Quantifying the Congruence between Air and Land Surface Temperatures for Various Climatic and Elevation Zones of Western Himalaya

The surface and near-surface air temperature observations are primary data for glacio-hydro-climatological studies. The in situ air temperature (Ta) observations require intense logistic and financial investments, making it sparse and fragmented particularly in remote and extreme environments. The temperatures in Himalaya are controlled by a complex system driven by topography, seasons, and cryosphere which further makes it difficult to record or predict its spatial heterogeneity. In this regard, finding a way to fill the observational spatiotemporal gaps in data becomes more crucial. Here, we show the comparison of Ta recorded at 11 high altitude stations in Western Himalaya with their respective land surface temperatures (Ts) recorded by Moderate Resolution Imagining Spectroradiometer (MODIS) Aqua and Terra satellites in cloud-free conditions. We found remarkable seasonal and spatial trends in the Ta vs. Ts relationship: (i) Ts are strongly correlated with Ta (R2 = 0.77, root mean square difference (RMSD) = 5.9 ◦C, n = 11,101 at daily scale and R2 = 0.80, RMSD = 5.7 ◦C, n = 3552 at 8-day scale); (ii) in general, the RMSD is lower for the winter months in comparison to summer months for all the stations, (iii) the RMSD is directly proportional to the elevations; (iv) the RMSD is inversely proportional to the annual precipitation. Our results demonstrate the statistically strong and previously unreported Ta vs. Ts relationship and spatial and seasonal variations in its intensity at daily resolution for the Western Himalaya. We anticipate that our results will provide the scientists in Himalaya or similar data-deficient extreme environments with an option to use freely available remotely observed Ts products in their models to fill-up the spatiotemporal data gaps related to in situ monitoring at daily resolution. Substituting Ta by Ts as input in various geophysical models can even improve the model accuracy as using spatially continuous satellite derived Ts in place of discrete in situ Ta extrapolated to different elevations using a constant lapse rate can provide more realistic estimates.


Introduction
Air temperature (T a ) is an important proxy for energy exchange between land-surface and atmosphere, making T a one of the most important parameters in climate research [1,2]. T a is generally gap, the glacio-hydrological community is further unsure of the significant role that spatiotemporally continuous satellite-derived surface temperatures can play as a substitute for spatially discontinuous T a observations. The land-surface temperature is more likely proxy of energy exchange between land-surface and atmosphere for phenomena which are more strongly linked to ground processes [27]. The main aim of the present study is to understand and quantify the statistically significant trends in T s − T a variation over a large spatiotemporal domain in Western Himalaya. Here, we start with providing a description of the study area, followed by data and used methods, and finally we discuss and conclude the main findings of the analyses.

Study Area
In the present paper, we analyze the relationship between daily and 8-day mean T a from 11 high-altitude weather stations (Table 1)  domain are completely missing for the Central and Western Himalaya, and (ii) owing to this research gap, the glacio-hydrological community is further unsure of the significant role that spatiotemporally continuous satellite-derived surface temperatures can play as a substitute for spatially discontinuous Ta observations. The land-surface temperature is more likely proxy of energy exchange between landsurface and atmosphere for phenomena which are more strongly linked to ground processes [27].
The main aim of the present study is to understand and quantify the statistically significant trends in Ts−Ta variation over a large spatiotemporal domain in Western Himalaya. Here, we start with providing a description of the study area, followed by data and used methods, and finally we discuss and conclude the main findings of the analyses.

Study Area
In the present paper, we analyze the relationship between daily and 8-day mean Ta from 11 highaltitude weather stations (Table 1)   The stations in the present study are located over a large elevation range above sea level (1587-4280 m) ( Figure 1). All the stations used in the analysis are located above 2100 m except for Srinagar which is located at 1587 m. These stations are located in three different Himalayan states of India namely Uttarakhand, Himachal Pradesh, and Jammu and Kashmir. In addition, we have also included one station (Shiquanhe) from the Chinese part of the region in the study (Figure 1). These stations represent various precipitation regimes of the region such as monsoon dominance, westerly dominance, the precipitation-transition zone from monsoon-to-westerly dominance, and orographic The stations in the present study are located over a large elevation range above sea level (1587-4280 m) ( Figure 1). All the stations used in the analysis are located above 2100 m except for Srinagar which is located at 1587 m. These stations are located in three different Himalayan states of India namely Uttarakhand, Himachal Pradesh, and Jammu and Kashmir. In addition, we have also included one station (Shiquanhe) from the Chinese part of the region in the study (Figure 1). These stations represent various precipitation regimes of the region such as monsoon dominance, westerly dominance, the precipitation-transition zone from monsoon-to-westerly dominance, and orographic precipitation-shadow zone. The topographic variations, i.e., altitudes and orography, among the Himalayan ranges not only govern the temperatures but also the precipitation [34]. Here, we aim to further untangle the degree of control of the altitude and orography in deciding the correlation between T s and T a in the Himalaya.

Air Temperature (T a )
The T a is generally observed at a height of about 2 m above the land surface. T a data observed at 11 different stations was used in the present study ( Table 1). The data for five stations from Global Historical Climatology Network (GHCN) which was acquired from National Centre for Environmental Information (NCEI), NOAA web portal (https://www.ncei.noaa.gov/) [35] had observed daily mean T a estimated using hourly or 6-hourly observations (Version 3). For the other six stations of Bhakra Beas Management Board (BBMB) and India Meteorology Department (IMD), the T a was calculated using the daily maximum and minimum observations due to unavailability of daily mean T a . This method of averaging the daily maximum and minimum temperatures for calculation daily mean temperature is widely used due to the instrumentation, logistic, and computational simplicity [36]. Although the method produces bias in the output due to inability to track the diurnal asymmetry [37], it has been used by considerable number of studies to make acceptable estimates requiring T a [36]. In order to understand the degree of bias for our study area, we compared the given daily mean T a with mean of daily maximum and minimum T a for the five stations of GHCN for which all the three parameters were available. The analysis showed that the RMSD was less than 1.62 • C with a very high correlation (R 2 > 0.96) for all the stations. In addition to the daily T a , we also calculated 8-day mean T a for comparison with the corresponding 8-day T s explained in next section. The observations were carefully checked for systematic and random errors before using it for further comparison. The stations used in the present study are distributed broadly over four different precipitation zones (Table 1, Figure 1). The precipitation varies significantly in these precipitation zones. The monsoon dominated, transition zone, westerlies dominated, and precipitation shadow zone receive >1500, 200-800, 600-800, and <150 mm total annual precipitation on an average, respectively.

MODIS Data
The daily, and 8-day night-and day-time T s from MODIS satellites on Terra (MOD11A1 and MOD11A2 available from February, 2000) and Aqua (MYD11A1 and MYD11A2 available from July, 2002) satellites [23,24] was downloaded from NASA Earthdata portal (https://earthdata.nasa.gov/) [38] and was used to calculate average of daily and 8-day T s ( Table 2). The remotely-sensed T s from MODIS (version 006) has been observed to have RMSD of less than 0.5 K in comparison to the in situ measurements of the T s [39] and therefore has been widely used for multiple scientific applications [18,19,22,[25][26][27][28]40]. The local time for the pass over the study area for Terra is around 10:30 and 22:30 and for Aqua is around 13:30 and 01:30 during day and night, respectively. The 8-day land surface temperature data MOD11A2 and MYD11A2 is a simple pixel wise average of all the respective MOD11A1 and MYD11A2 data collected during the 8-day period. The days with all the four observations, including the day and night-time measurements available from both Terra and Aqua were included in the comparison at both daily and 8-day scale. Equation (1) was used to compute the average of four MODIS observations during a day or 32 MODIS observations during an 8-day period (referred as T s in • C) from the pixel value corresponding to every station given in Table 1.
where, T T d = Terra day-time observation T T n = Terra night-time observation T A d = Aqua day-time observation T A n = Aqua night-time observation c = Constant for conversion from kelvin to Celsius (273.15) For every data point of daily and 8-day T s , two night-time and two day-time satellite observations were used. It moderated the calculated daily T s for further comparison with T a . Therefore, every data point of daily T s is average of four MODIS observations during that day and 8-day T s is average of 32 MODIS observations during that 8-day period. Since, the satellite observation from MODIS is unavailable in cloud cover condition and the calculated daily T s for comparison with T a can have large data gaps, we decided to also include MODIS 8-day T s in the analyses. For 8-day T s , the data available is comparatively more continuous due to correction of cloud contamination [39]. Although, the 8-day MODIS observations are more efficient in terms of temporal continuity, it compromises with the temporal resolution. Additionally, the number of data points available for comparison for 8-day average is significantly less than the dataset available for daily average. Thus, the average T s used in our analyses, and referred to hereafter, is essentially the average of four-times daily and 8-day MODIS T s observations and all the results should be considered accordingly. Therefore, based on these data limitations, we finally compared the average of four-times daily and 8-day MODIS T s observations with observed daily mean T a for five GHCN stations and with the average of observed daily maximum and minimum temperatures for the remaining six stations.

Statistical Analyses
We applied different statistical tools and tests to analyze the relationship between T s and T a . Firstly, the coefficient of correlation (r), coefficient of determination (R 2 ), standard error of regression (SE), and root mean square difference (RMSD) between T s and T a for all the stations was calculated. The SE is the standard deviation of the difference between two datasets while RMSD is the square root of mean of squared difference between two datasets. The R 2 explains the efficiency of the regression model. In other words, it is the degree to which the independent variable will be able to explain the dependent variable. During the analysis, the T a was considered to be the dependent variable (y) and T s was considered as the independent variable (x). The p-values for all the analyses were <0.01 at 99% confidence level. Additionally, we estimated these statistical parameters for all available data for different climate zones namely monsoon-dominated, transition, westerlies-dominated, and precipitation shadow and for all the stations. The value of modified R 2 which is adjusted for the number of predictors was observed to be around unadjusted R 2 and therefore was not shown in the table. The p-value for each of the analyses was found to be less than 0.01 at 99% confidence level showing the effect of predictors. In addition, we analyzed the variation in the magnitude of the coefficient of the difference between T s and T a observed after the multiple regression taking January as the base month. We also plotted the box and whisker plots for the daily difference between T s and T a to graphically represent the overall range of the data, median of the data, and distribution of the data in different quartiles.

T s vs. T a Relationship
We performed different statistical analyses to derive several first-hand conclusions regarding the relationship between T s and T a in the Himalayan region. The results show a strong relationship between observed daily mean T a and its respective daily mean T s in general for all the stations (R 2 = 0.77, RMSD = 5.9 • C, SE = 4.76, n = 11,101, p-value <0.01 at 99% confidence level) with variations corresponding to the altitudinal locations of the stations ( Figure 2 and Table 3). The strongest relationship between T a and T s at daily scale was observed for Shimla (R 2 = 0.94; RMSD = 1.5 • C, SE = 1.2 • C, n = 304, p-value <0.01 at 99% confidence level) and Mukteshwar stations (R 2 = 0.94; RMSD = 1.6 • C, SE = 1.2 • C, n = 355, p-value <0.01 at 99% confidence level) which are located on the southern slopes in monsoon-dominated precipitation regime. The coefficient of determination is considerable for all the stations (R 2 > 0.69, p-value <0.01 at 99% confidence level) at daily scale. Table 3. Summary of all the statistical tests used for analysis between T s and T a for all the stations, climate regimes and for all observations at daily and 8-day scale. (R 2 = Coefficient of determination; SE = Standard Error of Regression; RMSD = Root mean square difference).  The R 2 and RMSD for all the stations show slight improvement for 8-day average ( Figure 3 and Table 3). The relationship between Ta and Ts for 8-day average was also found to be strongest for Shimla (R 2 = 0.97; RMSD = 1.4 °C, SE = 0.96 °C, n = 55, p-value <0.01 at 99% confidence level) and Mukteshwar stations (R 2 = 0.96; RMSD = 1.2 °C, SE = 1.05, n = 63, p-value <0.01 at 99% confidence level). Overall, the Ta and Ts relationship was found to be stronger (R 2 = 0.96; RMSD = 5.7 °C, SE = 4.5, n = 3552, p-value <0.01 at 99% confidence level) at 8-day scale for all the stations as well. The regression equation for all the analyses was also given which can be used for estimating Ta for different climate regimes with continuity over large spatiotemporal domain using Ts (Table 3)  Overall observations The R 2 and RMSD for all the stations show slight improvement for 8-day average ( Figure 3 and Table 3). The relationship between T a and T s for 8-day average was also found to be strongest for Shimla (R 2 = 0.97; RMSD = 1.4 • C, SE = 0.96 • C, n = 55, p-value <0.01 at 99% confidence level) and Mukteshwar stations (R 2 = 0.96; RMSD = 1.2 • C, SE = 1.05, n = 63, p-value <0.01 at 99% confidence level). Overall, the T a and T s relationship was found to be stronger (R 2 = 0.96; RMSD = 5.7 • C, SE = 4.5, n = 3552, p-value <0.01 at 99% confidence level) at 8-day scale for all the stations as well. The regression equation for all the analyses was also given which can be used for estimating T a for different climate regimes with continuity over large spatiotemporal domain using T s (Table 3)  The number of data points available for 8-day analysis is significantly less in comparison to the daily analysis (Table 1). Additionally, the use of 8-day data gives spatiotemporal continuity due to correction of cloud contaminated pixels but poses a restriction on the frequency of comparisons. We decided to represent the analysis of the relationship between Ta and Ts at daily scale henceforth because there was very small improvement in the results observed after the use of 8-day data and the daily observation are crucial for different geophysical models as explained in the Introduction section. We noticed certain spatial patterns in the daily differences between Ts and Ta which are discussed in detail in the following paragraphs.

Altitudinal Relationship
First, we present the relationship between Ts-Ta by considering altitudinal positions of the stations. The variation in altitude affects the Ta due to difference in density of air which causes a reduced green-house effect in the higher reaches. The RMSD between Ts and Ta for stations has a direct correlation with the elevation of the station (Figures 2-4). Although, the RMSD increases systematically with increase in elevation in general, small variation in this trend is observed for northernmost stations (Skardu and Srinagar). The annual mean RMSD is strongly correlated to the elevation (R 2 = 0.74) in general (Figure 5a) except for two stations (Skardu and Srinagar), which even when located at comparatively low elevations show higher magnitude of RMSD (Figure 4). The R 2 is stronger for monsoon season (Figure 5b) in comparison to annual (Figure 5a) and summer season values (Figure 5c). The observed Ta was unavailable for Skardu for winter months (Figure 5d). Therefore, the R 2 is highest for winter when compared to monsoon, summer, and annual analysis. Furthermore, the magnitude of Ts-Ta is observed to be higher in summer months in comparison to the winter months for all the stations in general (Figure 4). The number of data points available for 8-day analysis is significantly less in comparison to the daily analysis (Table 1). Additionally, the use of 8-day data gives spatiotemporal continuity due to correction of cloud contaminated pixels but poses a restriction on the frequency of comparisons. We decided to represent the analysis of the relationship between T a and T s at daily scale henceforth because there was very small improvement in the results observed after the use of 8-day data and the daily observation are crucial for different geophysical models as explained in the Introduction section. We noticed certain spatial patterns in the daily differences between T s and T a which are discussed in detail in the following paragraphs.

Altitudinal Relationship
First, we present the relationship between T s − T a by considering altitudinal positions of the stations. The variation in altitude affects the T a due to difference in density of air which causes a reduced green-house effect in the higher reaches. The RMSD between T s and T a for stations has a direct correlation with the elevation of the station (Figures 2-4). Although, the RMSD increases systematically with increase in elevation in general, small variation in this trend is observed for northernmost stations (Skardu and Srinagar). The annual mean RMSD is strongly correlated to the elevation (R 2 = 0.74) in general (Figure 5a) except for two stations (Skardu and Srinagar), which even when located at comparatively low elevations show higher magnitude of RMSD (Figure 4). The R 2 is stronger for monsoon season (Figure 5b) in comparison to annual (Figure 5a) and summer season values (Figure 5c). The observed T a was unavailable for Skardu for winter months (Figure 5d). Therefore, the R 2 is highest for winter when compared to monsoon, summer, and annual analysis. Furthermore, the magnitude of T s − T a is observed to be higher in summer months in comparison to the winter months for all the stations in general (Figure 4).

Seasonal Relationship
The difference between mean monthly Ta and Ts (i.e., Ts-Ta) for the entire period of study shows high inter-monthly variability for all the stations except for the stations in monsoon-dominated regions (Figures 5 and 6). The mean monthly Ts is lower in comparison to mean monthly Ta for southern slopes (Figure 6a) and increases with increasing latitudes (Figure 6c) except for the stations in westerlies dominated areas (Figure 6d). The magnitude of difference between mean monthly Ts and mean monthly Ta is negative for the stations in monsoon-dominated areas and positive for the

Seasonal Relationship
The difference between mean monthly Ta and Ts (i.e., Ts-Ta) for the entire period of study shows high inter-monthly variability for all the stations except for the stations in monsoon-dominated regions (Figures 5 and 6). The mean monthly Ts is lower in comparison to mean monthly Ta for southern slopes (Figure 6a) and increases with increasing latitudes (Figure 6c) except for the stations in westerlies dominated areas (Figure 6d). The magnitude of difference between mean monthly Ts and mean monthly Ta is negative for the stations in monsoon-dominated areas and positive for the

Seasonal Relationship
The difference between mean monthly T a and T s (i.e., T s − T a ) for the entire period of study shows high inter-monthly variability for all the stations except for the stations in monsoon-dominated regions (Figures 5 and 6). The mean monthly T s is lower in comparison to mean monthly T a for southern slopes (Figure 6a) and increases with increasing latitudes (Figure 6c) except for the stations in westerlies dominated areas (Figure 6d). The magnitude of difference between mean monthly T s and mean monthly T a is negative for the stations in monsoon-dominated areas and positive for the stations in precipitation shadow and westerly-dominated regions. In the precipitation-transition zone, the difference is positive for summer months and negative for winter months except for Rakchham, the southernmost station of the transition zone ( Figure 5). For Rakchham, the T s − T a values are negative throughout the year similar to the stations in monsoon-dominated areas (Figure 6b). This might be a result of the added effect of humidity in the near-surface atmosphere and presence of snow on land surface which moderates the difference between T s and T a [42] throughout the year in monsoon-dominated regions. In the precipitation-transition zone, the difference is partly moderated by the presence of snow during winter months and partly humidity during summer months, particularly for the southernmost stations of the zone (Rakchham and Kalpa) (Figures 5 and 6b) which receive enough precipitation through both monsoon and westerlies. T s − T a values for the stations in westerly-dominated region are regulated mainly by the presence of snow during winters ( Figures 5 and 6d), which tends to cool the surface due to high albedo [42]. The T s − T a values for the stations in precipitation-shadow zone are significantly high and positive in magnitude throughout the year in comparison to all the other stations due to the perennial cold-arid atmospheric conditions (Figures 5 and 6c). This confirms the role of water cycle on this gradient and shows that in the absence of soil-atmosphere water cycle (dry conditions) the magnitude of the difference between T s − T a increases and is more positive.
Remote Sens. 2019, 11, x FOR PEER REVIEW 10 of 17 stations in precipitation shadow and westerly-dominated regions. In the precipitation-transition zone, the difference is positive for summer months and negative for winter months except for Rakchham, the southernmost station of the transition zone ( Figure 5). For Rakchham, the Ts − Ta values are negative throughout the year similar to the stations in monsoon-dominated areas ( Figure  6b). This might be a result of the added effect of humidity in the near-surface atmosphere and presence of snow on land surface which moderates the difference between Ts and Ta [42] throughout the year in monsoon-dominated regions. In the precipitation-transition zone, the difference is partly moderated by the presence of snow during winter months and partly humidity during summer months, particularly for the southernmost stations of the zone (Rakchham and Kalpa) (Figures 5 and  6b) which receive enough precipitation through both monsoon and westerlies. Ts -Ta values for the stations in westerly-dominated region are regulated mainly by the presence of snow during winters ( Figures 5 and 6d), which tends to cool the surface due to high albedo [42]. The Ts -Ta values for the stations in precipitation-shadow zone are significantly high and positive in magnitude throughout the year in comparison to all the other stations due to the perennial cold-arid atmospheric conditions (Figures 5 and 6c). This confirms the role of water cycle on this gradient and shows that in the absence of soil-atmosphere water cycle (dry conditions) the magnitude of the difference between Ts -Ta increases and is more positive. The comparison of Ts and Ta showed high inter-monthly variability throughout the study period. Therefore, we performed an additional analysis where we estimated the seasonal effect of each month on the difference between Ts and Ta (Figure 7) in reference to a base month. For this multiple regression analysis, January was considered as the base month since the Ts -Ta values in January were least for all the stations in general ( Figure 6). This analysis further corroborates the above-discussed aspect that the Ts -Ta coefficient values are larger in summer months in comparison to winter months ( Figure 7). Additionally, the difference in coefficient and RMSD is high for stations in precipitation shadow regions (Kaza, Shiquanhe and Losar) in comparison to the stations in monsoon-dominated areas (Shimla and Mukteshwar) (Figures 6 and 7). The comparison of T s and T a showed high inter-monthly variability throughout the study period. Therefore, we performed an additional analysis where we estimated the seasonal effect of each month on the difference between T s and T a (Figure 7) in reference to a base month. For this multiple regression analysis, January was considered as the base month since the T s − T a values in January were least for all the stations in general ( Figure 6). This analysis further corroborates the above-discussed aspect that the T s − T a coefficient values are larger in summer months in comparison to winter months (Figure 7). Additionally, the difference in coefficient and RMSD is high for stations in precipitation shadow regions (Kaza, Shiquanhe and Losar) in comparison to the stations in monsoon-dominated areas (Shimla and Mukteshwar) (Figures 6 and 7). To further corroborate the effect of seasonality and the presence of snow and humidity on the Ts -Ta values, we created monthly box and whisker plots of daily difference between Ts and Ta ( Figure  8). The whisker for the stations in precipitation shadow zone and transition zone is longer showing the high monthly variability of the difference value in comparison to the stations in monsoon and westerlies-dominated areas. This is due to the presence of snow during the winter and humidity in the atmosphere in summer regulating the difference between Ts and Ta in the monsoon dominated areas. Additionally, the size of the boxes are smaller for the stations in monsoon and westerliesdominated areas explaining the presence of maximum data points close to the median representing that throughout the year the difference between Ts and Ta is regulated by presence of snow or atmospheric moisture. On the contrary, the boxes for stations in precipitation shadow zones which receive significantly less precipitation throughout the year, are wider in size representing large variation in the difference between Ts and Ta throughout the year. The boxes for the stations in the southern part of the transition zone are smaller in summer and wider in winter showing the effect of humidity due to some influence of monsoon owing to their spatial closeness to the monsoondominated region. Besides, both the boxes and whiskers for the stations in north-eastern part of the transition zone, closer to the precipitation shadow zone, are wider in size showing the variability due to lack of both snow and humidity. To further corroborate the effect of seasonality and the presence of snow and humidity on the T s − T a values, we created monthly box and whisker plots of daily difference between T s and T a (Figure 8). The whisker for the stations in precipitation shadow zone and transition zone is longer showing the high monthly variability of the difference value in comparison to the stations in monsoon and westerlies-dominated areas. This is due to the presence of snow during the winter and humidity in the atmosphere in summer regulating the difference between T s and T a in the monsoon dominated areas. Additionally, the size of the boxes are smaller for the stations in monsoon and westerlies-dominated areas explaining the presence of maximum data points close to the median representing that throughout the year the difference between T s and T a is regulated by presence of snow or atmospheric moisture. On the contrary, the boxes for stations in precipitation shadow zones which receive significantly less precipitation throughout the year, are wider in size representing large variation in the difference between T s and T a throughout the year. The boxes for the stations in the southern part of the transition zone are smaller in summer and wider in winter showing the effect of humidity due to some influence of monsoon owing to their spatial closeness to the monsoon-dominated region. Besides, both the boxes and whiskers for the stations in north-eastern part of the transition zone, closer to the precipitation shadow zone, are wider in size showing the variability due to lack of both snow and humidity.

Discussion
The observed near-surface air temperature is one of the most important climate parameters used in different kinds of environmental studies particularly in Himalaya where the interaction between high elevation, climate, and cryosphere is highly significant and complex. It is extremely difficult to capture the spatial heterogeneity of the near-surface temperature [43] which is the primary forcing data for different glacio-hydrological models [3,4,[44][45][46]. It is also used as primary data for climate change assessment [47,48], agro-climatic [40], ecological [49,50], and socio-economic [51,52] studies. Our results present a freely available substitute for station recorded Ta with high temporal and spatial resolution. Conclusively, the Ts is highly correlated with Ta throughout the study area at both daily and 8-day scales. The correlation is highest at the stations located at Southern slope (Shimla and Mukteshwar) with significantly low RMSD in comparison to the stations located in the Eastern part (Losar and Shiquanhe). Although, the degree of congruence between Ts and Ta is slightly higher in the 8-day dataset (R 2 > 0.77) in comparison to the daily dataset (R 2 > 0.69), the number of data points available for comparison is significantly low. The overall RMSD improved by 0.2°C on an average by using the 8-day dataset. The largest improvement in RMSD was observed for Skardu (1.1 °C) but the number of data points available for correlation was significantly less than other stations. The overall SE improved by 0.38 °C except for Kalpa and Kaza for which it deteriorated by 0.02 and 0.08 °C, respectively. It is interesting to note that for Shiquanhe which is located in precipitation shadow zone and highest altitude among all the stations, shows largest improvement in SE (by 1.63 °C) and reduction in RMSD (0.2 °C).
The difference between Ts and Ta is primarily controlled by elevation, the land surface cover characteristics, and near-surface humidity. At higher altitudes, the thinner atmosphere shows lesser water holding capacity and the atmosphere saturates faster, thus allowing for lesser evaporation/sublimation in a given pressure-temperature scenario [53]. This puts a constraint on the limit of specific humidity in the high elevations and the comparatively lesser number of available water molecules in the near-surface atmosphere cannot trap the same amount of heat as those at lower elevations. This can provide a basis for the observed high values of Ts−Ta at the higher altitudes. The intercept of the regression between Ts and Ta shows increase for the stations in monsoon,

Discussion
The observed near-surface air temperature is one of the most important climate parameters used in different kinds of environmental studies particularly in Himalaya where the interaction between high elevation, climate, and cryosphere is highly significant and complex. It is extremely difficult to capture the spatial heterogeneity of the near-surface temperature [43] which is the primary forcing data for different glacio-hydrological models [3,4,[44][45][46]. It is also used as primary data for climate change assessment [47,48], agro-climatic [40], ecological [49,50], and socio-economic [51,52] studies. Our results present a freely available substitute for station recorded T a with high temporal and spatial resolution. Conclusively, the T s is highly correlated with T a throughout the study area at both daily and 8-day scales. The correlation is highest at the stations located at Southern slope (Shimla and Mukteshwar) with significantly low RMSD in comparison to the stations located in the Eastern part (Losar and Shiquanhe). Although, the degree of congruence between T s and T a is slightly higher in the 8-day dataset (R 2 > 0.77) in comparison to the daily dataset (R 2 > 0.69), the number of data points available for comparison is significantly low. The overall RMSD improved by 0.2 • C on an average by using the 8-day dataset. The largest improvement in RMSD was observed for Skardu (1.1 • C) but the number of data points available for correlation was significantly less than other stations. The overall SE improved by 0.38 • C except for Kalpa and Kaza for which it deteriorated by 0.02 and 0.08 • C, respectively. It is interesting to note that for Shiquanhe which is located in precipitation shadow zone and highest altitude among all the stations, shows largest improvement in SE (by 1.63 • C) and reduction in RMSD (0.2 • C).
The difference between T s and T a is primarily controlled by elevation, the land surface cover characteristics, and near-surface humidity. At higher altitudes, the thinner atmosphere shows lesser water holding capacity and the atmosphere saturates faster, thus allowing for lesser evaporation/sublimation in a given pressure-temperature scenario [53]. This puts a constraint on the limit of specific humidity in the high elevations and the comparatively lesser number of available water molecules in the near-surface atmosphere cannot trap the same amount of heat as those at lower elevations. This can provide a basis for the observed high values of T s − T a at the higher altitudes. The intercept of the regression between T s and T a shows increase for the stations in monsoon, transition, and westerlies-dominated areas. On the contrary, the stations in precipitation shadow zone show a sharp decrease in the intercept of the regression at high elevations (>3600 m). The slope of the regression between T s and T a is higher for stations in low elevation and precipitation-dominated areas (0.80-1.03) in comparison to the stations in high elevation and in transition-to-precipitation shadow zones (0.59-0.86). This observation is supported by a study which shows decrease in slope and degree of correlation in high elevation [54]. The high difference between T s and T a for the stations in dry atmosphere at high altitude may partially be due to the heat from the Sun and cooling of near-surface atmosphere due to heat exchange from surrounding air and temperature lapse rate [55]. The presence of more humidity moderates the difference between T s and T a in precipitation dominant areas. The difference between T s and T a is highest with positive magnitude when the land surface is snow-free and the near-surface atmosphere is dry. On the contrary, the T s and T a is negative and lower in magnitude when the land surface is covered by snow and/or atmosphere is more saturated with moisture regardless of high altitude. In addition to the elevation and precipitation regime, season was observed to have significant control over the difference between T s and T a . The summer months were observed to have a significantly higher effect on T a in reference to January, in general for all the stations. The inter-monthly variability was observed to be very high for year-round humidity-deficient transition zones and precipitation shadow zones in comparison to the monsoon-dominated and westerlies-dominated regions. It can be interpreted that the energy exchange between the surface and near-surface atmosphere in the precipitation dominant areas is more efficient in comparison to the precipitation deficient areas.
The lower magnitude of RMSD between T s and T a represents lower gradient of temperature between the land surface and near-surface air due to the cold bias caused by snow cover which protects the surface from warming because of its high albedo [42]. A possible contributing factor to this seasonal disparity can be the reported perpetually melting seasonal snow in Himalayan mountains [55] under the changing regional climate. The causal mechanism for this relationship deserves a separate detailed investigation. However, a possible cause of such observations can be linked to the fact that the diurnal temperatures even in the mid-winter months often cross the freezing point causing a certain degree of melting to prevail [56]. This can start a cascading event where during the preliminary warming phase, the average snowpack temperature reaches and stays at 0 • C isotherm until the melting typically starts within the snowpack prior to the ripening phase as meltwater is retained within the snowpack [57]. This meltwater may subsequently refreeze owing to the diurnal cycles of temperature and the latent heat released during this process can additionally warm the snow surface and the surrounding air, further minimizing the temperature difference [56]. In addition to the seasonal change in the land cover characteristics, the variation in humidity in the near-surface atmosphere is an important factor controlling the difference between T s and T a . It was recently proposed that the amount of moisture content on the land surface has a cooling effect on land surface temperature [40]. Thus, the precipitation regime in which a particular station is located can further provide us several clues regarding the observed variations in T s − T a values. These precipitation regimes have been previously characterized [34] and in the following discussion, we take a focused approach towards revisiting the T s − T a variations with respect to the respective precipitation scenarios.
All the statistical results and the regression equation between T s and T a have specific trends for particular climate setting and elevation which can be used to estimate T a using T s (Table 3) for glacio-hydrological and climate change studies in data-deficient Himalaya. For example, the RMSD ranges between 1.2-1.6, 2.6-5., 2.5-4.3, and 7.2-8.9 • C for the stations in monsoon-dominated, transition, westerlies-dominated, and precipitation shadow zones, respectively, for both daily and 8-day products. The slope and intercept of the regression equations between T s and T a are also similar for the stations in the same precipitation regime. The paper demonstrates different patterns of variation of T s − T a in different climate regimes within the region of study. Due to the inherent limitations of the available data, some of this analysis may be revised in the future by specific dedicated studies, in particular to asses if the relationships hold on daily scales and with what error bar. Some possible error sources for this analysis may come from the scarcity of the data, and the fact that we compared data from different instrumentation accuracies and cadences. T a is measured by three different organizations and two calculation methods are used for daily mean air temperature during different observation periods. The correlation of the instantaneous observation of T a in relation to satellite derived T s can be investigated by analyzing the diurnal variation of T a in relation to the time of pass of the satellite [58]. There are different parameters like wind speed and fractional vegetation which have additional effects on the difference between T s and T a , which have not been investigated in the present study [54,55] and can be interesting research questions for future investigations in the region.

Conclusions
Unavailability of reliable temperature observations with spatial continuity along with the extreme weather conditions and difficult terrain in the remoteness of Himalaya hampers our understanding of the cryosphere-climate coupling in these mountains. Here we attempt to compare remotely sensed T s with respect to in situ T a observations over different precipitation and altitudinal zones of the Western Himalaya. Although, there are several studies available from different parts of the globe attempting to estimate T a using T s or vice-versa using monthly or 8-day MODIS data, we provide an understanding of the spatiotemporal variability of the T a vs. T s relationship at diurnal scales. The results show a strong and statistically significant relationship between T s and T a in general with a spatiotemporal consistency, thus projecting satellite-derived T s as a viable alternative to the in situ T a for glacio-hydro-climatological studies. We also provide regression equations to facilitate modeling of gridded T a using corresponding T s for different regions of Western Himalaya. MODIS in combination with Sea and Land Surface Temperature Radiometer (SLSTR) onboard Sentinel-3 can provide better capability to overcome cloud gaps and ensuring spatiotemporal continuity for T s future studies in this direction.