Main

Himalayan glaciers have been losing mass consistently, with an acceleration in the last five decades, including at the iconic Mount Everest1,2,3. Whereas the increase in temperature over the last decades is recognized as a first-order control on melting4, how glaciers modulate the local climate remains unknown, also because of limited high-elevation stations5. This challenges both our understanding of ongoing changes and modelling skills to anticipate future changes6.

Here, we investigate the non-stationarity of the Himalayan climate (temperature, precipitation and wind) during the last three decades from the hourly to the seasonal scale. We use all existing daily observational climatic time series at high elevation and ERA5-Land reanalysis data. We take advantage of the longest time series of meteorological data at a glacierized elevation in Himalaya, the Pyramid Observatory Laboratory (on the southern slopes of Mount Everest, Nepal, 5,035 m above sea level (a.s.l.), off glacier, hereafter Pyramidoff) (Supplementary Fig. 1), which has continuously recorded hourly meteorological data since 1994. We also exploit the only five existing long-term climatic time series close to the main Himalayan ridge, located from 3,900 to ~4,500 m a.s.l. (Supplementary Table 1). We explore the physical processes behind the observed climate dynamics using three more stations along the southern slopes of Mount Everest, together with ozone data collected at Pyramidoff. Building on the agreement between ground observations and ERA5-Land reanalysis data, we extend our findings from Mount Everest to the entire Himalayan range, revealing an unknown picture of high mountain climate in the region.

Cooling trends of diurnal air temperature

In contrast to most regional and global records7, we find that the mean annual air temperature (Tmean) has been stationary at Pyramidoff during the last three decades (−0.002 ± 0.009 °C yr−1, P > 0.1, 1994−2020 period, Supplementary Table 2 and Supplementary Fig. 2, black line). This unexpected observation seems in contrast with the attribution of the accelerated glacier mass loss to increasing air temperature8. To reconcile this apparent discrepancy, we take advantage of the unique climatic dataset at Pyramidoff and analyse its diurnal temperature (Tmax) and nocturnal temperature (Tmin) separately, partitioning the year into two periods: the cold season from November to April and the warm season from May to October. We find that the Tmean trend has been brought to zero by a significant decrease in Tmax (−0.040 ± 0.020 °C yr−1, P < 0.01) during all months of the warm season. This decrease in Tmax started about 15 years ago (Fig. 1c,d, orange line), in contrast to Tmin, which has increased consistently, mainly during the cold season (+0.046 ± 0.019 °C yr−1, P < 0.01, Supplementary Fig. 3b, blue line). Tmax exerts a crucial control on both melting processes and katabatic wind generation, when Tmax is generally above +0 °C, while Tmin is mainly negative during the cold season and remains so despite the increase. Therefore, we focus on the non-stationarity of Tmax, which is the key control of glacier melt and its sensitivity to warming.

Fig. 1: Trend analysis for maximum 2 m air temperature in Himalaya during 1994–2020.
figure 1

a, Himalayan trend of ERA5-Land Tmax referred to the warm season (1994−2020). The map shows the location of the observational air temperature time series and the glacier mask used in ERA5-Land. Cold and warm colours represent decreasing and increasing trends, respectively, with associated Mann–Kendall (MK) test significance. The insert (i) provides a better picture of the ERA5-Land trends in the surrounding of the analysed ground stations. b, Elevation-dependence of Tmax SS for all cells in a (grey). The black points represent the cells located inside the glacier mask. c,e, Observational data (Pyramidoff) (c) and the correspondent pixel of ERA5-Land (e). The grids display the results of the MK test applied at the monthly scale and calculated from the beginning of the series to the given year. The colour bar represents the normalized Kendall’s tau coefficient µ(τ). The colour tones below −1.96 and above 1.96 are significant (α = 5%). On the right, the monthly Sen’s Slope (SS) and the significance levels for 1994–2020 (◦P < 0.1, *P < 0.05, ** P < 0.01, *** P < 0.001). d,f, Observational data (Pyramidoff) (d) and the correspondent pixel of ERA5-Land (f). The progressive µ(τ) (solid lines) and retrograde (dotted line) of the seqMK test (that is, calculated from the beginning or from the end, respectively, of the series to the given year) for the cold season (NDJFMA) (blue), the warm season (MJJASO) (orange) and for the entire year (black). For each year, below-zero lines indicate negative trends (calculated from 1994).

Source data

Full size image

We find negative Tmax trends at all the other existing stations close to the Himalayan glaciers: a similar trend exists at Phericheoff (4,260 m a.s.l.), in the same valley as Pyramidoff, which also excludes instrumental errors at single sites (Supplementary Fig. 4a); and negative Tmax trends occur at all the southernmost stations on the Tibetan Plateau9, mainly at the beginning of the warm season (May to July) (Supplementary Fig. 5, location in Fig. 1a). The stations with weaker declines in Tmax are farther from the glaciers than Pyramidoff. The consistency of the cooling trends across all available high-elevation stations lends confidence that this is a process typical of Himalayan glacierized regions beyond the Everest region.

Crucially, we find that the overall decreasing Tmax trend at Pyramidoff, particularly during the warm season, is correctly reproduced by ERA5-Land reanalysis (−0.026 ± 0.014 °C yr−1, P < 0.05, Fig. 1e,f), which reproduces the non-stationary patterns at the annual and monthly scale (Fig. 1c).

Regional cooling associated with the glacier mass

Given the correspondence between observations and reanalysis at Pyramidoff, we explore the warm season Tmax trend across the entire Himalaya (75° E to 91° E) based on ERA5-Land (Fig. 1a). The region includes glaciers that accumulate snow primarily during winter (predominantly in western Himalaya) and during the summer monsoon (eastern Himalaya) but are homogeneous from a morphometric point of view10. To describe the regional trends, we split the transect into four subregions (Lahaul-Spiti, Western, Central and Eastern Himalaya)3.

In agreement with the Pyramidoff record, a decrease in Tmax is evident in the entire region over the glacier surfaces (Fig. 1a). A similar pattern of decreasing trends is observed for Tmean (Supplementary Figs. 2, 3, 6 and 7). The decrease in Tmax is particularly strong in Lahaul-Spiti and in Central-Eastern Himalaya, characterized by the largest ice masses, and is less strong in Western Himalaya, where glaciers are less abundant8,11 (Fig. 1a).

Whereas an elevation-dependent warming (that is, stronger warming rate at higher elevations) has been suggested at the global level6, we find that glaciers are associated with a significant reduction of the local warming. In non-glacial conditions, we also observe a direct elevation-dependent warming, in particular above 4,000 m a.s.l. (up to +0.05 °C yr−1) (Fig. 1b). Cooling trends (up to −0.05 °C yr−1), however, are evident at all glacierized locations (black points in Fig. 1b) (P < 0.001 for the analysis of variance test on the means). At lower elevations, trends are positive, although less strong (mean +0.023 °C yr−1, P < 0.1) than those found at higher, non-glacierized elevations. The magnitude of the low-elevation trends in ERA5-Land corresponds well with the trend that we observe at 27 ground stations in Nepal below 2,000 m a.s.l. (mean +0.026 °C yr−1, P < 0.01, Supplementary Fig. 8).

Negative trends in Tmax are associated with glacier masses consistently across reanalysis and all available observations (Supplementary Fig. 5a–c). The smaller time window over which a trend is evident (beginning of the warm season) is probably due to a smaller influence of glaciers induced by the reduced extension of glacial masses and the larger minimum distance from the glacier fronts in the station surroundings (Supplementary Table 1 and Supplementary Fig. 9). On the contrary, we find no clear Tmax trends in the warm season (Supplementary Fig. 5d,e) for those stations without glaciers within a 20 km radius (Supplementary Table 1 and Supplementary Fig. 9).

Conceptual model of cooling and drying

The unexpected finding of cooling at high elevation close to glacier masses requires revisiting our understanding of glacier–atmosphere interactions. In the ablation season, the free atmosphere temperature (Tfa) on midlatitude glaciers is typically higher than the glacier surface temperature (Ts), which is close to 0 °C, since the excess energy at the ice surface is used for melting (Fig. 2a)12.

Fig. 2: Schematic diagrams explaining the air cooling observed in the surroundings of Himalayan glaciers.
figure 2

a,b, In both diagrams, an idealized case of wind interactions (vertical, katabatic and up-valley), that potentially dictate the along-flowline structure of on-glacier air temperature, is represented. a, Diagram for stationary conditions (subscript 0). Points I, II and III indicate locations of interest: point I0 represents the glacier zone, where Tg is more similar to Tfa and therefore less influenced by the glacier cooling effect; point II0 is the glacier part, where ideally the cooling is at its maximum; point III0 is the convergence zone between upward and downward winds, where the upshift of moist air masses induces the occurrence of precipitation. b, Diagram for non-stationary conditions (subscript 1). Currently, global warming increases the flux of sensible heat toward the glacier surface and thus enhances the cooling of near-glacier-surface air. Consequently, katabatic winds become more intense and capable of drawing further cold air masses from the higher elevations. As a result, the glacier cooling effect reaches downstream the periglacial environment (from II0 to II1), downshifting the convergence zone (from III0 to III1). Ts, temperature of glacier surface; Ta, temperature ambient; Tfa, temperature of free atmosphere; Tg, 2 m on-glacier temperature.

Full size image

Katabatic winds arise from adiabatic warming due to air subsidence and cooling of the near-surface air by sensible heat exchange with the glacier surface13. The interplay of these processes lowers the 2 m on-glacier temperature (Tg) (the so-called glacier cooling effect)14,15 giving rise to density gradients that drive the air down glacier.

Katabatic winds draw air masses from upper regions16 and we postulate an air convergence over upper parts of these glaciers, where strong air exchanges with higher atmospheric layers may occur17. This process generates a divergence of air masses along the northern and southern Himalayan valleys and causes further drying of the katabatic winds.

The variations of Tg along the glacier flowline are complex to model14,18. A growing body of studies, however, has shown stronger cooling effects at greater flow distances, until a maximum cooling point (from I0 to II0 in Fig. 2a)16. This point can be followed by an increase in temperature over the glacier tongues (III0 in Fig. 2a)18 due to up-valley afternoon winds19 counterbalancing the cooling effect of katabatic winds18. Recently, ref. 20 postulated that these moist, upward air masses counteract the cool, downwards katabatic winds, creating a local convergence. This process can force the warm and moist monsoonal air masses to uplift, inducing precipitation to occur near to, or below, the glacier fronts.

During non-stationary conditions, as in a global warming phase (Fig. 2b), higher free atmosphere temperature induces higher sensible heat exchange between the glacier surface and the near-surface air. On one hand, the additional heat is used to increase the melting rates, causing enhanced glacier mass losses. On the other hand, a further cooling of the surface air over glaciers generates more intense katabatic winds.

As a key consequence of the increased katabatic winds, the glacier cooling effect intrudes downstream into the periglacial environment, shifting the local convergence between up- and down-valley winds lower (Fig. 2b). Increased dry and cold winds push downwards the water vapour transported by the summer monsoon reducing moisture supply for precipitation over the upper parts of glaciers and thus reducing glacier accumulation.

Observations supporting the conceptual model

Measurements at the off-glacier stations downstream of Pyramidoff, Phericheoff (4,260 m a.s.l.) and Namcheoff (3,570 m a.s.l.) and at the on-glacier Changri-Nup (Changrion, 5,700 m a.s.l.) (Supplementary Table 3 and Supplementary Fig. 1; Methods) provide the observational evidence for our model (Fig. 2).

During the warm season, the mean 2 m air temperature recorded on glaciers is above 0 °C from 9:00 to 13:00 (local time) at Changrion (Fig. 3a), a condition favourable for the exchanging of sensible heat, which in turn allows the development of katabatic winds. As a result, from the early morning (9:00) the katabatic winds flow downward reaching the maximum speed during the first hours of the afternoon and decreasing in the evening (Fig. 3b), when the upward winds prevail19.

Fig. 3: Observations during the warm season on the south slopes of Mount Everest.
figure 3

a, Hourly mean of the 2 m air temperature at Changrion (thick line, mean; shadow area, standard deviation of the warm season). b, Wind rose and downward wind speed among hours of northerly (90°–270°) flow at Changrion (thick line, hourly mean; shadow area, standard deviation of the warm season). c, Boxplots of ozone concentration (n = 8,088, n = 646, n = 390) (i) and air temperature (ii) related to three classes of hourly wind speed at Changrion defined on the basis of the following criteria: wind direction between 90° and 270° (southward) and wind speed >1 m s−1. When one of these criteria was not satisfied, we included the observations in the ‘no event’ class (n = 7,578, n = 656, n = 395), the centre lines and box limits represent median values, 25th to 75th percentiles, respectively. d, Hourly lapse rates between meteorological stations located along the elevation gradient during 2011–2015 (map, Supplementary Fig. 1). A decreasing lapse rate denotes warming at higher elevation or cooling at lower elevation and vice versa. Note that the warming during the early morning at the station at higher elevation (Changrion/Pyramidoff at 9:00) is due to the different timing of exposure to solar radiation22. e, Trend analysis (1994–2020) of hourly downward wind speed (hollow) and air temperature (solid) at Pyramidoff. f, Hourly mean precipitation at the ground stations (ref. 20, revised).

Source data

Full size image

The exchange of air masses between the upper atmospheric layers and the glacier surface can be inferred from ozone (O3) data17,21. We find significantly higher ozone concentrations in correspondence to stronger downward winds at Pyramidoff (Fig. 3c1). When katabatic winds are stronger, temperatures are lower at Pyramidoff (Fig. 3c2), suggesting that air masses are drawn from the surrounding high-elevation reaches of the glaciers. This suggests that, during daytime, stronger katabatic winds are capable of transporting downward O3-rich air masses from high elevations to the glacier surfaces and the proglacial domain (Pyramidoff).

The diurnal down-glacier lowering of temperature is evident when analysing how the relationship between temperature and elevation varies during the day (Fig. 3d). We find midday lower temperature gradients until 4,260 m a.s.l. (Changrion/Pyramidoff and Pyramidoff/Phericheoff show lower temperature gradients between 13:00 and 16:00 and 10:00 and 16:00, respectively), which do not occur downstream at 3,560 m a.s.l. (Phericheoff/Namcheoff shows higher temperature at midday). Note that the higher temperature during the early morning at the station at higher elevation (Changrion/Pyramidoff at 9:00) is caused by the different timing of exposure to solar radiation22. An increase of katabatic winds, explaining the observed downstream cooling trend, is evident in the period of record. At Pyramidoff, wind speed from the early morning until the first hours of the afternoon increased during 1994–2020, with statistically significant increases from 9:00 to 10:00 (Fig. 3e). As a consequence, air temperature decreased mainly during the warmest hours of the day (that is, Tmax), with statistically significant decreases from 12:00 to 17:00.

Evidence of how the glacier cooling effect influences the precipitation amount at high elevation is shown in Fig. 3f. We observe a higher daytime (9:00–17:00) precipitation amount at Pyramidoff, than at the stations at lower elevations. The same elevation pattern was reported for Mount Everest and Langtang Valley20 and for Marsyandi Valley23. This is due to the local convergence between (cool, katabatic) down-valley and (warm, monsoonal) up-valley winds described in Fig. 2a.

Evidence of this process is provided by the substantial precipitation decrease (~50% with respect to the 1990s) at Pyramidoff for all months of the warm season (−8.9 ± 5.0 mm yr−1, P < 0.001, Supplementary Fig. 10a,b), during which 90% of the annual precipitation occurs24. The same decreasing trend was recorded at Phericheoff (Supplementary Fig. 4b), allowing one to exclude instrumental errors. ERA5-Land is able to reproduce the observed decreasing trends for some summer months (Supplementary Fig. 10c; Methods). Further evidence for the drying in the last decades stem from the evolution of high-elevated lakes; it was previously shown that possible variations in the lakes’ evaporation have a decidedly lower relative importance than that of precipitation25. Indeed, lakes fed only by precipitation show decreasing trends in area and volume from the mid-1990s both in the Mount Everest region25 and in the southern Tibetan Plateau26,27.

Across the entire Himalayan arc, the decreasing Tmax trends (Fig. 4a) correspond significantly with decreasing diurnal precipitation trends (P < 0.001). In Lahaul-Spiti, diurnal precipitation is decreasing (Fig. 4b), as it is in Central-Eastern Himalaya, although with a reduced magnitude and crucially only at high elevation (Fig. 4c). In contrast, in West Nepal, where glacier masses are smaller and their effect on the local climate reduced, the pattern is stationary or slightly increasing (Fig. 4d). It is worth noting that, in both Western and Eastern Himalaya (where decreasing trends are found), trends present a lower magnitude or opposite direction at lower elevations. This further confirms that the drying at high elevation is not caused by changes in large-scale atmospheric circulations.

Fig. 4: Himalayan trend of ERA5-Land diurnal (8:00–20:00) accumulated precipitation in the warm season (from May to October, during 1994–2020).
figure 4

a, The map reports the glacier mask used in ERA5-Land and the −0.01 °C yr−1 Tmax cooling trend isolines (Fig. 1). Cold and warm colours represent decreasing and increasing precipitation trends (% yr−1), respectively. The black line indicates their significance (P < 0.05). Three regions of interest have been selected to show the trend magnitudes for precipitation and temperature (from Fig. 1a) along the elevation gradient. b–d, The trend distribution is shown for Lahaul-Spiti (b), Western Himalaya (c) and Central-Eastern (d) subregions, respectively (blue points for precipitation trends, orange points for Tmax trends). The same trend, but considering both diurnal and nocturnal precipitation, is reported in Supplementary Fig. 21.

Full size image

Modelling the cooling and drying

In the recent decades of global warming (Supplementary Fig. 11), ERA5 shows a positive trend in daytime wind speed of katabatic winds (Supplementary Fig. 12b,c) flowing north and south of the Himalayan chain close to the glacier masses. During warmer days, the glacier nodes of ERA5-Land show more negative air temperature gradients (Supplementary Fig. 13b), resulting in enhanced diurnal downslope density winds (Supplementary Fig. 12c), sustained by the steep orographic gradients of the Himalayan chain (Supplementary Fig. 14). The same causal link between increased air temperature and katabatic wind emerged from high-resolution atmospheric simulations (Supplementary Fig. 15b3). As a result, a decrease in Tmax is evident in the entire Himalayan region in correspondence to the glacier surfaces of ERA5-Land (Fig. 1a).

In stationary conditions, ERA5-Land shows drying below the glaciers in the afternoon hours concomitant to a negative meridional wind gradient (southerly flow) anomaly (Supplementary Fig. 13d). On the contrary, in non-stationary conditions, a key consequence of the cooling (Fig. 1a) induced by increased katabatic winds (Supplementary Fig. 12), is the decreasing of the diurnal precipitation trends across the entire Himalayan arc (Fig. 4a). This is suggested to be caused by a downshift of the local convergence between up- and down-valley winds, which causes drying conditions at glacier elevations. This hypothesis is supported by our high-resolution atmospheric simulations which show this downshift of the convergence front (Supplementary Fig. 15b3).

Implications for climate change impacts in the Himalayas

Our study has provided evidence for a hitherto missed process in glacier–climate interactions at high elevations. We have described a glacier effect on local climate in the Himalaya, associated with global warming, which has lowered the daytime temperature close to the glacier masses. Potentially, glacier effects on local temperature could occur in other mountain chains, where glaciers can develop katabatic winds according to local climate regime, glacier size, slope and debris cover16,28.

The proposed conceptual model reconciles the apparent discrepancy between the observed local cooling and the accelerated glacier mass loss in Himalaya. While atmospheric warming is increasing glacier ablation, the lowering of the near-surface air temperature over the glacier surface and consequent enhancement of katabatic winds has shifted the extent of daytime cooling toward the lower reaches of the glaciers and into the proglacial domain. Whereas the local cooling could have partially protected these low glacier reaches from warming, it has further lowered the elevation of the local wind convergence and consequently precipitation has diminished at high elevation, implying a further negative effect on glacier mass balance. This precipitation decrease was proposed by ref. 29, after observing the glacier changes in the Mount Everest region over the last decades, although its relative impact is currently unknown.

In nearby periglacial areas, the effect we describe will have key implications as the decrease in air temperature has affected the elevation band between the mean elevation of glaciers (~5,400 m a.s.l.) (ref. 2) and the periglacial environment even below 4,500 m a.s.l (Fig. 1b). In those elevation bands, areas of permafrost may exist30 and their reaction to warming may be moderated by local cooling and drying during the summer, which drives permafrost thaw and associated geomorphic effects and hazards31 through increasingly deep seasonal thawing. Long-term observations for testing this hypothesis, however, are virtually absent30 and our study provides a compelling motivation to collect more high-elevation, long-term data to prove the new findings and their broader impacts.

Increased cooling during the early warm season in periglacial areas has the potential to reduce snowmelt rates and lengthen the snow season. However, at Pyramidoff, snow cover is not affected by this local cooling since during the warm season the snow cover is almost absent, as Tmax is largely positive at this elevation (Supplementary Fig. 16).

This phenomenon we discuss could substantially influence mountain ecosystems. It is well known that mountain plant species have shifted upslope in response to global warming. Unexpectedly, however, ref. 32 observed stable vegetation line elevation, reduction in greening (productivity loss) and decreasing recruitment during the last three decades in three glacierized areas in the Himalayas, providing a surprising, independent validation of our findings of local cooling and drying.

The process we highlighted is potentially of global relevance and may occur on any glacier worldwide where conditions are similar. We limited our analysis to the Himalaya, also considering the scarcity of data in high-elevation areas across the globe but future research should look into its existence in other regions of the world and into the morphological (for example, debris cover, local topography, total glacier area and glacier energy balance) and climatic factors that control it. Some of those elements have a simplified representation in ERA5-Land because of ERA5-Land coarse spatial resolution (or are absent, for example, debris cover). It is promising, however, that the reanalyses are able to represent the first-order controls and main processes leading to cooling and the generation of katabatic winds. Future research should focus on establishing the factors determining the occurrence, magnitude and downward effects of katabatic processes, the cooling they induce and their ability to change glacier mass balance across climates and regions.

Methods

Region of investigation and climate

The ground meteorological network belongs to the Pyramid Network24, which is located on the southern side of Mount Everest along the Khumbu Valley, central Himalaya. The land-cover classification shows that almost one-third of the territory is characterized by glaciers and ice cover, while less than 10% of the area is forested. Glacial surfaces are distributed from 4,300 to above 8,000 m a.s.l. These glaciers are identified as the summer-accumulation type, which are fed mainly by summer monsoon precipitation33. The climate in the South Asia and Himalayan region has a strong annual cycle, with the South Asian monsoon as a phase of this annual cycle. Regarding precipitation, the 20 yr measurements at Pyramidoff show that 90% occurs from June to September, while the probability of snowfall during these months is very low (4%). The annual cumulative precipitation at this elevation is 446 mm, with a mean annual temperature of −2.5 °C. The relatively low annual precipitation at this elevation is due to the elevation-dependent depletion of the moisture content resulting in a severe reduction of precipitation at higher elevations24.

As we show in Fig. 3 (ref. 34), in agreement with other authors35, the nocturnal maximum precipitation peak occurs most strongly during the summer monsoon. Furthermore, ref. 35 noted an asymmetry in diurnal winds, with daytime up-valley winds strongly predominating over weak nocturnal down-valley flow, with the daytime asymmetry being strongest during the monsoon.

A previous study36 has also documented the enhancement of nocturnal katabatic winds in the warm season over the Himalaya in recent decades (1979–2014), attributed to nocturnal clearer skies, which radiatively cooled the slopes, enhancing downslope winds.

We checked the possible relationship between the Interdecadal Pacific Oscillation37 and Tmax trend at Pyramidoff during the warm season (1994–2020). We did not find any significant relationship, although both series were decreasing in the last decades, thus excluding any possible influence from large-scale variability on the Tmax trend observed at Pyramidoff.

High-elevation climatic data (Nepal)

The first station (Pyramidoff) was established near the Pyramid Observatory Laboratory (5,035 m a.s.l.) in 1994. The main meteorological parameters (temperature, precipitation, relative humidity and wind) are recorded continuously (on an hourly basis), representing the longest time series existing at the high Himalayan elevations. Extra Automatic Weather Stations (AWSs) were installed along the Khumbu Valley in the following years. Metadata and location map are provided in Supplementary Table 3 and Supplementary Fig. 1. Wind, air temperature and precipitation sensors used in this study were supplied by LSI-Lastem (https://www.lsi-lastem.com/). All instrument details can be downloaded from http://geonetwork.evk2cnr.org.

Namcheoff station (3,560 m a.s.l.) is located in a forest-dominated landscape, Phericheoff (4,260 m a.s.l.) is slightly above the treeline (alpine tundra), Pyramidoff (5,035 m a.s.l.) is located at the mean elevation of glacial fronts33. The Changrion is a station located on the Changri-Nup Glacier (5,700 m a.s.l.) and is used in this study to investigate the relationship between katabatic winds and the glacier cooling effect. The Changri-Nup is a debris-covered glacier with a southeastern aspect (total length of ~4 km and a total area of ~2.7 km2). The accumulation zone of the glacier is a cirque surrounded by peaks reaching elevations >6,500 m a.s.l. Most of the ablation zone is covered by debris interspersed with supraglacial ponds and ice cliffs. The debris-covered portion of the tongue has a length of 2.3 km and an average width of 0.7 km, with a terminus located at 5,240 m a.s.l. Meteorological parameters were continuously monitored from 2011 to 2015, using an AWS installed on the debris-free surface of the glacier at 5,700 m a.s.l. AWS data were measured at 60 s intervals and averaged over a 30 min time period.

Data processing

We performed the data processing and quality checks according to the international standards of the World Meteorological Organization. The precipitation at all stations was measured using conventional buckets. Therefore, we only analysed the data collected in the warm season. Data discontinuity (for different variables) often occurred due to logistical problems at these elevations.

Data gap filling for temperature and precipitation time series

Pyramidoff has suffered a percentage of missing daily values of ~10% over the last 27 years, for both temperature and precipitation. In this study, we apply the same filling method (mainly quantile mapping) used for missing data in ref. 24 but extending the time series to 2020. The uncertainty introduced by the filling process on the Sen’s slope (SS) was estimated through a Monte Carlo uncertainty analysis24.

Low-elevation climatic data (Nepal)

Meteorological data in Nepal are collected and managed by the Department of Hydrology and Meteorology, Government of Nepal. We used daily Tmax and precipitation data series (1994–2020) covering an elevation range between 0 and 2,000 m a.s.l. Among these stations, we selected only stations with few missing data (<10%). The selected stations are homogeneously distributed in the country (Supplementary Fig. 17). The comparison with ERA5-Land data has been made considering 27 Tmax and 157 precipitation stations (Supplementary Figs. 8 and 18, respectively).

High-elevation climatic data (Tibetan Plateau)

The Chinese station data are provided by the China Meteorological Administration. The three stations close to the glacier masses have been recently discussed in ref. 9 (Nielaer, Pulan and Shiquanhe). The remaining two stations (Dingri and Gaize), located farther from glaciers, are used for comparison. Supplementary Table 1 provides some technical details on the stations, percentage of glacier cover included in a 20 km radius and minimum distance from the glacier fronts to each station. Supplementary Fig. 9 shows the relationship (distance and glacier cover) between those stations and the glacier masses. Trends analysis is shown in Supplementary Fig. 5.

Ozone data at high elevation (Nepal)

The Nepal Climate Observatory-Pyramid (NCO-P, 27° 57′ N, 86° 48′ E) is located at 5,079 m a.s.l., a few metres from Pyramidoff. As shown by ref. 38, the air-mass circulation at this measurement site is strongly influenced both by the synoptic scale circulation and the local mountain wind regime. The interaction between these components leads to the onset and decay of the summer monsoon and winter seasons38. At NCO-P, ozone-rich air masses indicate transport from the upper atmosphere39. Surface O3 measurements have been continuously performed over 2006–2015 with a UV-photometric analyser (Thermo Scientific, Tei 49C). During a maintenance campaign in February 2007, the instrument was compared against a travelling standard by GAW/WCC at EMPA (World Calibration Centre for Surface Ozone, Carbon Monoxide and Methane at the Swiss Federal Laboratories for Materials Testing and Research). The combined standard uncertainty (1 min basis) was ascertained to be less than ±1.5 ppbv in the range 0–100 ppbv.

The three classes of hourly wind speed at Changrion station (warm season) described in Fig. 3c were defined on the basis of the following criteria: wind direction between 90° and 270° (southward) and wind speed >1 m s−1. When one of the two criteria was not satisfied, we included the observations in the ‘no event’ class. The significance of the differences among the three classes was evaluated using the Dunn’s Kruskal–Wallis test (post hoc test for non-parametric comparison)40.

ERA5-Land reanalysis data

At present, among the reanalysis products with a high spatial resolution, there are three main products: ERA5-Land41 and two dynamical downscaling products. ERA5-Land is a reanalysis product derived by running the land component of ERA5 at increased resolution, which is enhanced globally to 9 km (0.1°) compared to 31 km (0.25°) (ERA5) or 80 km (0.70°) (ERA-Interim), whereas the temporal resolution is hourly and the simulations cover from 1950 to the present as in ERA5 (ref. 41). High Asia Reanalysis v.2 is a 10 km resolution dynamical downscaling of ERA5 performed using the Weather Research and Forecasting (WRF) model at present only available from 200442. Another dynamical downscaling is a 6.7 km resolution product derived by Climate Forecast System Reanalysis data using the WRF model36 available until 2014. These last two high-resolution dynamical downscaling products might have potentially improved the spatial representation of mountain precipitation; however, currently, they do not cover the entire period considered in this study. Therefore, we considered T and P data from the state-of-the-art ERA5-Land gridded climate reanalysis.

ERA5-Land is driven by atmospheric forcing derived from ERA5 near-surface meteorology state and flux fields. The meteorological state fields are obtained from the lowest ERA5 model level (level 137) and include air temperature, specific humidity, wind speed and surface pressure. The 2 m air temperature in ERA5-Land is a diagnostic output similar to ERA5 forecasts; that is, it is calculated with 10 m air temperature from ERA5 and the heat flux from ERA5-Land according to the Monin–Obukhov similarity theory. The surface fluxes include downward shortwave and longwave radiation and liquid and solid total precipitation. These fields are interpolated from the ERA5 resolution of about 31 km to ERA5-Land resolution of about 9 km using a linear interpolation method based on a triangular mesh.

In this regard, the Himalayan trends of ERA5 and ERA5-Land for Tmax were explored here considering the same analysis period (warm season during 1994–2020). We found that the decreasing pattern diagnosed by the two products (Fig. 1a and Supplementary Fig. 19 for ERA5-Land and ERA5, respectively) is the same even if ERA5 is characterized by a lower spatial resolution.

Moreover, we compared Tmax observed at the Pyramidoff and the corresponding ERA5-Land data in terms of correlation (Supplementary Fig. 20a) and bias (Supplementary Fig. 20b). Generally, the uncertainty of this reanalysis product was found to be low for all months of the year. Particularly, the bias results are constant for all considered years. This means that the non-stationarity is not influenced by possible biases during the analysed period.

Finally, ERA5-Land cannot be tested extensively at high elevations because of the lack of ground observations, while this is possible below 2,000 m a.s.l. for the entire area of Nepal. Therefore, we compared ERA5-Land with 157 ground stations (1994–2020 period). The relevant precipitation trends show good agreement (Supplementary Fig. 18).

Treatment of glaciers

Particularly important in this study is the representation in ERA5-Land of glaciers. As reported in ref. 41, the model formulation (as in ERA5) does not have an independent treatment of glaciers. Grid points with glaciers are assigned with a constant snow mass of 10 m. A threshold of 50% of a grid box covered by ice is used, below which the snow depth keeps the value computed by the snow scheme of the land model. Values above the threshold assign a snow water equivalent value of 10 m. This condition is used to avoid grid points near glaciers with large unrealistic snow depth that result from the interpolation from ERA5 fields to ERA5-Land41.

Precipitation and temperature

Currently, ERA5 is the most accurate of the existing reanalysis products for precipitation detection43. Others44 reviewed some recent studies showing the capability of ERA5 (both for temperature and precipitation) in complex terrain such as High Mountain Asia, where in situ meteorological observations are sparse and unevenly distributed. Among these studies, the performance of ERA5 precipitation over High Mountain Asia was analysed by ref. 45 and ref. 46. They found that ERA5 succeeds in reproducing the interannual and decadal variabilities of precipitation and reflecting the spatiotemporal patterns of precipitation over the whole Tibetan Plateau. Considering the temperature trends in mountains, recent studies, devoted to the comparison of different reanalysis products, point out that ERA5 performs well in capturing the non-stationarity of temperature during last decades on the Tibetan Plateau47.

In this study, a similar comparison was carried out in Nepal at low elevations (<2,000 m a.s.l.) for Tmax and precipitation (Supplementary Figs. 8 and 15, respectively). All these results provide robust evidence about the capability of ERA5 to correctly describe the large-scale trends. Regarding the high Himalayan elevations, Pyramidoff data have never been assimilated into ERA5 and this has not occurred for other high-elevated stations as well, considering their general absence at these elevations. Thus, any possible problem related to changes in station density or location over time should not have affected the local climatic trends. Therefore, Pyramidoff data are used here for validating the capability of ERA5-Land to simulate the non-stationarity of precipitation and temperature at the local level.

Trends analysis: the sequential Mann–Kendall test

In this study, the Mann–Kendall test (MK) was applied at the monthly scale (after daily data aggregation) to analyse the non-stationarity of observational and reanalysis data. This test is widely adopted to assess significant trends in hydrometeorological time series48. This test is non-parametric, thus being less sensitive to extreme sample values and is independent from the hypothesis about the nature of the trend, whether linear or not. The MK test verifies the assumption of the stationarity of the investigated series by ensuring that the associated normalized Kendall’s tau-b coefficient48, μ(τ), is included within the confidence interval for a given significance level (for α = 5%, the μ(τ) is below −1.96 and above 1.96). We used the SS proposed by ref. 49 as a robust linear regression allowing the quantification of the potential trends revealed by the MK. The significance level is established for P < 0.05. We defined a slight significance for P < 0.10. The uncertainty associated with the SS (1994–2013) is estimated through a Monte Carlo uncertainty analysis. Further details on this uncertainty are reported in ref. 24. In the sequential form (seqMK) μ(τ) the test is applied forward starting from the oldest values (progressive trend) and backward starting from the most recent values (retrograde trend). The crossing period allows us to identify the approximate starting point of the trend.