Impact of COVID-19-Related Trafﬁc Slowdown on Urban Heat Characteristics

: Governments around the world have implemented measures to slow down the spread of COVID-19, resulting in a substantial decrease in the usage of motorized transportation. The ensuing decrease in the emission of trafﬁc-related heat and pollutants is expected to impact the environment through various pathways, especially near urban areas, where there is a higher concentration of trafﬁc. In this study, we perform high-resolution urban climate simulations to assess the direct impact of the decrease in trafﬁc-related heat emissions due to COVID-19 on urban temperature characteristics. One simulation spans the January–May 2020 period; two additional simulations spanning the April 2019–May 2020 period, with normal and reduced trafﬁc, are used to assess the impacts throughout the year. These simulations are performed for the city of Montreal, the second largest urban centre in Canada. The mechanisms and main ﬁndings of this study are likely to be applicable to most large urban centres around the globe. The results show that an 80% reduction in trafﬁc results in a decrease of up to 1 ◦ C in the near-surface temperature for regions with heavy trafﬁc. The magnitude of the temperature decrease varies substantially with the diurnal trafﬁc cycle and also from day to day, being greatest when the near-surface wind speeds are low and there is a temperature inversion in the surface layer. This reduction in near-surface temperature is reﬂected by an up to 20% reduction in hot hours (when temperature exceeds 30 ◦ C) during the warm season, thus reducing heat stress for vulnerable populations. No substantial changes occur outside of trafﬁc corridors, indicating that potential reductions in trafﬁc would need to be supplemented by additional measures to reduce urban temperatures and associated heat stress, especially in a warming climate, to ensure human health and well-being.


Introduction
COVID-19 emerged in late 2019 and was declared a pandemic by the World Health Organization on 11 March 2020. In an effort to slow down the spread of COVID-19, strict measures were put in place by many governments around the world, aiming to reduce the frequency and duration of in-person interactions. These measures have drastically altered patterns of human activity. For example, [1] found that surface transport activity decreased by 50%, whereas industrial activity decreased by 35%.
Some of the changes brought about by COVID-19 are likely to become permanent-for example, several major cities are rededicating street space for pedestrians and cyclists, and

Climate Model and Urban Representation
The simulations in this study were performed using the limited-area version of the Global Environmental Multiscale (GEM; [27,28]) model, used for numerical weather prediction at Environment and Climate Change Canada (ECCC). It employs semi-Lagrangian transport and a (quasi) fully implicit time stepping scheme. In its fully elastic nonhydrostatic formulation [29], it uses a vertical coordinate based on hydrostatic pressure [30]. Condensation processes were computed by a double-moment microphysics scheme [31]. The land surface scheme used was CLASS v3.6 [32,33] and the urban scheme used was Town Energy Balance (TEB; [34]).
TEB is a physical scheme for urban energy budget based on a generalization of local canyon geometry. It is a single-layer urban canopy model. Buildings are assumed to be located along identical roads leading to the canyon structure. The lengths of the roads are considered to be far greater than their widths, any road orientation is possible, and all exist with the same probability. The buildings bordering the street canyon are assumed to have the same height and width within a model grid cell. In TEB, the urban morphology for each grid cell is defined in terms of the local building and road fractions, building height, building and canyon aspect ratio, and roughness length. The radiative characteristics such as albedos and emissivity for roofs, roads, and walls and thermal properties such as thermal conductivity and heat capacity of each roof, road, and wall layer are prescribed. TEB considers anthropogenic heat and moisture sources from combustion, split into two categories: traffic and industry. The traffic related fluxes are assumed to be released at the road level, while the industry fluxes are assumed to be released at the roof level.

Urban Climate Simulations and Influence of COVID-19 on Traffic
As mentioned in Section 1, one of the widely observed modifications of human behaviour due to the COVID-19 pandemic is a reduction in distance travelled. Prior to COVID-19, the average distance travelled by vehicles in Montreal was estimated at 43.19 million km per day, based on the citywide emissions of 3.567 × 10 9 kg CO 2 equivalent, an average consumption of 0.096 L of fuel per km driven, and the production of 2.357 kg of CO 2 per liter of fuel consumed [35]. This estimate is consistent with previous research based on origin-destination surveys [36].
The spatial distribution of vehicle traffic was estimated from GPS trajectory data obtained by the City of Montreal during 2016 and 2017 as part of the MTL Trajet project [37]. A total of 58,792 trips by car were analyzed for a total distance driven of 653,962 km over 24,414 h. This distance was distributed over the GEM grid by linking the GPS coordinates for each trip using straight lines and calculating the distance driven in each model grid cell. The prescribed anthropogenic heat fluxes due to traffic were obtained by scaling the fractional distance driven in each grid cell by the 43.19 million km driven per day and the production of 3.24 MJ sensible heat flux and 0.23 MJ latent heat flux per km driven [38]. The average sensible heat flux due to traffic is shown in Figure 1a, where highways and main roads can be clearly identified.
The temporal distribution of vehicle traffic was also estimated from the MTL Trajet data [37] using the start and end time of each trip. A traffic factor for each hour of the day was obtained (Figure 1b), revealing the morning (8 to 9 a.m.) and afternoon (5 to 6 p.m.) peaks of vehicular traffic. The highest value for the traffic factor (2.67) occurs during the afternoon peak, while the lowest (0.05) occurs overnight (4 to 5 a.m.). The variations in vehicular traffic at longer time scales (e.g., weekly and seasonal) are much smaller than this hourly variation and were omitted in this study for reasons of simplicity.
As the effect of decreased vehicular traffic on urban climate is likely to vary depending on the season, the urban climate simulations in this study cover the period from April 2019 to May 2020. The reference simulation (hereafter, GEM100) represents the business-as-usual case, with the anthropogenic heat fluxes due to traffic set to 100% of their reference values (Figure 1a). The reduced traffic simulation (hereafter, GEM20) represents the case where strict measures are in place to contain the spread of COVID-19, reducing traffic and its associated heat fluxes to 20% of their reference values. This 80% reduction is representative of the maximum expected reduction in distance travelled, which was estimated at an average of 76% during the 28-day period from 25 March 2020 to 21 April 2020 for the Montreal region, based on the city-level mobility data [39].
A third simulation (GEM_COVID) approximates the timing of the implementation of COVID-19 measures in Montreal, which occurred in mid-to-late March 2020. Until 16 March, GEM_COVID follows the GEM100 reference. From 17 March onwards, the anthropogenic heat fluxes due to traffic are set to 20% of their reference values (as in GEM20).
All three urban climate simulations were performed at a 250 m horizontal grid spacing and were driven at the lateral boundaries by a GEM simulation at a 0.025-degree (~3 km) horizontal grid spacing covering southern Quebec (Figure 1c,d), which was in turn driven by the ERA5 reanalysis data [40]. The land cover and urban morphology fields used in all three simulations were derived from high-resolution datasets and provided directly by ECCC. as-usual case, with the anthropogenic heat fluxes due to traffic set to 100% of their reference values (Figure 1a). The reduced traffic simulation (hereafter, GEM20) represents the case where strict measures are in place to contain the spread of COVID-19, reducing traffic and its associated heat fluxes to 20% of their reference values. This 80% reduction is representative of the maximum expected reduction in distance travelled, which was estimated at an average of 76% during the 28-day period from 25 March 2020 to 21 April 2020 for the Montreal region, based on the city-level mobility data [39]. A third simulation (GEM_COVID) approximates the timing of the implementation of COVID-19 measures in Montreal, which occurred in mid-to-late March 2020. Until 16 March, GEM_COVID follows the GEM100 reference. From 17 March onwards, the anthropogenic heat fluxes due to traffic are set to 20% of their reference values (as in GEM20).
All three urban climate simulations were performed at a 250 m horizontal grid spacing and were driven at the lateral boundaries by a GEM simulation at a 0.025-degree (~3 km) horizontal grid spacing covering southern Quebec (Figure 1c,d), which was in turn driven by the ERA5 reanalysis data [40]. The land cover and urban morphology fields used in all three simulations were derived from high-resolution datasets and provided directly by ECCC.

Remote Sensing Data
Land surface temperature (LST) products from satellite remote-sensing platforms are widely available and are potential sources for model validation, having comprehensive spatial coverage, comparable scale with models, and constant periodicity. The Moderate Resolution Imaging Spectroradiometer (MODIS) sensors onboard the Aqua satellite provide global coverage of LST over cloud-free regions two times per day. The satellite Sunsynchronous orbit of Aqua is designed to cross the equator at 13:30 and 1:30 (local solar time). The retrieved spatial resolution for LST is as high as 1 km. In this study, the MYD21A1D and MYD21A1N collection 6 products are used, which contain LST for the day and night passes of Aqua, respectively, available from mid-2002 until the present. These products use the physics-based Temperature Emissivity Separation (TES) algorithm to dynamically retrieve both the LST and emissivity simultaneously for three thermal infrared bands. Validation has shown consistent accuracies at the 1 • C level over all land surface types including vegetation, water, and deserts [12].

Model Validation
Urban climate simulations over several regions using GEM have been validated in previous studies [17,[41][42][43]. To assess whether GEM is able to capture the spatiotemporal variability of LST over the study region, the LST estimated by MODIS over heavily urbanized areas (urban fraction > 50%) is first directly compared to the LST modelled by GEM. The day-to-day comparison over the entire simulation period is shown in Figure 2, for the daytime (around 1 p.m. local time) and nighttime (around 2 a.m. local time) overpasses of MODIS. It can be seen that GEM is able to capture both the daytime and nighttime LST temperatures. The MODIS daytime LST temperatures are known for occasionally containing unrealistically high values, which is related to the larger differences with respect to GEM for the 1 p.m. overpass. The reduced frequency of the MODIS data during the cold season can be explained by the more abundant cloud cover during this season, which prevents MODIS from observing the surface. The influence of cloud cover can also be seen during the warm season, during which MODIS is unable to capture the occasionally colder daytime LST temperatures, as these occur during cloudy days. The differences between GEM and MODIS are consistent with the biases reported by previous urban modelling studies over the region [19,42].

Remote Sensing Data
Land surface temperature (LST) products from satellite remote-sensing platforms are widely available and are potential sources for model validation, having comprehensive spatial coverage, comparable scale with models, and constant periodicity. The Moderate Resolution Imaging Spectroradiometer (MODIS) sensors onboard the Aqua satellite provide global coverage of LST over cloud-free regions two times per day. The satellite Sunsynchronous orbit of Aqua is designed to cross the equator at 13:30 and 1:30 (local solar time). The retrieved spatial resolution for LST is as high as 1 km. In this study, the MYD21A1D and MYD21A1N collection 6 products are used, which contain LST for the day and night passes of Aqua, respectively, available from mid-2002 until the present. These products use the physics-based Temperature Emissivity Separation (TES) algorithm to dynamically retrieve both the LST and emissivity simultaneously for three thermal infrared bands. Validation has shown consistent accuracies at the 1 °C level over all land surface types including vegetation, water, and deserts [12].

Model Validation
Urban climate simulations over several regions using GEM have been validated in previous studies [17,[41][42][43]. To assess whether GEM is able to capture the spatiotemporal variability of LST over the study region, the LST estimated by MODIS over heavily urbanized areas (urban fraction > 50%) is first directly compared to the LST modelled by GEM. The day-to-day comparison over the entire simulation period is shown in Figure 2, for the daytime (around 1 p.m. local time) and nighttime (around 2 a.m. local time) overpasses of MODIS. It can be seen that GEM is able to capture both the daytime and nighttime LST temperatures. The MODIS daytime LST temperatures are known for occasionally containing unrealistically high values, which is related to the larger differences with respect to GEM for the 1 p.m. overpass. The reduced frequency of the MODIS data during the cold season can be explained by the more abundant cloud cover during this season, which prevents MODIS from observing the surface. The influence of cloud cover can also be seen during the warm season, during which MODIS is unable to capture the occasionally colder daytime LST temperatures, as these occur during cloudy days. The differences between GEM and MODIS are consistent with the biases reported by previous urban modelling studies over the region [19,42]. . For the MODIS data, darker colors denote that at least two thirds of the urban region has valid data, while lighter colors are used when between one third and two thirds of the urban region has valid data. . For the MODIS data, darker colors denote that at least two thirds of the urban region has valid data, while lighter colors are used when between one third and two thirds of the urban region has valid data.
Given that the UHI and its effects on human health are strongest during the warm season daytime, the magnitude of the UHI estimated by MODIS and modelled by GEM are compared ( Figure 3). Here, UHI is defined as the temperature difference with respect to the average over non-urbanized land areas (urban fraction < 1%) in the simulation domain. The higher daytime LSTs over urban areas clearly stand out in both MODIS ( Figure 3a) and GEM (Figure 3b), with the LST being around 10 • C higher in urban areas than in non-urban regions. Within urban areas, colder regions corresponding to greener areas (Figure 1d) can also be seen in both datasets. The same is true for water bodies (Figure 1d), which are much colder than urban and non-urban land areas in both datasets.
Atmosphere 2021, 12, 243 6 of 13 Given that the UHI and its effects on human health are strongest during the warm season daytime, the magnitude of the UHI estimated by MODIS and modelled by GEM are compared ( Figure 3). Here, UHI is defined as the temperature difference with respect to the average over non-urbanized land areas (urban fraction < 1%) in the simulation domain. The higher daytime LSTs over urban areas clearly stand out in both MODIS ( Figure  3a) and GEM (Figure 3b), with the LST being around 10 °C higher in urban areas than in non-urban regions. Within urban areas, colder regions corresponding to greener areas ( Figure 1d) can also be seen in both datasets. The same is true for water bodies ( Figure  1d), which are much colder than urban and non-urban land areas in both datasets. During cloudy days, GEM models a much weaker surface UHI (Figure 3c), consistent with the reduced differential heating of urban and non-urban areas by solar radiation. While data to directly validate the 2 m temperature UHI do not exist, it can be seen that GEM produces patterns resembling the surface UHI, both during clear-sky (Figure 3d) and cloudy (Figure 3e) conditions. In addition, the magnitude of the 2 m temperature UHI is substantially lower than the surface UHI and is consistent with previous literature [13,14].
In addition to the validation of LST presented above, the 2 m air temperature was validated using hourly data from eight observation stations in the study domain ( Figure  4a), available from ECCC. The data recorded at each station were compared directly to the 2 m temperature of the GEM grid cell in which the station was located. During the simulation period (April 2019-May 2020), the average bias across stations in GEM ranged from −0.3 °C to −1.4 °C, while the root mean square error varied between 2.0 °C and 2.6 °C, suggesting that GEM adequately captures both the mean and variability of the 2 m air temperature. The bias at each station was further analysed for each hour of the day and each season of the year (Figure 4b), revealing that GEM performs remarkably well for the During cloudy days, GEM models a much weaker surface UHI (Figure 3c), consistent with the reduced differential heating of urban and non-urban areas by solar radiation. While data to directly validate the 2 m temperature UHI do not exist, it can be seen that GEM produces patterns resembling the surface UHI, both during clear-sky (Figure 3d) and cloudy (Figure 3e) conditions. In addition, the magnitude of the 2 m temperature UHI is substantially lower than the surface UHI and is consistent with previous literature [13,14].
In addition to the validation of LST presented above, the 2 m air temperature was validated using hourly data from eight observation stations in the study domain (Figure 4a), available from ECCC. The data recorded at each station were compared directly to the 2 m temperature of the GEM grid cell in which the station was located. During the simulation period (April 2019-May 2020), the average bias across stations in GEM ranged from −0.3 • C to −1.4 • C, while the root mean square error varied between 2.0 • C and 2.6 • C, suggesting that GEM adequately captures both the mean and variability of the 2 m air temperature. The bias at each station was further analysed for each hour of the day and each season of the year (Figure 4b), revealing that GEM performs remarkably well for the summer daytime (from 6 a.m. to 6 p.m. local time) 2 m temperature, where the mean difference between the observed and modelled temperature was less than 1 • C for all hours and all stations. Averaging the bias at each station over a 30-day moving window (Figure 4c) shows that during most of the simulation, the bias stayed close to the mean, except during January 2020, where all stations except WVQ and WIZ (which are less urbanized) showed a stronger cold bias, possibly related to snow removal operations, which favor warmer temperatures but are not enabled in TEB for this study. The excellent performance demonstrated by GEM for the summer daytime is important, given that the strongest impacts of the UHI effect on human health (due to heat stress) occur during that time.
Atmosphere 2021, 12, 243 7 of 13 summer daytime (from 6 a.m. to 6 p.m. local time) 2 m temperature, where the mean difference between the observed and modelled temperature was less than 1 °C for all hours and all stations. Averaging the bias at each station over a 30-day moving window ( Figure  4c) shows that during most of the simulation, the bias stayed close to the mean, except during January 2020, where all stations except WVQ and WIZ (which are less urbanized) showed a stronger cold bias, possibly related to snow removal operations, which favor warmer temperatures but are not enabled in TEB for this study. The excellent performance demonstrated by GEM for the summer daytime is important, given that the strongest impacts of the UHI effect on human health (due to heat stress) occur during that time.

Modelled Impact of COVID-19 on Urban Temperatures
To assess the impact of the COVID-19 measures as implemented in Montreal, the 2 m air temperature at grid cells with heavy traffic is compared to the adjacent grid cells in the GEM_COVID simulation. The daily evolution of these differences is shown in Figure  5, where an abrupt decrease in the magnitude of these differences occurs in mid-March, caused by the activation of COVID-19 containment measures, modelled as an 80% reduction in heat fluxes associated with traffic. As expected, this decrease is largest for the grid cells with the heaviest traffic (above 25 W m −2 ).

Modelled Impact of COVID-19 on Urban Temperatures
To assess the impact of the COVID-19 measures as implemented in Montreal, the 2 m air temperature at grid cells with heavy traffic is compared to the adjacent grid cells in the GEM_COVID simulation. The daily evolution of these differences is shown in Figure 5, where an abrupt decrease in the magnitude of these differences occurs in mid-March, caused by the activation of COVID-19 containment measures, modelled as an 80% reduction in heat fluxes associated with traffic. As expected, this decrease is largest for the grid cells with the heaviest traffic (above 25 W m −2 ). COVID-19 measures could be implemented at any time during the year, and their effects on urban climate are likely to vary by season and be reflected in various extremes. To understand the potential impacts throughout the year, the 2 m temperatures in the GEM100 and GEM20 simulations are compared for each season (Figure 6a-d). This comparison reveals that the 2 m temperature during the afternoon traffic peak is up to 1 °C lower in the GEM20 simulation than in the GEM100 simulation for all seasons, with the differences being slightly larger in the autumn (SON) and winter (DJF) seasons ( Figure  6c,d). In all seasons, the differences in 2 m temperature are co-located with the road network, and the greatest differences occur for the grid cells with large heat fluxes from traffic ( Figure 1a). Considering that heat fluxes from traffic usually account for between one third and one half of the total anthropogenic heat flux [38], these results are broadly consistent with previous studies which have shown that the total anthropogenic heat flux can increase the local near-surface air temperature by ∼1-2 °C in summer and ∼2-3 °C in winter [44][45][46].
The grid cells where the sensible heat flux from traffic exceeds 10 W m −2 contribute close to one third of the total heat flux for the region. Given that these grid cells also exhibit the largest temperature differences (Figure 6a-d), their data were used to assess the temporal variations in the difference between the GEM100 and GEM20 simulations. The greatest variation follows the daily cycle of traffic, with the temperature difference exhibiting the same morning and afternoon peaks as the heat flux from traffic (Figure 6e), although somewhat attenuated, as the heat flux varies by a factor of about 50 during the day, while the temperature difference varies by a factor around 10. This attenuation can be explained by the heat storage of roads and buildings during the hours with the heaviest traffic load, heat which is subsequently released during hours with lower loads. In addition to the heat flux from traffic, the roughness of the urban region also plays an important role in controlling temperature differences at the local scale, with a lower roughness favouring a decreased vertical mixing, allowing enhanced near-surface warming due to the heat flux from traffic and thus leading to larger 2 m temperature differences (Figure 6f). COVID-19 measures could be implemented at any time during the year, and their effects on urban climate are likely to vary by season and be reflected in various extremes. To understand the potential impacts throughout the year, the 2 m temperatures in the GEM100 and GEM20 simulations are compared for each season (Figure 6a-d). This comparison reveals that the 2 m temperature during the afternoon traffic peak is up to 1 • C lower in the GEM20 simulation than in the GEM100 simulation for all seasons, with the differences being slightly larger in the autumn (SON) and winter (DJF) seasons (Figure 6c,d). In all seasons, the differences in 2 m temperature are co-located with the road network, and the greatest differences occur for the grid cells with large heat fluxes from traffic (Figure 1a). Considering that heat fluxes from traffic usually account for between one third and one half of the total anthropogenic heat flux [38], these results are broadly consistent with previous studies which have shown that the total anthropogenic heat flux can increase the local near-surface air temperature by ∼1-2 • C in summer and ∼2-3 • C in winter [44][45][46].
The grid cells where the sensible heat flux from traffic exceeds 10 W m −2 contribute close to one third of the total heat flux for the region. Given that these grid cells also exhibit the largest temperature differences (Figure 6a-d), their data were used to assess the temporal variations in the difference between the GEM100 and GEM20 simulations. The greatest variation follows the daily cycle of traffic, with the temperature difference exhibiting the same morning and afternoon peaks as the heat flux from traffic (Figure 6e), although somewhat attenuated, as the heat flux varies by a factor of about 50 during the day, while the temperature difference varies by a factor around 10. This attenuation can be explained by the heat storage of roads and buildings during the hours with the heaviest traffic load, heat which is subsequently released during hours with lower loads. In addition to the heat flux from traffic, the roughness of the urban region also plays an important role in controlling temperature differences at the local scale, with a lower roughness favouring a decreased vertical mixing, allowing enhanced near-surface warming due to the heat flux from traffic and thus leading to larger 2 m temperature differences (Figure 6f).
Substantial day-to-day variations in the temperature difference between the GEM100 and GEM20 simulations can be partly explained by variations in wind speed, with lower wind speeds favouring less mixing, meaning that heat from traffic remains near the surface, increasing the temperature differences (Figure 7a). The vertical temperature gradient in the lowest atmospheric layer also plays a role, with positive gradients (i.e., atmospheric inversions) being associated with larger temperature differences, as the increased atmospheric stability suppresses vertical mixing (Figure 7b). As inversions are much more common during winter than during summer, this also explains the temperature differences being larger in winter. Substantial day-to-day variations in the temperature difference between the GEM10 and GEM20 simulations can be partly explained by variations in wind speed, with lowe wind speeds favouring less mixing, meaning that heat from traffic remains near the sur face, increasing the temperature differences (Figure 7a). The vertical temperature gradien in the lowest atmospheric layer also plays a role, with positive gradients (i.e., atmospheri inversions) being associated with larger temperature differences, as the increased atmos pheric stability suppresses vertical mixing ( Figure 7b). As inversions are much more com mon during winter than during summer, this also explains the temperature difference being larger in winter.   Substantial day-to-day variations in the temperature difference between the GEM100 and GEM20 simulations can be partly explained by variations in wind speed, with lower wind speeds favouring less mixing, meaning that heat from traffic remains near the surface, increasing the temperature differences (Figure 7a). The vertical temperature gradient in the lowest atmospheric layer also plays a role, with positive gradients (i.e., atmospheric inversions) being associated with larger temperature differences, as the increased atmospheric stability suppresses vertical mixing ( Figure 7b). As inversions are much more common during winter than during summer, this also explains the temperature differences being larger in winter.

Modelled Impact of COVID-19 on Extreme Weather
As previously mentioned, increased heat stress in urban areas is widely recognized as the most important effect of the UHI on human health. Since heat stress is strongly linked to the near-surface air temperature, a hot hour is defined in this study as an hour where the 2 m temperature exceeds 30 • C. During the simulation period, the median grid cell in urbanized areas experienced 111 hot hours in the GEM100 reference, compared to 70 h in non-urbanized areas (Figure 8a). The reduction in the heat flux from traffic resulted in reductions of up to 20% in hot hours in the GEM20 simulation, over those grid cells with large traffic fluxes (Figure 8b). The grid cells where the sensible heat flux from traffic exceeds 10 W m −2 experienced on average 8 fewer hot hours in the GEM20 simulation, with the greatest reductions during the afternoon traffic peak, where high temperatures coincide with heavy traffic. Similar reductions are observed if hot hours are defined as hours when the humidex exceeds 35 (Figure 8c,d). The humidex combines temperature and humidity information in an attempt to better approximate heat stress [13,47]. Given that the largest reductions in heat stress occur within traffic corridors, the users of active transportation (i.e., pedestrians and cyclists) would be among the most benefitted.
that the largest reductions in heat stress occur within traffic corridors, the users of activ transportation (i.e., pedestrians and cyclists) would be among the most benefitted.
Freezing rain is major winter hazard, posing substantial dangers to surface transpor tation. Due to the UHI effect, urban regions usually experience reduced freezing rain com pared to non-urban regions. In the reference GEM100 simulation, the median grid cel received 12 mm of freezing rain during the simulation period (Figure 9a). Lower accumu lations are observed over highways and main roads due to the surfaces staying above °C, thus preventing the freezing of precipitation on impact. In the GEM20 simulation, th reduced heat flux from traffic results in cooler surfaces, increasing freezing rain accumu lation by up to 50% (Figure 9b). Given that these greater accumulations would occur du to substantially reduced traffic, the overall number of traffic incidents would likely b lower for the duration of COVID-19 measures [48], but the essential traffic would be ex posed to more hazardous road conditions. While road de-icing is routinely performed during freezing rain events, the expected increase in freezing rain accumulation due to reduced traffic needs to be taken into account when planning de-icing strategies unde reduced traffic conditions, in particular with respect to the required amount of de-icing agents (salt and abrasives).  Freezing rain is major winter hazard, posing substantial dangers to surface transportation. Due to the UHI effect, urban regions usually experience reduced freezing rain compared to non-urban regions. In the reference GEM100 simulation, the median grid cell received 12 mm of freezing rain during the simulation period (Figure 9a). Lower accumulations are observed over highways and main roads due to the surfaces staying above 0 • C, thus preventing the freezing of precipitation on impact. In the GEM20 simulation, the reduced heat flux from traffic results in cooler surfaces, increasing freezing rain accumulation by up to 50% (Figure 9b). Given that these greater accumulations would occur due to substantially reduced traffic, the overall number of traffic incidents would likely be lower for the duration of COVID-19 measures [48], but the essential traffic would be exposed to more hazardous road conditions. While road de-icing is routinely performed during freezing rain events, the expected increase in freezing rain accumulation due to reduced traffic needs to be taken into account when planning de-icing strategies under reduced traffic conditions, in particular with respect to the required amount of de-icing agents (salt and abrasives).

Summary and Conclusions
In this study, GEM simulations at a 250 m horizontal grid spacing were shown to be able to adequately capture the spatiotemporal variation in LST over Montreal and the surrounding areas when compared to estimates from MODIS, as well as the mean and variability of the 2 m air temperature at eight observing stations, performing especially well during summer daytime, when the strongest impacts of the UHI on human health occur. These urban climate simulations were then used to assess the impact of COVID-19 measures on urban temperatures, which here is assumed to occur through the substantial decrease observed in the usage of surface transportation, thus reducing the amount of heat emitted by traffic. The results show that an 80% reduction in the heat emitted by traffic corresponds to a decrease of up to 1 °C in the near-surface temperature for regions with heavy traffic. Weather conditions such as low near-surface wind speeds and temperature inversions in the surface layer were modelled to accentuate the contribution of traffic to the UHI, leading to the greatest reductions in urban temperatures. While modest, the reduction in near-surface temperatures is modelled to reduce hot hours by up 20% during the warm season, thus reducing heat stress for vulnerable populations, including pedestrians and cyclists. It is important to note that no substantial changes occur outside of traffic corridors. In the context of increasing temperatures due to global warming and potential adaptation/mitigation measures, these results indicate that potential reductions in traffic would need to be supplemented by additional measures to reduce the UHI effect and the associated heat stress.
Due to their computational cost and data availability considerations, GEM simulations were performed only for Montreal and its surroundings-a typical mid-latitude urban centre. Given the well-known physical mechanisms underlying the main findings of this study, most large urban centres around the globe are expected to exhibit a similar response to a reduction in surface traffic. This study only considers the direct effects of traffic-related emissions on urban temperatures. Indirect effects on temperature from aerosol and GHG reductions are likely to be close to negligible at the global scale, unless future emissions stay well below their pre-pandemic trajectory [20], but given their potential influence on the atmospheric radiation budget in the vicinity of urban regions, further study is merited.

Summary and Conclusions
In this study, GEM simulations at a 250 m horizontal grid spacing were shown to be able to adequately capture the spatiotemporal variation in LST over Montreal and the surrounding areas when compared to estimates from MODIS, as well as the mean and variability of the 2 m air temperature at eight observing stations, performing especially well during summer daytime, when the strongest impacts of the UHI on human health occur. These urban climate simulations were then used to assess the impact of COVID-19 measures on urban temperatures, which here is assumed to occur through the substantial decrease observed in the usage of surface transportation, thus reducing the amount of heat emitted by traffic. The results show that an 80% reduction in the heat emitted by traffic corresponds to a decrease of up to 1 • C in the near-surface temperature for regions with heavy traffic. Weather conditions such as low near-surface wind speeds and temperature inversions in the surface layer were modelled to accentuate the contribution of traffic to the UHI, leading to the greatest reductions in urban temperatures. While modest, the reduction in near-surface temperatures is modelled to reduce hot hours by up 20% during the warm season, thus reducing heat stress for vulnerable populations, including pedestrians and cyclists. It is important to note that no substantial changes occur outside of traffic corridors. In the context of increasing temperatures due to global warming and potential adaptation/mitigation measures, these results indicate that potential reductions in traffic would need to be supplemented by additional measures to reduce the UHI effect and the associated heat stress.
Due to their computational cost and data availability considerations, GEM simulations were performed only for Montreal and its surroundings-a typical mid-latitude urban centre. Given the well-known physical mechanisms underlying the main findings of this study, most large urban centres around the globe are expected to exhibit a similar response to a reduction in surface traffic. This study only considers the direct effects of traffic-related emissions on urban temperatures. Indirect effects on temperature from aerosol and GHG reductions are likely to be close to negligible at the global scale, unless future emissions stay well below their pre-pandemic trajectory [20], but given their potential influence on the atmospheric radiation budget in the vicinity of urban regions, further study is merited.

Data Availability Statement:
The data presented in this study are available on request from the corresponding author.