Field research in High Mountain Asia (HMA), which contains the largest glacierized area outside of the polar regions (Brun et al., 2017; Pfeffer et al., 2014), is extremely difficult. The region's high elevations result in low barometric pressure (and hence relatively low oxygen availability: Matthews et al., 2020a), and extreme weather, challenging both personnel and equipment during the assembling of high-quality, long-term observational datasets (Salerno et al., 2015; Shea et al., 2015; Khadka et al., 2022; Matthews et al., 2020a). However, there are strong incentives to overcome these difficulties. First, the region is an important freshwater source for communities downstream (Immerzeel et al., 2010). Second, the extreme weather, combined with steep terrain, places the region at a relatively high risk of hydrometeorological hazards (Miner et al., 2020). Establishing meteorological observation networks both helps improve understanding of water resource availability under a changing climate and supports the reduction of hydrometeorological risks.

Within HMA, the Khumbu region has attracted significant cryospheric research interest (Salerno et al., 2015; Perry et al., 2020; Wagnon et al., 2021; King et al., 2020; Thakuri et al., 2014, 2019; Sherpa et al., 2023; Potocki et al., 2022). Its extreme elevation, comprising the highest (Mount Everest) and fourth-highest (Lhotse) mountains on earth, as well as the highest-elevation glacier combined with its thriving tourism and mountaineering industry, enable altitudinal gradients in meteorological quantities to be assessed across the entire elevational range of HMA glaciers; drive high demand for accurate weather observations and forecasts (due to the high exposure of trekkers and mountaineers); and provide relatively strong logistical support for fieldwork.

Compared to elsewhere in HMA, the Khumbu region has already been relatively well instrumented with automatic weather stations (AWSs). For instance, an AWS was deployed at Pyramid (5035 m a.s.l.) by EvK2CNR in 1994, and their network was later extended up to South Col (in 2008) (Salerno et al., 2015, 2025). In 2010, the French National Research Institute for Sustainable Development (IRD France; GLACIOCLIM) and Nepalese partners began long-term weather measurements at Mera glacier up to 6352 m a.s.l. and up to 5450 m a.s.l. along the Changri Nup glacier nearby Khumbu Glacier (Fig. 1; Khadka et al., 2022). On Everest's northern slopes, AWSs were installed at the North Col (7028 m a.s.l.) and Ruopula Pass (6560 m a.s.l.) (Yang et al., 2011) though these ceased operation by 2008 and 2010, respectively. More recently, Chinese scientists reportedly installed eight AWSs up to near the summit but these data remain unpublished (The Himalayan Times, 2022) and the state of the network remains unknown.

Despite significant investments by numerous research groups, critical gaps persist in understanding HMA cryospheric processes, climate change signals, and data dependability. Weather forecasts for peaks above 6000 m a.s.l., which can have life-or-death consequences for mountaineers, also remain highly uncertain because of lack of ground-based observations for validation (Matthews et al., 2020a). Furthermore, a scarcity of high-quality, elevation-dependent datasets hinders efforts to refine regional climate models and downscale reanalysis products, which is necessary to understand longer-term, larger-scale climate change in this region (Khadka et al., 2022).

During the 2019–2025 period, our team of researchers installed and maintained AWSs in six different locations along Everest's southern slopes as part of the National Geographic and Rolex Perpetual Planet Everest Expeditions. These stations have since generated novel insights, including: seasonal oxygen availability at high elevations and climate-influenced changes in Everest's physiological summit height (Matthews et al., 2020b); sublimation dynamics at South Col Glacier (Sherpa et al., 2023); seasonal variability in cloud cover and visibility (Grey et al., 2022); extreme solar irradiance patterns (Matthews et al., 2020a); energy balance studies (Matthews et al., 2020a; Potocki et al., 2022; Brun et al., 2023); and monsoon behaviour at extreme altitudes (Perry et al., 2020; Khadka et al., 2021).

In this paper, we focus on describing the latest available open-access data (10.5281/zenodo.18849098; Khadka, 2026a). We first provide a summary of the network (Sect. 2), before outlining quality control (QC, Sect. 3). We conclude by summarizing the seasonal climatology at stations with the most complete records (Sect. 4), the correspondence between South Col AWS observations and ERA5 (Sect. 5), and by sharing new insights from the highest stations with more intermittent records.

The geographical locations of the different AWSs are shown in Fig. 1. The five stations were initially installed in the primary Khumbu valley from Phortse (3810 m a.s.l.; hereafter PH), Base Camp (5315 m a.s.l.; hereafter BC), Camp II (6464 m a.s.l.; hereafter CII), South Col (7945 m a.s.l.; hereafter SC), and Balcony (8430 m a.s.l.; hereafter BA) during the 2019 expedition (Matthews et al., 2020a). The BA station was toppled by extreme winds and anchor failure in January 2020, so in May 2022 a team of researchers and Sherpas returned to install a new AWS at Bishop Rock (8810 m a.s.l.; hereafter BR), remove the BA station, and upgrade the SC AWS (Matthews et al., 2022).

Data have been recorded using Campbell Scientific CR1000X dataloggers, at intervals ranging from 10 min to daily. However, here we present quality-controlled hourly datasets in Coordinated Universal Time (UTC). The respective variables and sensors include air temperature (T), relative humidity (RH), wind speed (WS), radiation variables (incoming (in) and outgoing (out) shortwave (SW) and longwave (LW) radiation), pressure (p), and precipitation (P) (Table 1). We also include data from experimental pitot wind sensors installed at the SC and BR in 2022.

Data gaps, sensor failures, and erroneous readings were common due to factors such as riming (icing of sensors), heavy snowfall partially or completely burying the upper stations, battery depletion, extreme temperature fluctuations, and communication failures. Acknowledging these challenges, we implemented, a multistage quality control (QC) procedure to identify and flag spurious data points (Fig. 2).

Our QC protocol takes a tiered approach, progressing from simple flagging to more complex, variable-specific corrections. The procedures applied to the raw dataset are as follows:

Data flagging: Where available, manufacturer-provided diagnostic data (Table S1) and internal sensor flags were used to identify and filter out physically implausible or erroneous observations.

Figure 3 presents the annual cycles of precipitation and temperature at PH and BC for December 2019–November 2022, together with monthly mean, minimum, and maximum temperatures based on all available observations at CII, SC, BA, and BR. Whilst this is a very short period for understanding climate variability, there is a clear seasonal pattern in precipitation and temperature at both sites. Across the network, the lowest temperatures generally occur during winter (December–February). Temperatures then rise progressively during the pre-monsoon season (March–May), coinciding with increasing incoming shortwave radiation and longer daylight duration (Fig. S4). The mean annual temperature was 4.1 °C at PH, -3.1 °C at BC, and -10.2 °C at CII. The mean monthly temperature amplitude was 6.7 °C (max 8.0 °C in March, min 4.6 °C in July) at PH, 6.5 °C (max 7.9 °C in March, min 4.9 °C in December) at BC and 6.4 °C (max 7.8 °C in March, min 3.6 °C in December) at CII, respectively. At SC, July was the warmest month, with a mean temperature of -12.2 °C and a mean diurnal range of 10.2 °C. February was the coldest month, with a mean temperature of -29.7 °C and a mean diurnal range of 4.4 °C. Precipitation at PH and BC also clearly delineates the seasonal cycle: winter is predominantly dry, precipitation gradually increases during the pre-monsoon, and the largest amounts occur during the June–September monsoon season, which accounts for 72 % and 77 % of the annual precipitation at PH and BC, respectively. Precipitation subsequently declines during the post-monsoon period as cooler and drier winter conditions become established.

Seasonal mean values, and elevation gradients between the PH and BC AWSs (separated by ∼ 1500 m) also reveal some interesting characteristics (Table 2). While these linear elevation gradients provide a useful baseline, it is important to note that the precipitation gradient across this elevational range is fundamentally non-linear, with the rate of change varying significantly between specific altitudes and locations. When calculated as a single linear gradient across this elevational range, the mean annual precipitation gradient was -107 mm km⁻¹, but it declines by almost 50 % during the monsoon to -56 mm km⁻¹. This contrast indicates that while the monsoon brings the highest total volume of water, non-monsoonal seasons (winter, pre-, and post-monsoon) feature much more dramatic relative precipitation drops with increasing elevation. Overall, the amount of precipitation measured at both observing sites (PH and BC) were higher than those reported at Pheriche (4200 m a.s.l.; 540 mm during November 2016–December 2020) and Pyramid (5050 m a.s.l.; 449 mm during 1994–2012 and 591 mm during November 2016–December 2020) (Khadka et al., 2022; Salerno et al., 2015). Similarly, between PH and BC, the mean temperature gradient (TG) is -4.8 °C km⁻¹ with least negative in winter (-4.5 °C km⁻¹) and most negative during the pre-monsoon season (-5.4 °C km⁻¹). When data from four stations with minimal data gaps (PH, BC, CII, and SC) are used to calculate linear and non-linear TGs, the resulting gradients are more strongly negative than those calculated using only the PH and BC AWSs. Both the seasonal and annual TGs derived from all four stations (Table 2) are more negative (pre-monsoon, monsoon, annual: -6.5, -5.7, and -6.0 °C km⁻¹, respectively) than the values previously reported by Salerno et al. (2015) and Graves et al. (2026), calculated using several stations in the Koshi basin between 1000–8000 m a.s.l. and Dudh Koshi basin between 2600–8400 m a.s.l. However, the TG calculated from all four AWSs (Table 2) is comparable at both seasonal and annual scales to the TG reported by Salerno et al. (2015). Crucially, we note that reported non-linearity in the temperature-altitude relationship in the Khumbu region (Khadka et al., 2021) limits the comparability of lapse rates calculated across different elevational ranges. Consequently, we follow Graves et al. (2026) and provide a non-linear equation for the temperature gradients between PH and SC (Table 2; Fig. S3).

Both incoming longwave (LWin) and shortwave radiation (SWin) exhibit strong seasonality and clear variation with altitude (Fig. S4). Both LWin and SWin reach their lowest values during winter when temperatures are coldest. As the pre-monsoon begins, rising air temperatures drive a steady increase in LWin, which remains high throughout the monsoon season. Notably, the maximum daily LWin at PH during the monsoon exceeds 350 W m⁻², while at SC it remains below ∼ 250 W m⁻². This contrast likely reflects both the lower atmospheric temperature at SC and differences in effective atmospheric emissivity, potentially associated with lower water-vapour content and reduced cloud cover at higher elevations.

SWin is lowest during mid-winter and gradually increases toward the pre-monsoon period. However, the influence of monsoonal cloud cover is weaker at higher-elevation sites (CII and SC) compared to lower-elevation stations (PH and BC), indicating reduced atmospheric moisture at higher altitudes and supporting the presence of a strong negative precipitation gradient with elevation (Graves et al., 2026).

Monitoring the meteorology at the highest altitudes in the Himalaya at elevations where glaciers are accumulating mass, and where mountaineers are most at risk from extreme weather provides the greatest test to observational efforts. Accordingly, there is considerable interest in using reanalysis data to extend brief in-situ records. However, to facilitate such extensions, it is necessary to establish the extent to which such reanalysis products can reproduce the observed meteorology (Khadka et al., 2022). ERA5, the fifth-generation global reanalysis produced by ECMWF, provides hourly atmospheric variables from 1940 onward at multiple pressure levels and at relatively high spatial resolution and is widely used as a reference dataset (Hersbach et al., 2020). While adjacent higher AWSs (BR and BA) suffer from highly limited data availability, SC AWS has longer observational record and remains operational. This is also the location of the highest-altitude camp on the main southern (Nepalese) Everest climbing route, where critical resources (e.g., tents and oxygen) are stored, and where climbers can spend considerable time. Consequently, we focus on comparisons between temperature, relative humidity and wind speed between the SC AWS and ERA5 reanalysis (Hersbach et al., 2020) data extracted from the nearest grid point at the 350 hPa pressure level (most representative of the SC: mean air pressure 2019–2021 =377 hPa). Our comparison is limited to air temperature – relevant for cold stress and with relatively high data completeness at the BA and BR sites.

During both 2019 and 2022, air temperature exhibited reasonably close co-variation between the SC observations and ERA5, with an R2>0.6 (Fig. 4 and Table 3). ERA5 systematically underestimates air temperature relative to the observations, although some of that reflects the higher altitude (lower air pressure) at 350 hPa relative to the SC data. Diurnal variability is also considerably greater in the SC dataset (Figs. S5 and S6), likely reflecting the influences of surface heat fluxes at SC, which are missing in ERA5 as this (higher altitude) pressure level is representative of the free atmosphere. To further assess the representativeness of ERA5 at high elevations, we compared ERA5 350 hPa temperatures with shorter observational records from BA (mean air pressure =354 hPa) and BR (mean air pressure =340.5 hPa). In 2019, the coefficient of determination (R2) between ERA5 and BA found 0.85, while in 2022 the R2 between ERA5 and BR was 0.56. Despite the shorter datasets, these results indicate that ERA5 temperatures at 350 hPa track observed variability in atmospheric conditions at these very high elevation sites.

The R2 values for AWS and ERA5 relative humidity were 0.49 and 0.57 in 2019 and 2022; however, the MAE and RMSE for the relative humidity are > 20 %. The 10-day running means of relative humidity from both the AWS observations and ERA5 clearly capture the seasonal transitions, with monsoon onset at SC occurring on 1 July 2019 and 14 June 2022 (Fig. 4).

In both years, wind speed exhibited the highest correlation between SC observations and ERA5 (R2=0.70). ERA5 consistently overestimated mean wind speeds at SC, with MAE values of 4.8 and 5.3 m s⁻¹ in 2019 and 2022, respectively (Table 3 and Fig. 4). This is again consistent with the ERA5 grid point being representative of the higher-altitude free atmosphere; frictional drag at the SC reduces the wind speed. We also observe faster winds in the post-monsoon in 2022 compared to 2019. Based on the generally good agreement between ERA5 reanalysis data and observations at the SC, we anticipate that the reanalysis could be used to gap-fill and extend intermittent meteorological records from Mount Everest's upper slopes – correcting the time-varying biases using empirical-statistical, or machine-learning based approaches (Van Wyk de Vries et al., 2024; Potocki et al., 2022).

The quality controlled dataset, including raw disdrometer and all available wind speed data, is available in CSV format from Zenodo (10.5281/zenodo.18849098; Khadka, 2026a). Selected data can also be downloaded from the National Geographic Society weather-data portal (https://www.nationalgeographic.org/projects/perpetual-planet/everest/weather-data/, last access: 15 July 2025). Additionally, the data processing code is available at 10.5281/zenodo.22108645 (Khadka, 2026b).

Observations from the SC (Fig. 4) highlight the potential for mean winds exceeding 30 m s⁻¹ and gusts above 60 m s⁻¹ that can be extremely dangerous for mountaineers by increasing the risk of being blown over and the probability of cold injury (Sherpa et al., 2023; Van Wyk de Vries et al., 2024). Elsewhere, atmospheric pressure data from the AWS network described here have shown that oxygen availability at the summit is highly variable on synoptic timescales, as it depends on the partial pressure of atmosphere. Consequently, the “apparent elevation” of Mt. Everest (i.e., how high the mountain would feel without supplemental oxygen) can vary by almost 750 m, and during the most challenging winter conditions an ascent without supplemental oxygen may be impossible (Matthews et al., 2020b). Accurately forecasting wind speeds and oxygen variability for the upper slopes of Mount Everest is therefore of high importance for increasing safety on the mountain. The data from the AWSs presented here provide the foundation to achieve such improvements. At a minimum, they can be used to assess uncertainty in the freely available forecasts widely used by expeditions (Matthews et al., 2020b); but they can also be used to train model output statistics including novel applications of machine learning to provide improved, hyper-local forecasts for identifying optimal climbing windows (Van Wyk de Vries et al., 2024).

The AWS data shared here can also be extremely useful for understanding cryosphere-climate interactions at the highest elevations within HMA. For instance, the archive has already been used to estimate the surface energy balances at the summit of Mt. Everest (Matthews et al., 2020a), and for the South Col Glacier (at ∼ 8000 m a.s.l.), revealing a high-altitude system that might be acutely sensitive to changes in effective precipitation due to extremely high levels of insolation and thus highly responsive to albedo variations (Matthews et al., 2020a; Potocki et al., 2022; Brun et al., 2023). However, significant uncertainties still remain, mainly because the observational record is short. These could be improved by extending the analysis with downscaled reanalysis products like ERA5.

The data here are useful not only for assessing cryosphere-climate interactions at extreme elevations but could be used to improve understanding of surface energy and mass balances within the ablation areas of the region's glaciers, and at lower elevations. The precipitation measurements, for instance, are particularly high quality, as they were collected behind double-Alter shields to reduce undercatch (which likely accounts for the greater accumulations relative to nearby stations without such shields: Pheriche and Pyramid), and hence offer a valuable opportunity to constrain mass inputs. More generally, when used in combination with the generally lower-altitude observations from the GLACIOCLIM and EvK2CNR networks, the data shared here can improve quantification of elevational gradients in key meteorological variables needed in distributed (e.g., glacier or hydrological) modelling assessments across up to approximately five vertical kilometres, and hence the entire glacierised elevational range of the Khumbu region.

Data across such a large elevational range could also provide valuable insight into regional-scale atmospheric dynamics. For instance, the timing and character of the monsoon's arrival – critical for determining the annual precipitation totals (Perry et al., 2020), and a key control on melt energy (Khadka et al., 2021; Matthews et al., 2020a) can be assessed right up to the upper accumulation area of the Khumbu Glacier. Important further work might improve understanding of the relevant physical drivers behind elevational differences in the monsoon's arrival, how they interact with local anabatic and potential katabatic forcing, and how they might change with further global warming (Salerno et al., 2023).

Ultimately, though, these data might be used in ways far beyond the illustrations above. The upper slopes of Mount Everest represented the last frontier for human exploration, reached successfully by humans 42 years after the South Pole was attained. Meteorological measurements from the highest reaches only began in 2022 (at the BR). We are still, therefore, at the very beginnings of exploring the weather of this extreme environment, and the list of questions that these data might help answer from disciplines beyond the climate sciences (e.g., biology: Dragone et al., 2023) might conceivably grow rapidly.