Satellite-based long-term spatiotemporal trends of wildfire in the Himalayan vegetation

Analysis of the spatiotemporal pattern of burned areas over time is necessary to understand how fire behavior in the Himalayan region has altered as a result of the complex climatic variables. The differenced Normalized Burnt Ratio (NBR) is calculated utilizing the cloud-based platform Google Earth Engine (GEE) to quantify the extent of burned regions. The spatial distribution of burnt areas in the Himalayan region over the last 21 years has been examined and correlated with climatic and edaphic factors in the current study. The area affected by forest fire has shown a direct correlation with the land surface temperature, but an inverse relationship with surface soil moisture, pre-fire precipitation, pre-fire Normalized Difference Vegetation Index (NDVI) and pre-fire Enhanced Vegetation Index (EVI). The p-value for 9 of the 20 regions in which the research area has been divided for the spatial analysis is less than 0.05, implying that the regression model is statistically significant. Trend analysis done using Mann–Kendall test and Theil–Sen estimator state the distinct trends of burnt area and other meteorological and edaphic parameters in the Western, Central and Eastern Himalaya. The assessment of burned areas aids forest managers in mitigating the impacts and managing the forest fires, as well as in the implementation of the restoration methods following a forest fire.


Introduction
Forest fires have a significant impact on ecosystem structure and function, as well as the landscape features (Chuvieco and Congalton 1989;Sunar and Özkan 2001;Gupta et al. 2018). The Himalaya is world's youngest mountain range, exposed to a variety of calamities, including forest fires. It is among the most unstable and fragile mountain areas of the world that is strongly affected by the drivers of global climate change (Shrestha et al. 2012). Forest fires occur more often in the western Himalayan region than in the eastern Himalayan region, owing to the fact that the eastern Himalaya receives more precipitation and is thus wet than the drier western Himalaya (Mukhopadhyay 2009). Forest fires have become more frequent and intense as a result of the large-scale spread of pine forests in the various parts of the Himalaya (Dobriyal and Bijalwan 2017). Analyzing the difference between the satellite images of the single band indices of a region, before and after the fire occurrence can be used to quantify the burn severity. This difference values indicates where the vegetation has been damaged as a result of the forest fire. Pre-fire and post-fire Normalized Burn Ratio (NBR) (Key and Benson 1999) images indicate the condition of the area before and after forest fire, respectively, and the difference between them helps in identifying the burned areas of the region (Gupta et al. 2018). The fire season in Himalayan region marks its beginning from February and continue till June (Babu 2019;ISFR 2016ISFR , 2019Satendra and Kaushik 2014). Forest fires, depending on their severity and spatiotemporal variation, can have long-term effects on the dynamics of the ecosystem and its climate (Ryan 2002).
Studies have been conducted using satellite images with different spatial resolutions to identify the burned areas (Miettinen et al. 2007;Veraverbeke et al. 2014;Chintala et al. 2017). Geospatial tools are effective in mapping the burn severity as the removal of vegetation, the presence of char on the leaves, changes in soil and fuel moisture content, relative humidity, and other factors lead to significant changes in the reflectance values that can be used to identify burnt areas (Jakubauskas et al. 1990). Since fire reduces the amount of chlorophyll in the leaves and the amount of moisture in the plant and atmosphere, reflectance values in the visible and near infrared (NIR) bands reduces, while reflectance in the short-wave infrared (SWIR) bands rises (White et al. 1996;Gupta et al. 2018) used hybrid (unsupervised and visual interpretation) classification algorithms using AWiFS (Advanced Wide Field Sensor) data from the RESOURCESAT-2 satellite for burnt area assessment in Uttarakhand following the fire episode of 2016. Rogers and Kearney (2004) used an index called Normalized Differenced Soil Index (NDSI) having same band combination as NBR, which reduces the spectral variability of water, vegetation and soil to a three-endmember model, for estimating their fractional representation in Thematic Mapper (TM) data. However, in the present study, data from Enhanced Thematic Mapper plus (ETM+) has been used which has two 8-bit gain ranges. Miettinen et al. (2007) assessed and compared the results of burnt area estimation from medium resolution satellite images of the Moderate Resolution Imaging Spectroradiometer (MODIS) and active fire hotspot data and found that MODIS burnt area method is unable to detect small burned areas caused by small-scale shifting cultivation activities in the study site, whereas the hotspot-based method gives more consistent estimations of the burnt area, for some sites although overestimated to some extent. Babu et al. (2018) mapped the forest fire burned areas using Sentinel-2 optical satellite datasets using NBR and Relativized Burn Ratio (RBR). These indices were calculated and compared using pre-fire and post-fire imageries. MODIS active fire locations were used for the validation, and it was found that RBR was more accurate than NBR in estimating the burn severity in low vegetation-covered areas.
Although remote sensing data processing has been an effective tool for mapping and identifying the forest fire affected regions but one of the major limitations being that the satellites with good spatial resolution generally do not have good temporal resolution. As a result, the damage caused by small fires is generally overlooked, and is only recognized when it impacts a larger region. The initial purpose of the NBR was to locate burned regions and provide a qualitative evaluation of the fire intensity. Many studies, however, demonstrate a strong correlation between field-measured burned areas and NBR-derived burned areas (Wagtendonk et al. 2004;Roy et al. 2006).
The pattern of spatial distribution of burned areas in the fire-affected areas in the Himalayan region is examined over the past 21 years in this study. The differenced NBR is used to estimate the burned areas using Landsat 7 optical images using the cloud-based platform called Google Earth Engine (GEE). The spatial pattern of burnt areas over time for 20 subregions in the Himalayan region was compared with various meteorological and edaphic indicators such as land surface temperature (LST), surface soil moisture (SSM), Normalized Difference Vegetation Index (NDVI) (Tucker 1979), and Enhanced Vegetation Index (EVI) (Huete et al. 2002) to see if there is any correlation. The trend and its magnitude for the burnt area and other parameters have been estimated using Mann-Kendall test and Theil-Sen estimator, respectively. The scope of the study includes the analysis of the longterm spatiotemporal pattern of the burnt area with climatic and edaphic parameters which are the dynamic parameters and are influenced by the climate change. Static parameters such as slope, aspect, elevation and relief are not considered in the study since analyzing the long-term trend is the main objective of the study. More parameters that may impact the burnt areas in any region can also be considered in this study. Burned areas must be assessed as soon as possible at the local or regional level so that economic, social, and environmental damages may be evaluated and mitigation and restoration actions can be carried out (Bahuguna and Singh 2002).
The study aims to understand the long-term spatiotemporal pattern of the burnt area in the entire Himalayan region spanning the Western Himalayan part in India, covering the Central Himalaya part in Nepal and the Eastern Himalayan part covering Bhutan and the northeastern part of India. This pattern has been correlated and analyzed with the long-term patterns of meteorological and edaphic factors such as NDVI, EVI, LST, precipitation and surface soil moisture which directly or indirectly influence the burn severity of the area along with statistical estimation of trend and magnitude. The study aims to highlight and the distinct patterns of forest fire impact due to the dynamic climatic and vegetation in western, central and eastern Himalaya.

Study area
The research area spans the entire Himalayan region, from Jammu and Kashmir to Arunachal Pradesh. Since the study area is extensive, it was divided into 20 small regions to better understand the spatial pattern of the forest fire burned areas. The sub divisions of the study area are as follows: Jammu and Kashmir, Ladakh, Himachal Pradesh, Garhwal region, Kumaon region, far-west Nepal, midwest Nepal, West Nepal, Central Nepal, East Nepal, Bhutan, West Bengal (only Himalayan part), Sikkim, Nagaland, Mizoram, Manipur, Assam, Meghalaya, Tripura, and Arunachal Pradesh. The Western Himalaya, Central Himalaya, and Eastern Himalaya regions make up these locations. The reason for dividing the entire area into smaller sections is that the meteorological conditions within these regions differ considerably (Shrestha et al. 2012). Rainfall pattern, moisture conditions, temperature, etc., vary from the western side to the eastern side and thus to analyze the spatiotemporal pattern of the burned areas and impact of different parameters on it, and the study area was divided into 20 regions whose details are given below clubbed into three headings.

Geographical extent
The Indian Himalayan Region (IHR) spans roughly around 5 hundred thousand square kilometers. It forms the country's northern boundary and includes Himachal Pradesh, Uttarakhand, Sikkim, Meghalaya, Tripura, Mizoram, Manipur, Nagaland and Arunachal Pradesh among the Himalayan states, two partly hill states as West Bengal and Assam, and two union territories, Ladakh and Jammu and Kashmir. Nepal is a country in Southeast Asia with a total land area of about 150,000 square kilometers and is part of the eastern Himalayan biodiversity hotspot, whereas Bhutan is a small country in Southeast Asia with sub-alpine Himalaya.

Climate
The Himalaya is a climate-controlling and regulating feature that prevents cold continental air from entering India from the north in the winter and southwesterly monsoon (rain-bearing) winds from losing most of their moisture before reaching the range northward due to its location and height (Tewari et al. 2017). On the Indian side, this results in substantial precipitation (both rain and snow) but dry conditions in Tibet. Nepal is divided into five climatic zones based on altitude, namely tropical and subtropical zone, sub-arctic zone, temperate zone, cold zone, and arctic zone. The climate of Bhutan is exceedingly variable due to its altitude, precipitation, and effects of the North Indian monsoon. Figure 1 depicts the entire study region, divided into 20 sections.

Vegetation
Tropical, subtropical, temperate, and alpine vegetation are the four types of vegetation found in IHR. Elevation and precipitation play a big role in determining their biogeographic zone. Pine, Oak, Rhododendron, Sal, Shisham, Teak and other tree species are common in the area. Nepal possesses a large variety of medicinal and aromatic plants, making it one of the world's richest biodiversity reserves. In the low elevations, subtropical and temperate deciduous mixed forests with primarily Sal can be found, while Chir pine dominates the subtropical-pine forests. Oak and Rhododendron dominate the landscape at moderate altitudes. Bhutan contains around 5400 plant species, with fungi accounting for a significant portion of the Bhutanese ecosystem.

Satellite datasets
In this study, cloud computing platform GEE has been used to retrieve freely available satellite data (optical) from the earth-observing satellite Landsat 7 with the Enhanced Thematic 1 3 Mapper (ETM+) sensor for a period of 21 years (2001-2021) for the entire Himalayan region to estimate the burnt areas. GEE is a cloud-based geospatial analytic platform that lets the users access, visualize, study, and export satellite images for any part of the world (Gorelick et al. 2017). It is a popular and valuable tool for researchers since it enables easy analysis of a vast amount of publicly available satellite data via its Application Programming Interface (API) (Shelestov et al. 2017).
For the study area and time period, datasets for meteorological and edaphic parameters such as land surface temperature, precipitation, surface soil moisture, NDVI and EVI were accessed (Table 1). The MOD13A2 product provides a 16-day NDVI and EVI composite with a spatial resolution of 1 km. For each year, these products were taken prior to the fire season. MODIS is a sensor on the Terra and Aqua satellites launched in 1999 and 2002, respectively. In every 1 to 2 days, Terra and Aqua MODIS pass over the entire globe, collecting data in 36 spectral bands. The 8-day MODIS Land Surface Temperature and Emissivity (MOD11A2) composite products have been used, which offer pixel wise Land Surface Temperature and Emissivity values with a spatial resolution of one kilometer. The global precipitation measurement (GPM) mission is an international satellite program that gives three-hour measurements of rain and snow around the world. The integrated multi-satellitE retrievals for GPM (IMERG) is an algorithm that integrates the data from all passive-microwave instruments in the GPM constellation to provide rainfall estimates. GPM data were obtained for each year's pre-fire season, i.e., December of the previous year and January of the present year, in order to examine its impact and correlation with the burnt severity of the area. The Soil Moisture Active Passive (SMAP) global soil moisture data provided by the National Aeronautics and Space Administration (NASA) and the United States Department of Agriculture (USDA) provides soil moisture information at a 10 km spatial resolution around the world. SSM data are available from 2015, and hence it was used for the fire season of 2015 to 2021. Table 1 contains information on the datasets utilized in the study.

Methods
Optical images from Landsat 7 satellite were used to estimate the burnt area for the last 21 years, i.e., from 2001 to 2021 by computing the differenced NBR. Satellite imageries have been retrieved using GEE. Forest masking has been done using NDVI thresholding method to exclude non-forest areas from the study, and then NBR was computed. NBR is a tool for detecting and extracting burned patches in large fire zones (Key and Benson 1999). The formula is similar to NDVI; however, it uses NIR and SWIR bands. The difference between pre-fire NBR and post-fire NBR, which are calculated for the day before the fire season and after or during the fire season, respectively, is referred to as differenced NBR. The formula for calculating NBR and dNBR is given as follows: The value of dNBR ranges between − 1 and + 1. The higher the dNBR value, the more severe the burn and the more damage the area has, whereas areas with negative dNBR values may suggest regrowth after a fire. The categorized value range of dNBR is shown in Table 2 Benson 2006, 1999;Escuin et al. 2008;Babu et al. 2018 ). Categorization of the dNBR values has been done following the literature (Key and Benson 1999) which states that the categorization is based on the evaluation of the amplitude of index response to different classes on field, direct correlation to field rated burn severity, and visual characteristics of derived images.
The data for each year was retrieved, and NBR was calculated for the entire Himalayan region. Similarly, data for other factors mentioned were also retrieved using the GEE platform for the study area. All of the datasets were processed in Geographical Information System (GIS) software (ArcMap 10.5) for further spatiotemporal pattern analysis.
For each year and each of the 20 regions, burned areas, mean LST, mean NDVI, mean EVI, mean SSM, and total precipitation were computed. The data were then exported as a spreadsheet so that the numeric values could be further examined. The methodology flow of this study is shown in Fig. 2. For statistical analysis, the correlation coefficient has been calculated to measure the degree of association between the burned regions and each of the factors considered in the study. Multiple regression analysis was used to determine the strength of the association between the dependent variable and the independent variables (Burgan and Aksoy 2022). Trend analysis was done by performing the nonparametric Mann-Kendall test, and its magnitude was estimated using Theil-Sen slope estimator (Neeti and Eastman 2011).

Results
As per the assessment of the burned areas region-wise for the Himalayan region over the last 21 years, the forests in the western and central Himalaya have been more devastated by forest fires than those in the eastern Himalaya. In comparison with the northeastern states of India and parts of Bhutan, the forest areas of Jammu and Kashmir, Himachal Pradesh, Garhwal and Kumaon regions of Uttarakhand, and parts of Nepal have significant amount of burned area. The burned area's pattern of changes was investigated using LST, precipitation, surface soil moisture, NDVI and EVI. A statistical study was also performed to determine the relationship between burned area and the parameters mentioned.

Burnt area assessment
For the last 21 years, from 2001 to 2021, Landsat 7 satellite data was used to assess burnt area for the Himalayan region subdivided into 20 regions: Jammu and Kashmir, Ladakh, Himachal Pradesh, Garhwal region, Kumaon region, far-west Nepal, midwest Nepal, West Nepal, Central Nepal, East Nepal, Sikkim, West Bengal, Bhutan, Assam, Meghalaya, Nagaland, Mizoram, Manipur, Arunachal Pradesh and Tripura. This subdivision of the research area was accomplished to better understand the burned region's spatial characteristics and pattern and how it relates to local weather and vegetation conditions. The burnt area pattern for the last 21 years for the entire Himalaya was also assessed; however, because of the huge size and differences in rainfall, temperature, moisture and vegetation in the western, central and eastern Himalaya, no specific pattern appeared for the entire study area. The maps for the study area (Himalayan region) are shown in Fig. 3, with the highlighted burnt regions detected using the dNBR, indicated by red patches. These maps indicate that the area of land burned varies by location and year. It has been found in the study that the western Himalaya has more burnt regions than the eastern Himalaya. On the various maps shown above, burned area of different parts of the study area is highlighted. To understand the spatial distributions of the fire-affected areas, raster images of the burned area are overlaid above the base map.

Spatiotemporal analysis
The burned area in each of the 20 locations has been extracted and quantified, and their spatial and temporal pattern was examined. Weather and vegetation characteristics such as precipitation, temperature, NDVI, EVI, and soil surface moisture patterns are also examined for the same locations. From 2001 to 2021, the graphs in Fig. 4 indicate the spatiotemporal pattern of burned areas as well as the parameters specified for each fire season.
The graphs show the relationship of the burnt area with the LST, pre-fire (December-January) total precipitation, NDVI and EVI for eight out of 20 regions of the study area. These areas are grouped as a representative of Western Himalaya (Jammu and Kashmir, Himachal Pradesh, Garhwal and Kumaon region), Central Himalaya (midwest Nepal) and Eastern Himalaya (Sikkim, Arunachal Pradesh and Tripura) to better understand the spatiotemporal pattern of the burnt area in these parts of Himalaya.
The results shows an inverse relationship between the burnt area and precipitation, as the amount of precipitation in any year's pre-fire season (December and January) increases, the area of the region affected by fire decreases, as seen in 2008 in Arunachal Pradesh, 2020 in the Kumaon region, and midwest Nepal, and if there is insufficient rainfall or snowfall in December and January, the amount of area damaged by forest fire increases, as seen in 2006 in Sikkim, 2007 in Tripura and 2006 in Arunachal Pradesh. The spatiotemporal pattern of the burnt area with the pre-fire precipitation is found to have similarity in groups of Jammu and Kashmir, Himachal Pradesh and Garhwal region, another group as Kumaon region and midwest Nepal and lastly in the Eastern Himalaya part depicted for Sikkim, Arunachal Pradesh and Tripura. In the case of surface soil moisture, a similar pattern is observed.
The graphs show the direct association between the burnt area and LST as the burnt areas are higher in years with higher than normal temperatures, i.e., when there are dry conditions, such as in 2004 in Himachal Pradesh, Garhwal and Kumaon region and in 2006 in Sikkim and Arunachal Pradesh. When there is a dip in LST in any year for any region, the area affected by forest fire also decreases, as illustrated in the graphs for Garhwal region, Kumaon region, midwest Nepal and Sikkim in the year 2020 in Fig. 4.
The variation in the burnt areas in the same region is also studied with the pattern of vegetation indices (NDVI and EVI). The NDVI and EVI are measured during the prefire season, that is, in December of the previous year and January of the current year. Both NDVI and EVI have shown an inverse relationship with burnt area. The burned area increases as the NDVI and EVI values decline slightly in the pre-fire or winter season, as seen in the years 2006 and 2010 in Arunachal Pradesh, 2007 in Tripura and 2005 in midwest Nepal. When the NDVI and EVI values rise, the burned area decreases, as seen in

Statistical analysis
The correlation coefficient for each parameter is determined by estimating the correlation of burnt areas with NDVI, EVI, LST, precipitation, and surface soil moisture. The correlation coefficient denotes the degree of positive or negative relationship between two variables. The values of correlation coefficient for each region are given in Table S1 (supplementary). The results show that for total precipitation, the majority of the correlation coefficient values are less than zero, indicating a strong negative correlation between burnt areas and precipitation, while for temperature, the majority of the values are greater than 0, indicating a strong positive correlation between burnt areas and land surface temperature. The correlation coefficient values for surface soil moisture, pre-fire NDVI, and pre-fire EVI also show a strong negative correlation with the burnt areas.
Burnt areas were used as the dependent variable in this study in the regression analysis, while temperature, precipitation, NDVI, and EVI were used as independent variables. Regression analysis is a tool for determining the relationship between two or more variables. As data on surface soil moisture was only available after 2015, it was not used in regression analysis. The R square value, adjusted R square value, and p-value for each region are shown in the Table S2 (supplementary). The results imply that the regression model for 9 of the 20 regions studied, namely Jammu and Kashmir, Ladakh, Himachal Pradesh, Garhwal region, far-west Nepal, midwest Nepal, west Nepal, central Nepal and Bhutan is statistically significant as the p-value is less than 0.05. The R square value and adjusted R square value for these regions are likewise greater in comparison with other regions. The aggregate impact of all the parameters on the dependent variable is represented by these statistical values. The causal relationship between the variables is not stated in the regression analysis.
For each parameter, residual plots have been plotted for each region. Residual plots are basically error plots; therefore, there should be no discernible pattern in them. Residual plots, like the plots illustrated in Fig. S1 (supplementary) should be random in character. These plots have a residual value or the error value, which is the difference between the predicted and observed values following the regression analysis. The plots with error values that are randomly distributed and have no discernible pattern are considered to be right. The values in the residual plots of land surface temperature, NDVI, EVI, and precipitation shown in the Fig. S1 are virtually evenly scattered above and below the origin and have no distinct pattern, i.e., they are randomly distributed. From the graphs shown in Fig. S1, it can be seen that the residual values for land surface temperature, total precipitation, EVI and NDVI for the regions of Western Himalaya and Central Himalaya, i.e., for Jammu and Kashmir, Himachal Pradesh, Garhwal region, Kumaon region and midwest Nepal are randomly distributed and do not show any distinct pattern, while for the regions in Eastern Himalaya, i.e., Sikkim, Tripura and Arunachal Pradesh, the residual values or the error values are showing some kind of pattern. Descriptive statistics information such as skewness, coefficient of variation and confidence intervals of the data used is given in the supplementary section (Table S3-S8).

Trend analysis
Over the last 21 years, trend of burnt area in the Himalayan region divided into 20 subregions was analyzed along with the trend of precipitation, land surface temperature, NDVI and EVI. The trend was analyzed using Mann-Kendall test and the magnitude of the trend was estimated using Theil-Sen slope. The maps in Fig. 5 show that for the regions in Western Himalaya, i.e., Jammu and Kashmir, Ladakh, Himachal Pradesh, Garhwal and Kumaon region show significant decreasing trend for the burnt area and non-significant decreasing trend for the land surface temperature. For the same region, pre-fire precipitation, NDVI and EVI showed the increasing trend for last 21 years.
For Central Himalaya, covering Nepal region has shown an overall decreasing trend for burnt area, land surface temperature and pre-fire precipitation and increasing trend for the pre-fire NDVI and EVI has been observed, while for the Eastern Himalaya covering Bhutan and all the northeastern states of India, decreasing trend for burnt area has been observed with the overall increasing trend for land surface temperature, pre fire precipitation, NDVI and EVI.

Discussion
The Himalayan range is home to some of the world's most diverse ecosystems. It is the world's youngest mountain range, and it is also the most susceptible to forest fires. It has a varying level of forest fire risk due to its broad area and differences in weather, vegetation and topographic features. The climatic perturbations in the region as a result of climate change have resulted in episodes of abnormally high temperatures which has resulted in significant forest fire episodes in the Himalayan range.

Burnt area assessment
As per the analysis of the pattern of burnt areas by region, parts of the Western and Central Himalaya, such as Jammu and Kashmir, Himachal Pradesh, Uttarakhand's Garhwal and Kumaon regions, and parts of Nepal, have experienced a large number of forest fires incidents in the last 21 years, whereas the eastern part of Himalaya has had less area affected by forest fires which can be attributed to the Himalayan belt's highly variable climate and rainfall patterns. Data on precipitation over the last two decades analysed in the present study indicates that the eastern Himalaya is wetter than the western Himalaya. Tropical wet evergreen and tropical moist deciduous forests can be found in the eastern Himalayan range, whereas coniferous forests and dry deciduous forests can be found in the western Himalayan range (Upgupta et al. 2015). Although Chir pine trees are naturally fire resistant but the pine needles are extremely combustible due to their high calorific value and the presence of resin ducts in them, which is one of the leading causes of forest fires in the western Himalaya (Gupta et al. 2018). Inaccessibility to the complex terrain, as well as migration of people from the Himalayan high regions, contributes to the severity of the fires in these locations (Babu et al. 2015). Villagers used to gather pine needles from the forest floors for their own use, which aids in the elimination of the fuel content, i.e., pine needles from the forest floors that catch fire even with a little flame. Since smaller fires are spotted at the local level, which are usually missed for large regions due to the constraints in spatial resolution of satellite images, so assessing burned areas at the local and regional levels is more useful than doing so for a broad area. This study involves the assessment of the burnt area for 20 administrative regions that together constitute the Himalayan range at the spatial resolution of 30 m, capturing the spatiotemporal damage caused by the forest fires both small and large in the region.

Spatiotemporal analysis
The regional assessment of burned areas and its analysis using various parameters shows that the pattern of forest fire in the Himalayan region has evolved over the last 21 years. Drought-like conditions, erratic rainfall patterns, heavy monsoonal rainfall but little winter precipitation, a decreasing trend in the diurnal temperature range, increasing population, climate change, changes in vegetation type in response to climate change, and other abnormalities in the climate pattern have made the Himalayan region much more vulnerable over time (Shekhar et al. 2010;Pandey et al. 2014;Sharma et al. 2017). Warmer and drier winters reduce the quantity of moisture in the air and fuels, increasing the intensity and severity of forest fires resulting in far more ecological damage (Negi and Kumar 2016). Coniferous trees that are quite flammable in nature are abundant in the forests of Uttarakhand, Himachal Pradesh, Jammu and Kashmir and Nepal which comprises the Western and Central Himalayan region (Girdhar 2017). The graphs in Fig. 4 illustrate that the dry winter season has had a significant role in the amount of area burned by forest fires. It also illustrates that in comparison with the western Himalayan states and Nepal, the northeastern states have received a significant amount of rainfall and thus distinct patterns are recognized in the western, central and eastern Himalaya. Pre-fire rainfall ensures a sufficient quantity of soil and fuel moisture content, which minimizes the severity of forest fires in any area (Babu et al. 2018).
Warmer climatic conditions result from a consistent increase in the daily mean minimum temperature (Shekhar et al. 2010). Results show how the rise in temperature throughout the fire season has increased the severity of forest fire in the regions. For all regions, LST has a direct relationship with burnt areas, whereas precipitation, NDVI, EVI, and soil moisture have shown an inverse relationship with fire severity and burned areas. The graphs in Fig. 4 also illustrate that the Eastern Himalaya is warmer than the Western Himalaya, which could be due to the lower latitudinal range of the eastern Himalaya (Dobriyal and Bijalwan 2017). The distinct pattern of the rise and fall of LST and burnt area can be seen in different regions comprising of the groups of Western, Central and Eastern Himalaya.
Surface soil moisture data over the last seven years indicates that low moisture content in the soil aids in the spread of fire, but high moisture content in the soil owing to rainfall and other factors inhibits the fire from spreading. The chlorophyll content of the leaves can be measured by NDVI, whereas the EVI value is a reflection of the canopy cover in the area (Lee et al. 2008). The use of NDVI for multi-temporal analysis can provide information regarding fire-affected areas. Graphs in Fig. 4 illustrates that NDVI and EVI values are slightly higher in Nagaland, Tripura, and other parts of the eastern Himalaya, while they are significantly lower in Uttarakhand's Garhwal and Kumaon region, parts of Nepal, and other parts of the western Himalaya. This is due to the fact that the eastern Himalaya receives more rainfall and has a different vegetation type (Jaiswal et al. 2002). Lower NDVI and EVI values indicate that the forest areas in the region have been degraded and are vulnerable to forest fires (Mamgain et al. 2022).

Relationship of the fire burnt severity with long-term climatic parameters
The correlation analysis of different parameters with burned area shown in Table S1 states that burned areas are negatively correlated with pre-fire season precipitation for most of the regions in the study area, i.e., low precipitation in the winter season leads to dry weather conditions, which increases fire intensity. Similarly, the burned area has a negative relationship with surface soil moisture. For all 20 regions, land surface temperature is found to be positively associated with burnt areas. High-temperature causes drier conditions, and minimal rainfall diminishes moisture in soil and vegetation, resulting in a significant fire risk in the region's forests (Bai et al. 2018). Burned areas have a substantial negative connection with NDVI and EVI.
Correlation coefficient (r) and coefficient of determination (R 2 ) have been widely used for model evaluation but these statistics are oversensitive to high extreme values (outliers) and insensitive to additive and proportional differences between model predictions and measured data (Legates and McCabe 1999;Moriasi et al. 2007). A p-value is calculated for each location using multiple regression analysis with burnt area as the dependent variable and other factors as independent variables. The p-value for 9 of the 20 regions in which the research study area is divided for the spatial analysis is less than 0.05, indicating that the regression model is statistically significant. These nine locations encompass the majority of the western Himalaya, implying that analyzing the spatial pattern of burnt areas in the Eastern Himalaya is substantially more complicated when all of the factors are considered. For multiple regression analysis, the adjusted R square value is used, which shows that the values for the regions in the Eastern Himalaya are quite low, while the values for the Garhwal region, parts of Nepal, and other parts of the Central and Western Himalaya are relatively high.
The error values, or residuals, are the differences between the actual and predicted values following a regression analysis. The residual plots, or error plots, plotted for regions of Western and Central Himalaya for each parameter reveal no discernible pattern, and the residual values are fairly uniformly distributed above and below the graphs' origin as shown in Fig. S1, while distinct pattern is visible for the regions of Eastern Himalaya.

Trend analysis
Analysis of the trend using Mann-Kendall test and Theil Sen slope estimator has revealed that the trends for the burnt area, land surface temperature, pre-fire precipitation, NDVI and EVI has significant variations in the complete Himalayan range. Distinct trends are observed in the Western, Central and Eastern Himalayan range as shown in Fig. 5. In the Western Himalayan region, burnt area has shown significant decreasing trend and land surface temperature has shown overall non-significant decreasing trend for last 21 years which can also be correlated to the changing trend of the meteorological and vegetation parameters as the pre-fire precipitation, NDVI and EVI has shown increasing trends for the region (Singh and Mal 2014). Burnt area showed overall decreasing trend for central and eastern Himalaya with increasing trends for winter mean NDVI and EVI. Although land surface temperature and winter precipitation showed overall decreasing trend for Central Himalaya, it showed overall decreasing trend for the Eastern Himalaya. Analysis of the trends at 95% confidence level shows that the land surface temperature and the pre-fire precipitation have not shown any statistically significant trend for the entire Himalayan range for last 21 years. The trends may vary when analyzed at sub-division (micro-scale) and regional (meso-scale) level as it has been found that region wise analysis may ignore the micro-scale spatial variations ).

Conclusion
A large amount of satellite images for a huge study area can be analyzed using a cloudbased platform like Google Earth Engine. Remote sensing and GIS tools assist in obtaining and interpreting data from remote and inaccessible places for a variety of parameters. Longterm pattern recognition of burned areas for smaller regions provides insight into how fire behavior has altered over time and in response to complicated climate trends. Although forest fire patterns differ by location, studying them over the entire Himalayan region provides an overview of forest fire behavior and intensity patterns. Because of its dry climatic conditions and persistent decrease in the diurnal range of temperature, the western Himalaya, despite being colder than the eastern Himalaya, has more forest fires and hence more loss of flora and fauna. Factors such as precipitation, temperature, soil moisture, NDVI and EVI have a substantial impact on burned areas. Precipitation and soil moisture have a negative association with burned areas, although land surface temperature has a positive correlation. Since the value range of NDVI and EVI is narrow and does not fluctuate largely, they have a moderate negative correlation with burned areas. This study also finds its relevance in understanding the changing trend of forest fires along with the variability in climate and will help in the better management of forest fires. The study shows the distinction in the spatiotemporal pattern of the burnt area between the Western Himalayan region, Central Himalayan region and the Eastern Himalayan region. A timely assessment of burned regions aids in estimating flora and fauna losses, as well as planning and reducing damage circumstances following forest fires. Forest managers will be able to manage and reduce future biodiversity loss due to forest fires utilizing spatial analysis of pattern of the burnt areas, fire severity, and progression in a given area. The result will help in development effective forest fire mitigation policies in the Himalayan region which is hampered due to insufficient information and scientific studies in the region.
Author contributions SM involved in conceptualization, methodology, data curation and writing-original draft. AR involved in conceptualization, methodology and writing-review and editing. HCK involved in methodology and writing-review and editing. PC involved in writing-review and editing.
Funding The authors declare that no funds, grants, or other support were received during the preparation of this manuscript.