Warming of the high-mountainous climate sensitive Jammu and Kashmir during the period 1980–2024

Abstract

This study examines the long-term surface temperature variability across Jammu and Kashmir using ground-based observations and reanalysis data during 1980–2024. The region shows a clear but spatially heterogeneous warming, with the strongest annual mean temperature (T_mean) rise at mid-elevation stations such as Bhaderwah (+ 0.3 °C/dec) and weak or insignificant trends at lower elevations like Jammu (about  − 0.1 °C/dec). Minimum temperature (T_min) shows the most rapid acceleration, by + 0.1 to 0.5 °C/dec at several mid-to-high elevation regions, whereas daytime maximum temperature (T_max) trends remain modest (about 0–0.2 °C/dec). These spatial and seasonal contrasts, along with enhanced warming at higher altitudes in specific seasons (e.g., pre-monsoon T_min up to + 0.6 °C/dec), indicate the presence of elevation-dependent warming (EDW) in these mountainous regions. The annual T_mean increases by 0.18 °C/km/dec and winter T_max shows the strongest altitude dependence of 0.43 °C/km/dec, whereas T_min exhibits no significant EDW across seasons. Multiple linear regression analysis suggests that wintertime altitude-dependent temperature trends are statistically associated with albedo-related surface processes, whereas the annual and seasonal increases in Tmin are more closely associated with atmospheric moisture and longwave radiative conditions. Overall, the region has warmed by up to nearly 1 °C in last two decades at several high-altitude locations (e.g., Pahalgam and Gulmarg), highlighting the sensitivity of the Himalayan environment to ongoing climate change with severe implications for high-altitude hydrology, cryosphere stability, and regional climate resilience.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-61302-2.

Introduction

Elevation-dependent warming (EDW) has become a major focus in contemporary climate research because temperature changes in high mountains often differ substantially from those in surrounding lowlands^1,2. A primary driver for EDW is the snow-albedo feedback, where warming accelerates snowmelt, exposing darker surfaces that absorb more solar radiation and further enhance warming at higher altitudes³. Water-vapor and cloud effects also contribute, as reductions in atmospheric water vapour at higher elevations weaken longwave absorption and alter cloud cover patterns, intensifying daytime warming and modifying radiative balance². Additionally, aerosol forcing, particularly changes in absorbing aerosols such as black carbon, enhances atmospheric heating and accelerates snow/ice darkening, whereas the minimum temperature is amplified through increased longwave interactions and boundary-layer stability changes^4,5. Land-use and vegetation changes, including deforestation and altered evapotranspiration, modify surface energy fluxes and can enhance or suppress EDW depending on regional conditions⁶. Additionally, glacier retreat, hydrological shifts, and ecosystem responses interact with atmospheric processes by altering surface roughness, moisture availability, and feedback loops that further reinforce elevation-dependent temperature trends⁷. Recent studies have further demonstrated that meteorological factors such as net radiation, temperature, and vapour pressure deficit strongly regulate crop water requirements and hydrological responses under drought and wet conditions, highlighting the critical role of vegetation–climate interactions in surface energy and water balance processes⁸.

Although it seems straightforward to evaluate EDW across elevation gradients, the task remains challenging due to data limitations. High-altitude meteorological stations with long-term observations are extremely sparse globally, and most available records originate from lower altitudes or valleys, which are strongly influenced by microclimatic factors such as cold-air pooling, slope orientation and shading^9,10. These factors can suppress or distort broader temperature trends, resulting in an observational network biased towards lower elevations and insufficient for detection of EDW.

Therefore, to supplement these gaps, previous studies often rely on satellite datasets, atmospheric reanalyses and climate models, but each of these has important constraints. Satellite products capture land-surface skin temperatures rather than near-surface air temperatures, and their utility in high terrain is reduced by cloud cover and short record lengths^11,12. Reanalysis datasets integrate multiple observational sources but are dominated by free-tropospheric data and are not fully homogenised for trend studies¹³. Climate models, from regional to global, have improved considerably, yet their ability to represent steep terrain, snow processes and land–atmosphere interactions remains limited^14,15. Despite these issues, many modelling studies consistently show amplified warming at higher elevations, particularly driven by snow–albedo feedback, cloud changes and shifts in atmospheric humidity^3,16.

Across major mountain systems, including the Rockies, Alps, Andes, East African highlands and the Tibetan Plateau, a substantial body of evidence points toward the presence of EDW, though with marked regional differences¹⁷. The signal is often stronger in minimum temperatures and during winter, highlighting the influence of nocturnal radiative processes and seasonal snow cover^18,19. Over the Tibetan Plateau, in particular, rapid warming has been widely documented, with several studies attributing elevation-linked amplification to reductions in snow depth, increased downward longwave radiation and changes in humidity²⁰. At the same time, some regions show weak or non-monotonic elevation responses, emphasising the role of local climate regimes, surface characteristics and observational uncertainties.

Recent studies have increasingly highlighted the emergence of EDW across the Indian Himalayan Region (IHR) above 3000 m, indicating that warming trends are often amplified at higher elevations and vary seasonally and spatially across different Himalayan sub-regions. For example, modelling and observational analyses over the IHR have shown enhanced warming above 3000 m (strongest during winter and post-monsoon seasons), particularly outside the monsoon season, associated with increased downward longwave radiation, snow–albedo feedbacks and changes in cloud and humidity patterns²¹. Similarly, studies over North-East India and the Eastern Himalayan Region (annual warming of about 0.16 °C/decade) have reported stronger warming trends in high-altitude states such as Arunachal Pradesh and Sikkim (post-monsoon and winter warming exceeding 0.2 °C/decade), suggesting a clear elevational influence on recent temperature changes²². These findings reinforce the growing evidence that EDW is an important feature of climate change over the Indian Himalaya (with stronger warming signals at higher elevations), although substantial regional heterogeneity remains. However, most existing studies have focused either on broad Himalayan domains or eastern Himalayan sectors, whereas the western Himalaya, particularly Jammu and Kashmir, remains comparatively underexplored due to sparse observations and complex topography.

The Jammu and Kashmir region, as shown in Fig. 1, located within the western Himalaya, is among the most climate-sensitive mountain environments in South Asia. Its steep elevation gradients, extensive ice/snow cover and reliance on snow- and glacier-fed rivers make it highly vulnerable to warming-related changes²³. Despite its importance, EDW has not been comprehensively assessed in this region because of sparse high-altitude observations and challenges associated with homogenising long-term temperature records. Determining whether warming intensifies with elevation in Jammu and Kashmir is essential for understanding future water availability, glacial melt, ecological shifts and downstream socio-economic impacts. Therefore, this study addresses this gap by quantifying elevation-dependent temperature trends across the region and evaluating their implications for climate resilience in the western Himalayas using long term temperature records from the India Meteorological Department (IMD) for the period 1980–2024.

Fig. 1.

Study region showing the altitude in kilometer (km) above mean sea level. Magenta marks indicate the locations of IMD temperature stations. The first three letters of each station name are labeled and correspond to the abbreviations used in Fig. 2, where the station elevations are also provided. All images are generated using Python version 3.13²⁸.

Data and methods

Data description

This study uses temperature data from ten surface observatories in the Jammu and Kashmir region, namely Banihal, Batote, Bhaderwah, Gulmarg, Jammu, Kokernag, Kupwara, Pahalgam, Qazigund and Srinagar (Altitudes in Table S1). These surface observatories are part of the operational network of India Meteorological Department (IMD) and are equipped with standardised meteorological instruments maintained according to IMD specifications and World Meteorological Organization (WMO) standards. Temperature observations are made manually using four precision thermometers—a dry bulb thermometer, a wet bulb thermometer, a maximum thermometer, and a minimum thermometer—all permanently installed within a Stevenson screen.

Station-wise monthly data availability is shown in Fig. S1. Although the majority of station-month combinations contain complete daily observations, some stations show periods of missing data, particularly during the early part of the record and at isolated intervals thereafter. The most notable gaps occur during the early 1980s at Jammu (JAM) and Gulmarg (GUL), around 1990 at Jammu (JAM), Kupwara (KUP), Banihal (BAN), Qazigund (QAZ), Kokernag (KOK), and Pahalgam (PAH), and intermittently at Batote (BAT), Bhaderwah (BHA), Kupwara (KUP), and Kokernag (KOK) during the 1990s and early 2000s. The post-2000 period generally shows improved data coverage across most stations, with only occasional missing months. Monthly mean T_max and T_min were calculated from the available daily observations, and seasonal and annual means were subsequently derived from the corresponding monthly values. Missing observations were not infilled, and all analyses were performed using the available observations only. Despite the presence of some data gaps, the overall temporal coverage is sufficient to support the long-term trend analyses presented in this study.

This study employs daily values of maximum temperature (T_max) and minimum temperature (T_min) recorded at each station, from which the mean temperature is calculated using the Eq. (1).

1

At IMD surface observatories, maximum and minimum temperature observations are recorded at fixed observation times. Both variables are reported in degrees Celsius (°C) to one decimal place of precision. The maximum thermometer contains mercury that rises as temperature increases and remains at the highest level reached due to a constriction in the tube, thereby recording the maximum temperature attained during the day. T_max is recorded and reported initially at 1200 UTC (1730 IST). At 0300 UTC (0830 IST) the next day, after the maximum temperature is again read and recorded, the thermometer is reset by shaking it vigorously to force the mercury back through the constriction into the bulb. The minimum thermometer, which contains a spirit-filled tube with a small dumbbell-shaped index marker, records the lowest temperature attained since the previous observation. T_min is recorded and reported at 0300 UTC (0830 IST). The minimum thermometer is reset at 0300 UTC (0830 IST) and at 1200 UTC (1730 IST) observation times. At each reset time, the thermometer is tilted bulb-upward until the index marker slides back to touch the top of the spirit column, ready to record the next minimum temperature. For climatological analysis and long-term trend studies, the daily maximum temperature and daily minimum temperature are both recorded and reported at 0300 UTC (0830 IST), ensuring a consistent 24-h observation period for both temperature extremes. This standardised procedure ensures consistent derivation of daily temperature extrema across all stations and throughout the entire observational period.

Following observations at the surface observatory, all meteorological data including T_max and T_min are recorded in the Monthly Meteorological Register (MMR), the standardised IMD record-keeping format. At the end of each month, the completed Monthly Meteorological Register (MMR) is forwarded to the Meteorological Centre (MC) or Regional Meteorological Centre (RMC) under whose jurisdiction each station falls. At the regional level, the data are scrutinised and subjected to quality checks and processing before being forwarded to the National Data Centre (NDC) at IMD Pune. This multi-stage transmission ensures that data are verified at the point of origin (surface observatory), checked for consistency and accuracy at the regional level, and finally archived at the national level.

Once validated and archived, the data are made available to users and researchers through the IMD Data Service Portal in standardised formats. All station data employed in this study were obtained directly from this official source, ensuring consistency, standardisation, and adherence to IMD protocols for data collection and storage. Daily records were provided in the standardised format of Surface Day Summary for the complete period spanning 1 January 1980 through 31 December 2024, encompassing 45 years of observations. The data used represent ground-based, operational measurements from IMD’s network of manual surface observatories. Additionally, IMD reports no changes in instrumentation or station relocation (Appendix 1, Supplementary Information).

Table S1 compares station elevations with the corresponding ERA5 grid-cell elevations. Elevation differences range from − 35 m to + 729 m, with the largest discrepancies occurring at several mountain stations where the complex Himalayan topography is convolved within the ERA5 grid²⁴. These differences can be due to grid-cell mean surface conditions in ERA-5 rather than point-scale station characteristics. Consequently, the ERA5-derived predictors used in the attribution analysis should be interpreted as representative of local-to-regional atmospheric and surface conditions rather than exact station-scale values.

Data homogenisation

The temporal homogeneity of the station temperature records is assessed using two widely applied change-point detection methods: the Pettitt test and the Standard Normal Homogeneity Test (SNHT). These tests were applied to the T_max, T_min, and T_mean series for the period 1980–2024 to identify potential discontinuities that could arise from non-climatic factors such as station relocation, instrumentation changes, or observational practice modifications. The results (Table S2) indicate that most station records did not show statistically significant breakpoints according to either test, suggesting generally homogeneous temperature series suitable for long-term trend analysis. Significant breakpoints were detected only in the T_min records of Batote and Bhaderwah, where both Pettitt and SNHT identified potential change points around 1994–1998. However, no corresponding station relocations, instrumentation changes, or other metadata-supported inhomogeneities were reported by the India Meteorological Department (IMD). Consequently, all station records were retained for the subsequent trend and elevation-dependent warming analyses.

Trend estimation

Long-term trends in annual and seasonal temperature time series (T_mean, T_max, and T_min) for the period 1980–2024 are quantified using Sen’s slope estimator and evaluated for statistical significance using the non-parametric Mann–Kendall (MK) trend test. These methods are widely used in climatological studies because they are robust against non-normal data distributions and less sensitive to outliers compared to ordinary least squares regression.

The Mann–Kendall test was applied to determine the presence and significance of monotonic trends in the temperature records. The null hypothesis (H₀) assumes that the data are randomly distributed with no monotonic trend, while the alternative hypothesis (H₁) assumes the existence of a significant increasing or decreasing trend over time. Statistical significance was evaluated at the 95% confidence level (p < 0.05).

The magnitude of the temperature trend was estimated using Sen’s slope estimator, which calculates the median slope between all possible pairs of observations within the time series. The Sen’s slope (Q) is expressed as Eq. (2):

2

where X_j and X_k represent temperature values at times j and k, respectively, with j > k. The final Sen’s slope estimate is obtained as the median of all individual slopes and is expressed in units of °C per year. To maintain consistency with standard climatological reporting, the resulting values were multiplied by 10 and reported as °C/decade.

Region of study

This study addresses a climatically and topographically complex area of the Western Himalaya, characterised by a significant elevation gradient ranging from around 1 km to over 6 km above mean sea level²⁵. The steep elevation gradient makes the region ideal to examine the impact of EDW¹⁹. We utilise surface temperature data from the India Meteorological Department (IMD) network, employing ten temperature stations, shown by magenta marks in Fig. 1, scattered throughout the study area. The elevation shown in Fig. 1 is obtained from the Earth Topography 1 arc-minute (ETOPO1) Global Relief Model developed by the National Oceanic and Atmospheric Administration (NOAA). ETOPO1 is a global Digital Elevation Model (DEM) that integrates terrestrial elevation and ocean bathymetry into a continuous topographic surface at a spatial resolution of 1 arc-minute, which is about 1.85 km at the equator²⁶. The importance of these stations is in their thorough sampling of the elevation gradient. The stations encompass a vertical range of 2.3 km, from Jammu (JAM) at an elevation of 367 m in the low-lying plains to Pahalgam (PAH) at 2740 m. This elevation disparity corresponds to a substantial temperature gradient, with mean temperatures declining from about 25 °C in Jammu to below 10 °C in Pahalgam. Several of these observational sites are observable as clusters at comparable altitudes. Srinagar (SRI, 1587 m) and Batote (BAT, 1587 m) are almost in the same elevation, whereas Gulmarg (GUL, 1600 m), Bhaderwah (BHA, 1613 m), Kupwara (KUP, 1615 m), and Banihal (BAN, 1630 m) constitute a compact high-altitude cluster ranging from 1600 to 1630 m. This distinctive arrangement facilitates a comprehensive analysis, allowing the differentiation of warming trends along the elevation gradient (e.g., JAM vs. PAH) from varying warming within the same altitudinal range, which is essential for isolating the altitude-dependent climate signal.

Elevation-dependent warming

The primary signal of EDW was quantified by analysing the linear relationship between station elevation and the observed long-term temperature trends. We used seasonal (Winter, DJF; Pre-monsoon (MAM); monsoon (JJAS); and post-monsoon (ON)) and annual (ANN) trends for mean, maximum, and minimum temperature (T_mean, T_max, T_min) for the 1980–2024 period as the dependent variables. Station elevation (in kilometers above mean sea level) is used as the independent variable. The strength and significance of EDW are determined using a simple Pearson’s correlation coefficient (R₁). A statistically significant, positive R₁ (e.g., for T_mean in winter) would indicate that temperature trends systematically increase with altitude.

Attribution using multiple linear regression (MLR)

We utilised a statistical framework based on partial correlation and Multiple Linear Regression (MLR) to analyse the physical drivers of the observed temperature changes, following previous studies. This methodology was designed to address an important statistical problem: mean temperature trends (T) at every station are a composite of the extensive, altitude-influenced EDW signal (E) and localised, site-specific physical factors, such as albedo, Surface short-wave (solar) radiation downwards (RSDS), Surface long-wave (thermal) radiation downwards (RLDS) and specific humidity (HUSS). A straightforward correlation between T and the drivers may be misleading, as the drivers themselves (e.g., albedo) are frequently highly associated with elevation.

Therefore, a three-step process was used to disentangle these effects:

1. Driver-Elevation Correlation (R₂): First, we assess the potential for confounding by calculating a Pearson’s correlation (R₂) between each driver trend (Albedo, RSDS, RLDS, and HUSS) and station elevation (E). A significant R₂ confirms that the driver is also elevation dependent. This step was included primarily as a diagnostic assessment to evaluate whether the observed driver–temperature relationships could partly arise from their shared dependence on elevation. Because the primary objective of the analysis was to isolate the residual local-scale variability after removing the influence of altitude, the subsequent interpretation focused on the elevation-detrended partial correlations and regression coefficients rather than on the raw R² relationships themselves. Table S3 shows the seasonal correlations between elevation and trends in the analysed climatic drivers, including surface albedo, downward shortwave radiation (RSDS), downward longwave radiation (RLDS), and specific humidity (HUSS).

2. Trend Estimation and Elevation detrending: To isolate the local physical signals, long-term annual and seasonal trends in temperature and driver variables were first estimated using Sen’s slope estimator applied to the corresponding station time series. For both the temperature trend (T) and each of the four driver trends (D), an Ordinary Least Squares (OLS) linear regression was then performed against station elevation (E). The residuals from these regressions (e.g., T_res = T - βE - c) represent the non-elevational component of the spatial variability in station-scale temperature and driver trends after removing the linear influence of altitude. Thus, the analysis is done on the spatial distribution of station trends (n = 10) rather than on the temporal evolution of the observations at individual stations. The resulting trend magnitudes are expressed in units of °C/decade.

3. Partial Correlation (R₃): We then calculated the Pearson’s correlation (R₃) between the elevation-adjusted residual temperature trend (T_res) and each driver trend (D_res). This residual correlation (R₃) quantifies the statistical association between the local-scale spatial variability in temperature trends and the corresponding variability in the drivers after removing their shared linear dependence on elevation.

Quantifying driver influence

Finally, to determine the relative importance of these local drivers, we constructed an MLR model (Eq. 3) for each season with the detrended temperature trend (T_res) was set as the dependent variable, and the four detrended and standardised driver trends (Albedo, RLDS, RSDS and HUSS) were used as the independent predictors:

3

The resulting standardised regression coefficients (β), quantify the contribution and relative influence of each physical driver in explaining the non-elevation-dependent (local) warming signal³. As the number of predictors is relatively large compared to the station sample size (n = 10), the MLR results are interpreted as exploratory statistical links rather than definitive causal factors.

Auxiliary data

Albedo, surface short-wave radiation downwards (RSDS), surface long-wave radiation downwards (RLDS), and specific humidity (HUSS) are obtained from the ERA5 reanalysis dataset produced by the European Centre for Medium-Range Weather Forecasts (ECMWF). ERA5 is generated using the Integrated Forecasting System (IFS Cy41r2) with 4-Dimensional Variational Data Assimilation (4D-VAR), providing consistent atmospheric and land-surface variables by combining model simulations with a wide range of in situ and satellite observations. The dataset has a horizontal resolution of approximately 0.25° × 0.25° with global coverage. Monthly mean values at one-month temporal resolution are used in this study. Albedo represents the fraction of incoming solar radiation reflected by the surface and plays a key role in regulating surface energy balance. In this study, the albedo predictor corresponds to the long-term trend in ERA-5 albedo extracted at each station location and therefore represents spatial variability in albedo trends rather than absolute albedo or direct measurements of snow-cover extent. RSDS describes the downward flux of incoming solar radiation at the Earth’s surface, whereas RLDS represents the downward atmospheric thermal infrared radiation reaching the surface. Specific humidity (HUSS) denotes the mass of water vapor per unit mass of moist air and is used to characterize atmospheric moisture availability. The high spatial consistency and temporal continuity of ERA5 make it suitable for climate, hydrological, and land–atmosphere interaction studies. A detailed discussion of the data is provided in Hersbach et al.²⁷. The auxiliary data from ERA5 corresponding to each IMD station is extracted from the nearest grid where the station is located. All analyses are performed using Python version 3.13²⁸.

Results

Temporal evolution and trends of T_mean, T_max, and T_min

Figure S2 shows the monthly time series of T_max and T_min across the stations for each year from 1980 to 2024. The decadal trends for maximum T_max show considerable regional variability. Daytime warming, which is statistically significant (p < 0.05), occurs mainly at specific mid-elevation valley regions, namely Srinagar (+ 0.346 °C/decade), Kupwara (+ 0.346 °C/decade), and Kokernag (+ 0.379 °C/decade). Nonetheless, this warming is not consistent over the elevation range; other mid-elevation sites such as Bhaderwah (1613 m) have minor (non-significant) cooling trends. Furthermore, the lowest station (Jammu, 367 m) in the network shows no notable daytime warming. This pattern indicates that T_max trends are influenced by highly localised causes, potentially surpassing a straightforward altitudinal lapse^29,30. Potential factors include microclimatic shifts in daytime cloud cover or radiative forcing due to fluctuating aerosol concentrations, wherein solar dimming from human-induced haze at lower-elevation locations such as Jammu could limit T_max increases, whereas cleaner, higher-altitude sites experience distinct radiative balances^31,32. Also, recent studies have shown that vegetation patchiness and associated soil moisture redistribution can substantially alter hydrological connectivity, near-surface energy partitioning, and local microclimatic conditions, thereby influencing temperature variability across heterogeneous landscapes³³.

The T_min trends indicate a significant and asymmetric warming signal during the night, which is also markedly localised. The most pronounced and statistically significant warming (p < 0.05) is recorded at Bhaderwah (+ 0.653 °C/decade) and Batote (+ 0.423 °C/decade), with notable warming being detected at the highest station, Pahalgam (+ 0.322 °C/decade). Conversely, some stations such as Jammu, Kupwara, and Kokernag exhibit minimal T_min trends. The notable heterogeneity, especially the significant warming in Bhaderwah and Batote, strongly suggests local rather than regional influences. The significant nighttime warming, particularly in conjunction with a negative T_max trend (as observed at Bhaderwah), leads to a significant reduction in the diurnal temperature range (DTR). This phenomena is usually linked with factors that impede outgoing longwave radiation, such heightened atmospheric moisture, nocturnal cloud cover, or specific land-use alterations, such as urban heat island impacts^34,35.

Figure 2 and Table S4 shows the highest T_max and T_min across stations for each year. We observe a considerable lack of statistically significant warming along the entire elevation gradient. Although many mid-elevation stations show positive trends (e.g., Kupwara, + 0.12 °C/decade; Kokernag, + 0.21 °C/decade) and others, such as Bhaderwah, show a cooling trend (− 0.22 °C/decade), neither of these trends show statistical significance (p > 0.05) at the 95% confidence level. The trends in annual T_min show a strong and statistically significant (p < 0.05) nighttime warming in majority of mid-to-high elevation stations, especially at Bhaderwah (+ 0.78 °C/decade), Banihal (+ 0.59 °C/decade), Kokernag (+ 0.47 °C/decade), and Batote (+ 0.35 °C/decade). The highest-altitude area, Pahalgam (2740 m), also shows notable nocturnal warming (+ 0.58 °C/decade). Nevertheless, a notable exception is the lowest-elevation station, Jammu (367 m), which shows a statistically significant cooling trend of − 0.45 °C/decade^36,37, whereas the unusual cooling in Jammu may be related to local land-use changes, such as the growth of irrigated agriculture³⁸. However, land-use and irrigation changes were not directly evaluated here, and therefore this explanation remains notional. Future studies incorporating land-use, irrigation, and urbanisation datasets are required to assess the relative importance of these factors.

Fig. 2.

Time series of highest maximum (T_max; left panel) and minimum (T_min; right panel) temperatures recorded at stations in the Western Himalayan region. Numbers in parentheses following each station name indicate elevation above mean sea level. Stations are arranged in ascending order of elevation, with Jammu representing the lowest and Pahalgam the highest. Numerical values in red (blue) denote decadal trends in maximum (minimum) temperature, and p values indicate statistical significance at the 95% confidence level.

Mean and trends of annual and seasonal surface temperature during 1980–2024

Figure S3 shows the mean of T_mean, T_min and T_max from each station during the period 1980–2024. The T_mean shows a distinct and persistent negative correlation with altitude. For instance, the lowest station, Jammu (JAM, 367 m), has the highest temperature, with an annual mean of roughly 21 °C. In contrast, the highest station, Pahalgam (PAH, 2740 m), shows the coldest climate, with an annual mean temperature of roughly 9 °C. The annual cycle appears at all stations, with the highest temperatures consistently seen during the monsoon (JJAS) and the lowest during winter (DJF). The seasonal amplitude, defined as the difference between JJAS and DJF, appears greatest at the lowest station, Jammu (about 28 °C compared to 14 °C), and diminishes progressively at higher elevations; for example, Gulmarg (GUL) shows a significantly reduced seasonal range within 2–5 °C.

T_max also follows a similar pattern when compared to T_mean, decreasing consistently with increasing elevation. Jammu (367 m) records the highest annual T_max (about 28 °C), exceeding 30 °C during the monsoon and pre-monsoon seasons. The highest-elevation stations, Gulmarg (1600 m) and Pahalgam (2740 m), have the lowest annual maximum temperatures, about 15–16 °C, respectively. At these higher altitude regions, the summer maximum temperature rarely exceeds above 20 °C. The monsoon season regularly shows higher T_max values at all stations. Nevertheless, the T_min most effectively shows the influence of altitude on climate. The lowest station, Jammu (367 m), shows an annual minimum temperature of about 18 °C, with its coldest winter temperatures around 9 °C, whereas the high-elevation stations experience freezing conditions. The annual minimum temperature for Gulmarg (1600 m) and Pahalgam (2740 m) is about 2 °C and 3 °C, respectively. In winter, T_min at these stations decreases considerably below freezing, averaging around − 4 to − 3 °C. This shows that the primary factor contributing to the low annual mean temperatures at high-altitude stations is the extreme cold of winter nights. The most significant diurnal temperature range is recorded at the lowest station (about 10–15 °C), Jammu, however the range is considerably limited at higher elevation locations (about 2–5 °C).

Figure 3 shows the trends of T_mean, T_min and T_max from each station during the period 1980–2024. The trends in T_mean are mainly positive throughout most mid-to-high elevation regions, however, show substantial variability. The most significant annual T_mean warming occurs at the mid-elevation station Bhaderwah (BHA), indicating a trend of around + 0.3 °C/decade. This yearly trend at BHA is driven by notably intense warming in pre-monsoon (about + 0.5 °C each decade) and winter (about + 0.4 °C/decade). Conversely, the low-elevation Jammu station shows an insignificant annual cooling in T_mean trend (about − 0.15 °C/decace), characterised by a notable winter cooling (about − 0.3 °C/decade) and an insignificant warming (0.3 °C/decade) in the pre-monsoon season. This pattern indicates that average warming varies with altitude and is likely influenced by asymmetric nighttime warming, which is closely associated with increases in surface longwave radiation^39,40.

Fig. 3.

Temperature trends at meteorological stations across the Jammu and Kashmir region during 1980–2024. The whiskers indicate the standard error of the trend, and the cyan star in the trend value denotes statistical significance at the 95% confidence interval. The stations are arranged in ascending order of elevation, with Jammu representing the lowest and Pahalgam the highest altitude above mean sea level. Here ANN is annual, DJF is December-January-February, MAM is March-April-May, JJAS is June-July-August-September and ON is October-November.

The trends in T_max show considerable variability and are often not statistically significant, with annual trends concentrated around 0–0.4 °C/decade at most locations. The low-elevation Jammu (JAM) and mid-elevation Bhaderwah (BHA) stations have marginal, statistically insignificant annual cooling trends (about − 0.1 to − 0.2 °C/decade). A significant positive T_max trend during pre-monsoon is seen across numerous stations, including Srinagar (SRI), Kokernag (KOK), and Pahalgam (PAH), which is about + 0.3 °C/decade. This seasonal trend strongly indicates snow-albedo feedback; reduced spring snow cover would lower surface reflectivity, increasing the absorption of incoming shortwave radiation and intensifying daytime warming^41,42. The lower Tmax trend at low-elevation regions such as Jammu may be associated with aerosol loading (solar dimming), which can reduce incoming shortwave radiation. However, aerosol variability was not explicitly analysed in this study, and this interpretation should therefore be regarded as a plausible hypothesis rather than a demonstrated mechanism³². The trends in T_min show a strong and significant disparate warming during nighttime across the region. The warming is most noticeable at the mid-elevation station Bhaderwah, showing a significant yearly trend of roughly + 0.7 °C/decade, with seasonal maxima in pre-monsoon and monsoon, which is about + 0.8 °C/decade. Several mid-to-high elevation stations, such as Batote, Banihal, and Pahalgam, show significant annual and seasonal nighttime warming, with trends varying from + 0.2 to + 0.4 °C/decade. This frequent nocturnal warming is consistent with enhanced longwave radiative trapping associated with increased atmospheric moisture, although the underlying mechanisms cannot be conclusively established from the present analysis. The added moisture retains more outgoing longwave radiation, hindering surface cooling at night and hence elevating T_min⁴³.

Discussion

Elevation dependent warming

T_mean shows a positive EDW relationship, albeit its statistical significance varies significantly by season, as shown in Table 1. The annual T_mean trend shows a positive but statistically significant EDW of 0.18 °C/km/decade (p = 0.01). The EDW signal shows strong seasonal variability, being most prominent and statistically significant in winter, where it shows a strong positive EDW of 0.35 °C/km/decade (p = 0.001) and a high correlation (R₁ = 0.84). A significant secondary EDW trend is also detected during the monsoon season (0.11 °C/km/decade, p = 0.05). Conversely, the pre-monsoon season shows no elevation dependency in mean warming (p = 0.99), suggesting that the determinants of EDW are considerably season specific. However, the post-monsoon season shows an EDW signal, which is moderate, about 0.15 °C/km/decade and is statistically significant at 90% CI (p < 0.1).

Table 1.

Elevation dependent warming (EDW) in °C/km/dec over the Jammu and Kashmir region. Here, R₁ is the Pearson’s correlation coefficient (correlated of Temperature trends to Elevation), p is the p value for statistical significance, ANN is annual, DJF is December-January-February, MAM is March-April-May, JJAS is June-July-August-September and ON is October-November. The bolded EDW values are significant at 95 CI (p < 0.05).

°C/km/dec	Tmean			Tmax			Tmin
	EDW	R 1	p	EDW	R 1	p	EDW	R 1	p
ANN	0.18	0.77	0.01	0.28	0.75	0.01	0.12	0.34	0.34
DJF	0.35	0.84	0.00	0.43	0.83	0.00	0.24	0.47	0.17
MAM	0.00	0.00	0.99	0.06	0.39	0.27	-0.01	-0.03	0.93
JJAS	0.11	0.64	0.05	0.10	0.32	0.37	0.12	0.33	0.35
ON	0.15	0.57	0.09	0.23	0.54	0.10	0.06	0.13	0.72

T_max shows a strong EDW signal, indicating a statistically significant trend of 0.43 °C/km/decade (p = 0.004) in winter. This suggests that winter days are warming significantly more rapidly at higher altitudes, which is usually aligned with strong snow-albedo feedback, whereas diminished snow cover reduces surface reflectivity and increases the absorption of shortwave radiation⁴⁴. The strong seasonal signal is additionally corroborated by a marginally significant trend in the post-monsoon season (0.23 °C/km/decade, p = 0.10). Nonetheless, this elevation dependence in T_max entirely dissipates during the pre-monsoon (p = 0.27) and monsoon seasons (p = 0.37), signifying the dominant role of the winter-time albedo mechanism. In addition to the winter, the annual T_max also shows an EDW signal, where the observed trend of about 0.28 °C/km/decade is statistically significant. Nonetheless, T_min shows a total lack of a statistically significant linear EDW in any season. The annual EDW for T_min is weak (0.12 °C/km/decade, p = 0.34), with higher insignificant p-values throughout all seasons, including winter (p = 0.17) and pre-monsoon (p = 0.93). The considerable nighttime warming found at specific stations, such as Bhaderwah in Fig. 3, indicates a localised phenomenon rather than a systematic process that scales linearly with elevation throughout the entire region, including site-specific increases in humidity (which enhance longwave radiation retention) or localised changes in land use, rather than by a broad, altitude-driven mechanism⁴⁵. In addition, the application of interpretable machine learning frameworks such as XGBoost coupled with SHAP analysis has proven effective in disentangling complex nonlinear interactions among vegetation structure, soil properties, and hydrological processes, providing valuable methodological insights for climate attribution studies⁴⁶.

To assess the robustness of EDW relationships presented in Table 1, an additional leave-one-out sensitivity analysis is performed. In this approach, the elevation–temperature regression is recalculated iteratively after sequentially removing one station at a time from the analysis. For each iteration, the EDW slope, correlation coefficient (r), and statistical significance (p) are recomputed for annual and seasonal T_mean, T_max, and T_min trends. This helps to evaluate whether the identified EDW signals are strongly influenced by any individual station, which is particularly important given the relatively limited station sample size (n = 10). The resulting range and variability of the recalculated EDW slopes are presented in Tables S5 and S6.

The sensitivity analysis demonstrates that the principal EDW signals identified in the results remain generally robust, particularly for annual and winter T_max and T_mean. For example, the annual T_max EDW slope remains consistently positive across all leave-one-out iterations, varying only from 0.196 to 0.345 °C/km/decade with a relatively small standard deviation (0.035), while the corresponding correlation coefficients remain strongly positive (R₁ = 0.47–0.84). Similarly, the DJF T_max EDW relationship remains stable, with recalculated EDW slopes ranging from 0.244 to 0.556 °C/km/decade and consistently positive correlations (R₁ = 0.53–0.87). The annual and DJF T_mean relationships are also stable, with low-to-moderate variability in the recalculated EDW slopes (standard deviations of 0.040 and 0.062, respectively). In contrast, weaker or statistically insignificant relationships, especially those involving T_min and non-winter seasons, show substantially higher variability and occasional sign reversals when individual stations are removed. This is particularly evident for ON T_min, where the recalculated EDW slopes range from − 0.195 to 0.147 °C/km/decade with the largest observed maximum slope change of about 0.25 °C/km/decade. Similarly, pre-monsoon T_min fluctuates between negative and weakly positive EDW values (− 0.148 to 0.030 °C/km/decade), indicating poor spatial robustness. These results suggest that the strongest annual and winter EDW signals represent comparatively stable regional-scale patterns, whereas the weaker T_min and transitional-season relationships are more sensitive to local station variability and should therefore be interpreted more cautiously.

The drivers of temperature variability in Jammu and Kashmir region of India

To deconstruct the complex drivers of the observed temperature trends, a Multiple Linear Regression was performed. A critical step in this methodology was to first detrend all variables (temperature and drivers) by removing the linear influence of elevation. This approach is essential as it isolates the large-scale, altitude-driven trend (the EDW signal, Table 1) from the more localised, site-specific physical processes. The partial correlations (R₃) and regression coefficients (β) therefore reveal the relative importance of these local drivers in explaining the temperature variance that is not explained by elevation alone. Figure 4 shows the Partial correlations (R₃) between detrended temperature and drivers after removing elevation effects. Table S7 provides the detailed statistical results for the MLR attribution model, which is designed to explain the local temperature trends after removing the linear influence of elevation. However, the regression analysis should be interpreted cautiously because the number of predictors is relatively large compared to the available station sample size (n = 10). Consequently, the regression coefficients may be sensitive to predictor covariance and local outliers. The MLR framework is therefore intended primarily as an exploratory statistical tool to identify plausible links between temperature variability and atmospheric or surface conditions, rather than as a definitive proof of causal physical mechanisms.

Fig. 4.

(Top) Partial correlations (R₃) between detrended temperature and drivers after removing elevation effects; magenta marks no EDW and * indicates p < 0.05. (Bottom) Standardized regression coefficients (β) showing the relative influence of albedo, RSDS, RLDS, and HUSS on temperature trends.

The winter EDW observed in T_max (Fig. 3) appears to be statistically associated with a combination of atmospheric and surface-related factors. During the winter season (DJF), Albedo shows the strongest positive partial correlation with temperature trends (R₃ = 0.47), followed by RLDS (R₃ = 0.31) and weak correlations for RSDS (R₃ = − 0.31) and HUSS (R₃ = 0.02). These results suggest that surface-related processes, particularly albedo variability, may play a more important role than atmospheric moisture in explaining the residual winter warming signal after removing the influence of elevation. Although snow–albedo feedback is a plausible physical mechanism in mountain environments, the present analysis only demonstrates a statistical association between temperature trends and albedo trends after removing the influence of elevation. Therefore, the results should be interpreted as evidence of albedo-related surface processes rather than as a direct and immediate effect of snow–albedo feedback³.

Simultaneously, the regression coefficients reveal that HUSS exhibits the strongest negative regression coefficient during DJF (β = − 0.88), while RLDS shows a positive coefficient (β = 0.75), followed by Albedo (β = 0.45) and a weak negative coefficient for RSDS (β = − 0.15). The positive RLDS coefficient suggests that enhanced downward longwave radiation may contribute to winter warming variability through increased radiative trapping under cloudy or moisture-rich atmospheric conditions^47,48. However, the contrasting signs between the partial correlations and regression coefficients indicate that interactions among atmospheric predictors remain important and may reflect covariance between humidity and radiative processes.

Importantly, the absence of statistically significant EDW in T_min does not contradict the presence of strong T_min warming at several stations. EDW describes the dependence of warming magnitude on elevation rather than the absolute warming trend itself. In this study, T_min warming appears spatially heterogeneous and is likely influenced by local atmospheric and radiative conditions that do not vary systematically with altitude. By contrast, the winter T_max trends exhibit a clearer elevational structure, resulting in a statistically significant EDW signal. The subsequent MLR analysis therefore focuses specifically on the residual, non-elevational component of the temperature variability after removing the linear influence of altitude. In this framework, the regression coefficients and partial correlations do not explain the EDW signal itself, but rather identify which local atmospheric and surface-related factors are statistically associated with the remaining station-to-station variability in warming.

Longwave radiative processes and atmospheric moisture nevertheless appear to be statistically linked with annual and non-winter temperature variability. Specific humidity (HUSS), an indicator of atmospheric water vapour, shows a partial correlation in the annual model of R₃ = 0.53 with a regression coefficient of 1.01. The consistent positive connection between HUSS and annual temperature variability suggests that atmospheric moisture may contribute to local warming through enhanced longwave radiative trapping. Nevertheless, the relatively small station sample size means that these relationships should still be interpreted cautiously as exploratory statistical associations. Localised increases in water vapour intensify the retention of outgoing longwave radiation, inhibiting nighttime cooling⁴⁹. Additionally, the downward longwave radiation (RLDS) shows a near zero annual partial correlation (R₃ = 0.00) and a negative regression coefficient of − 0.95, indicating that the influence of longwave radiation on local warming variability may depend on interactions with atmospheric moisture and cloud conditions⁵⁰.

Thus, the results suggest that, after removing the influence of elevation, local temperature variability across Jammu and Kashmir is associated with a combination of surface albedo variability, atmospheric moisture, and radiative fluxes, with their relative importance varying seasonally. However, the contrasting signs between several partial correlations and regression coefficients indicate that interactions among predictors remain important, and the derived regression coefficients should therefore be interpreted as relative statistical contributions rather than purely independent physical effects.

Additionally, a major limitation of the study is that the available station network has limitations in terms of elevation coverage and spatial distribution, particularly at very high elevations (> 3000 m), where observational records remain sparse in the Himalayan region. Most stations used in this study are within the mid-elevation range (500–1500 m), with comparatively limited representation at higher altitudes. Consequently, the identified EDW relationships should be interpreted primarily as representative of the available observational elevation range within Jammu and Kashmir rather than as a comprehensive characterisation of EDW across the entire Himalayan cryospheric domain. Future studies incorporating denser high-altitude observations and longer cryospheric records would help further constrain the elevational structure of regional warming.

Another major limitation is the multicollinearity among the selected variables in the MLR model. Given the relatively small sample size (n = 10 stations), additional diagnostics are performed to assess model robustness. Correlation matrices and Variance Inflation Factors (VIFs) reveal substantial covariance among several predictors, particularly RSDS, RLDS, and HUSS in some seasons (Table S8). This predictor interdependence likely contributes to differences between partial correlations and regression coefficients. Consequently, the standardised coefficients are interpreted as exploratory indicators of statistical association rather than definitive measures of independent physical influence or causality. This is common in regions with very limited monitoring stations. Therefore, future studies using a denser station network and larger sample size could further evaluate these relationships using more advanced attribution approaches such as ridge regression, principal-component-based methods, or bootstrap-based uncertainty estimation.

Conclusions

The present study provides a thorough analysis of temperature variability in the Jammu and Kashmir region, showing that the observed warming over these regions is not driven by a uniform, altitude-dependent phenomenon, but also due to the interactions of a stronger, localised, non-altitude-sensitive signal. Initially, we have identified a statistically significant EDW, which is highly particular, primarily occurring in the winter season and highest in the maximum temperature with a significant trend of + 0.43 °C/km/decade. The attribution analysis indicates that winter daytime warming is statistically associated with albedo-related surface processes, although the underlying mechanisms cannot be directly resolved from the present analysis. The regional, altitude-dependent warming identified in Tmax therefore appears linked to spatial variability in albedo trends together with seasonal radiative and atmospheric conditions, highlighting the sensitivity of high-altitude environments to ongoing climate change. The significant warming evident in the station data, particularly the strong asymmetric T_min warming observed annually at mid-elevation stations such as Bhaderwah (+ 0.6 °C/decade), is not fully explained by the regional-scale EDW relationship. In addition, no statistically significant linear EDW was identified for T_min across any season. The MLR analysis indicates that this T_min warming is primarily associated with local-scale atmospheric and radiative processes after removing the linear influence of elevation. Annual HUSS exhibits a moderate positive partial correlation together with the largest positive annual regression coefficient (β = 1.01), suggesting that atmospheric moisture may contribute to local warming variability through enhanced longwave radiative trapping and reduced nighttime cooling. RLDS shows a weaker annual relationship, with a near-zero partial correlation and a moderate negative regression coefficient (β = − 0.95), indicating that its influence may depend on interactions with other atmospheric and radiative variables. Given the relatively small station sample size and the potential covariance among predictors, these relationships should be interpreted cautiously as exploratory statistical associations rather than definitive causal mechanisms. In contrast, the winter EDW identified in T_max appears to be more strongly associated with surface albedo variability together with seasonal radiative and atmospheric processes. The stronger winter albedo relationship is consistent with the possibility that snow-cover-related surface feedbacks contribute to enhanced high-elevation winter warming. These results therefore suggest that EDW in T_max and local T_min amplification likely arise from partially distinct physical mechanisms. Furthermore, warming-induced alterations in vegetation cover, soil moisture redistribution, and hydrological connectivity may also affect sediment transport, SOC dynamics, and agricultural water stress, thereby influencing regional hydrological sustainability in fragile mountain environments⁵¹. Therefore, distinguishing between these processes is important for improving assessments of future mountain water-resource variability and cryospheric sensitivity under continued regional warming.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

GSG: Methodology, software, validation, formal analysis, investigation, data curation, visualization, writing—original draft. JK: Conceptualization, methodology, formal analysis, investigation, resources, writing—original draft, writing—review & editing, visualization, supervision. VMPC: Data curation, formal analysis, writing and reviewing—original draft.

Funding

This study did not receive any funding.

Data availability

IMD data is available via https://dsp.imdpune.gov.in/. ERA-5 data is available via https://cds.climate.copernicus.eu/datasets/. Additionally, all the data used in this analysis will be made available on request to the corresponding author.

Declarations

Competing interests

The authors declare no competing interests.