The Antarctic ice sheet is the largest reservoir of fresh water, holding a global sea-level potential of 58 m , and also acts as a reliable record of the recent climate e.g.. Due to its isolation, dry climate, and long austral winter, it also provides unique and often favourable locations for meteorological, astronomical, geophysical, and upper-atmosphere observations.

Reliable in situ measurements of meteorological quantities and components of the surface energy budget (SEB) are required for climate model evaluation on the Antarctic ice sheet e.g., to provide realistic atmospheric forcing for snow/firn models e.g., and to validate satellite remote sensing products e.g., but also to train machine learning algorithms e.g. to interpret ice core observations e.g. as well as to provide in situ estimates of the sub-seasonal surface mass balance SMB; e.g.. In addition, such in situ observations serve as a basis for process understanding, such as surface sublimation , surface melt quantification , formation and burial of meltwater lakes , blue ice formation , the formation of impermeable ice slabs , and the detection of climate trends e.g..

Despite the increasing need for reliable in situ measurements, automated weather station (AWS) observations are still scarce in Antarctica. The first continuous AWS observations started in 1978 , and since then, at most 146 AWSs were simultaneously recording near-surface meteorological variables across the continent . These AWSs were part of one of several networks, such as the Antarctic Meteorological Research Center (AMRC) network maintained by the University of Wisconsin-Madison , the Australian Antarctic Division (AAD) network , the Italian National Program of Antarctic Research (PNRA) network , the Chinese National Antarctic Research Expedition (CHINARE) PANDA network , the British Antarctic Survey (BAS) network, the Japanese Antarctic Research Expedition (JARE) network , the French Antarctic Program (Institut Polaire Francais-Paul Emile Victor, IPEV) network, and other similar networks maintained by different nations or organizations. These stations are shown in Fig.  and further described in . The number of AWSs recording the near-surface radiation balance and surface height change, allowing for the estimation of the SEB and SMB, is even much more limited. Since 1995, the Institute for Marine and Atmospheric Research Utrecht (IMAU), in cooperation with the Alfred Wegener Institute (AWI), the BAS, the Finnish Antarctic Research Program (FINNARP), the Norwegian Polar Institute (NPI), the United States Antarctic Research Program (USARP), the Swedish Antarctic Research Program (SWEDARP), the Royal Meteorological Institute of Belgium (KMI), and the KU Leuven, has been operating 19 AWSs in Antarctica, of which 16 measured the radiation components, allowing for a direct quantification of the entire SEB. Recently, presented a database containing SEB calculations at eight stations from the IMAU network where surface melt occurs.

In this work we describe the dataset of measurements from all these 19 AWSs that spans from 1995 to 2022. We describe the applied corrections and processing steps and also provide a single quality flag as well as the interpolated wind speed, air temperature, and specific humidity at standard heights of 10 and 2 m above the surface using similarity theory, which allows for an easier comparison with different datasets. Although the measurements from this dataset are also partly contained in the datasets from and , the data presented in this work have gone through an elaborate quality control and data correction strategy, which are specifically tailored for the unique combination of sensors and locations of the IMAU dataset. On the other hand, this dataset does not contain the calculation of SEB/SMB fluxes, which requires additional model calculations.

The period covered by the dataset is shown in Fig. , and the number of valid, non-flagged samples per variable and per station is given in Table . In total, 57 230 d (≈ 157 years) of data are available from 1995 through 2022, of which there were 31 183 d (≈ 85 station years, or 54 %) of data with both unflagged meteorological observations, sonic height ranger measurements, and all four components of net surface radiation (Fig. ). AWS09 has the longest time series (24.8 years), while AWS05 contains the longest time series of unflagged meteorological and SEB data (12.8 years). At AWS09, the outgoing long-wave radiation is missing between 2009 and 2016 due to a malfunctioning logger, while from 2017 onwards, the relative humidity is missing due to a malfunctioning sensor. This means that not more than 1 year of interpolated temperature data are available at AWS09 after 2009, since all variables need to be unflagged to reliably interpolate quantities using Eqs. ()–(). At the plateau sites AWS12 and AWS13, no pressure data is available due to a malfunctioning data logger, and no wind speed and wind direction data is available after 70 d of operation at AWS13 due to a malfunctioning sensor. Yet, despite the very cold and dry climate, most other variables including surface height are available for the entire period at AWS12 and AWS13. The type III AWSs have the highest success rate (95 %). Data gaps longer than several days often affect all the variables, due to either a malfunctioning logger or power supply (e.g. AWS14 during 2017) or the AWS being buried by snow (AWS11 in 2012). Long periods of long-wave radiation measurements are flagged at the plateau stations AWS09, AWS12, and AWS13 during winter, since the radiometers function well below their operation temperature range, but are not removed from the dataset.

The time averages of all measurements of incoming radiation, air temperature, wind speed, and surface height change are plotted as a function of station elevation in Fig. , and the averages of all variables per station and per season are given in Table . On average, annual downward long-wave radiation decreases with elevation from 230 W m⁻² at the ice shelf locations to 100 W m⁻² at the plateau stations. The decrease in incoming long-wave radiation is compensated by a similar increase in incoming short-wave radiation from about 130 W m⁻² near sea level up to 200 W m⁻² on the Antarctic plateau. The range of yearly averaged air temperature contained in this dataset is [-55; -15] °C. The lowest daily average temperature at a sensor height of -82.9 °C was recorded at AWS13 (Pole of inaccessibility) on 11 August 2010, and the highest daily averaged air temperature at sensor height of 10.2 °C was recorded at AWS18 (Cabinet Inlet, Larsen C ice shelf) on 26 May 2016. Temperatures above melting occur at 10 out of 19 stations, either on ice shelves (AWS04, 14, 15, 17, 18, 19) or close to ice shelves (AWS05), but also rarely at higher elevations on the grounded ice sheet (AWS06, 07, 16). The annual averaged wind speed at sensor height ranges between 3 and 8 m s⁻¹, with maximum daily averages ranging from 11.4 m s⁻¹ at AWS12 (Plateau Station B) up to 32.3 m s⁻¹ at AWS05 (Wasa/Aboa Camp Maudheimvida). The surface height change ranges from 92 mm yr⁻¹ of ice ablation at the Scharffenbergbotnen blue ice location AWS07 up to 2242 mm of snow accumulation per year at the Halvfarryggen ice rise location AWS11.

An example application of this dataset is presented that takes advantage of the correlation of the measured variables between nearby stations, in order to spatially interpolate point measurements or design the optimal AWS network. The daily averaged measurements of temperature, wind speed, and both downward radiation components for January are selected. The correlation of each of these four variables is then computed for all station pairs during the periods of overlapping data and only retained if at least 1 month of overlapping data are available. The correlation for each station combination for each variable is then regressed as a function of the distance between all station pairs in Fig. . The distances and correlations are also given in Tables  and . The largest correlation is found for air temperature (Fig. a, green dots), with a fitted e-folding distance of 1430 km, which is defined as the distance where the fitted correlation equals 0.37. This distance is consistent with and , who make use of different data. On the other hand, the correlation between stations for horizontal wind speed and for the two downward radiation components is lower (Fig. b), with an e-folding distance of about 500 km for all three variables. It must be noted that such correlations also depend on the region and on the season . The lower correlation for wind speed and downward radiation confirms the need for a denser array of stations for energy balance applications.

While the corrections presented in Sect.  are based on short experiments with benchmark datasets, the weather stations have been in operation in remote areas on the Antarctic ice sheet, which prevents a direct comparison of their long-term measurements with independent benchmark datasets. As such, the uncertainty in the final corrected parameters is estimated as follows. From 1 January through 28 October 2001, a type I and a type II station were simultaneously operating at AWS10, and from 23 January through 31 December 2015, a type II and a type III station were simultaneously operational at AWS14. The mean absolute difference (MD) and centred root-mean-square deviation (RMSD) of the overlapping 2 h (AWS10) and hourly (AWS14) parameters are then computed and reported in Table . In addition, we compare the measured horizontal wind speed by the type II station at AWS14 with overlapping sonic anemometer data taken from 6 January through 30 January 2011 during the experiment described by and at AWS05 with overlapping sonic anemometer and fine-wire thermocouple data from 15 January 2007 through 5 July 2008. These additional data are not part of this dataset. Finally, we compare both downwards radiation components measured by the type III station at AWS11 with independent reference data from the BSRN site at Neumayer from 14 February 2010 through 30 January 2019 , although these stations are located 80 km apart.

On average, the air temperature between stations differs by 0.76±0.91 °C (MD ± RMSD) for stations I and II and 0.25±0.45° for stations II and III, which exceeds the reported accuracy by the manufacturer (0.1 °C for types I and III, 0.3 °C for type II; Table ). We attribute this larger uncertainty to the different sensor heights, to the different sampling strategy, and to the uncertainty in the excess temperature correction due to the variable effectiveness of the radiation shields. The comparison with a fine-wire thermocouple at AWS05 also reveals recorded air temperature differences of 0.2±0.64 °C, which confirms that the sensor's performance is slightly different than reported by the manufacturer.

The difference in measured, corrected relative humidity between station types II and III (8 % ± 14 %) also exceeds the reported accuracy, which translates to specific humidity differences of 0.15 g kg⁻¹ ± 0.32 g kg⁻¹, which remains limited due to the low temperatures in this dataset. The differences are mainly caused by the relative large uncertainty in the hygrometer readings at low temperatures (minimum of -49.7 °C at AWS14) and different sensor types, but also by different sensor heights and different radiation shield types.

The horizontal wind speed does not substantially differ from the reported manufacturer accuracy, except when comparing types I and II, which we attribute to the different sampling strategy and to the different response and overspeeding of cup anemometers. Further, the cup anemometer at AWS10 was never replaced after its installation in 1995. The comparison between type II and sonic anemometer data at AWS05 reveals differences of 0.1±0.85 m s⁻¹, which may be interpreted as the uncertainty in measured wind speed by stations II and III in the polar conditions from this dataset. We attribute the differences from sonic anemometer readings to different heights; flow distortion; poor response of propeller anemometers at low wind speeds (lower than 1.5 m s⁻¹); and noise in the sonic anemometer readings due to, for example, precipitation, blowing snow in the sensor volume, or riming of the transducer heads.

The mean absolute difference of all four radiative components ranges from 1 to 8 W m⁻² between stations. The performance of the Anderaa 2770 sensor used to measure downwards short-wave radiation on the type I stations is lower than reported by the manufacturer and also lower than the CNR1/4 used on types II and III, with a RMSD value of 34 W m⁻² versus 19 W m⁻². On the other hand, we find that the corrected hourly CNR1/4 readings differ by less than 5 % of the daily averages, which is better than reported by the manufacturer and consistent with the result from . It must be noted that the location of AWS10 is more susceptible to frosting, which possibly causes larger downwards short-wave radiation differences compared to other locations, since different radiometers may behave differently under frosting conditions. Despite the 80 km horizontal separation and about 650 m elevation difference, downwards short-wave and long-wave radiation measured at AWS11 only differs from the benchmark measurements from the Neumayer BSRN site by 11±68 and 4±31 W m⁻², respectively, on average over a 10-year period . This further supports the high accuracy of the corrected CNR1/4 readings. Small average differences in measured short-wave components may still translate into relatively large error in broadband short-wave albedo, with RMSD values of 0.026 at AWS14 during 2015.

Finally, the intercomparison of the hourly surface height recorded by two different sonic height rangers reveals RMSD values of 0.08 m between the type I/II stations and 0.04 m between type II/III stations. These values are higher than reported by the manufacturer, which we attribute to the horizontal variability in the snow surface and to noise generated by secondary reflections.

The measurements in this dataset were taken in (or in the vicinity of) Dronning Maud Land and the Larsen C ice shelf; hence different datasets should be included for Antarctica-wide studies. This may include other AWSs e.g. or SMB compilations; surface meteorological observations from year-round crewed stations; or surface radiation observations from the BSRN network at Neumayer , Concordia , South Pole , and Syowa .

For future Antarctic AWS design, we recommend including measurements of the four radiation components and surface height change, which allow for SEB quantification including melt and therefore for a more complete evaluation of climate models.

The hourly dataset for all 19 AWSs is available at 10.1594/PANGAEA.974080 .

The code used to pre-process and correct measurement of temperature, long-wave radiation, and relative humidity for the specific instrument combinations used in this dataset is available at 10.5281/zenodo.15101447 . The code used to apply the other corrections, flag suspicious data, and compute new variables is available at 10.5281/zenodo.15058515 . The surface energy balance model used to compute the quantities at standard heights is available at 10.5281/zenodo.15082295 .

A dataset is presented that contains quality-controlled, continuous, hourly or 2 h measurements of meteorological quantities, net surface radiation components, and surface height change at 19 locations on the Antarctic ice sheet from 1995 through 2022. This dataset differs from other similar datasets containing these stations, such as and , in how quality control and data corrections were implemented, but also in the number of available recorded parameters. The temperature, humidity, and radiation data are corrected for commonly documented measurements errors; noise was both manually and automatically removed; and missing data are left empty. In addition, the temperature, humidity, and wind at standard heights, as well as the cumulative change in surface height, are provided when available. A simple quality flag is provided for further use in, for example, surface energy balance simulations. Despite the remoteness and harsh climatic conditions of these locations, the average success rate of the hourly data is 96 % for the temperature, 86 % for the wind speed, 71 % for the four components of net surface radiation, 96 % for the filtered measured height change, and 54 % for all of the hourly data being simultaneously available and unflagged.

Some of the measurements were transmitted to the GTS system and are therefore available to be assimilated in reanalyses products. The exact periods and stations for which this is the case are marked in Table . The other measurements have not been assimilated and thereby allow for an independent benchmark for reanalyses products. This dataset may also serve as basis for future work that requires in situ observations to study surface–climate interactions on the Antarctic ice sheet.