Kenora, Thunder Bay and

Timmins in Ontario. Given the close proximity, similar trends in wind speed and wind gust and even the peak hours when the maximum daily wind was recorded were similar in Schefferville and Wabush (Figures 2.2f and 2.2g), these two stations should be grouped together.

This study’s findings also somewhat in disagreement with the findings of Eichelberger et al. (2008) as their projections suggested that wind speed in eastern Hudson Bay would be increasing faster than western Hudson Bay. This study found that Churchill’s historical wind speed was increasing at a similar rate as Inukjuak and Kuujjuarapik, the two locations

64 representing eastern Hudson Bay. However, the projected increase in wind speeds is around 0.7 km/h at Inukjuak and Kuujjuarapik (Eichelberger et al., 2008), which were similar to Inukjuak and Kuujjuarapik’s historical increase (Table 2.2). Wind speeds at Schefferville and Wabush were anticipated to be unchanged or slightly faster (Eichelberger et al., 2008; Ma et al., 2016) but this study found an overwhelming reduction in historical wind speed analyzed with different metrics. Thus, there is a discrepancy in trends between historical records and future projections for the area around Schefferville and Wabush.

The greatest focus of safety is on Churchill Airport, where four of the months and three of its seasons had significantly increasing maximum daily winds. Churchill Airport is of high importance to other local airports as it serves as the transit hub for passengers and freight entering and leaving Kivalliq Region. Intense gust winds could potentially damage aircraft and airport infrastructure (Budd & Ryley, 2012). A European aviation climate impact study found that northwestern, central and eastern Europe would encounter more damages to infrastructure from increased winds and storms (Burbidge, 2017). In Canada, the Inuvik Airport much further northwest of Churchill had part of its roof partly torn off in a blizzard in 2012 (CBC, 2012).

Looking at the number of days with gust wind, Churchill and Inukjuak had significant increases in spring and fall while Baker Lake and Kuujjuarapik did not have significant trends. While surface winds are highly localized, the observed changes may tie in with sea ice conditions and distribution (van der Linden et al., 2017). Wind speed is reduced by the presence of sea ice as the sea ice imposes a drag on the wind. When Kowal et al. (2017) examined the changes in freeze-up and break-up dates in sea ice in that region they found that changes off the coast of Churchill and

Inukjuak were greater than off the coast of Kuujjuarapik. The significant changes in freeze-up and break-up dates coincide with the timing where significant increases in the surface wind were

65 observed. Increase in wind speed would lead to greater storm surges. Coupled with changes to wind direction and global sea level rise, storm surges could lead to capacity loss, delays, or rendering the airport unusable (Burbidge, 2017).

2.5.3 Calm Wind

Calm wind declined from the 1971-1990 to the 1991-2010 periods at five of the seven locations investigated (Figures 2.3 to 2.9), even at Schefferville and Wabush where average hourly and daily maximum wind speed were decreasing (Tables 2.1 to 2.4). Overall, the calm wind results were in agreement with another northern Canadian community’s study, which also found strong, significant decline in calm winds in each month (Gearheard et al., 2010). The percentage of calm wind belonging to true calm in this study (Table 2.5) was far lower than

Cook’s (2015) study at Adelaide, Australia. He found that true calms accounted for 94% of all calm observations in Adelaide whereas this study’s true calms only accounted for about one- third to two-thirds of total calm. Variation in geographic location may explain this difference.

Actual calm wind conditions could be less often than what was observed. As mentioned in

Chapter 2.3.5, hourly wind speeds were measured in the last one to two minutes of each hour.

Wind speed could not be measured if the wind happened to be calm over the last two minutes of the hour while present in the rest of the hour. Very light winds would also fail to be measured within the last two minutes if the force was not sufficient to push and rotate the cup anemometer.

Fewer calm wind conditions are more beneficial to wind power as it keeps the turbine rotating to generate electricity. Even though Churchill had the least frequent and Schefferville had the most frequent calm winds out of the seven locations in this study (Table 2.5), wind power potential at both locations plus Schefferville were not considered because these

66 communities already had its electricity generated by hydropower (Nalcor, n.d.). Nitchequon currently does not have energy demand. Inukjuak is most ideal to have wind generators since it experienced calm winds the least and had the lowest true calm percentage in the three remaining locations.

2.5.4 Wind Direction

The variability of wind direction is inversely related to the average wind speed across different types of surface on a small time scale (Mahrt, 2011). Since wind directional data in this study were recorded in a two-minute interval at the end of every hour, it would not reveal the relationship between wind direction and average wind speed in a way that Mahrt’s (2011) suggested. Only Baker Lake and Nitchequon showed a significant change in wind direction but some of the months had significantly faster mean wind speed while other months had lower

(Table 2.1). These changes were also far more unclear compared to other locations. Wind direction results from this study appeared to agree with other wind studies in Clyde River,

Nunavut and Sachs Harbour, Northwest Territories (Gearheard et al., 2010; Riedlinger & Berkes,

2001). People living at Clyde River claimed to observe shifting wind direction especially after

1990, yet wind direction analysis disagreed with their anecdote (Gearheard et al., 2010).

Aboriginals living in Sachs Harbour, Northwest Territories also noted a general change to wind direction (Riedlinger & Berkes, 2001). Surprisingly, annual maximum daily wind speed (Table

2.3) appeared to better explain the significant changes in wind direction because both locations had recorded less wind gust speed over time.

On a synoptic scale, this study has found that DP air masses from the Arctic polar region were reaching the Hudson Bay less frequently year-round (Leung & Gough, 2016). This

67 reduction was replaced by MP air mass originating from unfrozen waters further south. This study found additional evidence of such observation in the Hudson Bay region. At Baker Lake

(Figure 2.12a), southern winds of all magnitudes were becoming more frequent and at the expense of northern, northeastern and eastern winds. Wind speed < 15 knots (1 knot = 1.852 km/h) in the south and southwest increased by 1% to more than 60%. Western and northwestern winds from 15 to 25 knots were increasing by at least 30%. Winds from 5 knots or higher from the northeast and east declined by more than 30%. Churchill’s wind speed was increasing, as seen by an increase of wind speed recorded at > 5 knots in nearly all directions (Figure 2.12b).

Winds with speed > 20 knots from southwest, northwest to northeast at Churchill increased by over 30%. South and southwestern winds between 15 and 25 knots also increased by at least

30%. But winds > 25 knots in the southwest decreased by over 30%. At Inukjuak (Figure 2.12c), wind speeds of all magnitudes were increasing in the east, south and southwest. Interestingly, winds < 15 knots were decreasing in the northeast and southeast but winds above 15 knots at these directions were increasing. The overall winds from Kuujjuarapik were generally increasing from southwest to northeast in all magnitudes and winds < 10 knots were increasing from the northeast to south (Figure 2.12d). There was a strong decline for winds > 15 knots from the east, northeast and south. The actual change is more ambiguous due to the nearby terrain. Two mountains of approximately 200 m above sea level are located to the northeast and southeast of

Kuujjuarapik, with a river channel between the two mountains at about 90o east of the town

(Larouche, 1990). Winds would deflect around these mountains and channel towards the river while picking up speed due to the local orographic effects. The mountains appeared to have a strong influence and deflect wind direction when the winds were < 18 km/h (Larouche, 1990).

Katabatic winds flowing downslope from these mountains towards Kuujjuarapik were also a

68 possibility. This study revealed that eastern winds > 10 knots were declining and replaced by increase in winds at < 10 knots. Nonetheless, the western and southern onshore winds coming off

Hudson Bay were becoming more frequent. Nitchequon’s winds were harder to interpret and assess the trends given the change in recording method at the end of December 1970 (Figure

2.12e). Combining multiple sectors into one sector would raise the percentage change beyond the actual change. Factoring this issue into consideration, the winds appeared to be more frequently coming from the northeast, southeast, south and west. Finally, a consistent result for the

Schefferville and Wabush justifies their grouping (Figures 2.12f and 2.12g) as both sites face lower wind speed at > 20 knots in all directions. That decrease was offset by an increase in winds

< 10 knots at Schefferville but this increase was not as clear at Wabush. At Wabush, the wind seemed to concentrate at < 5 knots and between 10 and 15 knots while winds > 15 knots were getting less frequent.

It is likely that the local wind direction changes at Baker Lake and Nitchequon (Table 2.6) were more influenced by mesoscale and microscale than by synoptic scale weather patterns because significant wind direction changes would be reflected in other locations as well. This was particularly true for aircraft icing caused by different factors, including wind direction, that appeared locally but not on synoptic scale (Fernández-González et al., 2014). Wind direction on a local scale could impact public health. As some of these communities use diesel fuel for power and heating, localized lung irritation could occur if the wind blows the exhaust gas with nitrogen oxides, carbon monoxide and particulate matter from the diesel generators back into the community (Yan et al., 2019).

2.5.5 Aviation Perspective

Common intuition would suggest that greater wind speed might lead to more risky situations. However, slower wind speed does not necessarily equate to safer flying conditions, especially with very low speeds winds. As Beckwith (1985) explained, minimal winds are not ideal for aircraft to land and take off because there is less wind flowing over the wings that would generate less lift. During takeoff, pilots want to fly into the headwind to maximize the amount of wind over the wings for more lift. In landing, pilots also want to fly into the headwind to shorten the landing distance. Compounding this issue is the general increase of surface temperature, which lowers air density and further reduces lift generated by the wings during takeoff (Budd & Ryley, 2012). The results of this study suggest that the risk from low wind speed has become less of an issue as calm wind frequency went down at all of the locations

(Figures 2.3 to 2.9).

Higher wind speed combined with sea level rise could lead to more frequent storm surges and at greater intensity arising from higher storm surge height, particularly the airports located in coastal zones (Burbidge, 2017). That study mentioned that over 30 European airports will be vulnerable to sea level rise. At some of the airports, Burbidge (2017) stated that other supporting infrastructure such as ground transportation, electrical supply and IT system would be susceptible to storm surges caused by stronger winds. This demonstrated that airport infrastructure will require a holistic approach to ensure that different components of the airport would continue to function in strong wind conditions. In Northern Canada, past storm surges as high as 2.4 m above mean sea level occurred in Tuktoyaktuk, Northwest Territories (Harper et al.,

1988). In this study, Nitchequon, Schefferville and Wabush would be least vulnerable to storm surges as they are located inland. While Baker Lake is located on the western shores of the lake

70 of its namesake, it is approximately 300 km inland from Hudson Bay and therefore the community and the airport are not very vulnerable to storm surges. Churchill, Inukjuak and

Kuujjuarapik are located along the coastline of Hudson Bay. Inukjuak airport, however, is less vulnerable to storm surge than its community because the airport is located further inland.

Therefore, both the communities and airports of Churchill and Kuujjuarapik would be the most vulnerable in all of the seven locations in this study to high winds causing storm surge damage to infrastructure. Building codes will need to be revised to account for higher wind speeds and stronger storm intensity (Bartlett et al., 2003; Nguyen et al., 2018). Airports also need to anticipate low probability but high consequence events such as extreme gusts and tornadoes in their risk assessments (Cheng et al., 2013).

Unlike average wind speed, a gust is most dangerous when it travels in the same direction

(tailwind) or perpendicular (crosswind) to the direction of an airplane is about to land. Tailwind landings are dangerous because it can lead to a stall while pilots reduce the aircraft speed to land.

Major airports often have a second or third runway with a different orientation that is primarily used when the crosswind is present to lower the risk. In the seven airports analyzed in this study, only Churchill has a second runway (Table 2.9). Manasseh and Middleton (1999) identified that tailwind gusts are stronger than crosswind gusts at Sydney Airport in Australia. Wong et al.

(2006) found that the relative accident rate for crosswind in the US rose dramatically when the wind speed exceeded 33 km/h, which highlighted that high wind speed presented more risks than low wind speed.

Since aircraft must takeoff and land into the wind whenever wind is ≥10 knots, wind speed and direction are both taken into consideration to determine which end of the runway should be used if wind speed approaches 10 knots or higher. Below the 10-knot threshold,

71 aircraft have more flexibility and can takeoff or land in either direction of the runway. Using

Figure 2.12, aircraft flying to and from Baker Lake had more flexibility during northerly and easterly winds as winds ≥10 knots were decreasing. However, aircraft were expected to takeoff and land according to the Terminal Aerodrome Forecast (TAF) if the winds were southerly to northwesterly. Churchill was less affected by this constraint because it has 2 runways. However, the 70o side of the runway (Runway 07) would be used less frequently due to a decrease in >25- knot winds in that direction. At Inukjuak, since wind speeds >10 knots were at least 30% more frequently observed from 0o to 230o and the decline from 260o to 350o, the 70o side of the runway (Runway 07) would be used more often. At Kuujjuarapik, both ends of the runway would receive more usage if TAF forecasted wind direction to be between 200o and 30o. Winds from easterly to southerly direction would more likely allow the aircraft to land or takeoff at either end of the runway. There are no expected impacts at Nitchequon because the airfield was abandoned and no aircraft was authorized to land or takeoff from there. At Schefferville and

Wabush, planes were more likely to takeoff and land in either direction because wind speed >15 knots were declining by more than 60%.

Runways that do not align with prevailing wind direction may force pilots into “more dangerous and cognitively demanding crosswind landings” (Budd & Ryley, 2012). Two high profile crashes in the Netherlands and Canada were caused by a combination of crosswind and wind gust during landing. In December 1997, an aircraft carrying over 200 passengers crashed while attempting to land under intense crosswind at Amsterdam in the Netherlands (Koetse &

Rietveld, 2009). These researchers identified Amsterdam Schiphol airport’s runways to be wrongly aligned. Aside from an increased risk, non-optimized runway direction also reduced the productivity of the airport by reducing the maximum capacity of aircraft handled per hour. In

Canada, First Air’s flight 6560 crashed into a nearby hill and killed 12 people onboard while landing at Resolute, Nunavut in August 2011. While the cause of the crash was primarily attributed to pilot error, the Transportation Safety Board reported that wind was an aggravating factor by pushing the aircraft further off course during landing (Transportation Safety Board of

Canada, 2014). At Frankfurt Airport, Frech et al. (2007) examined the 30-year wind climatology and detected a general shift in wind direction that changed the number of times a particular runway was used for landing in order to reduce risk. In addition, sudden changes in wind directions could lead to unstable approaches or tailstrike (Chan, 2014), in which the rear part of the plane touches the runway during takeoff or landing and requires inspection and maintenance before the aircraft can return to service. Wan et al. (2010) concluded that the greatest wind variations in among locations examined in this study were during the spring and fall at Baker

Lake and during summer at Kuujjuarapik. Due to the nature of the airline scheduling in the

Canadian Arctic, passengers were exposed to multiple takeoffs and landings in each flight as the aircraft was flown to different communities along the way. Thus, the passengers were exposed to risks associated with wind gusts and crosswinds repeatedly on each flight from each takeoff and landing at different airports.

At two major airports in Canada, scientists used a new weather forecast system to nowcast conditions up to 6 hours, yet they still struggled to forecast vertical wind shear due to limitations with the model used (Issac et al., 2014). Since all of the airports in this study were not using this new forecasting system, it is believed that the forecast capability of these stations is substantially lower and even less accurate. Inukjuak airport would be the most vulnerable to any adverse weather conditions because almost all the arrivals and departures took place at around 2 pm. Since wind gust was most frequently observed at that time in Inukjuak (Figure 2.2c), this

73 could be a cause for concern. One way to mitigate this concern is to delay the departure flights slightly until the gust is reduced, especially in crosswind situations. While there are no scheduled flights in late night and early morning (Table 2.7), sometimes medevac flights took place overnight. It is important to note that airport scheduling was subject to change based on seasonal demands as well as business decisions. Tables 2.7 and 2.8 provided a snapshot of scheduled service on a weekday as an example at a particular time. Generally, there were more flights on weekdays than weekends at the airports in this region. Variation in customer demands and operational constraint, such as maintenance or mechanical issues, could lead to other types of aircraft not listed in Table 2.8 flying into these communities. Flight schedules also change seasonally and annually. Airlines could also start new routes or withdraw service at any time that would lead to alterations to the schedule and types of aircrafts used. The number of flights listed in Table 2.7 was a conservative estimate as it did not account for helicopter, cargo, military, charter, medical or private owners’ flights.

2.6 Conclusion

There were substantial changes to wind characteristics at the seven communities investigated in this study. Also, there was spatial variability between Hudson Bay and Labrador communities. Seasonal average wind speeds were significantly increasing at Churchill, Inukjuak

Kuujjuarapik and Nitchequon while significantly decreasing at Schefferville and Wabush between 1971-1990 and 1991-2010. Maximum wind speeds were significantly increasing at

Churchill, Inukjuak and Kuujjuarapik while significantly decreasing at Baker Lake, Nitchequon,

Schefferville and Wabush. Number of days with maximum wind speeds at or exceeding 28 km/h was significantly increasing at Churchill, Inukjuak and Nitchequon while significantly

74 decreasing at Schefferville and Wabush. The two Labrador communities analyzed in this study,

Schefferville and Wabush, appeared to encounter slower wind speeds under various analysis.

Communities around the Hudson Bay shores appeared to encounter greater wind speeds but their signals were more nuanced than in Labrador.

Hudson Bay communities often encountered highest wind speed around midnight and in the early afternoon while Labrador communities and Nitchequon encountered highest wind speed in the afternoon. All of the locations except Baker Lake and Nitchequon experienced fewer clam winds. Calm winds were encountered between 1.2% at Churchill to 14.6% at

Wabush. Wind speeds were generally increasing from all directions at Churchill and Inukjuak and decreasing from all directions at Schefferville and Wabush. Baker Lake and Nitchequon’s wind directions were significantly changing from the north towards the northwest. Higher wind speed in some locations appeared to disagree with the GCM model projections for the future but the wind direction trend was in agreement with future projections. Higher wind speed and less calm winds presents an opportunity for future wind power projects in the communities around the Hudson Bay region. However, it also brings challenges through stronger storms.

Wabush had the largest number of scheduled passenger flights and greatest varieties of turboprop aircraft. Inukjuak had five scheduled flight movements within an hour, at 2 pm, which was the highest number within an hour at all of the seven airports analyzed. Five of the seven airports were experiencing predominantly crosswind conditions. Higher wind speed combined with runways not aligned to prevailing wind direction would increase the risk of an accident.

From this, it was concluded that the two airports on the eastern shore of Hudson Bay, Inukjuak and Kuujjuarapik, face an elevated risk of aviation accidents. Kuujjuarapik airport would be at a greater risk due to the increased uncertainty posed by prevailing crosswind, the local orography

75 effects by the nearby mountains (Larouche, 1990) and the proximity of the airport to the community. Smaller aircraft are more susceptible to crosswind conditions. Potential risks also increase for flights with multiple stops. While Schefferville and Wabush airports’ runways are not aligned with the prevailing wind directions, their decreasing trend in wind speed would lower the risks posed by crosswinds. Lack of a crosswind runway may require amendments to the approach procedures of the affected airports (Burbidge, 2017) to reduce the risk caused by crosswind. Nitchequon airport was already closed by 1985 so the risk was not assessed. From the results of this study, it is suggested that the departure and arrival of scheduled flights should be moved to either before 1 pm or after 5 pm to minimize the risks posed by crosswind conditions with high wind speed. Higher maximum wind speeds at Churchill, Inukjuak and Kuujjuarapik suggested that these airports would be exposed to elevated risks on their infrastructure that were caused by wind damage. Buildings need to be reinforced to account for more frequent wind gusts and at higher intensities. Better pilot education and awareness campaign can potentially mitigate these risks.

2.7 Acknowledgement

We wish to thank Marcelo Ponce from SciNet HPC Consortium for his assistance with data computation on General Purpose Cluster (GPC) supercomputer. We would also like to thank

Lindsay Sutton from Canadian Meteorological Aviation Centre – West (CMAC-West) of

Environment and Climate Change Canada for her assistance with wind percentage change charts.

2.8 References

Agostinelli, C. & Lund, U. (2011) R Package ‘circular’: Circular Statistics (Version 0.4-7). Retrieved from https://r-forge.r-project.org/projects/circular

Arriaga, M., Cañizares, C. A., & Kazerani, M. (2013). Renewable Energy Alternatives for Remote Communities in Northern Ontario, Canada. IEEE Transactions on Sustainable Energy, 4(3), 661-670. Arriaga, M., Cañizares, C. A., & Kazerani, M. (2014). Northern Lights: Access to Electricity in Canada’s Northern and Remote Communities. IEEE Power and Energy, 12(4), 50-59. Bartlett, F. M., Hong, H. P., & Zhou, W. (2003). Load factor calibration for the proposed 2005 edition of the National Building Code of Canada: Statistics of loads and load effects. Canadian Journal of Civil Engineering, 30(2), 429-439. Beckwith, W. B. (1985). Aviation. In D. D. Houghton, Handbook of Applied Meteorology (pp. 945-977). New York: John Wiley. Budd, L., & Ryley, T. (2012). An International Dimension: Aviation. In T. Ryley & L. Chapman (Eds.), Transport and Climate Change (Transport and Sustainability, Volume 2) (pp. 39- 64). Emerald Group Publishing Limited. Burbidge, R. (2017). Climate-proofing the airport of the future. Journal of Airport Management, 11(2), 114-128. Canadian Armed Forces. (2015). Trenton Search and Rescue Region. Retrieved from http://www.forces.gc.ca/en/operations-canada-north-america-current/trenton-sar.page Chan, P. W. (2014). A tail strike event of an aircraft due to terrain-induced wind shear at the Hong Kong International Airport. Meteorological Application, 21, 504-511. CBC. (2012). Inuvik, N.W.T., airport roof damaged in blizzard. Retrieved from http://www.cbc.ca/news/canada/north/inuvik-n-w-t-airport-roof-damaged-in-blizzard- 1.1130330 Cheng, V. Y. S., Arhonditsis, G. B., Sills, D. M. L., Auld, H., Shephard, M. W., Gough, W. A., & Klaassen, J. (2013). Probability of Tornado Occurrence across Canada. Journal of Climate, 26(23), 9415-9428. Cheng, C. S. (2014). Evidence from the Historical Record to Support Projection of Future Wind Regimes: An Application to Canada. Atmosphere-Ocean, 52(3), 232-241. Cheng, C. S., Lopes, E., Fu, C., & Huang, Z. (2014). Possible Impacts of Climate Change on Wind Gusts under Downscaled Future Climate Conditions: Updated for Canada. Journal of Climate, 27(3), 1255-1270. Cleveland, W. S. (1981). LOWESS: A Program for Smoothing Scatterplots by Robust Locally Weighted Regression. The American Statistican, 35(1), 54. Cook, N. J. (2015). A statistical model of the seasonal-diurnal wind climate at Adelaide. Australian Meteorological and Oceangraphic Journal, 65(2), 206-232. Coumou, D., Lehmann, J., & Beckmann, J. (2015). The weakening summer circulation in the Northern Hemisphere mid-latitudes. Science, 348(6232), 324-327. Eichelberger, S., McCaa, J., Nijssen, B., & Wood, A. (2008). Climate Change Effects on Wind Speed. North American Windpower, 7, 68-72.

Environment Canada. (2015). Manual of Surface Weather Observations. Seventh Edition, Amendment 19. Gatineau, Quebec, Canada. Fernández-González, S., Luis Sánchez J., Gascón, E., López, L., García-Ortega, E. & Merino, A. (2014). Weather Features Associated with Aircraft Icing Conditions. The Scientific World Journal, 2014, 1-18. Francis, J. A., & Vavrus, S. J. (2015). Evidence for a wavier jet stream in response to rapid Arctic warming. Environmental Research Letters, 10(1), 014005. Frech, M., Holzäpfel, F., Tafferner, A., & Gerz, T. (2007). High-Resolution Weather Database for the Terminal Area of Frankfurt Airport. Journal of Applied Meteorology and Climatology, 46, 1913-1932. Gagnon, A. S., & Gough, W. A. (2005). Trends in the Dates of Ice Freeze-Up and Breakup over Hudson Bay, Canada. Arctic, 58(4), 370-382. Gagnon, A. S., & Gough, W. A. (2006). East–west asymmetry in long-term trends of landfast ice thickness in the Hudson Bay region, Canada. Climate Research, 32(3), 177-186. Gearheard, S., Pocernich, M., Stewart, R., Sanguya, J., & Huntington, H. P. (2010). Linking Inuit knowledge and meteorological station observations to understand changing wind patterns at Clyde River, Nunavut. Climatic Change, 100(2), 267-294. Gultepe, I., Zhou, B., Milbrandt, J., Bott, A., Li, Y., Heymsfield, A., . . . Cermak, J. (2015). A review on ice fog measurements and modeling. Atmospheric Research, 151, 2-19. Hamed, K. H., & Rao, A. R. (1998). A Modified Mann-Kendall Trend Test for Autocorrelated Data. Journal of Hydrology, 204, 182-196. Harper, J. R., Henry, R. F., & Stewart, G. R. (1988). Maximum Storm Surge Elevations in the Tuktoyaktuk Region of the Canadian Beaufort Sea. Arctic, 41(1), 48-52. Hanesiak, J. M., & Wang, X. L. (2005). Adverse-Weather Trends in the Canadian Arctic. Journal of Climate, 18, 3140-3156. Hooshangi, H. R. (2014). Feasibility study of wind-diesel hybrid power system for remote communities in north of Quebec. Journal of Advances in Clean Energy, 1, 84-95. Hori, Y., Gough, W. A., Butler, K., & Tsuji, L. J. S. (2017). Trends in the seasonal length and opening dates of a winter road in the western James Bay region, Ontario, Canada. Theoretical and Applied Climatology, 127(3-4): 753-760. IPCC. (2013). Climate Change 2013: The Physical Science Basis. Working Group I Contribution to the Fifth Assessment Report of the IPCC. Cambridge, United Kingdom: Cambridge University Press. IPCC. (2014). Climate Change 2014: Impacts, Adaptation, and Vulnerability. Working Group II Contribution to the Fifth Assessment Report of the IPCC Cambridge, United Kingdom: Cambridge University Press.

Isaac, G. A., Bailey, M., Boudala, F. S., Burrows, W. R., Cober, S. G., Crawford, R. W., … Fournier, M. (2014). The Canadian Airport Nowcasting System (CAN-Now). Meteorological Applications, 21(1), 30-49. Jenama, R. K., & Kumar, A. (2013). Bad weather and aircraft accidents – global vis-à-vis Indian scenario. Current Science, 104(3), 316-325. Kendall, M. G. (1975). Rank correlation methods. London, UK: Griffin. Klawa, M., & Ulbrich, U. (2003). A model for the estimation of storm losses and the identification of severe winter storms in Germany. Natural Hazards and Earth System Science, 3, 725-732. Koetse, M. J., & Rietveld, P. (2009). The impact of climate change and weather on transport: An overview of empirical findings. Transportation Research Part D, 14(3), 205-221. Kowal, S., Gough, W. A., & Butler, K. (2017). Temporal evolution of Hudson Bay Sea Ice (1971-2011). Theoretical and Applied Climatology 127(3-4): 753-760. Kulesa, G. (2002). Weather and Aviation: How does Weather Affect the Safety and Operations of Airports and Aviation, and How Does FAA Work to Manage Weather-related Effects? Washington, DC: US Department of Transport Center for Climate Change and Environmental. Larouche, P. (1990). Winter Offshore/Onshore Wind Differences in Southeastern Hudson Bay, Canada. Arctic, 43(1), 55-59. Leung, A. C. W., & Gough, W. A. (2016). Air Mass Distribution and the Heterogeneity of the Climate Change Signal in the Hudson Bay / Foxe Basin region, Canada. Theoretical and Applied Climatology, 125(3), 583-592. Lumley, T., Diehr, P., Emerson, S., & Chen, L. (2002). The Importance of the Normality Assumption in Large Public Health Data Sets. Annual Review of Public Health, 23, 151- 169. Ma, J., Foltz, G. R., Soden, B. J., Huang, G., He, J., & Dong, C. (2016). Will surface winds weaken in response to global warming? Environmental Research Letters, 11(12), 124012. Mahrt, L. (2011). Surface Wind Direction Variability. Journal of Applied Meteorology and Climatology, 50, 144-152. Manasseh, R., & Middleton, J. H. (1999). The surface wind gust regime and aircraft operations at Sydney Airport. Journal of Wind Engineering and Industrial Aerodynamics, 79(3), 269- 288. Mann, H. B. (1945). Nonparametric tests against trend. Econometricia, 13(3), 245-259. Mann, M. E., Rahmstorf, S., Kornhuber, K., Steinman, B. A., Miller, S. K., & Coumou, D. (2017). Influence of Anthropogenic Climate Change on Planetary Wave Resonance and Extreme Weather Events. Scientific Reports, 7, 45242. Mohsin, T., & Gough, W. A. (2010). Trend analysis of long-term temperature time series in the Greater Toronto Area (GTA). Theoretical and Applied Climatology, 101(3-4), 311-327.

Nalcor. (n.d.) Menihek. Retrieved from https://nalcorenergy.com/nalcor-operations/other- operations/menihek Natural Resources Canada. (2018). Front End Engineering and Design (FEED) Study Whapmagoostui - Wind Hybrid Power Plant. Retrieved from https://www.nrcan.gc.ca/energy/funding/current-funding-programs/eii/16146 Nguyen, T. S., Stathopoulos, T., & Tirca, L. (2018). Wind-induced shear and torsion in low-rise and medium-rise buildings: Provisions of National Building Code of Canada 2015. Canadian Journal of Engineering, 45(5), 339-350. Pejovic, T., Williams, V. A., Noland, R. B., & Toumi, R. (2009). Factors Affecting the Frequency and Severity of Airport Weather Delays and the Implictations of Climate Change for Future Delays. Transportation Research Record: Journal of the Transportation Research Board, 2139, 97-106. Pollon, C. (2017). What will it take to get Canada’s Arctic off diesel? Retrieved from https://www.revealnews.org/article/what-will-it-take-to-get-canadas-arctic-off-diesel Prowse, T. D., Furgal, C., Chouinard, R., Melling, H., Milburn, D., & Smith, S. L. (2009). Implications of Climate Change for Economic Development in Northern Canada: Energy, Resource, and Transportation Sectors. AMBIO: A Journal of the Human Environment, 38(5), 272-281. Quirk, W. A. (2013). Bush Pilot Way: Flying and Training in Alaska to Become the Best Pilot You Can Be. Anchorage, Alaska, United States of America: Publication Consultants. Reidlinger, D., & Berkes, F. (2001). Contributions of traditional knowledge to understanding climate change in the Canadian Arctic. Polar Record, 37(203), 315-328. Schäfer, A. W., & Waitz, I. A. (2014). Air transportation and the environment. Transport Policy, 34(1), 1-4. Sen, P. B. (1968). Estimates of the Regression Coefficient Based on Kendall's Tau. Journal of the American Statistical Association, 63(324), 1379-1389. Shepherd, M. (2018). Should Airplanes Land In 40-60 Mph Nor'easter Winds? Experts Weigh In. Retrieved from https://www.forbes.com/sites/marshallshepherd/2018/03/02/should- airplanes-land-in-40-60-mph-noreaster-winds-experts-weigh-in Steiner, N., Azetsu-Scott, K., Hamilton, J., Hedges, K., Hu, X., Janjua, M.Y., . . . Tallmann, R. (2015). Observed trends and climate projections affecting marine ecosystems in the Canadian Arctic. Environmental Reviews, 23(2), 191-239. Strefford, S. (2002). Weather and sailing. Weather, 57(6), 224-225. Theil, H. (1950). A rank-invariant method of linear and polynomial regression analysis. Proceedings of the Royal Netherlands Academy of Sciences Part I, Part II, Part III, 53, 386-392, 521-525, 1397-1412. Transportation Safety Board of Canada. (2014). Aviation Investigation Report A11H0002. Gatineau, Quebec.

80 van der Linden, E. C., Bintanja, R., & Hazeleger, W. (2017). Arctic decadal variability in a warming world. Journal of Geophysical Research: Atmospheres, 122, 5677-5696. Wan, H., Wang, X. L., & Swail, V. R. (2010). Homogenization and Trend Analysis of Canadian Near-Surface Wind Speeds. Journal of Climate, 23, 1209-1224. Weatherhead, E., Gearheard, S., & Barry, R. G. (2010). Changes in weather persistence: Insight from Inuit knowledge. Global Environment Change, 20(3), 523-528. Wong, D. K. Y., Pitfield, D. E., Caves, R. E., & Appleyard, A. J. (2006). Quantifying and characterising aviation accident risk factors. Journal of Air Transport Management, 12(6), 352-357. Yan, C., Rousse, D., & Glaus, M. (2019). Multi-criteria decision analysis ranking alternative heating systems for remote communities in Nunavik. Journal of Cleaner Production, 208, 1488-1497.

How does Fog and Low Visibility Affect Flying in Hudson Bay Region?

Andrew C.W. Leung1, William A. Gough1, Ken A. Butler2 1Department of Physical & Environmental Sciences, University of Toronto Scarborough 2Department of Computer & Mathematical Sciences, University of Toronto Scarborough 1265 Military Trail, Toronto, Ontario, Canada, M1C 1A4

Chapter 3: Fog and Low Visibility

3.1 Abstract

Fog and low visibility present a natural hazard for aviation in the Hudson Bay region.

Flights are often delayed or cancelled if fog is present since the airports in this region are less equipped to deal with low visibility operations. Sixteen communities on the eastern and western shores of Hudson Bay were selected for fog and low visibility statistical analyses. Both fog hours and ice fog hours were found to be in general decline, with some locations experiencing significant declines. Spatial asymmetries for fog and ice fog were observed among the various areas within the Hudson Bay region. Most of the northern locations in this study experienced statistically significant declines in fog hours while southern locations’ decline were not significant. Fog was significantly declining in some western Hudson Bay locations during spring and fall and in James Bay during winter and summer, but minimal trends in eastern Hudson Bay.

For ice fog hours, all of the locations in the western shore of Hudson Bay faced a significant decline in winter while only a third of the locations in eastern shores were found to be declining significantly during winter. Blowing snow, snow, ice and fog were the leading causes for reduced and low visibilities at the majority of the locations. Other factors such as rain contributed a minor role towards low visibility.

3.2 Introduction

Fog reduces visibility, making it difficult for pilots to find the runway to land (Hanesiak

& Wang, 2005). Reduced visibility accounted for half of all weather-related accidents in Canada

(Gultepe et al., 2015). The low visibility accident rates are lower in the US (21%) and India

(16%) (Jenama & Kumar, 2013). Low visibility is caused by a number of atmospheric conditions, including fog, heavy rain, blowing snow and haze. Ice and snow on airport surface were considered to be easier to respond to than fog or heavy wind as ice and snow can be readily removed. Fog can be dispersed by helicopter or jet engine, however, this is expensive and only applicable to thin layers of fog (Thornes et al., 2012). At San Francisco, known for its fog, flights delays and cancellations were needed to mitigate the risk caused by poor visibility (Koetse &

Rietveld, 2009). Cargo flights were most likely to be affected by fog as they tended to arrive at their destinations just before sunrise coinciding with the time of the highest fog formation probability (Baker et al., 2002). In the Hudson Bay region, flights often carry fuel, groceries or transporting medivac patients to medical facilities. Therefore, delayed and cancelled flights negatively affect the health and well-being of those living in these communities.

In addition, there have been fatal aircraft accidents in northern Canada that are attributed to poor visibility. Only a select few airports in the Hudson Bay region had the expensive precise navigation aid, Instrument Landing System (ILS), installed to allow aircraft to navigate around nearby hills and land in dense fog. Even with ILS installed, a plane crashed into the hill in

Resolute, Nunavut in 2011 and fog was recorded in the area at the time of the crash

(Transportation Safety Board, 2014). The crash claimed 12 lives, including a renown Arctic researcher.

Another risk occurs if there is already a plane on the runway and the landing or takeoff aircraft fails to spot that plane because of ground level fog, which can result in a collision. The

Tenerife disaster falls under this category that killed 583 passengers in March 1977 (Weick,

1990). Two Boeing 747s collided on the runway because neither aircraft nor the control tower operator could see each other due to heavy fog on the ground. This tragedy remains the highest death toll in a single aviation accident to this day. The third risk arises from aircraft flying through ice fog. The frost induced by ice fog was attached to its wings and disrupts the aerodynamics of the wings required to generate lift (Gultepe et al., 2015).

At major airports, poor visibility below 800 m hampers takeoff and landings (Thornes et al., 2012). Around the Hudson Bay region, airports are serviced by smaller aircraft and they require greater visibility to safely takeoff and land (Hanesiak & Wang, 2005). Helicopters, often used for search and rescue missions, require even greater visibility than smaller aircraft since they fly closer to the ground and are more vulnerable to severe weather at the surface (Thornes et al., 2012). Coastal communities along the Hudson Bay frequently encounters steam fog, which is formed when cold air moves over relatively warmer open waters (Lescop-Sinclair & Payette,

1995). The air just above the open waters was warmer than the cold air moving in by advection, causing the warmer air to cool and water vapour to condense. The water vapour evaporates into the air, leading to the formation of steam fog (Burbidge, 1951). In addition, ice fog is formed in extremely cold temperatures in winter when high relative humidity allows ice crystals to suspend in the air. Percentage frequency of fog and ice fog were also higher in summer than in winter at

Hudson Bay and the Canadian Arctic Islands (Burbidge, 1951; Maxwell, 1981). In the mid-

1940s, at Inukjuak, Quebec, 56% of fog occurred on days with westerly winds but only 20% occurred on days with easterly winds (Burbidge, 1951). The relationship between fog occurrence

85 and daily temperature has been established at Churchill airport in Manitoba, located in southwest

Hudson Bay (Gough & He, 2015). They found the seasonal variation in fog was influenced by sunrise. This would affect flights in the early morning before the fog has dissipated. Since weather and low visibility-related aviation accidents in northern latitudes are projected to increase by ten times (Gultepe et al., 2015), this is a cause for concern. If the frequency of fog was increasing, it could cause additional flight delays and cancelations. Since flights to Hudson

Bay communities usually make multiple stops in each direction, fog or restricted visibility at one or more airports could cause knock-on effects on subsequent legs.

Prior research was conducted to investigate adverse weather types at Baker Lake,

Churchill and Coral Harbour in the Hudson Bay region, all of which had 24-hour observation stations (Hanesiak & Wang, 2005). Hanesiak and Wang (2005) found that adverse weather conditions from 1953 to 2004 were rising but the causes were not due to an increase in blowing snow or fog events. In this study, the various causes of low visibility in the region were further analyzed. Factors causing low visibility include fog, ice fog, blowing snow, heavy rain, haze and others. The results determined how often the airport encountered low visibility conditions. A better understanding of visibility around the airports leads to improved safety, fewer flight delays and cancellations, and lower operating costs for airlines. Other transportation modes such as shipping, driving and snowmobiling also benefit from an enhanced understanding of local visibility that improves safety. With an expected increase of tourism in the Arctic region in North

America (Scott et al., 2004), frequency of low visibility could discourage potential tourists from visiting a particular destination (Førland et al., 2012).

This chapter seeks to investigate the following areas in this study:

1) Has fog and ice fog frequency at 16 communities around Hudson Bay region

changed between 1953 and 2014? Were there temporal trends?

2) If there is a change, what is the nature of this change?

3) Were there changes to restricted visibility over time?

3.3 Methods

3.3.1 Weather Conditions

Most airports in the Hudson Bay Region operate during the day as they do not have the flight frequency that merits operating the airport 24 hours a day. The normal operation starts at

7 am or 8 am and the airports are staffed for 10 to 12 hours on weekdays. Staffed operation hours are reduced on weekends and holidays, during poor visibility such as blizzard that results in early airport closure, or if the weather observer becomes ill. With the exception of medical evacuation

(medevac) flights, the airport is closed at night and no weather observations are typically made.

The hourly weather observer records various climatological parameters, such as temperature, precipitation, wind, cloud ceiling, visibility, significant weather and obstructions to visibility such as fog, rain, blowing snow and others. According to the most recent version of the

Environment and Climate Change Canada’s Manual of Surface Weather Observations

(MANOBS), fog is defined as small water droplets or ice crystals suspended in the air, reducing the visibility to 0.5 statute mile (0.8 km) or less at the surface (Environment and Climate Change

Canada, 2015). Definition for freezing fog was identical to fog, except that the temperature was between -0.1 and -30.0oC and the visibility was 0.5 statute mile (0.8 km) or less, or at temperatures below -30.0oC with physical evidence of ice accretion from fog and visibility was

0.5 statute mile (0.8 km) or less (Environment and Climate Change Canada, 2015). Fog could also be reported at below -30.0oC if no icing was observed. Prior to November 1, 1999, freezing fog was recorded as ice fog. Ice fog was a subset of fog and both were reportable from 0 to 6 statute miles (9.7 km) of visibility before November 1, 1999 (L. Baker, personal communication,

October 15, 2018). Ice fog could be reported at temperatures below freezing if there was no sign of icing. Starting from November 1, 1999, fog and freezing fog were only reported when the visibility was 0.5 statute mile (0.8 km) or less. The change in MANOBS definition was required because freezing fog contained supercooled liquid as part of is definition but supercooled liquid was very unlikely to exist at and below -30.0oC (ICAO, 2008a; ICAO, 2008b). The change also brought consistency in aerodrome observations within North America and developed consensus for best practices in weather reporting with other countries (ICAO 2008a). Since ice fog and freezing fog were recorded using similar definitions and that freezing fog replaced ice fog in aerodrome reports in 1999, their numbers were combined and treated as ice fog.

In MANOBS, the atmospheric phenomena that cause reduced and low visibilities include fog, ice fog, drizzle, rain, freezing drizzle, freezing rain, snow, snow pellets, ice pellets and blowing snow (Environment and Climate Change Canada, 2015). Drizzle is liquid water precipitation with the droplet diameter less than 0.5 mm. Rain is liquid water precipitation with the droplet diameter exceeding 0.5 mm or smaller but widely scattered droplets. Freezing drizzle is any drizzle which freezes upon impact with the ground or other objects. Freezing rain is any rain event which freezes upon impact with the surface. Snow is hexagonal frozen precipitation falling towards the surface. Snow pellets are spherically or conically-shaped opaque ice particles with a diameter of 2 to 5 mm. Ice pellets are irregular or spherically-shaped translucent ice particles with a diameter of 5 mm or less. Blowing snow is any snow being lifted by the strong

88 winds such that the horizontal visibility at eye level is 6 statute miles (9.7 km) or less. For the purpose of this study, snow and snow pellets were grouped together and treated as snow. Fog, ice fog and freezing fog were combined as fog for this part of the study. It is possible for the observer to attribute multiple weather phenomena to the same hour of observation.

Since fog tends to appear just before sunrise, airports that operate 24 hours a day in the region capture the presence of fog more accurately than the non-24 hour airports. Nonetheless, the observation records from the non-24 hour airport weather stations had its purpose for analyzing the hours at which flights arrive and depart. In addition, the records double as a tool to assess how other land-based activities (e.g. driving, hunting and trapping) and sea-based activities (e.g. fishing, sealift) are impacted by the presence of fog during the daytime.

3.3.2 Data Collection

Sixteen communities with airports in the Hudson Bay region were chosen for this study

(Figure 3.1). They were grouped, in a geographical manner similar to the Canadian Ice Service

Ice Regime subregions (Stewart et al., 2010), into eastern Hudson Bay and western Hudson Bay regions (Table 3.1). La Grande Riviere, Wemindji and Moosonee were grouped to James Bay region instead of eastern Hudson Bay. Seven of these sites operated 24-hours a day. The other nine stations operated only during the day. Hourly horizontal visibility conditions at ground level and the significant weather events that caused reduced and low visibilities were obtained from the study sites through Environment and Climate Change Canada’s Climate Data Online (CDO).

The sites were chosen as they were the only long-term weather stations located at airports in the

Hudson Bay and James Bay regions. Most of the non-24-hour weather observation stations had scarce meteorological or climatological research conducted at these locations because of the non-

89 uniform observation frequency and a lack of observation overnight which complicated the analysis. These settlements had relatively low population compared to other northern Canadian communities, causing them to be infrequently studied. Station records ranged from 21 to 62 years. The study period for each site was based on the most recent, longest period of records where observation frequency did not change (e.g. observation taken once every 3 or 6 hours became hourly observations, 24-hour weather station converted to daytime-only station, etc.), with the end year of 2014 (Table 3.1). Moosonee is the exception to this selection period criteria as the end year was 1992. This was due to periods of missing months in 1993 and 1994 followed by 15 years of no fog observation made on site. The long observation gap made it not ideal for analysis as the trend will be skewed and became potentially misleading (Wang, 2006) if data after 2009 were used.

The seasons are defined in climatological months: winter (December, January, February), spring (March, April, May), summer (June, July, August) and fall (September, October,

November). Annual is defined as the total for the four seasons, which includes the December of the previous year to be consistent with climatological seasons.

Figure 3.1: Location of airport sites used in this study.

Table 3.1: The length of study in the selected communities and their regional groupings, with asterisks (*) to indicate 24-hour stations. Region Location Study Period Length of study (years) Western Hudson Bay Baker Lake * 1963 – 2014 52 Chesterfield Inlet 1992 – 2014 23 Rankin Inlet * 1981 – 2014 34 Whale Cove 1985 – 2014 30 Arviat 1985 – 2014 30 Churchill * 1953 – 2014 62 Eastern Hudson Bay Ivujivik 1993 – 2014 22 Akulivik 1993 – 2014 22 Puvirnituq 1994 – 2014 21 Inukjuak * 1953 – 2014 62 Umiujaq 1993 – 2014 22 Kuujjuarapik * 1957 – 2014 58 Sanikiluaq 1988 – 2014 27 James Bay La Grande Riviere * 1977 – 2014 38 Wemindji 1993 – 2014 22 Moosonee * 1972 – 1992 21

Number of hours with fog and ice fog were counted and combined into seasonal and annual data to assess trends using the modified Mann-Kendall test to account for the possibility of autocorrelation in the trend by changing the variance of the data (Hamed & Rao, 1998). The significance level was adjusted based on the presence and degree of autocorrelation. Significance level was increased in datasets with positive autocorrelation, which increased the probability of accepting the null hypothesis. On the other hand, significance level was decreased in the data with negative autocorrelation, which decreased the probability of accepting the null hypothesis and increased the probability of accepting the alternative hypothesis. Durbin-Watson test was not considered as there was a possibility of inconclusive evidence for the presence of autocorrelation.

Hamed and Rao (1998)’s approach allows finer adjustment by allowing a correction factor such that even though the data could be autocorrelated, significant findings may still be found if the impacts of the autocorrelation on the significance level was miniscule. Annual number of hours of fog and ice fog were plotted in locally weighted scatterplot smoothing (LOWESS) time series graphs (Cleveland, 1981). The frequencies of reduced visibility (≤ 0.5 statute mile) and low visibility (≤ 0.25 statute mile), as defined by Transport Canada (2017), were examined. They are equivalent to ≤ 0.8 km and ≤ 0.4 km respectively. Hours of reduced and low visibility per year were also plotted in LOWESS time series graphs. The magnitudes of change for fog, ice fog, reduced visibility and low visibility were determined by Theil-Sen slope estimator (Theil, 1950;

Sen, 1968). Top three causes of reduced and low visibilities were calculated by examining the significant weather events that were present within the observed hour. These significant weather events were assumed to be the cause for reduced and low visibilities. Multiple significant weather events (e.g. snow and blowing snow) could be recorded by the weather observer in the

92 same hour and each factor was considered as one event for their corresponding category for that hour.

3.4 Results

3.4.1 Fog

Fog hours at each location are plotted in Figure 3.2. Fog had high spatial variability.

There was a general declining trend for fog hours per year at each study community across all seasons (Table 3.2). Autocorrelation was present in some locations and the significance levels were adjusted to account for positive and negative autocorrelations (Table 3.3). Almost all autocorrelations were positive autocorrelation, lowering the significance level of the trend.

However, fog hours in the summer in Arviat and Wemindji were negatively autocorrelated and thus increased the significance level of any trends. In Western Hudson Bay, the northern sites experienced the most significant decline (p < 0.05) in spring and fall. Chesterfield Inlet, Rankin

Inlet and Whale Cove experienced significantly less fog (p < 0.05) between 3 to 7 hours per year.

Arviat was the only location with significantly increased fog hours in summer (p < 0.01). In

Eastern Hudson Bay, Puvirnituq experienced significantly fewer fog hours on an annual scale while Kuujjuarapik had significantly less in spring and summer (both at p < 0.05).

(a) Akulivik (b) Arviat

(e) Churchill (f) Inukjuak

(g) Ivujivik (h) Kuujjuarapik

(i) La Grande Riviere (j) Moosonee

(k) Puvirnituq (l) Rankin Inlet

(m) Sanikiluaq (n) Umiujaq

(o) Wemindji (p) Whale Cove

Figure 3.2: Number of fog hours per year at (a) Akulivik, (b) Arviat, (c) Baker Lake, (d) Chesterfield Inlet, (e) Churchill, (f) Inukjuak, (g) Ivujivik, (h) Kuujjuarapik, (i) La Grande Riviere, (j) Moosonee, (k) Puvirnituq, (l) Rankin Inlet, (m) Sanikiluaq, (n) Umiujaq, (o) Wemindji and (p) Whale Cove.

Table 3.2: Trends in fog hours per year with autocorrelation present. Data which are autocorrelated are shaded in grey. Significance level: *** p < 0.001; ** p = < 0.01; * p < 0.05; ^ p < 0.10. Region Location Winter Spring Summer Fall Annual Western Baker Lake +0.08 -0.25 -0.38 -1.00* -1.33^ Hudson Chesterfield Inlet No change +0.27 -1.87^ -2.47* -4.00* Bay Rankin Inlet +0.09 -2.43** -1.91 -2.00* -6.93** Whale Cove -0.07 -0.70* -0.67 -1.57** -3.25** Arviat No change No change +1.30* +0.38 +1.29 Churchill No change -0.50 -0.23 -0.29 -1.21 Eastern Ivujivik +1.38 +0.10 -0.45 +1.00 +1.29 Hudson Akulivik No change -0.53 -0.58 -0.67 -1.71 Bay Puvirnituq -1.12 -2.21* -0.87 -1.75 -6.34* Inukjuak No change -0.67 -1.38 -0.94 -2.86 Umiujaq -0.53 -0.75 -0.69 No change -2.00 Kuujjuarapik -0.08 -0.96* -2.38* +0.59 -2.72 Sanikiluaq +0.17 -0.46 +1.67^ +0.15^ +1.26 James La Grande Riviere +0.82 -0.82 -2.23* -1.67* -5.80** Bay Wemindji -0.33* -0.69^ -0.75 -0.83 -2.85** Moosonee -0.22* -1.79** -4.46** -3.28* -8.71**

Table 3.3: Trends in fog hours per year with significance level adjusted for autocorrelation. Significance level: *** p < 0.001; ** p = < 0.01; * p < 0.05; ^ p < 0.10. Region Location Winter Spring Summer Fall Annual Western Baker Lake +0.08 -0.25 -0.38 -1.00* -1.33 Hudson Chesterfield Inlet No change +0.26 -1.87^ -2.47* -4.00* Bay Rankin Inlet +0.09 -2.43*** -1.91 -2.0* -6.93** Whale Cove -0.07 -0.70* -0.67 -1.57** -3.25** Arviat No change No change +1.30** +0.38 +1.29 Churchill No change -0.50 -0.23 -0.29 -1.21 Eastern Ivujivik +1.38 +0.10 -0.45 +1.00 +1.29 Hudson Akulivik No change -0.53 -0.58 -0.67 -1.71 Bay Puvirnituq -1.12 -2.21 -0.67 -1.75 -6.34* Inukjuak No change -0.67 -1.38 -0.94 -2.86 Umiujaq -0.53 -0.75 -0.69 No change -2.00 Kuujjuarapik -0.08 -0.96* -2.38* +0.59 -2.72 Sanikiluaq +0.17 -0.46 +1.67^ +0.15^ +1.26 James La Grande Riviere +0.82 -0.82 -2.23** -1.67* -5.80** Bay Wemindji -0.33* -0.69 -0.75** -0.83 -2.85** Moosonee -0.22* -1.79^ -4.46^ -3.28 -8.71^

3.4.2 Ice Fog

Ice fog hours were plotted in Figure 3.3. Unlike fog, there was an overall decline for ice fog at almost all locations. There was a slight increase in ice fog at Baker Lake, Churchill,

Kuujjuarapik, Inukjuak from approximately 1975 to 1985 followed by a downward trend. On an annual scale, there were significant declines in ice fog hours with autocorrelation in eastern

Hudson Bay and almost all western Hudson Bay locations (Table 3.4). There was no trend for ice fog in summer due to small sample size at each location. Winter was found to be the season with the greatest decline in ice fog, but positive autocorrelation was present in each of the locations except Arviat. There were less significant trends and positive autocorrelation for ice fog hours in spring. Negative autocorrelation was observed in spring at Chesterfield Inlet and Whale Cove on an annual scale. In James Bay communities, no ice fog trend was observed at Wemindji as it was not present in 15 out of 23 years chosen in this study. Significant decline was observed on an annual scale (p < 0.10) and in winter (p < 0.05) at La Grande Riviere and Moosonee. La Grande

Riviere also observed a somewhat significant increase for ice fog in fall (p < 0.10).

After adjusting the significance level to account for autocorrelation (Table 3.5), each community in western Hudson Bay experienced significantly less ice fog (p < 0.05) in winter and all but Chesterfield Inlet experienced significantly less (p < 0.05) on an annual scale. Only two communities in eastern Hudson Bay had significantly less ice fog (p < 0.05) in winter and four communities in spring (p < 0.05). The magnitude of decline greater in western Hudson Bay than eastern Hudson Bay. The decline was the weakest in James Bay.

(a) Akulivik (b) Arviat

100

(e) Churchill (f) Inukjuak

(g) Ivujivik (h) Kuujjuarapik

101

(i) La Grande Riviere (j) Moosonee

(k) Puvirnituq (l) Rankin Inlet

102

(m) Sanikiluaq (n) Umiujaq

(o) Wemindji (p) Whale Cove

Figure 3.3: Number of ice fog hours per year at (a) Akulivik, (b) Arviat, (c) Baker Lake, (d) Chesterfield Inlet, (e) Churchill, (f) Inukjuak, (g) Ivujivik, (h) Kuujjuarapik, (i) La Grande Riviere, (j) Moosonee, (k) Puvirnituq, (l) Rankin Inlet, (m) Sanikiluaq, (n) Umiujaq, (o) Wemindji and (p) Whale Cove.

103

Table 3.4: Trends in ice fog hours per year with autocorrelation present. Data which are autocorrelated are shaded in grey. Significance level: *** p < 0.001; ** p = < 0.01; * p < 0.05; ^ p < 0.10. N/A = not applicable due to small sample size across the years. Region Location Winter Spring Summer Fall Annual Western Baker Lake -5.13*** -2.13*** No change -0.44*** -8.03*** Hudson Chesterfield Inlet -0.43* +0.17 No change No change -0.57 Bay Rankin Inlet -9.17*** -2.10* No change -0.36* -10.32*** Whale Cove -1.67*** -0.20 No change No change -2.00*** Arviat -1.82*** -0.47* No change No change -2.50*** Churchill -1.83*** -0.06 No change No change -1.76*** Eastern Ivujivik -3.33* -1.20* No change No change -4.25** Hudson Akulivik -1.00* -0.27* No change No change -1.00^ Bay Puvirnituq -2.46^ -1.47* No change No change -4.13* Inukjuak -1.00** -0.40* No change No change -1.20** Umiujaq -0.94* -0.29^ No change No change -1.50** Kuujjuarapik -1.00*** -0.03 No change No change -1.04*** Sanikiluaq -2.65** -0.54 No change No change -3.19** James La Grande Riviere -0.86*** +0.04 No change +0.19* -0.47^ Bay Wemindji N/A N/A N/A N/A N/A Moosonee -0.17* No change No change No change -0.45**

Table 3.5: Trends in ice fog hours per year with significance level adjusted for autocorrelation. Significance level: *** p < 0.001; ** p = < 0.01; * p < 0.05; ^ p < 0.10. N/A = not applicable due to small sample size across the years. Region Location Winter Spring Summer Fall Annual Western Baker Lake -5.13* -2.13*** No change -0.44** -8.03* Hudson Chesterfield Inlet -0.43* +0.17* No change No change -0.57 Bay Rankin Inlet -9.17*** -2.10 No change -0.36^ -10.32* Whale Cove -1.67* -0.20 No change No change -2.00*** Arviat -1.82*** -0.47* No change No change -2.50** Churchill -1.83** -0.06 No change No change -1.76*** Eastern Ivujivik -3.33 -1.20* No change No change -4.25** Hudson Akulivik -1.00 -0.27* No change No change -1.00 Bay Puvirnituq -2.46 -1.47 No change No change -4.13 Inukjuak -1.00* -0.40* No change No change -1.20^ Umiujaq -0.94 -0.29* No change No change -1.50* Kuujjuarapik -1.00*** -0.03 No change No change -1.04** Sanikiluaq -2.65 -0.54 No change No change -3.19^ James La Grande Riviere -0.86*** +0.04 No change +0.19^ -0.47^ Bay Wemindji N/A N/A N/A N/A N/A Moosonee -0.17* No change No change No change -0.45**

104

3.4.3 Reduced and Low Visibilities

Reduced visibility (≤ 0.8 km) trends at each location were presented in Figure 3.4 and low visibility (≤ 0.4 km) trends were presented in Figure 3.5. Arviat experienced the highest percentage of reduced and low visibility during the study period (Table 3.6). While the Arviat airport was operating, there was a 10% probability that the visibility was below 0.8 km and 5.3% that the visibility was below 0.4 km. Western Hudson Bay experienced a higher likelihood of the airport encountering reduced visibility (3.3 to 10.0%) and low visibility (1.7 to 5.3%) than

Eastern Hudson Bay (2.0 to 3.8% and 0.9 to 2.4% respectively). James Bay region had the least likelihood of encountering reduced visibility (0.6 to 1.9%) and low visibility (0.2 to 0.9%).

(a) Akulivik (b) Arviat

105

(e) Churchill (f) Inukjuak

106

(g) Ivujivik (h) Kuujjuarapik

(i) La Grande Riviere (j) Moosonee

107

(k) Puvirnituq (l) Rankin Inlet

(m) Sanikiluaq (n) Umiujaq

108

(o) Wemindji (p) Whale Cove

Figure 3.4: Time series trend for reduced visibility (a) Akulivik, (b) Arviat, (c) Baker Lake, (d) Chesterfield Inlet, (e) Churchill, (f) Inukjuak, (g) Ivujivik, (h) Kuujjuarapik, (i) La Grande Riviere, (j) Moosonee, (k) Puvirnituq, (l) Rankin Inlet, (m) Sanikiluaq, (n) Umiujaq, (o) Wemindji and (p) Whale Cove.

109

(a) Akulivik (b) Arviat

110

(e) Churchill (f) Inukjuak

(g) Ivujivik (h) Kuujjuarapik

111

(i) La Grande Riviere (j) Moosonee

(k) Puvirnituq (l) Rankin Inlet

112

(m) Sanikiluaq (n) Umiujaq

(o) Wemindji (p) Whale Cove

Figure 3.5: Time series trend for low visibility (a) Akulivik, (b) Arviat, (c) Baker Lake, (d) Chesterfield Inlet, (e) Churchill, (f) Inukjuak, (g) Ivujivik, (h) Kuujjuarapik, (i) La Grande Riviere, (j) Moosonee, (k) Puvirnituq, (l) Rankin Inlet, (m) Sanikiluaq, (n) Umiujaq, (o) Wemindji and (p) Whale Cove.

113

Table 3.6: Historical percentage of the reduced and low visibilities during study airports’ operational hours. Region Location Reduced visibility Low visibility (≤ 0.8 km) (≤ 0.4 km) Western Hudson Bay Baker Lake 5.1% 3.3% Chesterfield Inlet 6.7% 3.7% Rankin Inlet 6.2% 2.2% Whale Cove 4.2% 2.1% Arviat 10.0% 5.3% Churchill 3.3% 1.7% Eastern Hudson Bay Ivujivik 2.4% 1.6% Akulivik 2.0% 0.9% Puvirnituq 3.2% 1.5% Inukjuak 2.5% 1.3% Umiujaq 3.8% 2.4% Kuujjuarapik 3.1% 1.6% Sanikiluaq 3.7% 1.7% James Bay La Grande Riviere 1.9% 0.9% Wemindji 0.7% 0.3% Moosonee 0.6% 0.2%

Baker Lake, Arviat, Churchill, Inukjuak and Moosonee had positive autocorrelation

while Rankin Inlet had negative autocorrelation for reduced visibility (Table 3.7). After adjusting

for autocorrelation, the frequency of encountering reduced visibility was significantly decreasing

(p < 0.05) at Baker Lake, Churchill, Kuujjuarapik, La Grande Riviere and Moosonee while less

significantly (p < 0.10) at Rankin Inlet. The overall decrease for sites with significant change

was approximately 2 to 6 hours of reduced visibility per year. Significantly increasing frequency

of reduced visibility (p < 0.05) was found at Ivujivik, Akulivik, Puvirnituq and Sanikiluaq while

less significantly (p < 0.10) at Arviat. Sites with significant increase experienced about 1.5 to 4

additional hours of reduced visibility conditions per year.

For the time series on the frequency of low visibility over time, Arviat, Churchill and

Akulivik had positive autocorrelation (Table 3.7). Adjusted for autocorrelation, low visibility

conditions were less significantly occurring at Baker Lake, Churchill, Kuujjuarapik, La Grande

114

Riviere and Moosonee at p < 0.05 and Whale Cove at p < 0.10. About 1 to 6 fewer hours per year of low visibility was observed at these locations. Ivujivik was the only site with a significant increase of low visibility conditions at p < 0.01 while Arviat and Puvirnituq also experienced more often low visibility at p < 0.10. These three sites experienced about 1 to 4 additional hours of low visibility per year.

Table 3.7: Trends for change in number of reduced (≤ 0.8 km) and low (≤ 0.4 km) visibility hours with significance level before and after adjusted for autocorrelation. Shaded cells indicated the presence of autocorrelation. Significance level: *** p < 0.001; ** p = < 0.01; * p < 0.05; ^ p < 0.10. Region Location Reduced Reduced Low visibility Low visibility visibility (with visibility (with (adjusted for autocorrelation) (adjusted for autocorrelation) autocorrelation) (hrs/year) autocorrelation) (hrs/year) (hrs/year) (hrs/year) Western Baker Lake -6.29*** -6.29*** -6.13*** -6.13*** Hudson Chesterfield +1.85 +1.85 -0.14 -0.14 Bay Inlet Rankin Inlet -1.73 -1.73^ -0.30 -0.30 Whale Cove -0.46 -0.46 -1.20^ -1.20^ Arviat +6.46** +6.46^ +4.19* +4.19^ Churchill -2.40*** -2.40*** -1.32*** -1.32*** Eastern Ivujivik +2.08* +2.08* +2.20** +2.20** Hudson Akulivik +1.50* +1.50* +0.50 +0.50 Bay Puvirnituq +4.30* +4.30* +1.46^ +1.46^ Inukjuak -3.05 -3.05 -1.42 -1.42 Umiujaq +2.53 +2.53 +1.50 +1.50 Kuujjuarapik -1.95*** -1.95*** -1.75*** -1.75*** Sanikiluaq +2.56*** +2.56*** +0.57 +0.57 James La Grande -2.57*** -2.57*** -1.57*** -1.57*** Bay Riviere Wemindji +0.20 +0.20 +0.33 +0.33 Moosonee -2.44*** -2.44* -0.67** -0.67**

The top three causes for reduced visibility (≤ 0.8 km) and low visibility (≤ 0.4 km) at each of the community are reported in Table 3.8. In western Hudson Bay, blowing snow

(blizzard) was the leading factor in northern communities while fog was the leading factor at

115

Churchill, southernmost community in this region. In eastern Hudson Bay, fog was the leading factor for reduced and low visibility at Ivujivik, the northernmost community in this region, while blowing snow and rain were the second and third most common factor. Three eastern

Hudson Bay communities south of Ivujivik share the same restricted visibility factors as those in western Hudson Bay. Starting from Sanikiluaq and Umiujaq southwards, the leading factor became fog. Rain at Umiujaq was the second leading factor, accounting for 24% to 25% for all hours with reduced and low visibilities. In the James Bay region, rain accounted for 40% of reduced visibility and 59% of low visibility while fog accounted for 50% and 64% respectively.

At Moosonee, snow was the leading cause for reduced visibility (47%) followed by fog (33%).

However, fog was the leading cause for low visibility (62%) while snow only accounted for 26%.

Ice pellets was one of the top three leading cause in western Hudson Bay but not in eastern

Hudson Bay nor James Bay. Other factors such as haze from forest fires did not contribute to severe restrictions on visibility except a few hours during the entire study period.

Table 3.8: Top three causes and their proportions for reduced and low visibilities at study airports. Region Location Reduced visibility (≤ 0.8 km) Low visibility (≤ 0.4 km) Western Baker Lake 1. Blowing snow (55%) 1. Blowing snow (60%) Hudson 2. Fog (16%) 2. Fog, Snow and Ice Bay 3. Snow and Ice pellets pellets (each tied at (each tied at 14%) 13%) Chesterfield 1. Blowing snow (46%) 1. Blowing snow (49%) Inlet 2. Fog (31%) 2. Fog (32%) 3. Snow (14%) 3. Snow (10%) Rankin Inlet 1. Blowing snow (48%) 1. Blowing snow (51%) 2. Fog (22%) 2. Fog (21%) 3. Ice pellets (16%) 3. Ice pellets (16%) Whale Cove 1. Blowing snow (43%) 1. Blowing snow (47%) 2. Fog (23%) 2. Fog (23%) 3. Snow (16%) 3. Snow and Ice pellets (tied at 13%)

116

Arviat 1. Blowing snow (39%) 1. Blowing snow (38%) 2. Snow (30%) 2. Snow (30%) 3. Ice pellets (15%) 3. Ice pellets (14%) Churchill 1. Fog (45%) 1. Fog (50%) 2. Blowing snow (37%) 2. Blowing snow (34%) 3. Ice pellets (9%) 3. Ice pellets (7%) Eastern Ivujivik 1. Fog (35%) 1. Fog (35%) Hudson 2. Blowing snow (27%) 2. Blowing snow (26%) Bay 3. Rain (21%) 3. Rain (22%) Akulivik 1. Blowing snow (49%) 1. Blowing snow (58%) 2. Snow (31%) 2. Snow (28%) 3. Rain (17%) 3. Rain (13%) Puvirnituq 1. Blowing snow (38%) 1. Blowing snow (38%) 2. Snow (30%) 2. Snow (28%) 3. Fog (26%) 3. Fog (26%) Inukjuak 1. Snow (34%) 1. Snow (32%) 2. Blowing snow (29%) 2. Blowing snow (30%) 3. Fog (27%) 3. Fog (29%) Umiujaq 1. Fog (38%) 1. Fog (39%) 2. Rain (24%) 2. Rain (25%) 3. Snow (20%) 3. Blowing snow and Snow (tied at 18%) Kuujjuarapik 1. Fog (36%) 1. Fog (40%) 2. Snow (26%) 2. Snow (23%) 3. Blowing snow (24%) 3. Blowing snow (23%) Sanikiluaq 1. Fog (45%) 1. Fog (48%) 2. Blowing snow (29%) 2. Blowing snow (29%) 3. Snow (18%) 3. Snow (17%) James Bay La Grande 1. Rain (40%) 1. Rain (59%) Riviere 2. Snow (33%) 2. Snow (19%) 3. Blowing snow (18%) 3. Blowing snow (12%) Wemindji 1. Fog (50%) 1. Fog (64%) 2. Snow (26%) 2. Snow (13%) 3. Blowing snow (13%) 3. Blowing snow (12%) Moosonee 1. Snow (47%) 1. Fog (62%) 2. Fog (33%) 2. Snow (26%) 3. Blowing snow (11%) 3. Blowing snow (11%)

117

3.5 Discussion

3.5.1 Fog and Ice Fog

There was a strong signal for reduction of fog and ice fog in the Hudson Bay and James

Bay regions on seasonal and annual scales, with the greatest significant declines for fog on an annual scale and ice fog in winter and annually (Table 3.3 and Table 3.5). A decline of 3 to 7 fog hours per year was approximately equal to a decline of 1% to 1.5% annually. The decline, in percentage, was more difficult to determine for ice fog due to the change in observation standards (ICAO, 2008b). Presence of positive autocorrelation in ice fog suggested that ice fog hours, especially in winter and annually, appeared to be highly dependent from one year to the next year (Table 3.4). Positive autocorrelations for fog were also found at some western Hudson

Bay and James Bay communities in spring, summer and annual scales (Table 3.2). With autocorrelation present in fog and especially in ice fog, this suggests that these events, as a whole, extended beyond hourly and daily scales, perhaps to a larger scale, low frequency variations in the atmosphere. Surprisingly, negative autocorrelations were also found in a few analysis. Fog at

Rankin Inlet in spring, Arviat and Wemindji in summer and ice fog at Chesterfield Inlet in spring and Whale Cove on the annual scale had negative autocorrelation, which suggested that a large number of fog/ice fog events in a particular year was followed by very few fog/ice fog in the subsequent year. The significance values were lowered in negative autocorrelations because the effective sample sizes increased to account for the effects of negative autocorrelation on the dataset. This information would be useful for estimating whether a community would be prone to ice fog and, to a lesser extent, fog in the upcoming year by examining the observed data from the previous year. As that the autocorrelations were found on annual scale and each value was the sum of fog or ice fog hours per year, the presence of autocorrelation was more likely the result of

118 a physical process, such as low frequency large scale flow variation rather than a statistical artefact.

Reduction in fog was found as well in Hanesiak and Wang (2005), in which both studies shared Baker Lake and Churchill as one of the study sites and their study period was from 1953 to 2004. In their study, they performed homogenization on study sites to remove known step changes due to change in observer or observation methodology. They found fog was reducing significantly (p < 0.05) at Baker Lake on annual, spring, summer and fall. This finding slightly differed from the current work where only fall had significant decline trends (Table 3.3).

Hanesiak and Wang (2005) also identified a significant declining trend (p < 0.05) at Churchill in fall but this study did not identify any trends on annual and seasonal scales. They summarized that southeastern stations in the Arctic tended to show a decline in fog in all seasons except fall while this study showed regional differences and geographic locations appear to be a greater factor than seasons. This study found that northern sites in western Hudson Bay had significant decline (p < 0.05) in spring, fall while southern sites in this region either had no change or even a significant increase (p < 0.01) in summer at Arviat. In Eastern Hudson Bay, only two southern sites (Kuujjuarapik and Sanikiluaq) experienced significant changes (p < 0.10) in fog hours.

Kuujjuarapik experienced less in spring and summer while Sanikiluaq experienced more in summer and fall. In the James Bay region, La Grande Riviere experienced a decline in summer and fall, Wemindji had less fog in winter and summer, and Moosonee had less fog in all seasons except fall. The difference between this study and Hanesiak and Wang (2005) could be explained by data period and how the data was handled. This study used an additional 10 years of data that was not available to Hanesiak and Wang (2005) back then. This study used the raw observation count of fog by the airport observers while their study transformed the raw observation through

119 the homogenization process. Homogenization was not possible in this study because it works on the premise that step-changes from change in observation method, change in instruments or station re-location are known in advance from metadata. Without metadata, homogenization efforts were very difficult even with change-point analysis being performed. These methodology differences might lead to different significance levels and conclusions drawn from the same starting point. Apparently, seasonality differences in fog frequency were only present in Hudson

Bay and Labrador, but not observed in the rest of Canada (Wang, 2005).

A reduction in ice fog may be attributed to general warming in the region, lowering the frequency of liquid water droplet frozen to the surface or forming ice crystal in the air. Relative humidity can easily reach saturation level when the temperature is very cold (Gultepe et al.,

2015). A reduction in fog may be caused by warmer temperature which requires a narrower difference in temperature between dry bulb and the dew point to achieve the same amount of relative humidity in colder temperatures. Since the vapour pressure of water increases at an exponential rate in higher temperature, relative humidity would be lower even if air temperature and dew point temperature were raised at the same linear rate as a result of a warmer surface.

This is consistent with a study by Ye (2009) in northern Russia. Her results revealed that warmer air temperature was the most important factor in the reduction of fog in the cold environment.

However, Tjernström et al. (2015) documented a moist air mass moving northwards in the East

Siberian Sea and caused dense fog to form due to a temperature inversion at the surface, but their study cautioned that a single event might not be applicable to all circumstances. The air mass pattern in the Hudson Bay region is trending towards more southern, warm air mass and replacing northern, cold air mass (Leung & Gough, 2016). The dynamics of air mass is consistent with the decline in fog due to warmer air from the south which could contain more

120 water vapour in general than cold air from the north. As explained earlier, the air would be drier even if the dry bulb and dew point temperatures increased at the same rate. Burbidge (1951) also found that there was more fog at Inukjuak from onshore wind than offshore.

Hudson Bay can also be a major source for advection fog or sea fog at any coastal community during spring to fall when it is not frozen (Hanesiak & Wang, 2005; Gough & He,

2015). Advection fog is formed when warm air from the south is cooled by wind blowing northward over cold Hudson Bay water while sea fog is formed when cold Arctic air moves south and heated up by relatively warmer Hudson Bay water. With delayed fall freeze-up of sea ice, if more sensible heat than latent heat was released into the atmosphere, the dry bulb temperature would rise quicker than the dew point. This would lower the relative humidity and less chance for fog to form. Ye (2009) suggested that future warming in northern Russia and

Siberia would result in less radiative fog but higher advective fog at locations near large bodies of water. However, this study disagreed with the suggestion by Ye (2009) because the communities in this study all experienced a decline in fog and ice fog. Even with polynya around

Sanikiluaq, ice fog decreased significantly by 1 to 1.5 hours per year at Kuujjuarapik and

Umiujaq (p < 0.05) and less significantly but still declining (p < 0.10) by 3.2 hours per year at

Sanikiluaq (Table 3.5). These three communities are closest to this polynya. However, the polynya might explain why fog rather than blowing snow was the biggest factor when visibility was below 0.8 km (Table 3.8). Another explanation was that the polynya provided open water and generated dense fog in these communities. Both are equally likely because the visibility formed by ice fog is highly variable (Gultepe et al., 2015). This idea was also supported by

Hanesiak and Wang (2005) as they found that fog in the Arctic was highly variable by seasons and by locations. This evidence suggested that a combination of factors, including radiative and

121 advective forces, are affecting the formation frequency of fog in the study area of this research paper.

The longer ice-free season in Hudson Bay due to earlier break-up and later freeze-up of sea ice (Kowal et al., 2017) should have provided a longer duration for fog formation in spring and fall due to advection from the Bay (Gough & He, 2015), yet fog hours continued to drop significantly in these seasons in western Hudson Bay in this study and in Hanesiak and Wang

(2005)’s study at Baker Lake for all seasons and Churchill in winter, spring and summer. This might suggest that the reduction in fog by lower relative humidity was not fully offset by an increase in fog due to longer ice-free period. The lack of significant trend for winter in Hudson

Bay was not unexpected as there was little moisture available while the Bay was ice covered

(Gough & He, 2015). Coincidentally, stations that did not observe a clear downward trend were the stations that had the longest period of record and 24-hour observation stations. Between 1975 and 1985, these 24-hour stations reported about three to four times the number of ice fog than the average number before 1975. Ice fog observations appeared to be captured by both 24-hour stations as well as non-24-hour stations as there were no noticeable trend differences between these types of stations. Other than Wemindji which had almost no ice fog hours and Moosonee which had a limited record ending in 1992, all of the remaining 14 stations had a declining trend in ice fog observation since 1995 (Figure 3.3). On an annual scale, 24-hour stations experienced -

0.45 to -10.32 hours of ice fog per year while non-24-hour stations experienced -0.57 to -4.25 hours per year (Table 3.3). Fog is more nuanced because most of the 24-hour stations as well as some non-24-hour stations which began records in the 1980s recorded a dramatic increase of fog from 1980 to 1990 (Figure 3.2). The pattern was not as clearly observed at stations which began records in the 1990s. Interestingly Moosonee and Rankin Inlet actually recorded far fewer fog

122 events during the same period yet both are 24-hour stations. On an annual scale, 24-hour stations encountered a trend of -1.21 to -8.71 fog hours per year while non-24-hour stations encountered

+1.29 to -6.34 fog hours per year (Table 3.5). The magnitude of the decrease is understandably lower at non-24-hour stations because they did not observe weather at night and in the early morning such that ice fog occurring at these times would not be recorded. They also observe less frequently and therefore had less opportunity to capture fogs and ice fogs. Since fog and ice fog were observed to be declining at both 24-hour and non-24-hour stations, it follows that these events were occurring less frequently in daytime and nighttime. Trends for reduced and low visibilities (Table 3.7) were equally impacted by this observation frequency issue.

This study found different regimes for fog, ice fog and reduced visibility percentages between the eastern and western sides of Hudson Bay and James Bay. Some locations in western

Hudson Bay showed a decline for fog in spring and fall, eastern Hudson Bay had minimal significant trends while James Bay had a significant decline in winter and summer. Ice fog trends also showed temporal differences. Western Hudson Bay recorded significantly fewer ice fog in winter and spring. Eastern Hudson Bay recorded a significant decline in spring and James Bay observed significantly fewer ice fog only in winter. This is similar to the sea ice thickness asymmetry in Hudson Bay (Gagnon & Gough, 2006). Further evidence of weather asymmetry can be found in Wang (2005). In this study, while Kuujjuarapik’s expected long-term average frequency for encountering blowing snow from 1953 to 2004 in fall was 2.8%, Baker Lake and

Churchill in western Hudson Bay region had a much higher frequency, at 10.9% and 6.2% respectively. The average frequency for blowing snow in winter was 20.8% at Baker Lake, 11.4% at Churchill and 5.4% at Kuujjuarapik. The blowing snow frequency was less than 1% in spring at the aforementioned locations. While Wang (2005) found that there was more blowing snow in

123 both fall and winter in western Hudson Bay than in eastern Hudson Bay, the reverse was true for fog in summer. Kuujjuarapik had a long-term average of 16.2% for fog in summer, much greater than Baker Lake’s 3.8% and Churchill’s 8.7%. This evidence further shows that other weather patterns also exhibited east-west asymmetry.

3.5.1.1 Uncertainties in Fog and Ice Fog Observation

The mandated change of fog and ice fog definitions plus the shift of reporting ice fog to freezing fog in MANOBS in 1999 was required for Canada’s aviation meteorological observation to remain consistent with other countries and its effects could not be avoided. This change that brought in freezing fog was due to International Civil Aviation Organization

(ICAO)’s desire for due diligence (L. Baker, personal communications, October 15, 2018). The change lowered the number of observations for fog and ice fog after 1999 as the surface horizontal visibility required for reporting either event was lowered from under 6 statute miles

(9.7 km) to under 0.5 statute miles (0.8 km), thus narrowing the range where fog and ice fog could be reported by the human observer. It skewed the trends for both observations after 1999.

However, the impact from changes in MANOBS appeared to be less than anticipated.

Trends in fog appeared to be more affected by the change in MANOBS visibility requirements than ice fog (Figure 3.2). After 1999, increase in fog hours were observed at two stations (Ivujivik and Wemindji). No trends were observed after 1999 at three stations (Arviat,

Churchill and Puvirnituq) and no data was available at Moosonee due to lack of consistent data after 1992. Declining trends were observed at ten stations (Akulivik, Baker Lake, Chesterfield

Inlet, Inukjuak, Kuujjuarapik, La Grande Riviere, Rankin Inlet, Sanikiluaq, Umiujaq and Whale

Cove). Of these ten stations reporting declining fog hour trends, only four locations (Akulivik,

124

Chesterfield Inlet, Sanikiluaq and Umiujaq) appeared to be affected by the change in MANOBS as the declining trend began after 1999. The other six locations recorded declining fog hours before the change in MANOBS in 1999. Thus, declining fog hours prior to 1999 would be likely due to warmer temperature that led to a decline in relative humidity (Ye, 2009) while the decline after 1999 would be attributed to warmer temperatures and change in observation procedures.

For ice fog (Figure 3.3), Sanikiluaq appeared to have an increasing trend prior to 1999 and reversed into a decreasing trend after 1999. Moosonee had no data available after 1992 and

Wemindji had no ice fog trend due to it being next southernmost location after Moosonee, which suggested that the temperature was possibility too warm for ice fog to occur at Wemindji. The remaining locations experienced less ice fog even before the MANOBS reporting requirements for ice fog changed in 1999. Interestingly, under WMO Manual No. 306 published in 1995, freezing fog became a mandatory reported observation whether or notice accretion was occurring or not between 0 and -30.0oC due to safety concerns (L. Baker, personal communications,

October 15, 2018). Yet there was no increase in ice/freezing fog observation after 1999 despite the requirement for fog to be reported. This evidence suggested that the decreasing observation for freezing fog due to warmer temperature or reducing the visibility range where freezing fog was reported was greater than the increase in freezing fog reports by mandatory reporting requirements. Therefore, similar to the trends in fog hours, the declining trends for ice fog prior to 1999 was due to warmer temperature which led to unfavourable conditions for the formation of ice fog. The continued decline of ice fog after 1999 was due to warmer temperatures and to a lesser extent, change in MANOBS and switchover from reporting as ice fog to freezing fog.

Weather conditions were measured and archived each hour along with other atmospheric variables such as temperature, humidity, pressure, and visibility in the form of Meteorological

125

Terminal Aviation Routine Weather Report (METAR). However, MANOBS stipulated that weather should be measured more frequently if the weather met one of the thresholds which were linked to the beginning or the end of significant weather events such as thunderstorm, freezing fog or high winds (Environment Canada, 2015). These additional observations were reported as Special observations (SPECI) in between the hourly observations. Since SPECIs were irregular observations and often took place not at the top of the hour, they were not systemically archived and made publicly available for long term analysis. Therefore, frequency of fog and ice fog were higher than the results presented if the duration of the fog or ice fog was shorter than an hour and did not take place at the top of the hour.

An additional factor that causes fog observation to be underreported was human error during the reporting step. The interface for human observers to report present weather was shown in Figure 3.6. The dropdown menu was sorted in alphabetical order. Occasionally, the observer would incorrectly choose the first item in the letter F category for weather as fog. This led to funnel cloud (FC) being reported as the present weather of the hour rather than fog (FG). If the observer failed to notice the error and issued a correction within that hour, the database would archive the weather conditions as funnel cloud instead of fog. As a result, some of the funnel cloud hourly reports were actually fog. This factor affects ice fog/freezing fog far less as the code for freezing fog is FZFG.

126

Figure 3.6: Illustration of the human observer’s present weather conditions reporting interface.

3.5.2 Visibility

In the three regions investigated, western Hudson Bay had the highest percentage of encountering reduced and low visibilities during the airport’s operating hours while James Bay had the lowest (Table 3.6). There were mixed signals for reduced visibility and low visibility trends within each region (Table 3.7). Similar to fog and ice fog, negative autocorrelation was present for reduced visibility at Rankin Inlet which suggested that the hours with reduced

127 visibility would fluctuate from many in one year to very few in the next year and then back to a lot of hours. It is not clear why this is the case.

Visibility studies in this region were scant and mostly related to tourism in the Arctic region, particularly in Scandinavia (Førland et al., 2012; Denstadli & Jacobsen, 2014; Stewart et al., 2010). Poor visibility not only affects the safe operations of aircraft but also the tourism sector. Denstadli and Jacobsen (2014) described that having poor visibility diminished the satisfaction of tourists as they negatively affected sightseeing and taking photos of the site. The cruise ship industry planned to visit Chesterfield Inlet, Rankin Inlet, Arviat, Churchill in western

Hudson Bay and Inukjuak in eastern Hudson Bay in 2006, 2008 and 2009 (Stewart et al., 2010), bringing much-needed tourism spending to these communities. This research seemed to suggest that fog hours per year would be largely unchanged as the magnitude was minimal at these locations (Table 3.7). Arviat was having roughly one additional hour of fog per year while other locations visited by the cruise ships had up to two fewer hours of fog per year. The visibility conditions were also not in favour for tourists heading to Hudson Bay region. All of the communities except those in James Bay had at least 2.0% of observed visibility less than 800 m.

This contrasted with Arctic tourist destinations in Norway where six out of seven locations studied had much better visibility (Førland et al., 2012). In their study, they examined the observed frequency for visibility less than 1 km, which was more relaxed than this study’s criteria, 800 m, and yet those six Norwegian locations had less than 1.4% of observation meeting their criteria. In addition, tourists in Canada also had to pay much more to travel to the Arctic than other Arctic countries. Mallory et al. (2018) found that it costed 50% more to travel from the low Arctic to high Arctic within Greenland comparing to travelling from Denmark to

Greenland. However, in Canada, it was 900% more expensive to travel from Halifax to Resolute

128

Bay, Nunavut via Ottawa than from Halifax to Vancouver, which is roughly equal distance for both flights. Hudson Bay communities had to compete against similar Arctic destinations that had better weather (visibility) and costed less to visit. They were also prone to adverse weather conditions as snow and blowing snow were often two of the top three weather variables causing reduced or low visibility (Table 3.8). But tourists could mitigate these factors by travelling to

Canadian Arctic destinations in the summer months instead of visiting in winter.

Further south of Hudson Bay, Hori et al. (2018) interviewed indigenous population near

Moosonee on winter roads usage and their study participants described snowstorms and blizzards in the region to be weaker and lasted shorter. Another group of researchers (Wendler & Shulski,

2009) examined Fairbanks, Alaska. They determined that extreme cold days (< -40oC) were declining by 6 days over the last 100 years and extreme cold days was associated with the formation of ice fog which led to lower visibility. Having less extreme cold days would lead to less ice fog formation, which in turn reduces the frequency of lower visibility. These studies seemed to be in agreement with a decline in hours with ice fog, reduced visibility and low visibility.

Visibility trends appeared to be highly variable depending on the location (Table 3.7).

Two communities (Baker Lake and Churchill) in western Hudson Bay, one community

(Kuujjuarapik) in eastern Hudson Bay and two communities in James Bay (La Grande Riviere,

Moosonee) experienced a significant decline (p < 0.05) in the frequency of reduced visibility.

However, three communities in the northern part of eastern Hudson Bay (Ivujivik, Akulivik and

Puvirnituq) experienced a significant increase in the frequency of reduced visibility. Similar results could be seen for low visibility. Same communities that encountered reduced visibility less often were also facing less low visibility. The significance levels for these findings were

129 strengthened to p < 0.01 or better. On the other hand, only Ivujivik in eastern Hudson Bay faced a higher frequency of low visibility.

As described by Rae (1954), the probability of fog decreased when the temperature was below 0oC and then the probability increased starting from -28.9oC to -34.4oC. Below -40oC, ice fog dominated but it should not impact airport operations greatly. At Resolute, Nunavut, visibility dropped to less than 0.8 km if the reported weather was blowing snow and the wind speed exceeded 48.3 km/h (Rae, 1954). Hanesiak and Wang (2005) found that 30 to 40% of blowing snow events resulted in visibility of 0.8 km or less. Since Transport Canada defined reduced visibility as horizontal eye-level visibility as less than 0.8 km (Transport Canada, 2017), study results by Rae (1954) and Hanesiak and Wang (2005) could be important for visibility analysis as they aligned with the current definitions for reduced visibility. Using Rae (1954)’s findings, visibility could be expected to be less than 0.8 km if blowing snow and wind speed exceeded 48 km/h.

3.5.2.1 Uncertainties in Visibility Records

Similar to fog and present weather conditions, visibility was measured and reported at the top of each hour as METAR. For visibility, SPECIs were issued whenever visibility crossed one of the threshold values (0, 0.4, 0.8, 1.2, 1.6, 3.2 and 4.8 km). Similar to the under-reporting of fog, restricted visibility conditions were not systemically archived and made publicly available for long term analysis if it did not take place during the window for METAR observation.

Therefore, frequency of reduced and low visibilities were higher than the results presented if the duration of the poor visibility was short.

130

3.5.3 Weather Monitoring

Aside from MANOBS changes, an external, non-climatic, factor was identified as a potential source for influencing fog, ice fog and visibility observations at Rankin Inlet.

According to the weather station inspection report from May 26, 1985 (Environment Canada,

1985), an inspector made the following comment:

“Visibility chart – although having been signed off by a previous inspector, the staff complained of the lack of accurate markers on the chart. Using airport diagrams, topographic maps and much effort, a vsby chart agreeable to the staff was prepared.”

Based on this information, it is very likely that the staff lacked the proper tools to assess visibility accurately. The correction was made and as a result, the accuracy of visibility improved.

Low visibility appears to be more accurate because the distance is much closer (400 m) and less prone to deceive observer's eyes while distance further away may appear to be within the definition of reduced visibility (800 m) but in fact could be over this threshold. Therefore, some of the significant declines for fog (Figure 3.2) and ice fog (Figure 3.3) at Rankin Inlet could be attributed to inaccurate observation and the trend caused solely by climate change would be less than the results from this study. Visibility observations (Figures 3.4 and 3.5) at Rankin Inlet prior to 1985 would also be affected. It is unclear whether similar observation issues were present at other stations, but the possibility for this as a source for error cannot be ruled out (Hanesiak &

Wang, 2005). Their research noted that visibility arose from fog and blowing snow and had an error rate of 10%. Nonetheless, each of the community studied has very little industry and population due to their remoteness, which allows them to be good study sites to monitoring fog frequency caused by environmental change and anthropogenic climate change while minimizing other anthropogenic factors such as air pollution from vehicle emissions, intense human activity and urban heat island from urban environment (Bokwa et al., 2018; Chen et al., 2018).

131

While almost all of the airport weather stations in Hudson Bay were used in this analysis, the study duration was often truncated at many sites due to a switch in observation frequency or missing hourly weather conditions. In Gough and He (2015) paper, they presented a way to detect the presence of fog in winter, spring and summer by examining the difference between mean hourly temperatures and the mean temperature calculated by the average of the daily maximum and daily minimum temperatures. Due to the radiative properties of fog, it elevates the minimum temperature of the air just before sunrise. This approach can be suitable to detect fog at sites that use automated weather stations (AWOS) that currently do not report significant weather (e.g. Enndai Lake, Nunavut located west of Arviat), or current sites that have human observer to record weather conditions during the day and complimented by an AWOS that measures a small set of variables (temperature, relative humidity, wind speed, wind direction and station pressure) during the night when the observer is not on duty, such as Chesterfield Inlet and

Wemindji. It could also backfill data from Moosonee from 1995 to 2010 to bridge the observation gap where no fog observation was done or locations with sparse overnight observations in earlier years (e.g. Baker Lake from 1953 to 1962 and Inukjuak from 1953 to

1976).

3.5.4 Flight Safety

Any of the factors that cause low visibility impedes safe aircraft navigation during takeoff, landing and taxiing. Gultepe et al. (2015) stated that ice fog posed a greater risk to aviation than snow because ice fog could adhere to the surface of the aircraft better than snow in a cold environment. Less ice fog would lead to less frequent ice build up on the wings and the fan blades of the engines. Accidents in the US, particularly take-off overruns and crashes after

132 takeoff, occur more often when the visibility was below 0.3 km (Wong et al., 2006). Pilots also described that they had to taxi at half to a third of the normal speed under low visibility conditions and feared that they would not see obstacles such as other aircraft or fuel trucks

(Andre, 1995). Low visibility requires greater separation distance between departing and arriving aircraft. These safety precautions lead to delays or flight cancellations. From these evidence, it was concluded that fog and ice fog frequencies were decreasing and that risks presented by these factors alone will be reduced under warmer conditions in the area from climate change. Since ice fog was reported as freezing fog after 1999 (ICAO 2008a), the declining trends for freezing fog after 1999 were also partly attributable to warmer temperatures in this region. Except for several important transfer points which operate their airports 24 hours a day, the delays are unlikely to cause capacity issues at the airports studied in this research due to the low volume of traffic they receive. However, the overall visibility conditions in this region remain largely unchanged.

Blowing snow, snow, ice and fog remained hazards to flying into and out of these communities.

These meteorological events can cause flight delays and cancellations that lead to a backlog of cargo and passengers at the transfer airports until the weather improves.

3.6 Conclusion

Ice fog was declining rapidly by 1 to 10 hours per year at the sixteen airports during the study period. Fog was also declining mostly at western Hudson Bay and James Bay airports by 3 to 7 hours per year. Both of these trends appeared to be linked to warmer air temperature in the region caused by climate change. James Bay airports had the lowest frequency in encountering reduced and low visibilities, followed by eastern Hudson Bay. Western Hudson Bay airports had the greatest number of hours with reduced and low visibilities while the airports were operating.

133

There was no consistent trend for reduced visibility as some locations experienced significantly more hours with reduced visibility while some locations experienced significantly fewer. The trend was clearer for low visibility. Most locations had fewer hours of low visibility and thus flight safety would generally improve. The leading causes for reduced and low visibilities were fog, blowing snow and snow. Since fog and ice fog were declining, the risks presented by these factors would also decline. However, non-fog factors continued to present risks during reduced and low visibility conditions which could affect both aircraft and airport operations.

This study demonstrated the use of weather data from non-24 hour stations for climatological research. It showcased the suitability of using these stations which are common at many northern Canadian airports for research purposes. Using hourly weather conditions and visibility data, it improved the understanding of historical occurrences for fog, ice fog, reduced visibility and low visibility at these communities which are not frequently studied. The research findings on fog and visibility were not only beneficial to aviation but also applicable to other transportation modes and activities. Driving, snowmobiling and boating would become safer as visibility improves. As tourists often desire destinations with good visibility and less fog, tourism and cruise industry also benefits from the general increase in visibility. Self-subsistent activities such as hunting and fishing would also be safer.

3.7 Acknowledgements

We would like to thank John Macphee and Lorne Baker of Environment and Climate

Change Canada for their assistance in providing ICAO documents and identifying the changes in

MANOBS for fog and freezing fog observations.

134

3.8 References

Andre, T. A. (1995). Information requirements for low-visibility taxi operations: What pilots say. In R.S. Jensen & L.A. Rakovan (Eds.), Proceedings of the Eigth International Symposium on Aviation Psychology (pp. 484-488). Columbus: Ohio State University. Baker, R., Cramer, J., & Peters, J. (2002). Radiation Fog: UPS Airlines Conceptual Models and Forecast Methods. Paper presented at the 13th Conference on Applied Climatology and the 10th Conference on Aviation, Range, and Aerospace Meteorology, Portland, Oregon, United States. Retrieved from https://ams.confex.com/ams/13ac10av/techprogram/paper_39165.htm Bokwa, A., Wypych, A., & Hajto, M. (2018). Impact of Natural and Anthropogenic Factors on Fog Frequency and Variability in Krakow, Poland in the Years 1966–2015. Aerosol and Air Quality Research, 18, 165-177. Burbidge, F. E. (1951). The modification of continental polar air over Hudson Bay. Quarter Journal of the Royal Meteorological Society, 77(333), 365-374. Chen, X., Li, X., Yuan, X., Zeng, G., Liang, J., Li, X., … Chen, G. (2018). Effects of human activities and climate change on the reduction of visibility in Beijing over the past 36 years. Environment International, 116, 92-100. Cleveland, W. S. (1981). LOWESS: A Program for Smoothing Scatterplots by Robust Locally Weighted Regression. The American Statistican, 35(1), 54. Denstadli, J. M., & Jacobsen, J. K. S. (2014). More Clouds on the Horizon? Polar Tourists’ Weather Tolerances in the Context of Climate Change. Scandinavian Journal of Hospitality and Tourism, 14(1), 80-99. Environment Canada. (1985). Station inspection report. Environment and Climate Change Canada. (2015). MANOBS – Manual of Surface Weather Observations. 7th edition, Amendment 19. Førland, E. J., Steen Jacobsen, J. K., Denstadli, J. M., Lohmann, M., Hanssen-Bauer, I., Hygen, H. O., & Tømmervik, H. (2013). Cool weather tourism under global warming: Comparing Arctic summer tourists' weather preferences with regional climate statistics and projections. Tourism Management, 36, 567-579. Gagnon, A. S., & Gough, W. A. (2006). East–west asymmetry in long-term trends of landfast ice thickness in the Hudson Bay region, Canada. Climate Research, 32(3), 177-186. Gough, W. A., & He, D. (2015). Diurnal temperature asymmetries and fog at Churchill, Manitoba. Theoretical and Applied Climatology, 121(1-2), 113-119. Gultepe, I., Zhou, B., Milbrandt, J., Bott, A., Li, Y., Heymsfield, A., . . . Cermak, J. (2015). A review on ice fog measurements and modeling. Atmospheric Research, 151, 2-19. Hamed, K. H., & Rao, A. R. (1998). A Modified Mann-Kendall Trend Test for Autocorrelated Data. Journal of Hydrology, 204, 182-196.

135

Hanesiak, J. M., & Wang, X. L. (2005). Adverse-Weather Trends in the Canadian Arctic. Journal of Climate, 18, 3140-3156. Hori, Y., Gough, W. A., Tam, B., Tsuji, L. J. S. (2018). Community vulnerability to changes in the winter road viability and longevity in the western James Bay region of Ontario’s Far North. Regional Environmental Change, 18(6), 1753-1763. International Civil Aviation Organization. (2008a). Aerodrome Meteorological Observation and Forecast Study Group (AMOFSG) Seventh Meeting Study Notes No. 12. Montreal, Canada. International Civil Aviation Organization. (2008b). Aerodrome Meteorological Observation and Forecast Study Group (AMOFSG) Seventh Meeting Summary of Discussions. Montreal, Canada. Jenama, R. K., & Kumar, A. (2013). Bad weather and aircraft accidents – global vis-à-vis Indian scenario. Current Science, 104(3), 316-325. Koetse, M. J., & Rietveld, P. (2009). The impact of climate change and weather on transport: An overview of empirical findings. Transportation Research Part D, 14(3), 205-221. Kowal, S., Gough, W. A., & Butler, K. (2017). Temporal evolution of Hudson Bay Sea Ice (1971–2011). Theoretical and Applied Climatology, 127(3-4), 753-760. Lescop-Sinclair, K., & Payette. S. (1995). Recent Advance of the Arctic Treeline Along the Eastern Coast of Hudson Bay. Journal of Ecology, 83(6), 929-936. Leung, A. C. W., & Gough, W. A. (2016). Air mass distribution and the heterogeneity of the climate change signal in the Hudson Bay/Foxe Basin region, Canada. Theoretical and Applied Climatology, 125(3), 583-592. Mallory, M. L., Gilchrist, H. G., Janssen, M., Major, H. L., Merkel, F., Provencher, J., & Strøm, H. (2018). Financial costs of conducting science in the Arctic: examples from seabird research. Arctic Science (in press). Maxwell, J. B. (1981). Climatic Regions of the Canadian Arctic Islands. Arctic, 34(3), 225-240. Rae, R. W. (1954). Meteorological activities in the Canadian Arctic. Arctic, 7(3/4), 119-128. Scott, D., McBoyle, G., & Schwartzentruber, M. (2004). Climate change and the distribution of climatic resources for tourism in North America. Climate Research, 27, 105-117. Sen, P. B. (1968). Estimates of the Regression Coefficient Based on Kendall's Tau. Journal of the American Statistical Association, 63(324), 1379-1389. Stewart, E. J., Tivy, A., Howell, S. E. L., Dawson, J., & Draper, D. (2010). Cruise Tourism and Sea Ice in Canada’s Hudson Bay Region. Arctic, 63(1), 57-66. Theil, H. (1950). A rank-invariant method of linear and polynomial regression analysis. Proceedings of the Royal Netherlands Academy of Sciences Part I, Part II, Part III, 53, 386-392, 521-525, 1397-1412.

136

Thornes, J., Rennie, M., Marsden, H., & Chapman, L. (2012). Climate Change Risk Assessment for the Transport Sector. London, United Kingdom: Department for Environment, Food and Rural Affairs. Tjernström, M., Shupe, M. D., Brooks, I. M., Persson, P. O. G., Prytherch, J., Salisbury, D. J., … Wolfe, D. (2015). Warm-air advection, air mass transformation and fog causes rapid ice melt. Geophysical Research Letters, 42(13), 5594-5602. Transport Canada. (2017). Canadian Aviation Regulations 300 Series – Aerodromes and Airports. Gatineau, Quebec. Transportation Safety Board of Canada. (2014). Aviation Investigation Report A11H0002. Gatineau, Quebec. Wang, X. L. (2006). Climatology and trends in some adverse and fair weather conditions in Canada, 1953-2004. Journal of Geophysical Research: Atmospheres, 111(D9), D09105. Weick, K. E. (1990). The Vulnerable System: An Analysis of the Tenerife Air Disaster. Journal of Management, 16(3), 571-593. Wendler, G., & Shulski, M. (2009). A Century of Climate Change for Fairbanks, Alaska. Arctic, 62(3), 295-300. Wong, D. K. Y., Pitfield, D. E., Caves, R. E., & Appleyard, A. J. (2006). Quantifying and characterising aviation accident risk factors. Journal of Air Transport Management, 12(6), 352-357. Ye, H. (2009). The influence of air temperature and atmospheric circulation on winter fog frequency over Northern Eurasia. International Journal of Climatology, 29(5), 729-734.

137

Analysing Historical and Modelling Future Soil Temperature at Kuujjuaq, Quebec (Canada): Aviation implications

Andrew C.W. Leung, William A. Gough, Tanzina Mohsin Department of Physical & Environmental Sciences, University of Toronto Scarborough 1265 Military Trail, Toronto, Ontario, Canada, M1C 1A4

138

Chapter 4: Soil Temperature

4.1 Abstract

Climate modelling is commonly used to project future temperature using various emission scenarios for a wide range of applications. In this work the impact of climate change on soil temperatures at Kuujjuaq, Quebec is assessed. First, long-term historical soil temperature records (1967-1995) are statistically analyzed to provide a climatological baseline for soils at 5 to 150 cm depths. Next, the nature of the relationship between atmospheric variables and soil temperature are determined using a statistical downscaling model (SDSM) and NCEP, a climatological data set. SDSM was found to replicate historic soil temperatures well. This relationship is then used to project soil temperatures for the remainder of the century using climate model output (Canadian Second Generation Earth System Model; CanESM2). Three climate projection scenarios were used from the IPCC AR5, RCP 2.6, 4.5 and 8.5. This study found that the soil temperature at this location may warm up at 0.9 to 1.2oC per decade at various depths. Annual soil temperatures at all depths are projected to rise to above 0oC for 1997-2026 period for all climate scenarios. Winter soil temperatures, however, are projected to remain below 0oC, suggesting a continuation of winter freezing.

139

4.2 Introduction

The Arctic experiences one of the fastest warming climates on Earth (Graversen et al.,

2008). In Canada, many indigenous peoples live in remote Northern communities without year- around road access. These communities rely heavily on air transportation to fly in goods, produce and passengers. Most soil temperature measurement stations in Canada, especially in the north, showed a warming trend (Qian et al., 2011). Subsurface warming could damage airport infrastructures such as runways, control towers, terminal buildings and fuel tanks, especially for airports underlain by permafrost. Kuujjuaq, formerly known as Fort Chimo, in northern Quebec was chosen as the study area because it has long-term soil temperature data (1967-1995) and it was located on airport property. Kuujjuaq is the administrative capital of the local Kativik regional government. Kuujjuaq lacks road access to other communities, but is a major transfer hub to other communities in the region, highlighting the importance of air service at this airport.

It is also an entry-exit point for cruise ship passengers heading to the Northwest Passage (Stewart et al., 2010). The permafrost extent at Kuujjuaq has been identified as discontinuous during the study period (Natural Resources Canada, 1985) and this suggests that the airport infrastructure may be vulnerable to melting permafrost as soil temperatures warm because the infrastructure did not anticipate climate change at the time it was built (Boucher & Guimond, 2012). In addition to these pragmatic concerns, Olelke and Zhang (2004) describe soil temperature measurements as an important and sensitive climate indicator because this variable reflects the integrated impact of climate processes such as surface air temperature, snowfall, evaporation rate and soil moisture variation.

There is a general lack of subsurface temperature measurements in Canada’s north, resulting in observation data gaps for analysis (Beltrami et al., 2003; Beltrami et al., 2006).

140

Therefore, subsurface temperature analysis has relied on related proxy variables. For example, one study used Stefan’s equation and mean air temperature to estimate subsurface temperatures and permafrost conditions through freezing degree-days in the present and future emission scenarios in northern Canada (Tam et al., 2015). A different study coupled water and heat transport into a dynamic global vegetation model to project soil temperatures in circumpolar region (Jiang et al., 2016). Another study used the relationship between mean annual air temperature and the existence of permafrost to map the extent of permafrost in Iceland and

Nordic countries (Harris et al., 2009). Gough and Leung (2002) examined permafrost in the

Hudson Bay region using a frost number analysis that linked surface air temperature to permafrost conditions and concluded soil moisture played a critical role in determining thermal thresholds. The coupling of air and soil temperatures appeared to be rather site-specific within

Canada (Beltrami & Kellman, 2003).

One study did use soil temperature records to directly measure the observed soil temperature changes but it did not conduct future soil temperature projections (Qian et al., 2011).

This study showed that most soil temperature measuring sites in Canada were located in southern regions. They were often measured at Agriculture and Agri-Food Canada and Reference

Climatological Stations (RCS). A soil temperature projection study was conducted in three sites in southern Quebec (Houle et al., 2012). Using the SRES A2 scenario, Houle et al. (2012) projected that the soil would warm by 1.1oC to 1.9oC by 2050 and 1.9oC to 3.3oC by 2080 depending on the soil depth. In their study, higher temperature increases were observed in summer and there was no clear spatial trend among the three sites. While such borehole measuring sites do exist in more northern regions of Canada, their measuring durations were often too short to perform climatological analysis. From Environment and Climate Change

141

Canada’s climate archives, Kuujjuaq was the only site with long-term soil temperature records near the Hudson Bay region suitable for climate analysis. Baker Lake and Schefferville, two of the sites investigated in Chapter 2 for wind analysis, had short observation records at 16 years and 3 years respectively. Other long-term soil measuring locations were further south in the

Quebec region, further west at Fort Smith in Northwest Territories or in the far north at Clyde

River and Resolute in Nunavut.

For this study, the research objectives for this work are to explore the temporal trends of soil temperatures in the Kuujjuaq historical record and to assess the impact of climate change on soil temperatures over the next century. Specifically, the following goals are addressed:

1) to assess temporal trends of the historical soil temperature record of Kuujjuaq,

Quebec and their relationship to concurrent air temperature,

2) using statistical downscaling to develop a robust relationship between soil

temperatures and a range of atmospheric variables, and

3) using the relationships developed to project soil temperatures at Kuujjuaq from 1997

to 2086.

4.3 Methods

4.3.1 Site Characteristics

Kuujjuaq, Quebec (Figure 4.1) is located at the western shore of the Koksoak River, which flows north and empties into Ungava Bay. The airport is located southwest of the community. There was no vegetation on top of the soil temperature measuring site when the earth thermistors were installed (Figure 4.2). Adjacent to the soil temperature site were other weather instruments such as an anemometer and a Stevenson screen, and two-storey buildings

142 such as the radio office and hydrogen building for radiosonde launches. During the installation of the earth thermistors at this site, a weather station inspection report on October 9, 1966 stated that “(a)t five feet deep, solid rock was hit and ten foot level could not be reached, therefore no data will be available for the 300 cm level” (Environment Canada, 1966). From a weather station inspection report in May 1972, the inspector described the surrounding areas as being characterized by gneiss rocks and the subsurface soil was frozen in permafrost with an active layer thawed at the surface (Environment Canada, 1972). The report mentioned that the site was surrounded by lichen woodland forest with tamarack, black spruce, willow and showy mountain ash trees. Vegetation transitioned into tundra about 30 km north of the station.

Figure 4.1: Site Location: Kuujjuaq, Quebec (58o05’42’’N, 68o25’20’’W).

143

Figure 4.2: Photographs taken during the installation of earth thermistors by Environment Canada’s staff at Kuujjuaq (formerly known as Fort Chimo) on October 14, 1966.

144

4.3.2 Data Collection

Soil temperature records at Kuujjuaq were obtained from Environment and Climate

Change Canada’s Climate Data Online (CDO). Daily data were available from January 1967 to

August 1995 at 5, 10, 20, 50, 100 and 150 cm depths. According to an Environment Canada’s undated weather station report, soil measurement program ceased at this location in September

1995 when the weather station service was contracted out. It was believed that this decision was made as part of the “program review” by the federal government in 1994 to 1995 to lower the expenditure of each department in an attempt to balance the budget (Kroeger, 1996). Soil temperatures were measured by Northern Electric Type 14B potted thermistor probes buried at those depths and had a corrected accuracy of 0.1oC (Kenneth Devine, personal communication,

September 22, 2017). While soil temperatures were recorded twice daily, this study only used the morning (8:00 am local time) soil temperature measurements due to less missing data in the record (Qian et al., 2011) and because afternoon (4:00 pm local time) measurements were not taken at 50, 100 and 150 cm depths (Environment Canada, 1984).

Surface air temperature at 2 m from the same time period was also downloaded from

CDO. The daily mean air temperature was derived from the average of highest and low hourly temperatures of each day, from 06:01 Universal Time Coordinated (UTC) to 06:00 UTC the next day. Hourly and daily air temperatures continue to be available to the present day.

4.3.3 Historical Data Analysis

To address the first research goal, historical soil temperature trends were averaged into annual and winter time series for all six levels. Winter was defined as spanning from November to April and summer as May to October given the northern location of this site. Trend analysis

145 was done with the Mann-Kendall test (Mann, 1945; Kendall, 1975) and the rates of changes were determined by the Theil-Sen slope estimators (Theil, 1950; Sen, 1968). Any significant p-values influenced by autocorrelated data were adjusted by modifying the variance of the data (Hamed &

Rao, 1998). In this modified Mann-Kendall test, a correction factor was calculated to determine the effective number of observations to account for the effects of autocorrelation on the significance level. The variance in a time series dataset with negative autocorrelation was reduced while variance in positive autocorrelation was increased (Hamed & Rao, 1998). Finally, the soil temperatures are correlated as a time series to the locally collected air temperature using a Pearson r analysis (Pearson, 1895).

4.3.4 Statistical Downscaling

To address the second research goal, the Statistical Downscaling Model – Decision

Centric version 5.2 (SDSM - DC) software, a tool created by Wilby et al. (2002) for modelling surface weather conditions such as temperature and precipitation was used. The software provided tools to determine which atmospheric variables from NCEP/NCAR reanalysis were best correlated to a predictant. The first step in the use of SDSM – DC was to determine if the observed data (soil temperatures at various depths) can be accurately reproduced by a combination of atmospheric variables taken from the NCEP/NCAR reanalysis gridded data

(Kalnay et al., 1996). NCEP/NCAR reanalysis data uses surface, satellite and other data to generate a representative value for the average air temperature change within the locations of the grid based on past data. Since Kuujjuaq airport has the longest observed dataset of all airports in its vicinity, NCEP/NCAR reanalysis data are strongly linked to Kuujjuaq’s historical air temperature records. Kuujjuaq’s soil temperature data was split into two halves. The first half

146

(1967 to 1980) was used for calibration and the second half (1981 to 1995) was used for verification. The calibration process involves selecting soil temperature at each depth in

Kuujjuaq and compared with the NCEP/NCAR reanalysis variables to determine how well those variables were linked to soil temperature with Pearson r correlation (Pearson, 1895). Top four variables with the highest Pearson r correlation were chosen as the components of the model for the validation process. The weather generator in the SDSM software creates the soil temperature model by using the selected four variables for the verification period. In the verification stage, the model’s soil temperature was compared with the observed soil temperature in the same period. Once the verification was satisfactory, the selected variables and the full period of historical soil temperature data (1967 to 1995) was used in the software’s weather generator component to project future conditions in conjunction with climate model output.

The modelling efficiency formula (MEF; Equation 4.1) evaluates the performance of the

SDSM model output of daily soil temperature data based on selected variables and compare that to the observed average (Stow et al., 2003). MEF generates a value between -1 and +1. A positive MEF value closer to +1 indicates that the modelled values were a close match with the observed values. A value of zero indicates that the model’s predicted value was no better than the observed average value. A negative value indicates the observed average value was better than the model’s individual predicted values. This validation exercise is repeated for all depths.

n n 2 2 (Oi  O)  (Pi  Oi ) i1 i1 MEF  n (4.1) 2 (Oi  O) i1 where n = number of observations; 푂푖 = observed value; 푂̅ = mean of observed values; 푃푖= predicted value

147

4.3.5 Climate Projections

The climate projections were generated using the relationships between soil temperatures and atmospheric variables in the previous section. However, the model would be projecting air temperature and not soil temperature in the SDSM – DC software. Therefore, it is necessary to establish the relationship between soil temperature and atmospheric variables from Chapter 4.3.4 and use that relationship to evaluate how future soil temperature will change due to warmer air temperature.

The climate model chosen to provide the atmospheric variables was the second generation Canadian Earth System Model (CanESM2). CanESM2 is a model developed by

Environment and Climate Change Canada’s Canadian Centre or Climate Modelling and Analysis

(CCCma) for IPCC Fifth Assessment Report (AR5). It captures the observed temperature variability in the 20th century in the Arctic environment reasonably well (Chylek et al., 2011). In addition, CanESM2 uses the Canadian Land Surface Scheme (CLASS) as its land-surface model, which contains soil temperature input (Verseghy, 2000), and the CanESM2 model output includes soil temperature as one of the variables (Environment Canada, 2017). The regional climate model driven by CanESM2, the Canadian Regional Climate Model (CanRCM4), did not have soil temperature as one of its output variables. According to Chylek et al. (2011)’s assessment, the CanESM2 model is also an improvement over the previous models developed by

CCCma because older models overestimated the Arctic warming rate by two to three times.

The calculation for the rate of change in the future is divided into two parts. In the first part, the soil temperature was assumed to change at the same linear rate in the future as in the historic trend. In the latter part, additional changes were calculated by using the Localizer Tool to determine the future projections for the change in air temperature at 2 m in different RCP

148 scenarios over the three projection periods (1997-2026, 2027-2056 and 2057-2086) under

CanESM2 model within the GCM grid (University of Toronto Scarborough, 2018). Change in air temperature was calculated by determining the difference between the historic and projected future mean air temperature. The air temperature change was converted into a percentage and this percentage was applied onto soil temperature. Changes of the future soil temperatures were assumed to be the same as in the historical period for the same depth. The relationships established in Chapter 4.3.4 were applied to climate model data (replacing the NCEP data used in the calibration process) for three 30-year time periods, ending on years 2026, 2056 and 2086.

This is a common approach for modelling future temperatures and a standard for SDSM projection studies.

Each of the future soil temperature depths was projected using three IPCC AR5’s

Representative Concentration Pathway (RCP) scenarios (RCP 2.6, RCP 4.5 and RCP 8.5).

Created for IPCC AR5, RCPs are four greenhouse gas concentration trajectories that allow climate projections without assumptions on population growth or economic development

(Meinshausen et al., 2011). This allows for a wide range of government policies, decision pathways and fossil fuel usage. RCP superseded previous climate change projection scenarios that were based on IPCC’s Special Report on Emission Scenarios (SRES), in which SRES had specific scenarios for population growth, economic development and fossil fuel usage.

RCP 2.6 scenario represents a greenhouse gas emission mitigation scenario, with emissions peaking at 2020 and decline rapidly afterwards (Meinshausen et al., 2011). RCP 4.5 scenario represents greenhouse gas emissions peaking at around 2040 and then decline while

RCP 8.5 scenario represents continuous rising global greenhouse gas emissions in the 21st century. The model output by the software generated 30 years of future daily soil temperatures

149 for 20 simulations at each of the six depths (5, 10, 20, 50, 100 and 150 cm). The soil temperatures for each day for each level were averaged across the 20 simulations to derive an averaged daily soil temperature. Then, thirty years of averaged daily soil temperature values were averaged to create a single representative value for that 30-year projection period. The steps were repeated for winter projections by using data from November to April, where the air temperature was below 0oC during these months, within the annual projection dataset.

4.3.5.1 Assumptions

A number of assumptions were made in the process of generating future soil temperatures.

It was presumed that the predictor variables with the strongest correlation with the historical daily soil temperature would also be strongly correlated in the future. Moreover, while the selected predictor variables would be strongly correlated for the entire study period, other predictor variables could be better correlated in winter or particular months. Aside from assuming the historical warming rate would form as the baseline trend, the changes to air temperature under the three RCP scenarios was assumed to have the same percentage of effects on the soil at each depth. There was no RCP 6.0 scenario available for CanESM2 and it is believed that the projection for RCP 6.0 would fall between RCP 4.5 and RCP 8.5. The variability of soil temperature at each depth in the future was assumed to be identical to the historical level. Snow depth in winter and any vegetation such as grass and shrubs growing on top of the soil were ignored even though both factors influenced the radiative forcing by reducing the incoming solar radiation that reached the surface of the soil. Snow depth in the future was also assumed to be held constant even though the depth was decreasing during the historical period at this location (Environment Canada, 2018). The Pearson r correlation would not consider the thermal lag, particularly at deeper depths. The temperature in deeper soils had

150

less magnitude and more distant from the fluctuations in air temperature. Moisture level, soil

type, organic matter contents and distribution within the soil were assumed to remain constant in

the future even though these variables would change over time. The study also speculated that

the instrument remained accurate and that regular maintenance on the instrument were performed

without disturbance to the soil.

4.4 Results

4.4.1 Historical Analysis

The statistical analyses for the period of 1967 to 1995 for the six soil temperature depths

are reported in Table 4.1. The analyses include average annual and average temperatures,

variability and temporal trends. The temporal trends were done using the Mann-Kendall test.

Theil-Sen slope estimators determined the magnitudes of the rates of changes. The significance,

p-values, was corrected for autocorrelation as appropriate. As noted above “winter” was defined

as November through April. Table 4.2 reports a similar analysis for the concurrent air

temperatures at 2 m measured at the same location.

Table 4.1: Historic average soil temperature and the rate of change over time for soil temperature at depths from 5 to 150 cm from 1967 to 1995. Bolded numbers indicate trends that are significant at p-values less than 0.10 (^), 0.05 (*) and 0.01 (**). Soil Depth Mean Trend (oC/decade) Annual Winter Annual Winter 5 cm -2.1oC -11.1oC +1.2* +1.9* 10 cm -2.4oC -10.8oC +1.1* +1.7^ 20 cm -1.9oC -9.9oC +1.3** +1.4^ 50 cm -1.4oC -7.9oC +1.0* +1.2 100 cm -1.6oC -6.2oC +0.9^ +0.8 150 cm -1.9oC -4.2oC +1.1** +1.0

151

Mean annual soil temperature at 5 cm was -2.1oC and the temperature was becoming

slightly warmer with increasing depth to 50 cm (Table 4.1). Below this depth, soil temperature

became colder. For winter, the soil temperature closest to the surface was the coldest and became

warmer with depth. All soil depths, except 100 cm, were warming significantly at p < 0.05 on an

annual scale and 100 cm depth warmed at a lower significance level, p < 0.10. However, only 5

cm was warming significantly at p < 0.05 in winter. Both 10 cm and 20 cm warmed at a

significance level of p < 0.10. Surface soil at 5 cm is warming at about 1.2oC per decade

annually. In winter, the warming rate was faster, at about 1.9oC per decade at the same level. At

all of the depths other than 100 cm and 150 cm, winter’s warming rate was larger than the annual

rate.

The results of the Pearson r correlation between coincident air and soil temperatures are

presented in Table 4.2. The relationship is strongest for the upper soil levels and for annually

averaged data as expected. The transfer energy into and out of the soil is determined by soil type

and moisture content. This leads to the muting of the surface forcing and thermal lags. Hence the

lowest correlations occur for the daily winter date at the 150 cm depth. However, the upper 50

cm correlates well with the surface forcing and this bodes well for the SDSM modelling in the

next section.

Table 4.2: Pearson r correlation between air temperature at 2 m and soil temperatures. Soil Depth Pearson r coefficient (Daily) Pearson r coefficient (Monthly average) Annual Winter (November Annual Winter (November to April) to April) 5 cm 0.845 0.453 0.955 0.746 10 cm 0.847 0.452 0.951 0.733 20 cm 0.848 0.450 0.946 0.722 50 cm 0.819 0.391 0.906 0.597 100 cm 0.721 0.268 0.797 0.407 150 cm 0.455 0.087 0.514 0.147

152

4.4.2 Statistical Downscaling

Using SDSM, historical soil temperature data (1967-1995) was statistically modelled using a suite of variables taken from the NCEP data set. The simulations included those created using three, four and six NCEP variables and four CanESM2 variables to test the effects in varying the number of variables used as well as comparing the performance between NCEP and

CanESM2. The top four atmospheric variables for all soil depths are surface specific humidity, mean air temperature at 2 m, 500 hPa geopotential height and specific humidity at 850 hPa. The linkage to the 2 m atmospheric temperature is expected given the results of the previous section

(Table 4.3). In most cases, the CanESM2 variables had a stronger correlation with the soil temperature than NCEP’s corresponding variables. While NCEP’s specific humidity at 850 hPa had a stronger correlation with soil temperature than CanESM2’s corresponding variable, the difference in correlation between the two models was very small and less than the correlation differences in other variables where CanESM2 had a stronger correlation. The strength of the relationship between the variables and soil temperature decreases as soil depth increases. Figure

4.3 shows the results of various SDSM simulations compared to observations for the 5 cm depth.

The reproduction of observed temperatures by the SDSM modelling is particularly good for the summer months but the models consistently underrepresent by a few degrees Celsius for the winter months (Figure 4.3a) suggesting that surface processes such as snow cover and phase change within the soil may not be captured by the model. The monthly variance pattern was also reproduced, but the greatest difference was found in March (Figure 4.3b). Differences between the number of variables used in the NCEP model or between the NCEP and CanESM2 models were barely distinguishable as seen in Figure 4.3. It was noted by Hassan and Harun (2012) that there was no standard rule for selecting which predictor variables to use and the process itself

153 was described as difficult and tricky. Since CanESM2’s predictor variables were shown to be better correlated to soil temperature than NCEP’s in most cases and at most depths, CanESM2 model was chosen to repeat statistical downscale modelling at other depths.

Comparisons between observed and CanESM2 modelled soil temperature at 10 to 150 cm depths were shown in Figures 4.4 to 4.8. At 10 cm, observed temperatures were higher than modelled in all the months and the difference between observed and modelled were greatest in winter. The model produced a similar variance in temperature as the observed from April to

October. Over-estimating the coldness of soil temperature in the CanESM2 model was also found in all of the months at 20 to 150 cm depths, with winter having the greatest overestimation while summer had the least. In general, the variability of the modelled output was similar to the observed data’s variability, which is an indication that the model’s variability was able to reproduce the historical variability. The variance in the historical data is mainly attributed to natural variability of the temperature in the soil. The variances of observed 5, 100 and 150 cm were greater than the modelled in winter but 5 cm had less variance in the observed than the modelled in summer. The variance of modelled 150 cm temperature approached 0oC2 from June to November but the observed was much higher in July and August. The variances of observed

10 to 50 cm were generally higher in the modelled data than the observed in most of the months.

154

Table 4.3: Pearson r correlation between the four selected variables and observed soil temperature, with bolded numbers indicating the model variable having a stronger correlation on the soil temperature than the other model. Soil Pearson correlation (r) Depth Surface specific Mean 2 m air 500 hPa Specific humidity humidity temperature geopotential height at 850 hPa NCEP CanESM2 NCEP CanESM2 NCEP CanESM2 NCEP CanESM2 5 cm 0.825 0.828 0.825 0.821 0.696 0.756 0.700 0.699 10 cm 0.826 0.832 0.828 0.830 0.693 0.752 0.700 0.697 20 cm 0.812 0.824 0.806 0.825 0.673 0.732 0.689 0.687 50 cm 0.817 0.833 0.811 0.831 0.658 0.711 0.694 0.682 100 cm 0.728 0.758 0.722 0.763 0.557 0.601 0.618 0.599 150 cm 0.466 0.504 0.462 0.525 0.314 0.337 0.390 0.365

155

(a)

(b)

Figure 4.3: Comparison between the observed monthly 5 cm soil temperature and the predicted values from the CanESM2 and NCEP models for (a) mean and (b) variance.

156

(a)

(b)

Figure 4.4: Comparison between the observed monthly 10 cm soil temperature and the predicted values from the CanESM2 model for (a) mean and (b) variance.

157

(a)

(b)

Figure 4.5: Comparison between the observed monthly 20 cm soil temperature and the predicted values from the CanESM2 model for (a) mean and (b) variance.

158

(a)

(b)

Figure 4.6: Comparison between the observed monthly 50 cm soil temperature and the predicted values from the CanESM2 model for (a) mean and (b) variance.

159

(a)

(b)

Figure 4.7: Comparison between the observed monthly 100 cm soil temperature and the predicted values from the CanESM2 model for (a) mean and (b) variance.

160

(a)

(b)

Figure 4.8: Comparison between the observed monthly 150 cm soil temperature and the predicted values from the CanESM2 model for (a) mean and (b) variance.

161

The fidelity of the modelling to observations is measured using Equation 4.1 (modelling efficiency formula, MEF). These are reported in Table 4.4. The performance of the modelling is consistent through to 100 cm and decreases below 100 cm. As all MEF values were above 0, the individual modelled values across all depths were a close match with the observed values and were better than using the observed average value as a representative for their respective depths.

Table 4.4: Modelling Efficiency (MEF) calculation for the CanESM2 model output. Modelled Depth MEF value (-1 to +1) 5 cm 0.83 10 cm 0.83 20 cm 0.84 50 cm 0.85 100 cm 0.81 150 cm 0.66

4.4.3 Projections

4.4.3.1 Annual

Future climate projections were conducted at the various depths sampled, first on all 12 months (Figure 4.9) followed by just winter months (Figure 4.10). In the projections that used all

12 months, soil temperature exceeded 0oC in the first projection period (1997 to 2026) in each of the CanESM2 model’s RCPs at all depths. During this period, there were little temperature differences between the three RCP trajectories. By the second projection period (2027 to 2056), projected RCP 8.5 soil temperature increased the greatest and deviated from RCP 2.6 and RCP

4.5 projections. RCP 4.5 projected a slightly warmer soil temperature than RCP 2.6. In the third projection period (2057 to 2086), the warming of RCP 8.5 increased at a greater rate than between the first and second projection periods. The warming rate for RCP 4.5 stayed approximately the same and RCP 2.6 slowed down.

162

(a) 5 cm

(b) 10 cm

163

(d) 50 cm

164

(e) 100 cm

(f) 150 cm

Figure 4.9: Annual soil temperature projections from 1997 to 2086 for depths at (a) 5 cm, (b) 10 cm, (c) 20 cm, (d) 50 cm, (e) 100 cm and (f) 150 cm.

165

4.4.3.2 Winter

In winter months, projections consistent with the local definition of winter, November to

April were considered. Winter projections (Figure 4.10) showed the same patterns as annual projections. Winter soil temperatures were projected to remain below 0oC at all depths in each scenario with the exception of RCP 8.5 at 150 cm by the third projection period (2057 to 2086).

The warming rate for RCP 8.5 scenario in winter appeared to be at a linear rate while RCP 2.6 and RCP 4.5 scenarios warmed at a slower rate than RCP 8.5 across the three projection periods.

(a) 5 cm

166

(b) 10 cm

167

(d) 50 cm

(e) 100 cm

168

(f) 150 cm

Figure 4.10: Winter soil temperature projections from 1997 to 2086 for depths at (a) 5 cm, (b) 10 cm, (c) 20 cm, (d) 50 cm, (e) 100 cm and (f) 150 cm.

4.5 Discussion

4.5.1 Historical Trends and Model Fitting

In this study, the long-term trends of soil temperature at Kuujjuaq were assessed by using an in-situ measurement record. Between 1967 and 1995, annual soil temperature warmed by 0.87 to 1.32oC per decade and winter soil temperature warmed by 0.84 to 1.85oC per decade (Table

4.1). Soil at 5, 10, 20, 50 and 150 cm depths showed significant warming with p < 0.05 and 100 cm depth showed a statistical significance at p < 0.10. The soil temperature was strongly positively correlated with air temperature (Table 4.2). The difference in the correlation between air temperature and soil temperature at various depths were mainly attributed to the thermal capacity of soil that acted as a buffer against radiative forcing from the surface. As the depth

169 increased, the buffer capacity was higher. In addition, time lag was also responsible for different response time to the surface warming or cooling. Interestingly, Jean (1991) conducted a soil temperature analysis at Kuujjuaq from 1967 to 1989 and found that the increase in soil temperature at all depths did not appear to be associated the increase in air temperature. He suggested that environmental conditions other than air temperature were responsible for the increase in mean soil temperature. But he also mentioned that short time series analysis might not capture the variability and significant trends and inferences drawn from the results should be treated with caution. An additional six years of soil temperature records in daily and monthly average scales as well as linkage with the GCMs during the model validation process strengthened the validity of this study’s claims that there was a strong correlation between soil and air temperatures at all depths. The Pearson correlation coefficients between air and soil temperatures at Kuujjuaq were also similar to other discontinuous permafrost regions in the

Arctic (Oelke & Zhang, 2004).

The number of variables beyond four and the use of NCEP vs. CanESM2 did not substantially improve the model accuracy, however, CanESM2 was better correlated to soil temperature than NCEP in most cases (Table 4.3). While surface air temperature’s correlation with soil temperature at 5 cm was established without using model data as illustrated in Table 4.2, other variables with high correlation in Table 4.3 could be explained by local weather. Surface specific humidity is linked to the short term of past weather. If the surface humidity is high, then the soil is likely to be moist and caused by recent precipitation events (rain or snow).

Precipitation events are almost always accompanied by clouds which lower the solar radiation reaching the surface and reduced the surface heating. Moreover, some of the energy is used as latent heat of evaporation. Drier soils at the surface allow greater temperature increase even with

170 the same amount of energy. Specific humidity at 850 hPa is also linked to clouds and cloud cover is directly linked to a reduction in solar radiation heating on both the air and soil close to the surface. The strong correlation between soil temperature and 500 hPa geopotential height appeared to be counterintuitive at first, since 500 hPa geopotential height is linked to large-scale mid-troposphere circulation (Skinner et al., 2002) and distant from the surface. Upon closer examination, the synoptic scale circulation patterns occurring at 500 hPa geopotential height are responsible for air mass and frontal movements which in turn affects mesoscale phenomenon such as high-pressure systems, squall lines and convective storms. They are all associated with the variation in solar radiation, cloud cover, precipitation, snow cover and air temperature at the surface (Jean, 1991). Therefore, all of the top four variables are related to surface heating.

The model performed reasonably well as demonstrated by positive MEF values across all depths (Table 4.4). The model performed particularly well from April to November in all soil depths (Figures 4.3a to 4.8a). The historical air temperature warming at Kuujjuaq is consistent with temperature warming trends in the Hudson Bay Region (Leung & Gough, 2016). Leung and

Gough (2016) and this study showed that the largest temperature warming occurred during winter.

Comparing with Tam et al. (2015)’s study, this study provides a more accurate baseline and future temperature for soil because this study used in-situ measurements rather than using air temperature and Stefan’s equation to determine the ice thickness level. Stefan’s equation requires the thermal conductivity of the soil as an input, which would require a site visit to obtain a soil sample for lab analysis, while this study does not require this variable. Coupling the air and soil temperature also ensured that the relationship between the two variables was directly measured

171 rather than indirectly linked by calculating frost factor from Stefan’s equation to determine the permafrost conditions, which could not generate an actual soil temperature value.

4.5.2 Projections

Under projected warming (Figure 4.9), the annual soil temperature would be above 0oC in the first projection period (1997-2026). In the second projection period (2027-2056), the warming rate will be similar for RCP 2.6 and RCP 4.5 scenarios while RCP 8.5 would have the greatest warming rate. The warming rate for RCP 8.5 in the third projection period (2057-2086) is further enhanced while RCP 4.5’s warming rate will be progressing at approximately the same rate as between the first and second projection periods. RCP 2.6 scenario’s warming rate slows down in the third projection period, which is largely explained by a reduction in global greenhouse gas emissions after 2020 (Meinshausen et al., 2011). A combination of reduction in fossil fuel consumption, government policy implementations to drastically reduce greenhouse gas emissions, improved efficiencies in existing technologies (e.g. automobiles and heating) and new technologies (e.g. carbon capture and storage) would lead to reduced radiative forcing in the atmosphere and cause lesser warming starting from second projection period. Winter projection trends in different RCP scenarios were similar to the annual’s corresponding RCP trends (Figure

4.10). For RCP 4.5 and RCP 8.5 scenarios, Phillips et al. (2014) suggested that the slowdown in the warming rate across different scenarios may be explained by soil drying. They found that under RCP 4.5 and RCP 8.5, 15 CMIP5 models showed soil temperature warming at a rate of 10% less than air temperature. They also hypothesized that soil drying leads to reduced thermal conductivity and heat capacity in the soil. However, after soil was dried, there would be more warming from sensible heat than latent heat as evaporation decreased.

172

At the 20 cm depth, the Zhang et al. (2005) simulations suggested that the soil warming rate was 0.8oC on a national level yet their model output indicated that the soil actually got cooler (-0.6oC) at Kuujjuaq during the 20th century. The cooling in this area was cited as an exception to the general warming but shallow layers of soil had increased by up to 2oC since the mid-1990s (Nelson, 2003). Unfortunately, this study site measurements stopped in 1995 and could not offer evidence to support this claim. This study did find that the 20 cm soil increased significant (p < 0.01) at 1.3oC per decade during the study period (Table 4.1). The strong correlation between air and soil temperatures close to the surface was not unexpected since the radiative forcing from the surface contributes to the warming of those soils. Deeper soils are less likely to receive surface energy by heat transfer. Another interesting observation was that the winter soil temperatures at 5 and 10 cm were warming faster than the annual trend. This showed that the magnitude of warming was greater from winter (November to April) than in summer

(May to October) at Kuujjuaq even though about 23 cm of snow cover was present during the winter period and that the snow cover reduced the heat conduction transferred from air to soil.

There are two possible explanations to explain this observation. The presence of soil layers could result in a delay in the heat transfer process, leading to a thermal lag. Another possibility was the latent heat effect in the soil that slowed the rising of soil temperature due to phase change, known as the zero-curtain effect (Mühll et al., 1998).

This study’s projections also provided more information on permafrost degradation rate than Tam et al. (2015). Tam et al. (2015) determined the future permafrost conditions based on the relationship of air temperature and frozen ground based on Stefan’s equation. This study coupled the air temperature with soil temperature with atmospheric variables for projections.

173

Instead of outputting a ratio that determines the coverage of the permafrost, this study provided a projection for the actual soil temperature.

The observed difference between soil temperature and air temperature in Canada varies greatly and highly location dependent, with some sites showing the increase in air temperature was less than that of soil temperature while others showed greater warming in soil temperature than air temperature (Zhang et al., 2005). At Kuujjuaq, Zhang et al. (2005) found that the difference between the changes in soil temperature was greater than air temperature by 0.1 to

0.2oC between the 1901-1910 and the 1986-1995 periods. Other variables that affect the relationship between air and soil temperatures include thermal conductivity and the type of soil at the site. Wen et al. (2003) studied the accuracy of remote sensing at Tibetan Plateau by comparing the results from the microwave remote sensing algorithm with the actual soil temperature measurements from 4 to 200 cm in those sites. They found the average difference between estimated and actual soil temperature was around 0.5oC. Gómez et al. (2016) used a regional climate model and remote sensing on soil temperature up to a 150 cm depth to simulate extreme heat events in Valencia, Spain. Given that some of the soil observation sites in Canada began recording data as early as 1958 (Qian et al., 2011), it is believed that these records can be joined with remote sensing data to create a longer observation period. This is especially important since many soil observation sites in Canada, particularly north of 60o, were too short to be used to create reliable climate analysis. Satellite data is not available in earlier years, making in-situ data collection the only source for those years. The measurements taken from the ground can be used to extend a location’s soil temperature regime backwards. Due to the soil observation program shutting down by 1995 at Kuujjuaq and by the early 2000s across most

174 stations in Canada, remote sensing can extend the availability of soil temperature data to create longer records and allow more accurate projections based on additional soil temperature data.

Houle et al. (2012)’s study in southern Quebec achieved a correlation coefficient of ≥0.96 with the observed mean monthly temperature in their model output. This study’s CanESM2 model performed better, as it achieved a correlation coefficient of >0.99 between monthly mean modelled and observed temperature. Several notable differences between Houle et al. (2012) and the current study were identified. While both of the studies were conducted in Quebec, one major difference is the distance between the sites involved. This study site was over 1000 km north of their northernmost site. Another major difference was the lack of incorporating snow data into this study’s model. Lack of snow in this model introduced inaccuracy due to the difference between air and soil temperatures because of snow cover (Zhang et al., 2018). Nonetheless, the

CanESM2 modelled output still performed much better than Houle et al. (2012)’s model.

According to the Climate Normals at Kuujjuaq, the 30-year average snow depth between 1961-

1990 was 23 cm (Environment Canada, 2018). The average snow depth dropped to 19 cm between 1971-2000 and then to 17 cm between 1981-2010. This model’s accuracy would continue to improve as the snow depth at this location was decreasing. The lack of trees and organic layer at Kuujjuaq may have also unintentionally improved the model performance as it did not account for this variable. This was explained by Zhang et al. (2018) as they found that the response of air temperature warming on soil temperature was quicker at climate stations which had low vegetation, such as Kuujjuaq, than at densely forested locations. The zero-curtain effect is not captured by this model as the SDSM program did not factor in the phase change of water in the soil during the overall warming nor the freeze-up process at the beginning of each winter.

175

4.5.3 Uncertainties

4.5.3.1 Observation Uncertainties

One of the major uncertainty around soil temperature projections in northern Canada was the scarce number of soil temperature monitoring sites. Most of the soil monitoring sites were located in southern parts of Canada (Qian et al., 2011) and the existence of those sites was primarily for agriculture purposes. Currently, soil temperature continues to be measured at many of these agriculture sites on an hourly scale. The current data has finer resolution than the historical data used in this study, where morning soil temperature was considered to be more complete than the afternoon temperature and the morning temperature was used as a representative of the temperature for the entire day (Qian et al., 2011). This presents an uncertainty over the recorded temperature at the site. The current measuring instrument is believed to be more accurate than the older versions. The current temperature probe, Campbell

Scientific CSI 109-L has a tolerance of ±0.1°C at 25oC and the accuracy drops as temperature deviates away from 25oC. This instrument reports the temperature up to one-hundredth of the decimal place. The older Northern Electric Type 14B thermistors that were used in this study had an accuracy of ± 0.1oC with unknown temperature range for accuracy and the thermistors reported the values up to one-tenth of the decimal place. Unfortunately, the accuracy of soil data dropped when an installation of instruments in December 1978 (Environment Canada, 1978) resulted in the recorded value to be rounded to nearest 0.5oC at all depths until the end of the study period in 1995. However, Jean (1991) noted that there were no significant changes in observation or site methods for soil measurements at Kuujjuaq. It was unclear what caused a decrease in the observation accuracy.

176

Repairs to the thermistors would disturb the soil as the process involved digging the soil to retrieve the instrument from the ground and infill the soil after repairs were completed.

Digging into the ground in northern sites also posed a challenge as it was common to encounter permafrost or bedrock before reaching the desired depth. Bedrock prevented Kuujjuaq from having soil temperature at 300 cm (Environment Canada, 1966). Vegetation growth on top of the soil reduced the solar radiation from reaching the ground. Vegetation also reduced the evaporation rate by convection as it slowed the wind speed at the top surface of the soil.

Majorowicz and Minea (2015) suggested that the heat flow at Kuujjuaq was between 44 to 49 mW/m2. However, the site conditions were likely compacted soil or gravel as the instrument was in the vicinity of the airport. Nonetheless, it was believed that the thermal diffusivity at Kuujjuaq was much higher than forested locations with moss or leaf litter as the organic layers in the forest were better insulators.

Soil moisture was not one of the variables measured at this site but it was directly related to soil temperature and evaporation (Miralles et al., 2012). The coupled relationship between soil moisture and temperature would allow monitoring indirectly from satellite and extend the data availability beyond 1995. A dataset with longer study period would improve the validation process and more robust in the accuracy of future projections. But as Beltrami and Kellman

(2003) pointed out, the coupling between air and soil temperature in Canada was site-specific.

Having direct soil temperature measurements on site would always be superior over drawing inferences from models through soil moisture and temperature coupling (Miralles et al., 2012) or

Stefan’s equation to calculate frost factor (Tam et al., 2015).

177

4.5.3.2 Modelling and Projections Uncertainties

The CanESM2 model was an improved model over older generations of Canadian GCMs.

Chylek et al. (2011) found that CanESM2 was more in agreement with observed air temperatures in the Arctic from 1900 to 1970 but it overestimated the warming after 1970. This raised a question on whether the warm bias would persist into the future. CanESM2 model incorporates the Canadian Land Surface Scheme (CLASS) which uses three depth layers at 0-10 cm, 10-35 cm and 35-410 cm (Verseghy, 2000). Slater and Lawrence (2013) described the third CLASS layer was subjected to large temperature fluctuations because this layer has a shallow upper bound while the lower bound has low latent heat sink due to bedrock. This could explain some of the issues seen in winter where there was a cold bias for the model at 5 cm and a warm bias for depths at 10 cm and deeper. However, deficiencies in the land-surface model were not limited to

CanESM2. As an example, HadGEM2 and GFDL’s land models minimized the insulating effects of snow and caused cold bias in the soil (Slater et al., 2001; Koven et al., 2013). This demonstrated that cold bias could stem from a variety of reasons in different models but ultimately affected the projection of soil temperatures that had an impact of permafrost distribution. Under the present-day conditions, not using snow depth in the SDSM process introduced uncertainty into the projections. As the snow depth diminishes in the future, the uncertainties related to snow depth will decrease over time and can result in greater temperature fluctuations in deeper soil. The SDSM approach also failed to reproduce the general effects of zero-curtain effects or thermal lag in the deeper soil in both the historical and future time periods since the method was designed for climate projections on surface weather and not for soil. All of the SDSM predictors were above-surface variables, which introduced additional uncertainties for projecting below-surface parameters. Air temperature and surface precipitation were not prone to

178 zero-curtain effects or thermal lag but soil temperature would often be affected by both factors.

The zero-curtain effect is especially problematic to measure at this site due to the coarse resolution of measuring instrument, especially after December 1978, as described in the previous section. A mathematical model may be required to account for zero-curtain effects on soil temperature.

There was a major difference between this study and some of the permafrost studies published in the 1990s in this region. In Wang and Allard (1995) and Allard et al. (1995), these studies used Kuujjuaq and Iqaluit’s air temperature records to analyze permafrost distributions and soil temperatures at Salluit, Kangiqsujuaq and Quaqtaq which are located northwest of

Kuujjuaq. Both studies suggested a recent cooling of air and soil temperatures in this area. Wang and Allard (1995) observed a cooling rate in air temperature at 0.02oC per year over a 40-year period from 1947 to 1988 in Kuujjuaq. Allard et al. (1995) found that the soil was cooling at

0.05oC per year at the three sites and the cooling trend was observed at first 5 to 7 m of soil at

Salluit. Using the observed air and soil cooling trends as the premise, Wang and Allard (1995) modelled the future soil temperatures to 2044 at Salluit and found that the active layer thickness would decrease due to the continuing cooling trends at the surface. The observed trends in Wang and Allard (1995) and Allard et al. (1995) contrast greatly with more recent research in the same area such as this study, Boucher and Guimond (2012) and Doré et al. (2012). This study found significant warming at the top 1.5 m of the soil (Table 4.1). Contrary to Wang and Allard (1995) and Allard et al. (1995), more recent studies by Fortier et al. (2011), Boucher and Guimond

(2012) and Doré et al. (2012) documented photographic evidence of permafrost degradation that was causing damage to the service road and drainage systems at the airports. Unlike Wang and

Allard (1995) which suggested surface cooling in the future, this study found that warming

179 would continue in the future at all the depths (Figures 4.9 and 4.10). Wang and Allard (1995)’s results also contrasted with the CanESM2’s GCM output for soil temperature (Environment

Canada, 2017). Therefore, Wang and Allard (1995) and Allard et al. (1995) drew different conclusions in their studies based on the trends in the time period that they studied and newer research appeared to have invalidated their findings.

The CanESM2, without SDSM, produced historical soil temperature at this site poorly.

Using January 1967’s monthly mean soil temperature as an example, the observed temperature was -23.4oC at 5 cm and -22.3oC at 10 cm. Without SDSM, CanESM2’s historical soil temperature at first depth layer (0-10 cm) within Kuujjuaq’s gridded cell was -7.9oC while the downscaled 20-ensemble mean was -15.2oC for 5 cm and 10 cm were -14.8oC. The results demonstrated that SDSM vastly improved the historical CanESM2 estimation of soil temperature at this location. Further analysis showed that sometimes CanESM2 gridded data were slightly more accurate than SDSM in the summer. In August 1995, the monthly observed mean soil temperature was 12.0oC at 5 cm and 11.2oC at 10 cm. The gridded data estimated the temperature to be 11.9oC between 0 and 10 cm while SDSM’s 20-ensemble mean was 10.8oC at

5 cm and 10.0oC at 10 cm. However, in the 2 months prior (June 1995), the observed soil was

10.3oC at 5 cm and 8.3oC at 10 cm while the SDSM modelled 8.5oC at 5 cm and 6.9oC at 10 cm.

Both of the SDSM modelled values were better than gridded data’s temperature at 0.9oC.

With the expected warming in the region, organic matter and vegetation such as moss would alter the surface layers of the thermal conductivity and hydraulic properties of the soil and this needed to be captured by models (Burke et al., 2018). The decomposition rate of organic matters in soil was projected to increase in the future, partly caused by the increase in soil temperature (Todd-Brown et al., 2014). However, most Earth system models perform poorly in

180 permafrost regions due to the lack of permafrost carbon dynamics in the current models (Todd-

Brown et al., 2014). Weather station inspection report in 1972 confirmed coniferous trees were growing in the surrounding area of Kuujjuaq (Environment Canada, 1972), indicating that the site was favourable for grass to grow at the surface of the soil where the measurement took place.

But existing site management practice dictated active trimming of grass and removal of shrubs growing on top of the measuring site (Environment Canada, 1984). As the temperature warms at

Kuujjuaq and the climate allows more vegetation to grow in the surrounding areas, this raised a question whether the soil temperature records located on a barren surface will be truly representative of the surrounding areas with trees and shrubs. In addition, Burke et al. (2018) highlighted the importance of a model in capturing the feedback cycle of additional CO2 emission through melting permafrost because the effects would also affect the salinity and humidity of the soil, with the change in the latter variable strongly linked to changes in soil temperature (Miralles et al., 2012).

4.5.4 Implications

4.5.4.1 Implications to the Environment

With the projected warming of both air and soil temperatures in the area, this would lead to drier soil (Phillips et al., 2014). Lower soil moisture could cause the soil to be more susceptible to drought conditions because the water, in liquid form, could evaporate more easily than ice (Drijfhout et al., 2015). Thawing of the permafrost would lead to higher decomposition rate (Todd-Brown et al., 2014) of existing organic materials that were previously frozen in ice.

The decomposition process would lead to additional greenhouse gases being released into atmosphere, further contributing to a positive feedback cycle. Indirectly, the melting permafrost

181 would lead to more methylmercury being released into the aquatic ecosystem as the released carbon would provide suitable conditions for microbes to convert inorganic mercury into methylmercury (Yang et al., 2016). The bioaccumulation and magnification of methylmercury would present challenges for aquatic animals and humans in the Arctic region.

4.5.4.2 Implications to Airports

Common impacts of degrading permafrost to airport infrastructure include disruption to the drainage system and the thaw settlement of the runways, taxiways and access roads (Boucher

& Guimond, 2012; Oldenborger & LeBlanc, 2015). At the nearby Tasiujaq airport, a study was conducted to assess the effectiveness of various methods that enhanced surface cooling or heat removal to lower the impacts of permafrost degradation on the runway (Doré et al., 2012). They found that snow removal and heat drains with pipes installed at the shoulders of the runway were cost-effective measurements to preserve the permafrost underneath the runway. However, rising air temperature and diminishing snow depth at Kuujjuaq (Environment Canada, 2018) may have unintended consequences of increasing rare occurrences of frostquakes, which causes popping sound and cracks on the ground if the ground is saturated with water as examined further in detail in Chapter 5 (Leung et al., 2017). Warmer temperature led to days with temperature above

0oC more often particularly in the fall and spring and thinner snow cover lessen the moderating temperature effects on soil. Both factors were crucial for frostquake because it required large temperature swings to occur. Active removal of snow cover in an attempt to allow rapid subsurface cooling to preserve permafrost could lead to even greater fluctuation in soil temperature and more prone to formation of frostquakes. But it is believed that enhanced surface cooling of the runway will more than offset the drawbacks from cracks caused by frostquakes as permafrost would damage the airport infrastructure on a wider scale. Another study conducted at

182

Iqaluit Airport in Nunavut also stated that surface snow cover as the greatest factor in the permafrost thaw and snow-covered edges along the taxiways and access roads would be most vulnerable to thaw settlement (Ghias et al., 2017). One of those taxiways was moved after multiple repairs failed to rectify the permafrost subsidence (Kloepfer, 2017). The Nunavik region

(including Kuujjuaq and Tasiujaq) communities were characterized as low to moderately susceptible to hazards posed by climate change (Nelson et al., 2002). Given that Kuujjuaq is relatively close to Tasiujaq, the challenges and mitigation solutions to the runway integrity at

Kuujjuaq is expected to be similar to Tasiujaq. As a result, aviation infrastructure at Kuujjuaq is expected to face higher maintenance cost and more frequent inspection and repairs. On top of that, mitigation strategies will involve additional costs. Responding to these challenges is a slow and challenging process due to the remoteness of these communities. For example, Iqaluit

Airport’s construction materials had to be carefully planned and ordered in advance to take advantage of lower shipping costs from summer sealift over delivering them by air (Kloepfer,

2017). Kuujjuaq and many villages in Nunavik have sea access through Hudson Bay and Hudson

Strait, which makes them ideal candidates to ship supplies by summer sealift to address climate change impacts and implement adaptation strategies.

4.5.5 Long-term Monitoring Network

This research would not be possible without the existence of a long-term soil temperature stations in a cold environment. Climate research relies on long-term observation records to accurately establish historical baselines in which the future projections are based on. The consequences of short-term government priorities such as budget cuts (Kroeger, 1996) on environmental monitoring programs might not be felt until years or decades later. The soil monitoring at Kuujjuaq might not appear to be useful at the time when the program review

183 determined it to be low impact or significance, but this research demonstrated its usefulness in the present day when Arctic warming becomes a global concern. Historical data in the Arctic are scarce, highlighting the usefulness of sites such as Kuujjuaq that provided a baseline for comparison with recent data. Lack of long-term monitoring led by the government cannot be offset by ad-hoc weather stations or boreholes records such as those at Salluit, Kangiqsujuaq and

Quaqtaq northwest of Kuujjuaq (Allard et al., 1995) due to agency needs or university research projects as the sites are not intended for long-term data collection in this region (Jean, 1991) and are primarily tied to the funding cycle. Even though these ad-hoc stations tended to be maintained, the sensors were often not calibrated (Meyer & Hubbard, 1992) and thus the data quality from these stations was uncertain and questionable. Moreover, government weather stations had the advantage of having longer time series and met the standards set out by the

World Meteorological Organization (Jean, 1991) which allow data to be exchanged globally and used in meteorological forecast, research and GCMs. Ad-hoc weather stations could supplement but not replace the data collection done by government weather stations. Therefore, consistent and stable funding to various monitoring networks is needed for reliable atmospheric and geophysical data collection.

4.6 Conclusion

This study used in-situ, long-term soil temperature records at Kuujjuaq and found significant warmings of 0.9 to 1.3oC per decade on an annual scale and 0.8 to 1.9oC in winter at 5 to 150 cm depths. Soil at deeper depths had a lesser day-to-day variability and warmed slower than surface soils. SDSM modelled the mean and variance of historical soil temperature data accurately, particularly from April to October and in soils closer towards the surface. Variables

184 that had the strongest correlation to historical soil temperature at this location were surface specific humidity, mean 2 m air temperature, 500 hPa geopotential height and specific humidity at 850 hPa. SDSM approach produced results that are close to the observed data and are robust as shown by the very small model errors. Also, the differences in results between the models generated from SDSM used to predict future was also less than the model error. By 2026, soil temperature projections suggested that the annual mean will be above 0oC in all of the RCP scenarios. Winter mean soil temperature from November to April would remain below 0oC. Soil temperatures would continue to rise by 2056 and 2086. The warming rate was linked to the greenhouse gas emission levels. Implications of the warming include possible cracks forming on the runway, damaging fuel tanks or compromise the structural integrity of terminal buildings that support the operation of the airport due to the deteriorating permafrost and freeze-thaw events each year.

Future directions of this study would involve in improving the accuracy of winter projections. This could be achieved by performing another model validation process exclusively for winter months (November to April) and identify atmospheric variables which were best correlated to soil temperature in winter or a specific month to remove the winter bias in the model. In addition, if new soil temperature data at this location becomes available, the data should be compared with the model to evaluate the performance of modelled future data against actual measurements. Projections should also be attempted by using other climate models or with an ensemble approach. Finally, the SDSM process should be applied at another location with long-term soil temperature records to ensure that the method is applicable elsewhere.

This study established a strong linkage between soil temperature and CanESM2 model output through statistical downscaling. It also demonstrated that a novel usage of SDSM

185 software was applicable to future soil temperature projections. This approach showed that using the SDSM approach on CanESM2 model, the downscaled historical soil temperature were more accurate than the CanESM2 output for historical soil temperature at this location. The tool performs reasonably well in summer as well as in soil up to 100 cm deep. Through this study, it illustrated the importance of long-term government-funded supplementary network data such as soil temperature to monitor climate change in the north.

4.7 Acknowledgements

We would like to thank Kenneth Devine for providing information on soil measurement methods, site characteristics and photos through the station inspect reports. We would also like to thank Kinson Leung for his assistance in statistical downscaling in SDSM software.

4.8 References

Allard, M., Wang, B., & Pilon, J. A. (1995). Recent Cooling along the Southern Shore of Hudson Strait, Quebec, Canada, Documented from Permafrost Temperature Measurements. Arctic and Alpine Research, 27(2), 157-166. Beltrami, H., & Kellman, L. (2003). An examination of short- and long-term air–ground temperature coupling. Global and Planetary Change, 38(3-4), 291-303. Beltrami, H., Gosselin, C., & Mareschal, J. C. (2003). Ground surface temperatures in Canada: Spatial and temporal variability. Geophysical Research Letters, 30(10), 1499. Beltrami, H., Bourlon, E., Kellman, L., & González-Rouco, J. F. (2006). Spatial patterns of ground heat gain in the Northern Hemisphere. Geophysical Research Letters, 33(L06717). Boucher, M., & Guimond, A. (2012). Assessing the Vulnerability of Ministère des Transports du Québec Infrastructure in Nunavik in a Context of Thawing Permafrost and the Development of an Adaptation Strategy. In Guy Doré & Brian Morse (Eds.), Cold Regions Engineering. Quebec City: American Society of Civil Engineers.

Burke, E. J., Chadburn, S. E., Huntingford, C., & Jones, C. D. (2018). CO2 loss by permafrost thawing implies additional emissions reductions ot limit warming to 1.5 or 2oC. Environmental Research Letters, 13(2), 024024.

186

Chylek, P., Li, J., Dubey, M. K., Wang, M., & Lesins, G. (2011). Observed and model simulated 20th century Arctic temperature variability: Canadian Earth System Model CanESM2. Atmospheric Chemistry and Physics Discussions, 11, 22893-22907. Doré, G., Ficheur, A., Guimond, A., & Boucher, M. (2012). Performance and Cost-Effectiveness of Thermal Stabilization Techniques Used at the Tasiujaq Airstrip. In Guy Doré & Brian Morse (Eds.), Cold Regions Engineering. Quebec City: American Society of Civil Engineers. Drijfhout, S., Bathiany, S., Beaulieu, C., Brovkin, V., Claussen, M., Huntingford, C., … Swingedouw, D. (2015). Catalogue of abrupt shifts in Intergovernmental Panel on Climate Change climate models. Proceedings of the National Academy of Sciences of the United States of America, 112(43), E5777-E5786. Environment Canada. (1966). Station Inspection Report. Environment Canada. (1972). Station Inspection Report. Environment Canada. (1978). Station Inspection Report. Environment Canada. (1984). Canadian Climate Normals 1951 – 1980 Volume 9: Soil Temperature, Lake Evaporation, Days with Blowing Snow, Hail, Fog, Smoke/Haze, Frost. Ottawa, Canada: Environment Canada. Environment Canada. (2017). CanESM2 / CGCM4 Model Output Variables. Environment Canada. (2018). Canadian Climate Normals. Fortier, R., LeBlanc, A., Yu, W. (2011). Impacts of permafrost degradation on a road embankment at Umiujaq in Nunavik (Quebec), Canada. Canadian Geotechnical Journal, 48(5), 720-740. Ghias, M. S., Therrien, R., Molson, J., & Lemieux, M. (2017). Controls on permafrost thaw in a coupled groundwater-flow and heat-transport system: Iqaluit Airport, Nunavut, Canada. Hydrogeology Journal, 25(3), 657-673. Gómez, I., Caselles, V., Etrela, M. J., & Niclòs, R. (2016). Impact of Initial Soil Temperature Derived from Remote Sensing and Numerical Weather Prediction Datasets on the Simulation of Extreme Heat Events. Remote Sensing, 8(7), 589. Gough, W.A., & Leung, A. (2002). Nature and fate of Hudson Bay permafrost. Regional Environmental Change, 2(4), 177-184. Graversen, R. G., Mauritsen, T., Tjernström, M., Källén, E., & Svensson, G. (2008). Vertical structure of recent Arctic warming. Nature, 451, 53-56. Hamed, K. H., & Rao A. R. (1998). A modified Mann-Kendall trend test for autocorrelated data. Journal of Hydrology, 204(1-4), 182-196. Harris, C., Arenson, L. U., Christiansen, H. H., Etzelmüller, B., Frauenfelder, R., … Mühll, D. V. (2009). Permafrost and climate in Europe: Monitoring and modelling thermal, geomorphological and geotechnical responses. Earth-Science Reviews, 92(3-4), 117-171.

187

Hassan, Z., & Harun, S. (2012). Application of Statistical Downscaling Model for Long Lead Rainfall Prediction in Kurau River catchment of Malaysia. Malaysian Journal of Civil Engineering, 24(1), 1-12. Houle, D., Bouffard, A., Duchesne, L., Logan, T., & Harvey, R. (2012). Projections of Future Soil Temperature and Water Content for Three Southern Quebec Forested Sites. Journal of Climate, 25, 7690-7701. Jean, M. (1991). Évolution de la température du sol à Kuujjuaq, Québec 1967-1989 [Evolution of soil temperature at Kuujjuaq, Quebec 1967-1989]. Le Climat, 9(2), 39-48. Jiang, Y., Zhuang, Q., Sitch, S., O’Donnell, J. A., Kicklighter, D., Sokolov, A., & Melillo, J. (2016). Importance of soil thermal regime in terrestrial ecosystem carbon dynamics in the circumpolar north. Global and Planetary Change, 142, 28-40. Kalnay, E., Kanamitsu, M., Kistler, R., Collins, W., Deaven, D., Gandin, L., … Joseph, D. (1996). The NCEP/NCAR 40-year reanalysis project. Bulletin of the American Meteorological Society, 77(3), 437-471. Kendall, M. G. (1975). Rank correlation methods. London, UK: Griffin. Kloepfer, C. (2017). An airport public–private partnership in the Canadian Arctic. Journal of Airport Management, 11(2), 136-146. Koven, C. D., Riley, W. J., & Stern, A. (2013). Analysis of Permafrost Thermal Dynamics and Response to Climate Change in the CMIP5 Earth System Models. Journal of Climate, 26(6), 1877-1900. Kroeger, A. (1996). Changing Course: The Federal Government’s Program Review of 1994-95. In A. Armit & J. Bourgault (Eds.), Hard choices or no choices: assessing program review (pp. 21-28). Toronto, Canada: The Institute of Public Administration of Canada. Leung, A. C. W., & Gough, W. A. (2016). Air mass distribution and the heterogeneity of the climate change signal in the Hudson Bay/Foxe Basin region, Canada. Theoretical and Applied Climatology, 125(3-4), 583-592. Leung, A. C. W., Gough, W. A., & Shi, Y. (2017). Identifying Frostquakes in Central Canada and Neighbouring Regions in the United States with Social Media. In M. Leitner & J. Jokar Arsanjani (Eds.), Citizen Empowered Mapping (pp. 201-222). Cham, Switzerland: Springer Nature. Majorowicz, M., & Minea, V. (2015). Geological, Economical and Environmental Assessment of Combined Geothermal Power and Heat Generation in Québec, Canada. Proceedings of World Geothermal Congress. Melbourne, Australia. Mann, H. B. (1945). Nonparametric tests against trend. Econometricia, 13(3), 245-259. Meinshausen, M., Smith, S. J., Calvin, K., Daniel, J. S., Kainuma, M. L. T., Lamarque, J- F., …van Vuuren, D. P. P. (2011). The RCP greenhouse gas concentrations and their extensions from 1765 to 2300. Climatic Change, 109, 213.

188

Meyer, S. J., & Hubbard, K. G. (1992). Nonfederal Automated Weather Stations and Networks in the United States and Canada: A Preliminary Survey. Bulletin of the American Meteorological Society, 73(4), 449-457. Miralles, D. G., van den Berg, M. J., Teuling, A. J., & de Jeu, R. A. M. (2012). Soil moisture- temperature coupling: A multiscale observational analysis. Geophysical Research Letters, 39(21), L21707. Mühll, D. V., Stucki, T., & Haeberli, W. (1998). Borehole temperatures in alpine permafrost: a ten year series. Paper presented at the meeting of PERMAFROST – Seventh International Conference, Yellowknife, Northwest Territories, Canada. Natural Resources Canada (1985). 5th edition of the National Atlas of Canada. Nelson, F. E., Anisimov, O. A., & Shiklomanov, N. I. (2002). Climate Change and Hazard Zonation in the Circum-Arctic Permafrost Regions. Natural Hazards, 26, 203-225. Nelson, F. E. (2003). (Un)frozen in Time. Science, 299(5613), 1673-1675. Oelke, C., & Zhang, T. (2004). A model study of circum-Arctic soil temperatures. Permafrost and Periglacial Processes, 15(2), 103-121. Oldenborger, G. A., & LeBlanc, A. M. (2015). Geophysical characterization of permafrost terrain at Iqaluit International Airport, Nunavut. Journal of Applied Geophysics, 123, 36- 49. Pearson, K. (1895). Notes on regression and inheritance in the case of two parents. Proceedings of the Royal Society of London, 58, 240-242. Phillips, C. L., Torn, M. S., & Koven, C. D. (2014, December). Response of Soil Temperature to Climate Change in the CMIP5 Earth System Models. Paper presented at the meeting of American Geophysical Union, Fall Meeting 2014, San Francisco, United States. Qian, B., Gregorich, E.G., Gameda, S., Hopkins, D., & Wang, X.L. (2011). Observed soil temperature trends associated with climate change in Canada. Journal of Geophysical Research: Atmospheres, 116, D02106. Sen, P. B. (1968). Estimates of the Regression Coefficient Based on Kendall's Tau. Journal of the American Statistical Association, 63(324), 1379-1389. Skinner, W. R., Flannigan, M. D., Stocks, B. J., Martell, D. L., Wotton, B. M., Todd, J. B., … Bosch, E. M. (2002). A 500 hPa synoptic wildland fire climatology for large Canadian forest fires, 1959-1996. Theoretical and Applied Climatology, 71(3-4), 157-169. Slater, A. G., & Lawrence, D. M. (2013). Diagnosing Present and Future Permafrost from Climate Models. Journal of Climate, 26(15), 5608-5623. Slater, A. G., Schlosser, C. A., Desborough, C. E., Pitman, A. J., Henderson-Sellers, A., Robock, A., … Zeng, Q. (2001). The Representation of Snow in Land Surface Schemes: Results from PILPS 2(d). Journal of Hydrometeorology, 2(1), 7-25. Stewart, E. J., Tivy, A., Howell, S. E. L., Dawson, J., & Draper, D. (2010). Cruise Tourism and Sea Ice in Canada’s Hudson Bay Region. Arctic, 63(1), 57-66.

189

Stow, C. A., Roessler, C., Borsuk, M. E., Bowen, J. D., & Reckhow, K. H. (2003). Comparison of Estuarine Water Quality Models for Total Maximum Daily Load Development in Neuse River Estuary. Journal of Water Resources Planning and Management, 129(4), 307-314. Tam, A., Gough, W. A., & Xie, C. (2015). An Assessment of Potential Permafrost along a North-South Transect in Canada under Projected Climate Warming Scenarios from 2011 to 2100. The International Journal of Climate Change: Impacts and Responses, 6(2), 1- 18. Theil, H. (1950). A rank-invariant method of linear and polynomial regression analysis. Proceedings of the Royal Netherlands Academy of Sciences Part I, Part II, Part III, 53, 386-392, 521-525, 1397-1412. Todd-Brown, K. E. O., Randerson, J. T., Hopkins, F., Arora, V., Hajima, T., Jones, C., … Allison, S. D. (2014). Changes in soil organic carbon storage predicted by Earth system models during the 21st century. Biogeosciences, 11(8), 2341-2356. University of Toronto Scarborough. (2018). Climate Lab at the University of Toronto Localizer. University of Toronto Scarborough. Retrieved from http://climatechange.utsc.utoronto.ca Verseghy, D. L. (2000). The Canadian Land Surface Scheme (CLASS): Its History and Future. Atmosphere-Ocean, 38(1), 1-13. Wang, B., & Allard, M. (1995). Recent Climatic Trend and Thermal Response of Permafrost in Salluit, Northern Quebec, Canada. Permafrost and Periglacial Processes, 6(3), 221-233. Wen, J., Su, Z., & Ma, Y. (2003). Determination of land surface temperature and soil moisture from Tropical Rainfall Measuring Mission/Microwave Imager remote sensing data. Journal of Geophysical Research: Atmospheres, 108(D2), ACL2-1-ACL2-10. Wilby, R. L., Dawson, C. W., & Barrow, E. M. (2002). SDSM – a decision support tool for the assessment of regional climate change impacts. Environmental Modelling & Software, 17(2), 145-157. Yang, Z., Fang, W., Lu, X., Sheng, G-P., Graham, D. E., Liang, L., … Gu, B. (2016). Warming increases methylmercury production in an Arctic soil. Environmental Polution, 214, 504- 509. Zhang, Y., Cheng, W., Smith, S. L., Riseborough, D. W., & Cihlar, J. (2005). Soil temperature in Canada during the twentieth century: Complex responses to atmospheric climate change. Journal of Geophysical Research, 110, D03112. Zhang, Y., Sherstiukov, A. B., Qian, B., Kokelj, S. V., & Lantz, T. C. (2018). Impacts of snow on soil temperature observed across the circumpolar north. Environmental Research Letters, 13, 044012.

190

Identifying Frostquakes in Central Canada and Neighbouring Regions in the United States with Social Media

Andrew C.W. Leung, William A. Gough, Yehong Shi Department of Physical & Environmental Sciences, University of Toronto Scarborough 1265 Military Trail, Toronto, Ontario, Canada, M1C 1A4

191

Chapter 5: Frostquakes

5.1 Abstract

Following the ice storm of December 2013 in southern Ontario, the general public heard noises that resembled falling trees and reported these occurrences on social media. These were identified as a rare phenomenon called cryoseism, or more commonly known as frostquakes.

These occurrences became the first large-scale documented frostquakes in Canada. Using meteorological metrics, this study was able to forecast two subsequent frostquake events in

January 2014 that coincided with reports on social media. In total, six more episodes of frostquakes as well as their locations were identified in January and February of 2014. Results showed that in central Canada, frostquake occurrences ranged from Windsor, Ontario to the west to Montreal, Quebec to the east and from Niagara Falls, Ontario to the south to North Bay,

Ontario to the north. In the United States, the reports came from states bordering the Great Lakes and the New England areas. Two frostquake clusters were identified, one in and around the

Greater Toronto Area and the other in eastern Wisconsin. Frostquakes were most frequently heard at nighttime. The use of social media as an observation network was critically assessed, including the possibility of false positives and population bias. Frostquake had the potential to generate cracking on runways and other airport infrastructure. This study demonstrates that rare phenomena such as frostquakes can be identified and assessed using data gathered through social media.

192

5.2 Introduction

Frostquakes, also known as cryoseism, are relatively rare weather phenomenon. They occur after sudden freezing of the ground under specialized conditions and are characterized by a

“boom” or “cracking” noise that resembles falling trees. Sometimes, a small tremor is also reported (Lacroix, 1980). Occurrences are infrequent (Nikonov, 2010; Barosh, 2000) and recurrences can be delayed by decades or longer.

In southern Ontario, frostquakes were first heard on the night of December 24, 2013, just after the ice storm of 2013 (December 20–22, 2013). The general public reported these noises on various social media platforms such as Twitter, Facebook and online discussion boards. Some described the noise as similar to someone banging their fist against the wall or a gunshot (Allen,

1993). Many individuals reported on social media that they were asleep and were woken up by the noise. A number of them mentioned that their pets became startled when the noise began.

According to media outlets, some people called the police believing that someone was firing a shotgun or that their house was being broken into. Meteorologists from local media in Toronto,

Ontario identified that the noises were likely the result of frostquakes and elaborated on the antecedent processes that gave rise to them. Seismic events were quickly ruled out as seismic stations in Canada did not find any seismic waves in this area that night. Meteors were also ruled out.

After reading the term online or hearing it in the news, the public appeared to be looking for more information and turned to Wikipedia (Figure 5.1). Prior to December 25, 2013, the page on cryoseism received about 300 views on average per day (searches for the terms “ice quake” and “frostquake” on Wikipedia are redirected to the cryoseism page). Just after the first wave of frostquakes in southern Ontario, the number of page views on cryoseism spiked up to 2772 views

193 on December 25 and then 6363 views on December 26. After that, visits to this page subsided but still were well above the average prior to the first wave. When frostquakes returned during the night of January 2 to 3, 2014, the news media once again covered the event in the newspaper, on the radio, TV and online. Because of the attention generated by these various publications, the

English Wikipedia page drew over 21,000 views on January 3. The public became conscious of the noise so that when frostquakes occurred once again on January 6 to 7 the cryoseism entry on

Wikipedia also drew over 40,000 views, the highest daily visit in that page’s history. These page views are considered to be a conservative number since the traffic statistics software did not include mobile views into the total, which accounts for roughly 30% of all page views (Heilman

& West, 2015).

Figure 5.1: Number of non-mobile page views on cryoseism article on English Wikipedia. * indicates dates with missing page view data.

Prior to 2013, the only frostquake officially reported in Canada occurred near a seismic station in Sadowa, Ontario (44.8o N, 79.2o W) on January 18, 2000 and the occurrence was

194 recorded on a fortuitously closely located seismometer (Natural Resources Canada, 2016). Eight probable frostquakes from 1870 to 1898 in New Brunswick were identified by Burke (2004).

A search of the scientific literature produced scant results. Lacroix (1980) examined frostquakes and their intensity in the New England area up to 1979. He also identified that frostquakes were frequent in January. Barosh (2000) reported additional frostquakes in the New

England area, including the damage caused by related cracking. The cracks, as documented by

Barosh (2000), could damage infrastructure such as buildings and runways (Boucher & Guimond,

2012). Fujita and Sleep (1991) confirmed three frostquakes and four probable ones from 1872 to

1922 in Michigan. Allen (1993) monitored and recorded frostquake activities with seismographs in Sebago Lake region of Maine during the winter of 1990 to 1991. Burke (2004) found three likely frostquake events in eastern Maine on top of the eight events in New Brunswick. Nikonov

(2010) examined events spanning 1803 to 1908 in Eastern Europe. He identified three critical factors were required for the formation of frostquakes: moist soil, low to no snow cover and a sudden drop in temperature that exceeded -20oC.

A major reason why frostquakes appear so infrequently in the scientific literature is their relative infrequency and difficulty in detection, it is a largely unstudied phenomenon. While networks have been set up to detect earthquakes, frostquakes are too localized and infrequent to be effectively monitored in a similar fashion. Past occurrences rely on anecdotal information such as journals or newspaper reports (Burke, 2004), the social media of the time. With the advent of the internet and the contemporary social media, frostquakes have the potential to be reported more readily and thus researched. Gathering data through social media has been used in natural science and earth science research in the past (Hyvärinen & Saltikoff, 2010; Ogden,

2013). For example, the US Geological Survey used Twitter to improve its earthquake

195 monitoring response time (Earle et al., 2011). In Europe, forest fires are usually detected by remote sensing but also augmented by citizens contributing volunteered geographic information

(VGI) in the forms of blogs, tweets and photos (De Longueville et al., 2010). Other climatology- related observations that benefitted from VGI include assessing the availability of outdoor skating rinks due to warmer winters in Canada (Robertson et al., 2015), floodings and storm surges caused by Hurricane Sandy in New York in 2012 as well as tornado damage in Oklahoma in 2013 (Middleton et al., 2014). However, this is the first time that frostquake data are gathered using VGI.

The questions that this chapter seeks to answer are as follows:

1) What were the climate conditions of the January 18, 2000 Sadowa frostquake?

2) What was the geographical range and associated climate conditions of frostquakes in

Canada and US during the winter of 2013–2014?

3) What role can social media play in detecting frostquakes?

5.3 Methods

Two approaches were taken to investigate the frostquakes in Ontario during the winter of

2013–14, climate data analysis and social media reporting. This study limited the analysis period to begin with the frostquakes that occurred on January 2/3, 2014. While frostquakes were first heard on the night of December 24, 2013, reports for possible frostquakes from that night were not included for several reasons. First, in southern Ontario, over 1 million houses lost electricity as a result of the ice storm that was a precursor to the formation of frostquakes. Power was not restored to some homes until a week later. There would be an inherent bias of under-reporting or no reporting towards those who lost power since they could not go online to report their

196 observations. Second, because of the ice storm, many trees and branches had fallen. These noises that appeared to be coming from frostquake could actually be trees falling down under the weight of the ice and could have been mistakenly identified as frostquakes or vice versa. This could lead to false positive reports. Third and finally, the term frostquake was not familiar to

Canadians. Only a few Canadian TV media outlets in the Toronto area ran online stories on

December 25–26 about the booming sound by mentioning the term frostquake. The term was not publicized until another round of frostquakes on January 2, 2014 as noted above.

5.3.1 Climate Data Analysis

The criteria established by Nikonov (2010) was used to analyze local weather data. These metrics included saturated soils, low to no snow cover and a rapid drop of temperature to below

-20oC. The January 18, 2000 frostquake reported at Sadowa, Ontario was first analyzed and then for the frostquakes that occurred during the winter of 2013–14.

For the January 18, 2000 frostquake event, Muskoka Airport (44°58' N, 79°18' W) weather station data was used for climatological analysis. The weather station is approximately

27 km away from Sadowa. The daily temperature (minimum, mean and maximum), precipitation

(rain and snow) and snow depth on ground were examined.

For the winter of 2013–14, fifteen Canadian weather stations from Environment and

Climate Change Canada’s Climate Data Online (Figure 5.2a) and data from six American weather stations from National Ocean and Atmospheric Administration’s Climate Data Online were used (Figure 5.2b). Station selection was based on the spatial range of frostquake reports and the number of reports mentioned in that area. Similar to the event in 2000, the daily temperature, precipitation and snow on ground were analyzed.

197

198

Figure 5.2: Weather stations in (a) Canada and (b) the US chosen for temperature and snow depth analysis.

Precipitation is reported using “trace” as a measure and this value needs to be quantified.

For Canadian stations, trace amount of daily rainfall (< 0.2 mm/day) was given a value of 0.1 mm/day and trace amount of daily snowfall (< 0.2 cm/day) was given a value of 0.07 mm/day

(Mekis & Vincent, 2011). The adjustment for trace amounts of precipitation in United States stations was more problematic due to different measuring equipment and different definition of trace precipitation that stems from its use of imperial units. Unlike Canada, US weather stations

199 use standard rain gauges to measure the amount of snowfall (Doesken & Judson, 1997). In Yang et al. (1998), trace amount of daily rainfall (< 0.01 in/day; 0.254 mm/day) was assigned the same value (0.1 mm/day) as Canadian weather stations. For trace amount of daily snowfall (< 0.1 in/day; 0.254 cm/day), Sugiura et al. (2003) suggested an assigned value equal to a quarter of its measuring limit, which is 0.025 in/day or 0.0635 mm/day. This assigned trace daily snowfall value for American weather stations was almost identical to the value given to Canadian weather stations.

Handling trace snow depth was more problematic. For the US, substantive inconsistencies exist in terms of how trace amounts of snow depth were interpreted and recorded among airport weather stations and volunteer stations (Doesken & Judson, 1997). In addition, while all Canadian airport and volunteer weather observers record snow depth at or around 6 am daily, some US volunteer stations record snow depth in the early evening while US airport weather stations report snow depth at midnight. In both Canada and the US, snow depth is rounded to the nearest whole unit of measurement (cm for Canada, inch for the US). Thus, a snow depth of 0.5 cm to 1.0 cm is reported as 1 cm. For Canadian stations, snow depth below 0.5 cm is described as “trace”. Since snow depths of 0.1 cm to 0.4 cm were considered equally likely to occur, the average value of 0.25 cm was assigned. This approach is identical in principle with that used by Mekis and Vincent (2011). Similarly, for US stations, trace snow depth between 0.1 in and 0.4 in was given a value of 0.25 in, which is equal to 0.635 cm.

5.3.2 Social Media

The first approach was to analyze social media reports, particularly from Twitter and produce maps of frostquake reports for each frostquake episode during the winter of 2013–14.

200

To identify frostquake events on Twitter, the search terms “frostquake”, “cryoseism” and “ice quake” were used. All frostquake reports from January 2 to February 28, 2014 were examined.

Location was identified as the city mentioned in an individual’s tweet or post if given. Otherwise, the city location that an individual associated with on their user profile was assumed to be the location where the frostquake occurred. For date and time, the timestamp of the tweet or post was assumed to be the time of occurrence if the user mentioned that they just heard the noise just prior to posting. If this was not the case and the individual specified the approximate time of the noise, then that time was used as a proxy for the actual frostquake occurrence. Finally, if the individual did not specify the time the frostquake was heard, only the date was assigned to the location of the report. Those that did not specify locations were not included in the study.

Additional locations were obtained from a user-generated online Google Map

(https://www.google.com/maps/d/viewer?mid=zId7WwTT0PPk.kmYXHjIndA-w), which solicited social media users to collaboratively mark when and where they heard the frostquake.

This online map’s URL was also linked in multiple news media’s online versions of the story and encouraged the readers to add their reports. Results from all crowdsourced information were sorted by date and grouped by individual towns and cities. The reports were cleaned by examining obvious plotting errors on the user-generated Google Map. Locations were removed if the points were plotted in the middle of a large waterbody (e.g. Lake Ontario).

201

5.4 Results

5.4.1 Climate Data Analysis

5.4.1.1 January 18, 2000

The only frostquake officially confirmed by seismic record was the one from Sadowa,

Ontario on January 18, 2000 at 6:55 pm. According to Natural Resources Canada (2016), that night was very cold and 12 frostquakes were recorded within a two-hour period. Coincidentally, individuals from Skowhegan, Maine also reported frostquakes around the same time, on January

14 to 15, 2000 (Maine Geological Survey, 2016). Upon examining the weather conditions for both locations at that time, both had above 0oC temperature two days prior to a quick drop in temperature. The temperature drop on January 16, 2000 was quite large, from 0oC to -25oC in one day. On the day of the frostquake, Sadowa had 8 cm of snow cover on the day of frostquake while Skowhegan had none. The way that water entered the soil was also different between these locations. At Skowhegan, rain was recorded two days prior to the frostquake. But at Sadowa, a rain event occurred seven days before the frostquake occurred. The increase in soil moisture was likely caused by the melting of the snow cover on the ground, as the snow cover reduced from 11 cm to 8 cm. Therefore, it appeared that the saturated soils required for frostquakes could be the result of either rainfall or melting snow on the ground. Since the events in Sadowa and

Skowhegan happened within the same week, it was believed that the spatial variability was mainly caused by the particular temperature, rainfall and snow depth at the respective locations.

5.4.1.2 Winter of 2013–14

Temperature graphs for Canadian and American weather stations are shown in Figure 5.3.

Thawing followed by a quick drop in temperature was observed in the following periods at

202

Canadian stations: December 19–22, December 26–29, January 3–6, January 9–14, January 16–

17 and February 18–23. Similar observations at the American stations were found on December

26–29, January 9–15 and February 17–23.

Snow depths are presented in Figure 5.4. Most stations in southern Ontario had less than

30 cm of snow on the ground from December 2013 to January 2014. All of the American stations saw a decrease in snow cover after January 6–12. On average, almost all of these American stations had less than 20 cm of snow depth on average.

203

Figure 5.3: Temperature at weather stations in (a) Canada and (b) the US during the winter of 2013–14.

204

Figure 5.4: Snow depth at weather stations in (a) Canada and (b) the US.

5.4.2 Social Media

Overall, there were 2301 frostquake reports recorded through social media (Table 5.1).

Maps of frostquake reports for three events with the highest number of reports were generated:

January 2/3, January 6/7 and January 20–22 (Figure 5.5). Over 2100 public reports were recorded from these three events alone and the majority were from Canada (Table 5.1). Spatial

205 analysis showed that regions with high population density (Greater Toronto Area) also experienced the highest number of reports. For the January 2/3 event, most reports came from

Toronto and Brampton, Ontario (Figure 5.5a). Virtually all of the reports were from Ontario, though there were two reports from Wisconsin and one each from Indiana and New York State.

For January 6/7 event, the highest number of reports came from Toronto and around the Green

Lake area in Wisconsin (Figure 5.5b). There were also reports from Montreal, Quebec (n=3),

Montague, Prince Edward Island (n=1) and St. John’s, Newfoundland and Labrador (n=1). From the United States, there were multiple reports from Indiana, Ohio, Michigan, Vermont and Maine.

Prior to this event, Colorado, Iowa and Virginia did not experience frostquake. For the January

20–22 event, Toronto and Newmarket, Ontario had the highest number of reports (Figure 5.5c).

One frostquake report originated in Minnesota, a new state for reporting. There were 55 reports not classified because the public only specified the location and did not include the time or date of the event.

206

Figure 5.5: Plots of reported frostquake locations on (a) January 2/3, (b) January 6/7 and (c) January 20–22 of 2014. Dot sizes and colours are scaled to the number of reports in each community. Larger dots represent more reports from a particular town or city.

Using all of the gathered reports from Figure 5.5, the total counts from each location on various dates were combined to create a density report map (Figure 5.6). The density report map

207 identified an actual cluster of reports by taking in higher population in urban areas into consideration. The density report values in each community were calculated by adding the total number of frostquakes reported on Jan. 2/3, 6/7 and 20–22 of 2014 then dividing by the population of the community. The population of each community was based on Statistics

Canada’s 2011 Census and U.S. Census Bureau’s 2010 Census data. In total, there were 236 communities that experienced frostquakes during those periods. Two clusters of reports around the Toronto region and eastern Wisconsin region were found.

Figure 5.6: Density report map for frostquakes. Individual values are classified by the number of reports per 10,000 individuals in the community.

Based on Google Map reports, Twitter and Facebook posts, a temporal distribution of the timing of the frostquakes was created (Figure 5.7). The most common time when frostquakes

208 were reported was at night. In Canada, most reported hearing frostquake during the overnight period, especially between 1 am to 3 am (Figure 5.7a). In the US, most reported that they heard frostquakes between 7 pm and 11 pm (Figure 5.7b).

Figure 5.7: The local time in (a) Canada and (b) the US at which the public reported to have heard a frostquake.

209

Table 5.1: Breakdown of date ranges with frostquake and the number of reports within each range. Dates with Number of reports frostquake Canada US Total January 2–3 878 4 882 January 6–7 824 158 982 January 13–15 5 0 5 January 20–22 261 10 271 January 23–25 7 11 18 January 26–29 9 16 25 February 1–3 7 18 25 February 5–7 4 32 36 February 8–12 6 6 12 February 17 0 1 1 February 22–23 4 2 6 February 25–28 27 11 38 Total 2032 269 2301

5.4.3 Coincidence of Frostquake Reporting and Weather Conditions

For the winter of 2013–14 the climate data analysis indicated for southern Ontario, dramatic temperature drops for the following dates: December 19–22, December 26–29, January

3–6, January 9–14, January 17–20, January 24–27, February 15–19 and February 26–28 (Table

5.2). Frostquakes are reported in Table 5.1 for days where there was a large difference between maximum temperature and minimum temperature and often after temperature passed through the melting point of ice. Since Toronto had the most reports in the overall period, the weather station located in downtown Toronto (Toronto City) was used as the representative for the region. For

January’s frostquakes, the dates that had most public reports (January 2/3, January 6/7, January

20–22) all had a large drop in temperature and the minimum temperature was below -20oC.

Other dates with frostquake events also had a considerable drop in temperature but the minimum temperature did not drop below -20oC.

210

Table 5.2: Daily maximum, minimum and the difference between the maximum and minimum temperatures for the Toronto City weather station in downtown Toronto. Dates in bold indicate that over 10 reports were recorded in those periods. Date Max. Temperature (oC) Min. Temperature (oC) ∆Temperature (oC) 1/1/2014 -8.4 -14.5 6.1 1/2/2014 -14.3 -19.2 4.9 1/3/2014 -7.1 -22.3 15.2 1/4/2014 0.3 -7.2 7.5 1/5/2014 1.4 -1.7 3.1 1/6/2014 2.4 -15.8 18.2 1/7/2014 -15.8 -22.2 6.4 1/8/2014 -7.7 -16.2 8.5 1/9/2014 -2.9 -11.8 8.9 1/10/2014 4.1 -4.0 8.1 1/11/2014 7.3 2.9 4.4 1/12/2014 3.4 1.3 2.1 1/13/2014 7.4 1.9 5.5 1/14/2014 5.3 1.3 4.0 1/15/2014 2.4 -3.5 5.9 1/16/2014 -0.4 -3.6 3.2 1/17/2014 2.3 -1.2 3.5 1/18/2014 -0.8 -6.9 6.1 1/19/2014 -1.9 -7.2 5.3 1/20/2014 -1.8 -16.5 14.7 1/21/2014 -14.2 -19.8 5.6 1/22/2014 -12.1 -20.5 8.4 1/23/2014 -12.2 -17.4 5.2 1/24/2014 -6.8 -17.7 10.9 1/25/2014 -2.8 -13.8 11 1/26/2014 -3.8 -14.6 10.8 1/27/2014 -3.7 -16 12.3 1/28/2014 -11.6 -18.6 7.0 1/29/2014 -9.2 -16.5 7.3 1/30/2014 -0.4 -10.6 10.2 1/31/2014 1.5 -2.5 4.0 2/1/2014 1.4 -1.1 2.5 2/2/2014 0.9 -6.1 7.0 2/3/2014 -2.4 -9.7 7.3 2/4/2014 -3.9 -10.5 6.6 2/5/2014 -5.2 -11.7 6.5 2/6/2014 -6.5 -13.2 6.7

211

2/7/2014 -8.5 -13.9 5.4 2/8/2014 -8.6 -14.2 5.6 2/9/2014 -7.2 -12.1 4.9 2/10/2014 -5.7 -12.0 6.3 2/11/2014 -8.8 -13.6 4.8 2/12/2014 -5.2 -14.1 8.9 2/13/2014 -1.6 -9.0 7.4 2/14/2014 0.7 -4.4 5.1 2/15/2014 -2.4 -11.9 9.5 2/16/2014 -6.9 -12.8 5.9 2/17/2014 -2.2 -14.0 11.8 2/18/2014 0.7 -4.2 4.9 2/19/2014 7.8 -1.4 9.2 2/20/2014 2.6 -1.0 3.6 2/21/2014 5.1 0.7 4.4 2/22/2014 4.3 -0.6 4.9 2/23/2014 1.4 -4.7 6.1 2/24/2014 -4.3 -8.8 4.5 2/25/2014 -5.2 -10.4 5.2 2/26/2014 -8.2 -14.2 6.0 2/27/2014 -8.3 -16.0 7.7 2/28/2014 -6.9 -17.7 10.8

5.5 Discussion

5.5.1 Weather Conditions

Nikonov (2010) concluded that moist soil, low snow cover and a sudden drop in temperature from above to below freezing were the variables required for frostquakes to occur.

However, there is a paucity of observations of frostquakes to explore the nature of the precursors.

Using social media reported frostquakes and coincident climate data, this presents an opportunity to study frostquakes in more detail. Using the social media data gathered in this study, a parallel study conducted by Battaglia and Changon (2016) found that seasonal frost and Arctic air mass outbreak could have also contributed to frostquakes.

212

The weather station data showed that these conditions were met during the study period.

The snow depth was shallow when the temperature went from below 0oC to above (Figure 5.4).

A few days later, thawing stopped when the temperature quickly dropped and in some locations, by up to 20oC within 24 hours (Figure 5.3). During and shortly after the temperature plunged, the public started to hear the noise or feel the shaking from the frostquakes.

In this study, it was speculated that these steps need to have a specific sequence and this was documented by using local weather data. For example, if a sudden drop in temperature that results in freezing occurred before the soil became moist or before the snow cover was reduced to low levels, a frostquake would not have happened because the ground would have sufficient space for the soil to freeze and therefore not cause any cracking noise. Therefore, the only realistic sequence would be a location first having low to no snow cover which allowed the soil to become moist. After that, the temperature drop must greatly exceed the insulation effect of the snow cover so that the temperature in the soil will quickly drop and water molecules inside the soil will freeze and expand.

5.5.2 Social Media

It is a commonly used research method to utilize VGI data gathered from the general public. Very often, these studies provide online forms for the public to fill out in a structured manner. This ensures the completeness of the data often by asking the user to select from a list of pre-determined options. In contrast, data collection from social media and Google Map often faces bigger challenges. The submitted information is considered “free-flowing” because the user chooses the level of detail and in an unstructured manner that sometimes requires follow-up prior to analyzing the data. The main benefits of using social media for data gathering is a larger

213 sample size that does not require the public to fill out a web form to report their findings. Instead of waiting for the public to engage with the scientists, social media allows scientists to reach out and obtain data directly from the public. In this study, the spatial data production of frostquakes can be classified as “bottom-up, amateur and asserted” (Cinnamon, 2015) as almost all of the data points were generated by the general public on social media or Google Map rather than by authoritative experts.

Hyvärinen and Saltikoff (2010) listed the service provider’s terms and conditions, data retention policy, privacy and copyright as the biggest challenges for collecting meteorological observations from social media. Their study analyzed user-submitted photos that were uploaded to Flickr (an online photo depository) and identified meteorological events at a specific time and location from the photos. This study is believed to have fewer issues related to content removal policy, privacy and copyright. The location data was collected in near real-life time, which circumvented the issue with the deletion of older materials. Hyvärinen and Saltikoff (2010) indicated that Twitter only stored one week of tweets before deletion. Twitter has since modified its content removal policy to keep tweets in perpetuity unless deleted by the user. In addition, there are third-party providers who store tweets on a particular topic or keyword for future retrieval. However, some frostquake tweets were deleted by the user in rare instances (<0.1% of all reports) by comparing the differences between the search results from Twitter with the third- party’s. Privacy was not a concern for this study. Users chose to opt-in to post the details publicly by themselves. All reports were aggregated from a town or city into one location thus resulting in a coarse resolution of the locations. Even though some users enabled their GPS locations while tweeting, the coordinates were imprecise or sometimes obviously inaccurate and

214 unreliable on Twitter. For example, on a few occasions, the coordinates identify the user’s location to be in the middle of Lake Ontario, 500 m away from the shorelines of Toronto.

While this is a novel approach of using social media for scientific data collection, there are some potential issues with using Twitter and Facebook to gather data. It is difficult to pinpoint the exact time and location when the frostquake occurred. Twenty percent of the reports gathered did not specify the approximate time of occurrence. The number of reports is dependent on the population of a city and citizens who use social media, thus skewing towards larger cities like Toronto. An issue exists for towns and cities with different area sizes. Towns with smaller areas were, in theory, less likely exposed to frostquakes as there were less land for frostquake to occur. On the other hand, regardless of land area, a single frostquake occurring in a densely populated area could trigger multiple reports. But obtaining a precise postal code for the location of the report is more challenging than a town or city name due to privacy concerns. Another under-reporting problem arose from those who heard the noise but decided not to report it on social media. Furthermore, there is an age disparity for social media users. A survey conducted in 2009 in the US found that 75% of young adults from age 18 to 24 had a social media accounts whereas only 7% of those aged 65 and above had an account (Lenhart, 2009). Therefore, reports gathered from different social media were more likely to be coming from teenagers and young adults rather than older adults or seniors. This study’s age bias was somewhat lessened when mainstream media included the link to the Google Map frostquake reporting system in their online news stories. Privacy setting on social media accounts also suppressed some reports.

While Twitter’s tweets were set to public by default, Facebook’s posts were private by default.

60% of teenagers reported that they set their profiles (along with their posts) to private and only visible to their friends (Madden et al., 2013). Hence, frostquake reports were less likely to be

215 found on Facebook because of the users’ privacy settings. On the other hand, false positives are not uncommon on social media (Hyvärinen & Saltikoff, 2010) and this was particularly true for frostquake since the only identification was a banging noise and light tremor but usually no physical observation can be made and in some cases a report could be completely fictitious.

Some obvious false positive frostquake reports were identified. Some users reported hearing frostquake noises from places such as San Diego which did not have frost on the ground at that time of the year. There were a number of well-intentioned reports from the Pickering, Ontario area on the morning of January 21, 2014. The volunteers later corrected themselves after discovering that the shaking and noise were the result of a nearby wastewater treatment plant explosion and not by frostquake. These reports were removed from the analysis. A few plausible locations such as Atlanta, Georgia and Denver, Colorado were kept after examining the climatological conditions on January 6–7, 2014 and found the conditions to be possible for frostquake to have occurred (Figure 5.5b).

The nature of the reports on social media between Canada and the US appeared to be different. The reported frostquake time in the US tends to be in the evening period where most people were still awake (Figure 5.7b). They might have heard the noise and decided to mention it on the internet. On the other hand, the majority of Canadian reported frostquakes took place during the overnight period after most people went to bed (Figure 5.7a). It is likely in this case that the public was woken up by the noise and decided to share their experience. In a number of reports, the public said that their pets were woken up by the sound or felt the jolt and the pets woke up their owners. A lot of Canadians mentioned in their tweets that they were surprised by the sound and in some cases, they were delighted to finally hear it after their friends shared a similar experience with them. There were several explanations which explained why more

216 frostquakes were heard at night than during the day. Night time temperature tends to be colder and a quick temperature drop appears to be a requirement for frostquakes to occur. Another reason to explain the temporal difference is that it is quieter at night and people are less active, which makes it easier for the frostquake to be heard or felt. On a greater temporal scale, more frostquakes were happening in January than in February (Table 5.1) and this appeared to be in agreement with Lacroix (1980).

On social media, the public stated that the sound appeared to be coming from the roof even though the cause of the noise was the expansion of ice within the soil. At the time of their reports, these individuals were in various types of buildings (detached houses, apartments). It is unclear how the vibration sound resonated through different building materials (e.g. wood, concrete) and propagate to upper-level floors in apartment buildings. In addition, the frostquakes in 2013–2014 occurred in highly populated areas. This was noticeably different from reports from New Brunswick and New England where most of the people who heard the frostquakes were living in farmhouses in rural communities (Allen, 1993; Barosh, 2000; Burke, 2004).

5.5.3 Frostquake Clusters

Given that urban centres like Toronto and Montreal have large populations, their total number of reports are not unexpectedly high when compared to suburban and rural areas (Figure

5.5). However, this study did not observe a cluster of reports around Montreal and that reports from Montreal area only appeared on January 6/7, 2014 (Figure 5.5b). The density map showing a cluster in Toronto (Figure 5.6) is not surprising given that there was extensive media coverage and social media presence. It was also the first area that received prominent attention given that many people believed to have experienced it just after the ice storm of 2013. The spike in

217

Wikipedia traffic to cryoseism article (Figure 5.1) on December 26, 2013 and January 3, 2014 were very likely to be coming from people in the Toronto cluster because there were only four reports coming from the US up until January 3, 2014 (Table 5.1). The Toronto cluster consists of four of the top ten largest municipalities in Canada (Brampton, Hamilton, Mississauga and

Toronto) plus a number of suburban towns and rural villages. While most people heard or felt the frostquake were living in detached homes, some of them, especially in downtown Toronto, heard it inside their apartments. Despite the large population size, Hamilton and Mississauga were placed 28th and 30th percentile respectively on the density map while Toronto was placed at the

57th percentile and Brampton was placed at the 73rd percentile. There was a mix of low and high density reports from other communities within the Toronto cluster. However, the density map showed that Montreal had a very low number of reports per 10,000 individuals.

The Wisconsin cluster, in contrast, is very different from the Toronto cluster. Most of the communities in the Wisconsin clusters had very small population but a high density of reports. In fact, the top nine communities with the highest density reports were all from the Wisconsin cluster and seven of the nine communities had multiple reports of frostquakes. Yet, none of these nine communities had a population over 1500 and practically everyone lives in detached housing in these rural villages and towns. One thing in common between the Toronto and Wisconsin clusters was that both areas were reported extensively by mainstream media. In Wisconsin’s case, the attention was drawn to The Weather Network (2014)’s video of a Wisconsin farmer discovering a crack about 30 m long and 0.2 m deep after hearing a booming noise which was attributed to frostquake. The video caused a heightened awareness of this phenomenon and was the subject of local discourse. It led to a positive feedback loop of more awareness leading to more reports, and that, in turn, led to even greater awareness. Similar cracks were also present in

218 two instances in Maine and Massachusetts (Allen, 1993; Barosh, 2000). In Allen (1993)’s study, the frostquakes were heard and felt between 7–9 pm, which is consistent with this study’s temporal analysis of when frostquakes were most likely to occur in the United States (Figure

5.7b). The timing was quite similar to the event in Sadowa, Ontario at 6:55 pm (Natural

Resources Canada, 2016) and the event in Rothesay, New Brunswick in 1884 from 9:30 pm–

10:30 pm (Burke, 2004). However, the Canadian events in this study took place much later in the night (Figure 5.7a). Also, it was impossible to explain why the traffic to Wikipedia’s cryoseism article did not show a noticeable spike after the third major event on January 20–22, 2014

(Figure 5.1).

Using Figure 5.6 to compare with existing literature, it was determined which provinces and states were new to frostquakes. In Canada, there were no frostquakes reported in Quebec,

Prince Edward Island and Newfoundland and Labrador prior to this study. Likewise, in the

United States, the states that did not have reported frostquakes before January 2014 were

Colorado, Georgia, Illinois, Indiana, Iowa, Pennsylvania and Virginia. Almost all of the

Canadian provinces and US states which experienced the first frostquake took place on Jan. 6/7.

5.5.4 Aviation Impacts

Although the study area of the frostquake study was further south, several inferences related to soil temperature and ground conditions can be drawn and linked back to northern

Canadian airports. Andersland and Al-Moussawi (1987) demonstrated that higher stress in soil was achieved in colder temperatures in a shorter cooling period and that cracks could be formed by the rapid freezing. Barosh (2000) presented photos of cracks caused by frostquakes which were similar to those formed by frost heave. A crack that was several metres long and more than

219

0.5 m deep was discovered after numerous frostquakes in Calgary on March 4, 2014 (CTV,

2014). In addition, a Wisconsin farmer discovered a long crack on his driveway that was caused by frostquake (The Weather Network, 2014). Frostquake cracks in photos appeared to be similar to ice wedges. However, ice wedges form in permafrost regions whereas frostquakes could, but not necessarily, occur in areas with permafrost. As demonstrated in Chapter 4.5.4, airports are more frequently exposed to cracks formed by permafrost degradation. Warmer temperature would lead to more frequent days with temperature above 0oC, which could increase the possibility of ice wedges or frostquakes in these communities and greater chance of crack formation by either event. This could further impact the runway, access road or other infrastructure at the airport.

5.6 Conclusion

This research demonstrated that yet another type of rare weather phenomenon like frostquake can be monitored through social media. Data collected via user-submitted entries on

Google Map, Facebook posts and Twitter tweets were found to be generally reliable and carries scientific value. The VGI reports of frostquakes were linked to concurrent weather conditions.

Frostquake forecasts can be issued based on data from weather forecasts. Moisture level in soil was qualitatively linked to air temperature, precipitation and snow depth. Shallow snow cover, high amounts of rain or freezing rain combined with rapid temperature drop from above 0oC to freezing temperature were the conditions required for frostquakes. Through the use of social media, the greatest number of frostquake locations to date were collected and used to identify two frostquake clusters. Three Canadian provinces and seven US states were discovered to have their first ever reported frostquake during the study period. This study showed that citizen

220 science projects could augment data collection, both in high and low population density areas, for rare weather phenomenon. While the impacts of frostquakes to aviation were quite low, there were records of cracks that were caused by frostquakes. Such cracks have the potential to cause additional damage to airport infrastructure.

5.7 Acknowledgements

We thank Ashley King for creating the initial user-generated Google Map that served as the collaborative VGI mapping platform for the public to report their observations.

5.8 References

Allen, R. P. (1993). A Study Of Cryoseisms (“Frostquakes”) In The Sebago Lake Region, Maine. In: Symposium on the Application of Geophysics to Engineering and Environmental Problems, 1993:415–430. Andersland, O. B., & Al-Moussawi, H. M. (1987). Crack formation in soil landfill covers due to thermal contraction. Waste Management and Research, 5(1), 445-452. Barosh, P. J. (2000). Frostquakes in New England. Engineering Geology, 56(3-4):389–394. Battaglia, S. M., & Changon, D. (2016). Frost Quakes: Forecasting the Unanticipated Clatter. Weatherwise, 69(1), 20-27. Boucher, M., & Guimond, A. (2012). Assessing the Vulnerability of Ministère des Transports du Québec Infrastructure in Nunavik in a Context of Thawing Permafrost and the Development of an Adaptation Strategy. In Guy Doré & Brian Morse (Eds.), Cold Regions Engineering. Quebec City: American Society of Civil Engineers. Burke, K. B. S. (2004). Historical Seismicity in the Central Highlands, Passamaquoddy Bay, and Moncton Regions of New Brunswick, Canada, 1817-1961. Seismological Research Letters, 75(3):419–431. Cinnamon, J. (2015). Deconstructing the binaries of spatial data production: Towards hybridity. The Canadian Geographer, 59(1):35–51. CTV. (2014). Possible epicenter of frost quake found in northwest schoolyard. Retrieved from https://calgary.ctvnews.ca/possible-epicenter-of-frost-quake-found-in-northwest- schoolyard-1.1722711

221

De Longueville, B., Annoni, A., Schade, S., Ostlaender, N., & Whitmore, C. (2010). Digital Earth’s Nervous System for crisis events: real-time Sensor Web Enablement of Volunteered Geographic Information. International Journal of Digital Earth, 3(3):242– 259. Doesken, N. J., & Judson, A. (1997). A Guide to the Science, Climatology, and Measurement of Snow in the United States. Fort Colins, CO: Colorado State University Department of Atmospheric Science. Earle, P. S., Bowden, D. C., & Guy, M. (2011). Twitter earthquake detection: earthquake monitoring in a social world. Annals of Geophysics, 54(6):708–715. Fujita, K., & Sleep, N. H. (1991). A re-examination of the seismicity of Michigan. Tectonophysics 186(1-2): 75-106. Heilman, J. M., & West, A. G. (2015). Wikipedia and Medicine: Quantifying Readership, Editors, and the Significance of Natural Language. Journal of Medical Internet Research, 17(3):e62. Hyvärinen, O., & Saltikoff, E. (2010). Social Media as a Source of Meteorological Observations. Monthly Weather Review, 138(8):3175–3184. Lacroix, A. V. (1980). A short note on cryoseisms. Earthquake Notes, 51(1):15–20. Lenhart, A. (2009). Adults and Social Networking Websites. Pew Research Internet Project. Retrieved from http://www.pewinternet.org/2009/01/14/adults-and-social-network- websites Madden, M., Lenhart, A., Cortesi, S., Gasser, U., Duggan, M., Smith, A., & Beaton, M. (2013). Teens, Social Media, and Privacy. Pew Research Internet Project. Retrieved from http://www.pewinternet.org/2013/05/21/teens-social-media-and-privacy Maine Geological Survey. (2016). Reports of earth shaking in Maine possibly due to cryoseisms. Retrieved from http://www.maine.gov/dacf/mgs/hazards/earthquakes/quake-cryolist.htm Mekis, E., & Vincent, L. A. (2011) An Overview of the Second Generation Adjusted Daily Precipitation Dataset for Trend Analysis in Canada. Atmosphere-Ocean, 49(2):163–177. Middleton, S. E., Middleton, L., & Modafferi, S. (2014). Real-time Crisis Mapping of Natural Disasters using Social Media. IEEE Intelligent System, 29(2):9–17. Natural Resources Canada. (2016). Frequently Asked Questions about Earthquakes (FAQ). Retrieved from http://www.earthquakescanada.nrcan.gc.ca/info-gen/faq-eng.php Nikonov, A. A. (2010). Frost Quakes as a Particular Class of Seismic Events: Observations within the East-European Platform. Izvestiya, Physics of the Solid Earth, 46(3):257–273. Ogden, L. E. (2013). Tags, Blogs, Tweets: Social Media as Science Tool? Bioscience, 63(2):148. Robertson, C., McLeman, R., & Lawrence, H. (2015). Winters too warm to skate? Citizen- science reported variability in availability of outdoor skating in Canada. The Canadian Geographer, 59(4):383–390.

222

Sugiura, K., Yang, D., & Ohata, T. (2003). Systematic error aspects of gauge-measured solid precipitation in the Arctic, Barrow, Alaska. Geophysical Research Letters, 30(4):1192– 1195. The Weather Network. (2014). Wisconsin: Ice quake causes property damage. Retrieved from http://www.theweathernetwork.com/news/articles/wisconsin-ice-quake-causes-property- damage/19445 Yang, D., Goodison, B. E., Ishida, S., & Benson, C. S. (1998). Adjustment of daily precipitation data at 10 climate stations in Alaska: Application of World Meteorological Organization intercomparison results. Water Resources Research, 34(2):241–256.

223

Chapter 6: Conclusion

This thesis examined four topics that affect aviation in the Hudson Bay region and related cold weather phenomena in central Canada. Using historical climate and weather data, the thesis successfully addressed all four research objectives set out in Chapter 1.

For historical wind patterns at seven locations in Hudson Bay area and northern Quebec

(research objective #1), Chapter 2 found that average wind speed and daily maximum wind speed were increasing from 1971 to 2010 at the five airports around the Hudson Bay and central part of northern Quebec. Churchill and Inukjuak recorded significant increase of average daily wind speed by 0.6 to 1.5 km/h in March to May during the study period while Kuujjuarapik’s wind speeds were approaching significant levels by 0.3 to 0.5 km/h in the same months. These three locations had a significant increase in wind speed in spring. However, wind speed and daily maximum wind speed were significantly declining at Schefferville and Wabush at the border of eastern Quebec and western Labrador. Wind speeds at Schefferville and Wabush were declining at about 1 km/h in all the seasons and annually between the 1971-1990 and 1991-2010 periods.

Wabush saw the greatest and significant decline in maximum daily wind speed (wind ≥ 28 km/h) at about 2 to 3 km/h in all of the monthly, seasonal and yearly scales. Highest wind speed of each day occurred most often from midnight to 2 am and from 1 pm to 5 pm. Calm winds were observed in 1 to 9% at the Hudson Bay locations. Churchill had the lowest hours of calm wind while Wabush had the most. Churchill also had the highest number of incidental calm winds, where calm was reported in between hours with measurable wind, and Wabush had the highest number of true calm winds where winds were not blowing for consecutive hours. Wind directions were significantly changing from northerly to northwesterly at Baker Lake and

Nitchequon. This study found that Inukjuak, Kuujjuarapik, Nitchequon, Schefferville and

224

Wabush’s runways were not aligned with the prevailing wind direction. Many of the flights at these airports were scheduled to arrive or depart when the wind speed was the greatest, exposing the pilots in small, turboprop aircraft to potentially dangerous crosswind conditions. While the potential risk would rise from increased wind speed, it also highlighted opportunity for these communities to transition from fossil fuel for electricity to renewable energy generated by wind power. Increasing wind speed also challenged the general assumption in future climate models that wind speed in this area would decline.

For fog and low visibility conditions (research objective #2), Chapter 3 showed that spatial variability for fog, ice fog and restricted visibility was present in this area. Ice fog was declining significantly at 1 to 10 hours per year at the Hudson Bay airports and at about 0.5 hours per year at James Bay region airports, just south of Hudson Bay. The decline in ice fog was explained by a general warming of air temperature in the region. Decline in fog frequency was also observed but the trend was less noticeable at Hudson Bay airports. However, fog was noticeably decreasing at a rate of 3 to 9 hours per year at the three James Bay airports. The decline in fog was explained by rising air temperature which lowered the relative humidity.

Positive autocorrelations were found in many cases for fog and ice fog, suggesting that these weather conditions were frequently present in these communities each year. Reduced visibility was encountered in 0.6 to 2% of the airport’s operational hours in James Bay, 2 to 4% in eastern

Hudson Bay and 4 to 10% at western Hudson Bay. Low visibility was encountered during the operational hours in 0.2 to 1% in James Bay, 1 to 2% in eastern Hudson Bay and 2 to 5% at western Hudson Bay. Hours of reduced visibility was changing by +4 to -6 hours and +2 to -6 hours for low visibility at some locations. Fog, snow and blowing snow were the top factors that led to a reduction in visibility.

225

Chapter 4 analyzed the baseline soil temperature at Kuujjuaq, Quebec. The historical annual mean soil temperature at 5 to 150 cm depths from 1967 to 1995 was -1.4 to -2.4oC

(research objective #3). Annual mean soil temperature was increasing significantly at all depths at a rate of 0.9 to 1.3oC per decade. Soil temperature in winter (November to April) ranged from

-11.1oC at 5 cm depth to -4.2oC at 150 cm depth. In winter, the soil temperature was warmer in deeper soil. The mean soil temperature in winter was increasing at 0.8 to 1.9oC per decade. Using statistical downscaling on CanESM2 model variables, the study successfully validated the modelled data against the historical data. It showed that statistical downscaling on CanESM2 often produced a more accurate historical soil temperature at a finer resolution with more depth layers than the GCM. The method established a strong connection between soil temperature and atmospheric variables on local scale (surface temperature, surface humidity and humidity at 850 hPa) and synoptic scale (500 hPa geopotential height). Under CanESM2 simulations, the annual mean soil temperature across all depths would rise to above 0oC by first projection period (1997 to 2026) in all RCP scenarios. Soil temperature continues to warm in second and third projection periods (2027 to 2056 and 2057 to 2086 respectively). The warming under RCP 2.6 scenario will slow down by second and third projection period. Soil temperature in RCP 4.5 scenario will warm greater than RCP 2.6 but slower than RCP 8.5. In winter, with the exception of 150 cm depth in RCP 8.5 projection by third projection period, soil temperatures at all depths and three

RCP scenarios would remain below 0oC.

In Chapter 5, the environmental conditions that caused frostquakes were identified

(research objective #4). Moist, saturated soil combined with minimal snow cover and rapid drop of temperature below 0oC led to reports of frostquakes on social media in central Canada and nearby US states. The cracking sound or mild tremor were heard or felt most often during the

226 evening and overnight hours. This study recorded over 2300 reports of frostquake in January to

February 2014. Greater Toronto Area and eastern Wisconsin were identified as having the most frostquake reports. Using weather conditions from forecasts, this study successfully predicted frostquakes on two days in January 2014. It demonstrated the use of social media to augment scientific data collection in densely and sparsely population areas.

Through these chapters, this thesis provided a better understanding of weather and climate conditions related to aviation in northern Canada. They demonstrated the importance of various less-commonly used observations in hourly (wind speed, wind direction, visibility, present weather observation) and daily (soil temperature, snow depth) formats in the Canadian

Climate Data Online archive. It highlighted the continuous need of open data movement for data that has already been gathered by the government but underutilized because they were not made available publicly such that the public were not made aware that such data were available to be used. Advertising the availability of datasets and allowing them to be accessed easily will facilitate and increase the data being utilized. This research also showed the increasing demand to data mine for analysis.

Wind speed was increasing at many Hudson Bay airports and would present a challenge and greater risk when the wind direction was not aligned with the runway. Wind direction also appeared to be shifting at some locations and that many airport runways were not aligned with the prevailing winds during the initial construction. Since most of these airports cannot expand due to topography, lack of flat lands in the vicinity or lack of funds, improve pilot training on greater awareness and bolster search and rescue equipment at these airports would alleviate some of the risks. The results of this research can be used to identify suitable communities that are favourable for wind power generation and lower their reliance on fossil fuel that are shipped

227 from southern Canada. This effort will contribute to a reduction of greenhouse gas emissions and improve air quality in these communities.

Risks associated with fog and ice fog were declining as they were less frequently encountered. But reduced and low visibility trends were more nuanced as different locations exhibited different trends. There also appeared to be no consistent trend for reduced and low visibility over the years, which suggested that non-fog events could be offsetting a reduction of fog and ice fog that led to obstruction to visibility. The findings for reduced and low visibility frequencies would be beneficial to airlines, airport operations and tourism companies. Airlines could minimize the cost associated with flight delays and cancellations by scheduling their flights to avoid the times where the probability of low visibility was the greatest. Fog research at the airport was also applicable to other transportation modes such as travelling by car, snowmobile or boat. It could also be applied to other activities such as hunting and fishing, which are important aspects in the aboriginal and Inuit cultures. Tourists would better understand that these destinations were becoming more favourable in summer due to less fog. The study also demonstrated the suitability of using of non-24 hour stations for analysis. This would allow additional climatological, meteorological and operational research at these communities at a local scale.

Soil temperature was rising in the latter part of the 20th century and will continue to warm in the future. This presented an issue not just for airports but also for many infrastructures in the

Canadian Arctic that relied on permafrost as foundations. The disappearing permafrost also has implications on the Arctic environment. This study presented the importance of long-term soil temperature monitoring in cold environments in Canada. It also demonstrated a new approach of

228 conducting a climate change impact assessment by using statistical downscaling on the

CanESM2 model to project future soil temperature at this airport under three RCP scenarios.

Finally, frostquakes were linked to causing cracks on the ground. These cracks could damage runways or affect the integrity of other infrastructure and buildings which would lead to increased repairs and maintenance cost in these communities. Frostquake study also demonstrated the use of social media for tracking recent weather phenomenon. From these research findings, it provided a greater insight into the impacts of climate change on aviation in

Hudson Bay and northern Canada. It also demonstrated the additional costs and risks associated with climate change in the region. This illustrated the need to consider the impacts from different weather effects (increase in wind speed, shift in wind direction, change in visibility and fog conditions, increase in soil temperature) an integrated manner to safeguard the airport from the effects of climate change through adaptation and mitigation.

229