Remote Sensing of Local Warming Trend in Alberta, Canada during 2001–2020, and Its Relationship with Large-Scale Atmospheric Circulations

: Here, the objective was to study the local warming trend and its driving factors in the natural subregions of Alberta using a remote-sensing approach. We applied the Mann–Kendall test and Sen’s slope estimator on the day and nighttime MODIS LST time-series images to map and quantify the extent and magnitude of monthly and annual warming trends in the 21 natural subregions of Alberta. We also performed a correlation analysis of LST anomalies (both day and nighttime) of the subregions with the anomalies of the teleconnection patterns, i.e., Paciﬁc North American (PNA), Paciﬁc decadal oscillation (PDO), Arctic oscillation (AO), and sea surface temperature (SST, Niño 3.4 region) indices, to identify the relationship. May was the month that showed the most signiﬁcant warming trends for both day and night during 2001–2020 in most of the subregions in the Rocky Mountains and Boreal Forest. Subregions of Grassland and Parkland in southern and southeastern parts of Alberta showed trends of cooling during daytime in July and August and a small magnitude of warming in June and August at night. We also found a signiﬁcant cooling trend in November for both day and night. We identiﬁed from the correlation analysis that the PNA pattern had the most inﬂuence in the subregions during February to April and October to December for 2001–2020; however, none of the atmospheric oscillations showed any signiﬁcant relationship with the signiﬁcant warming/cooling months.


Introduction
The world is continually experiencing warming (i.e., incremental rise in temperature over a long period of time) at an unprecedented rate [1]. This has been occurring over the decades as noticed from the steady increase in the global annual mean temperature anomaly [2]. Warming rates are nearly double in the northern high latitudes compared to the global average [3,4], Canada being the most impacted country [5]. It has been attributed to cumulative net CO 2 emissions, non-CO 2 radiative forcing agents (i.e., methane, aerosol, nitrous oxide, etc.), and anthropogenic activities such as industrialization and land-use/land-cover changes across the globe [6,7]. Warming at the global level has not been occurring at the same pattern at a local scale [8]. Incremental local temperature (i.e., local warming) has adverse effects on humans, natural resources, and the economic development of society. For instance, Quebec (an Eastern Canadian province) experienced high temperature that killed over 70 people in July 2018, while on the western coast of the analyzed at the natural subregional level. On the other hand, Yan et al. determined the correlation coefficients of the atmospheric parameters with the mean LST anomaly considering entire North America as 71 grid units of 900 km * 900 km (810,000 km²). It would not be appropriate for the subregions of Alberta, where even Alberta (area of 661,848 km²) itself is smaller than one grid they covered. Considering the limitations and issues identified from the literature, the overall objective was to comprehend the local warming trend and its driving factors in the natural subregions of Alberta using MODISacquired monthly LST data at 5.6 km spatial resolution. The specific objectives were: (i) quantification of monthly and annual warming trends in the 21 natural subregions of Alberta using the Mann-Kendall test and Sen's slope estimator on the day and nighttime MODIS LST time-series images; and (ii) correlation analysis of both day and nighttime LST anomalies of the 21 subregions with North American oscillations, i.e., SST anomaly, AO, PDO, and PNA patterns. Note that we excluded the NAO index from the analysis because it has a strong effect only on winter weather in Europe, Greenland, northeastern North America, North Africa, and Northern Asia [36], whereas Alberta is in the northwestern part of North America.

Description of the Study Area
We selected Alberta, a western Canadian province, as the study area (see Figure 1). It is between latitudes 49 and 60°N and longitudes 110 and 120°W with an area of 661,848 km 2 . According to Statistics Canada, the population of Alberta increased from 2,974,807 in 2001 [37] to 4,067,175 in 2016 [38] and reached an estimated 4,436,258 as of 01 January, 2021 [39]. The major cities of Alberta are Calgary, Edmonton (Capital), Red Deer, and Lethbridge (see Figure 1), with a population of 1,361,852; 1,047,526; 106,736; and 101,324, respectively, as of 1st July 2020 [40]. The ecology of the area is geographically divided into six natural regions and 21 subregions (see Figure 1) based on the landscape patterns, vegetation, soil types, and physiographic features [41]. The distribution of these features is due to the combined influence of climate, topography, and geology in the area [42,43]. In terms of the climate of the study area, the mean temperature varies from −7.1 to 6 °C with long cold winters and short summers, and mean annual precipitation is 510 mm [41]. The elevation ranges from 150 to 3650 m above mean sea level [41].

Data Used and Their Preprocessing
We used three datasets in this study, including: (i) MODIS monthly LST image product (MOD11C3 Version 6) at 5.6 km for 2001-2020 acquired from the National Aeronautics and Space Administration (NASA); (ii) North American atmospheric oscillation indices, i.e., large-scale anomalies that influence the variability of the atmospheric circulations, from NOAA's National Centers for Environmental Information (NCEI); and (iii) shapefiles of the provincial boundary of Alberta and Alberta natural subregion obtained from geospatial data repository of the Government of Alberta. In the case of atmospheric oscillation indices, we used monthly tabular anomaly data of Niño 3.4 SST, AO, PDO, and PNA patterns for 2001-2020. We chose these four indices for their potential influences in the study area, including the following reasons: (i) the SST anomalies above and below the threshold of +0.5 • C and −0.5 • C cause El Niño and La Niña phenomena, respectively; (ii) the AO is characterized by winds circulating counter clockwise around the Arctic at around 55 • N latitude; (iii) PDO is defined by the pattern of SST anomalies in the North Pacific basin of 20 • N; and (iv) the positive phase of the PNA pattern is associated with above-average temperatures over western Canada and United States.
We preprocessed the acquired MODIS LST image in HDF formats to subset and mosaic for the study area in the NAD83 UTM Zone 12 North projection system. Next, we extracted layers of the LST and quality indicators for both day and nighttime. We used the quality layers to check for any low-quality pixels or gaps in the LST layers caused due to cloud or aerosol presence in the atmosphere during the image acquisition. The Bit flags in the quality indicator layer aided in interpreting the LST layers' quality assurance. We excluded those pixels where the average error for LST was greater than or equal to 3 K in the quality indicator layer. Note that such excluded pixels were negligible in the study area, i.e., 0.005 and 0.006% for day and nighttime images, respectively.

Mapping Local Warming Trend
We applied two non-parametric statistical tests, i.e., Mann-Kendall [44,45] and Sen's slope estimator [46], for determining the spatial and temporal (time-series) trends in the monthly and annual day and nighttime LST time series for both the entire study area (Alberta) and each of the 21 natural subregions. Moreover, we performed pixel-level trend analysis for mapping the spatial dynamics in the study area. We adopted these nonparametric statistical tests because they are independent of data distribution, insensitive to outliers, suitable for skewed data, and widely used for the trend analysis of atmospheric data [47,48].

Mann-Kendall Test
This test compared each value in the time series preceding it in sequential order, where the null hypothesis indicated no trend, and the alternative was a trend. To implement the statistical test, we calculated the sum of all the counts in the time series (i.e., statistics S) using Equation (1) [23].
where the number of points is n, the values in the time series i and j are x i and x j , respectively (considering j > i), and the sign function is sgn x j − x i calculated using Equation (2) [23]. For large n, the S tends to normality with the variance, var(S), calculated using Equation (3) [23].
where the length of time series and the number of tied values are n and m, respectively, and t i is the number of ties of the extent i. Here, a tied value indicates a set of points with the same value. Further, we used the standardized test statistics Z S [49] to determine the nature of the trends in the time series (i.e., increasing and decreasing trends with positive and negative Z S values, respectively) with the significance levels (p) at 90, 95, and 99% confidence (i.e., p equals to 0.10, 0.05, and 0.01, respectively). For the calculation of Z S (considering n > 10), we used Equation (4) [23].

Sen's Slope Estimator
We applied Sen's slope estimator (Q i ) to determine the magnitude of warming trend (i.e., slope) of each LST value in the time series for the N pairs of data using Equation (5) [23].
where x j and x k are the values of data pairs at times j and k (considering j > k). Q i ranks from the smallest to the largest, including the positive Q i values for increasing or upward trends (warming), the negative values for decreasing or downward trends (cooling), and the zero values for no-trends (no change) in the LST values of the time series.

Correlating Atmospheric Oscillations
To determine the influence of North American atmospheric circulations in the study area and each of the 21 natural subregions, we performed a correlation analysis between anomalies of LST (monthly and annual for both day and nighttime) and the atmospheric oscillation indices (i.e., PNA, PDO, AO, and SST Niño 3.4). We calculated Pearson's correlation coefficient (r) using Equation (6) [50].
where x i and y i are the values of x and y variables, respectively; x and y are the mean values of the x and y variables, respectively; and n is the number of observations.

Analysis of Local Warming Trend at the Natural Subregion Scale
Monthly analysis revealed that May was having the most significant warming trends (0.15 to 0.32 • C/yr) in daytime LST for all the natural regions except Grassland, and two subregions from each of the Rocky Mountain (Subalpine and Montane) and Parkland (Foothills Parkland and Central Parkland) regions (see Table 1). In addition, we observed significant warming trends in June, August, and December for the Northern Mixedwood (0.08 • C/yr), Athabasca Plain (0.20 • C/yr), and Northern Fescue (0.01 • C/yr) subregions, respectively. All the warming trends were statistically significant at 95% confidence, except the Alpine subregion (90% confidence). Conversely, we found significant cooling trends for the Grassland region in July (−0.14 to −0.34 • C/yr) and November (−0.31 to −0.36 • C/yr) except for Mixedgrass and Northern Fescue subregions, respectively. Furthermore, the Northern Fescue subregion showed a cooling trend in April. In the Parkland region, we noticed a significant cooling trend in July for the subregions of Foothills Parkland and Central Parkland, August for Central Parkland, and November for Peace River Parkland. Only one subregion (i.e., Dry Mixedwood) in the Boreal Forest region also showed a significant cooling trend in November. In the case of annual analysis, we did not find any significant warming trends, rather only cooling trends (−0.07 to −0.14 • C/yr) in the subregions of Montane (Rocky Mountains), Mixedgrass, Northern Fescue, and Foothills Fescue (Grassland), and Foothills Parkland and Central Parkland (Parkland).
In the case of monthly nighttime LST, we found a nearly similar pattern of warming trends with different magnitude (0.12 to 0.19 • C/yr) in May for all the natural regions, except Grassland and Canadian Shield, Montane subregion of the Rocky Mountain region, Foothills Parkland and Central Parkland in Parkland, and Athabasca Plain of Boreal Forest (see Table 2). In August, we observed it in Montane, Mixedgrass, Foothills Fescue, Foothills Parkland, Central Parkland, and Peace-Athabasca Delta subregions. In contrast, we noticed significant cooling trends in October for the Dry Mixedgrass and November for Dry Mixedgrass, Northern Fescue, and Northern Mixedwood subregions. However, we did not see any significant warming or cooling trends in the annual nighttimes. In general, we found warming trends occurred in the study area (Alberta) in May for both day and night at the rate of 0.14 and 0.11 • C/yr, respectively. From the linear regression model of the slopes of daytime and nighttime LSTs of subregions (see Figure 2), we discerned that they had an acceptable correlation (R 2 = 0.58), where the daytime is associated with an additional 0.60 of nighttime (i.e., nighttime was 60% cooler than the daytime).
The pixel-level spatial trend analysis revealed that daytime in May, June, July, August, November, and annual (see Table 3 and Figure 3a-f) and nighttime in May, June, August, and November (see Table 3 and Figure 3g-j) showed significant trends in at least 10% coverage of the study area with over 90% confidence levels. Among these, May was the significant month for the warming trends in both day and night with an area of 46.75 and 44.61%, respectively. This occurred in the 13 natural subregions in the northeastern, northern, northwestern, and western natural regions (Figure 3a,g), i.e., in most of Boreal Forest, Canadian Shield, Foothills, and the northern part of the Rocky Mountains. In the Rocky Mountain region (mostly Alpine and Subalpine subregions), area for the warming trend at night was more pronounced than for the day. In addition, we found areas of a significant warming trend in June and August with 7.90 and 7.43%, respectively, during the day scattered in the north and northeast (see Figure 3b,d), while it was 14.02 and 23.01%, respectively, at night scattered over the study area (see Figure 3h,i). We also found cooling trends of daytime during July, August, and November in the eastern, southeastern, and southern parts of the study area mostly in the Grassland and Parkland regions (see Figure 3c-e) and some areas in the centre and west under the Foothills and Boreal Forest regions in November (see Figure 3e). The nighttime cooling trend in November showed the spatial distribution scattered from the northwest to southeast through the centre in the study area (see Figure 3j).

Correlation Analysis of LST Anomalies with the Atmospheric Oscillations
We presented the correlations of the day and nighttime LST anomalies in the natural subregions with the PNA, PDO, AO, and SST (Niño 3.4) patterns in Tables 4-7 Table 4). We also found acceptable and satisfactory positive correlations (r = 0.51 to 0.62) in January for the subregions of Alpine, Subalpine, Upper Foothills, Lower Foothills, and Dry Mixedwood during the day and Alpine and Subalpine during the night (see Table 4).
In the case of the PDO, we observed acceptable to good positive correlations with the LST anomalies (r = 0.52 to 0.74) in all eight natural subregions of the Rocky Mountains, Foothills, and Parkland regions during both the day and night in April, including the entire Alberta (see Table 5). In April, all subregions of Grassland showed acceptable to good positive correlation during night, where some subregions of Boreal Forest showed acceptable during the day or night. We also found above the acceptable positive correlations in the Alpine and Subalpine subregions for January and Alpine and Northern Mixedwood for annual during the day and night, in addition to Foothills Fescue for March during the day and Lower Boreal Highlands for April during the night (see Table 5).
We noticed the acceptable to good correlations between the LST anomalies of the subregions and AO index were negative in April (r = −0.51 to −0.74) and positive in June (r = 0.50 to 0.77) (see Table 6). The negative correlations were in the Rocky Mountain, Foothills (except Lower Foothills at night), Grassland (except Dry Mixedgrass and Northern Fescue during the day), and Parkland (except Peace River Parkland during the day and night and Central Parkland at night) regions during both the day and night. In contrast, we observed positive correlations in June during the night for all subregions with the exception of Alpine, Subalpine, Peace-Athabasca Delta, and Kazan Upland, while during the day for the Alpine, Subalpine, Upper Foothills, Central Mixedwood, Lower Boreal Highlands, Upper Boreal Highlands, Peace-Athabasca Delta, Boreal Subarctic, and Kazan Upland subregions (see Table 6). In addition, we found positive correlations during nights in August for Peace River Parkland and Boreal Subarctic and during days in October for Lower Boreal Highlands and Peace-Athabasca Delta subregions. The correlations of the entire Alberta were negative (r = −0.51) during the day in April and good positive (r = 0.77) at night in June (see Table 6).
In most of the cases, the correlation between the LST anomalies and Niño 3.4 SST indicated a weak relationship, except in April with five subregions (Alpine, Subalpine, Montane, Upper Foothills, and Foothills Parkland) for the day and two (Alpine and Subalpine) for the night, June with four subregions (Dry Mixedgrass, Mixedgrass, Northern Fescue, and Dry Mixedwood) for the day, and July with one subregion (Northern Fescue) for the day (see Table 7). In addition, the entire Alberta showed an acceptable positive correlation (r = 0.57) in June during the day.

Discussion
LST is a major element in weather, climate, and natural (ecological) environment research. In this study, we used monthly day and nighttime LST in finding local warming and cooling trends in the natural regions and subregions of Alberta. Results reveal that both warming and cooling trends occurred in the study area. The warming trends were observed in the northern and western regions and subregions (see Figure 3a-d,g-i), and the cooling trends in the south and southeast (see Figure 3c-f,j). The extent and magnitude of the warming trends were more pronounced than cooling. A similar warming trend in the northern part of Alberta was also reported in another study [8]. This warming was mostly in the subregions of the Rocky Mountain and Boreal Forest regions (see Figure 3a,g). The Rocky Mountains are characterized by glaciers, scattered vegetation, non-vegetated steep sloping bedrocks, and receiving the highest amount of snowfall [41]. Warming in such cold Rocky Mountains with high elevations (the highest peak is Mount Robson in Canadian Rockies with 3954 m above mean sea level [53]) in comparison to other land cover at the same latitude would probably be due to the global climate change [54][55][56]. This region is more sensitive to climate change (particularly increasing global annual mean temperature [2]), where the warming rates are nearly double compared to the global average [3,4]. Such warming possibly accelerates the mechanism of heating surface related to elevation-dependent warming by decreasing snow/ice albedo, changing water vapour and releasing latent heat, and changing surface water vapour and radiative flux [28,30,54,57]. In addition, global warming is likely responsible for observing the warming trends in the Boreal Forest region in this study. It is impacting the ecosystem variables, such as the early onset of spring due to warming [58,59] leading to the onset of early runoff in March and April [60,61] that changes vegetation phenology of the region by wetting the soils early [62]. Wet soils retain more heat than dry soil, and the mechanism of heating land surface requires some additional time (time lag) [63,64]. This is probably the reason why results show the pronounced warming in May (for both day and nighttime) after the onset of runoff in March and April.
In contrast, the cooling trends we observed in the south during July and August (see Figure 3c,d) would probably be due to the increasing and intensified irrigation in the agricultural area during the growing season in the subregions of prairies of Southern Alberta [65] and an increment of July precipitation [66]. In fact, the irrigation area increased by 13.4% in the province during 2000-2019 [65]; that might be the reason for the declining LST trend we found in the agriculture-dominant area in the south and southeastern Alberta. Irrigation and increased precipitation help wetting and moistening soils to increase the latent heat flux between the irrigated agricultural land and atmosphere to decrease the surface temperature [67,68]. While the moist and wet soils cool the surface by evaporation (from soils) and evapotranspiration (from vegetation) mechanisms using the solar radiation during the longer daytime, it retains surface heat when the mechanism ceased during shorter nighttime. This is a probable cause of having warming trends (with low magnitude) at summer nighttime (see Figure 3h,i). In addition, we observed significant cooling trends in the month of November (see Figure 3e,j) that coincided with other studies reported for the same area for the latter half of the last century [69,70]. Such cooling trends could be associated with the decrease in winter precipitation (including November) in Alberta [71], where less precipitation would potentially cause cooler winter temperatures in some places in Canada [72]. Overall, we found declining LST in the late spring and winter (especially October, November, and February) in most of the subregions; however, those were not significant (see Tables 1 and 2). The decreasing and increasing trends of LST (in most cases not significant) over the months in a year neutralized the significant trends in annual LST for both day and nighttime, except daytime declining trends observed in three subregions of Grassland, two of Parkland, and one of Rocky Mountain regions (see Table 1).
In addition, large-scale atmospheric circulations (i.e., teleconnections patterns) might have significant influence in causing local warming [9,33]. Correlation analysis between monthly LST and teleconnections patterns (i.e., PNA, PDO, AO, and SST Niño 3.4) for the period 2001-2020 showed that the PNA pattern had the most influence in the regions and subregions while moderate from PDO and AO and very little from SST (Niño 3.4). However, we did not find significant correlations of these teleconnections with the months that showed warming and cooling trends during the period 2001-2020. The most pronounced warming trend in the northern and western subregions was observed in May, where the PNA pattern showed the most influence during October to December and February to April. This is obvious because the PNA pattern is one of the most prominent teleconnections with low-frequency variability in the extra-tropical Northern Hemisphere that appears in fall through early spring in Western Canada [32,33]. Moreover, the positive polarity of the PNA pattern in the Canadian prairies is supposed to cause warmer temperatures in winter [73], and therefore, warmer associated land surface; however, we observed a decreasing LST trend in November. In contrast, the least influence we observed was the SST Niño 3.4 because of its concentration between latitudes 5 • N-5 • S and longitudes 120-170 • W [74], where Alberta is between latitudes 49 and 60 • N and longitudes 110 and 120 • W. In addition, we did not find any significant relationships of LST trends with PDO due to its potential association with SST [75]. Moreover, AO has more influence in the central and eastern parts of Canada during winter [76], and therefore, we did not find significant influence from it as the study area is located in Western Canada.

Conclusions
We demonstrated, in this study, the use of MODIS-derived monthly LST in mapping local warming and cooling trends in the 21 natural subregions of Alberta during 2001-2020 and its relationship with large-scale atmospheric circulations. For deriving the spatial extent and magnitude of the local warming and cooling trends in the study area, we applied the Mann-Kendall test and Sen's slope estimator. In addition, in finding the relationships of the warming trend with the atmospheric circulations, we correlated the LST anomaly of each subregion with the anomalies of PNA, PDO, AO, and Niño 3.4 SST teleconnection patterns (i.e., atmospheric oscillations). We found significant warming trends in most of the subregions of the Rocky Mountain and Boreal Forest regions in May during both day and nighttime. A significant cooling trend was observed during the daytime of July and August in the south and southeastern subregions of the Grassland and Parkland regions. In November, we also noticed a significant cooling trend. However, we did not find a significant warming trend in the annual LST, except a daytime declining trend in six subregions. In addition, analysis revealed that the PNA pattern (among the atmospheric oscillations used in this study) had the most influence in the subregions during February to April and October to December. However, none of the atmospheric oscillations showed any significant relationships with the monthly warming trend during 2001-2020. Since the influence of atmospheric oscillations is prevailing and appreciating for a longer period, a longer timescale of LST would be critical in building their relationships. Consequently, we would like to suggest further study using remote-sensing sensors that could provide LST for a longer period, e.g., NOAA satellites and passive microwaves. The outcomes of this study would be useful for policymakers and stakeholders at the local level to develop warming adaptation and mitigation strategies suitable for a sustainable environment. Researchers can also use this study to understand the changing pattern of monthly warming in the context of climate change.