Air pressure is a key atmospheric variable used to interpret weather patterns and calculate air density. Barometric pressure (in hPa) is measured in the fibreglass reinforced polyester logger enclosure using a CS100/Setra 278 barometer. The logger enclosure adjusts to ambient pressure through a pressure-compensating plug. The barometer manufacturer reports a measurement accuracy of ± 2 hPa within the -40 to +60 °C temperature range (Table ). AWSs equipped with an OTT Lufft WS401 use an internal digital barometer based on MEMS (micro-electromechanical systems) technology and provides reliable data over a pressure range of 300–1100 hPa. The accuracy is ± 0.5 hPa within the temperature range of 0 to 40 °C, and ± 1.5 hPa outside that range.

Air temperature is measured to quantify surface energy balance, melt conditions, and boundary-layer atmospheric variability. In our current AWS setups, we use one of three types of thermometers depeding on the locations.

Air temperature (in °C) is measured inside a fan-aspirated radiation shield using the Rotronics setup described in . The Rotronics setup refers to the combined air-temperature and humidity measurement system built around the Rotronic HygroClip probe family (HC2/HC2A-S3), installed inside a fan-aspirated radiation shield. The primary temperature sensor is a PT100 probe, which has an accuracy of ± 0.1 °C (Table ). A secondary air temperature reading is obtained from the HygroClip temperature/humidity sensor, also housed in the aspirated shield. This sensor also has a manufacturer-stated accuracy of ± 0.1 °C but needs more frequent recalibration than the PT100 .

The OTT Lufft WS401 is a compact, multi-parameter weather sensor that measures air temperature, relative humidity, air pressure, and liquid precipitation using an integrated tipping-bucket mechanism. It features a ventilated radiation shield for improved temperature and humidity measurements. The thermometer in the OTT Lufft WS401 is a capacitive sensor designed for air temperature measurement. It has a measurement range of -50 to +60 °C, with an accuracy of ± 0.2 °C at 20 °C. The sensor is enclosed in an aspirated shield.

AWSs equipped with the Vaisala HMP155E sensor measures air temperature (in °C) using a Platinum Resistance Thermometer (PT100) inside a fan-aspirated radiation shield from Rika. The measurement range is -80 to +60 °C, with accuracy varying with temperature. Specifically, for the range of -80 to +20 °C, the accuracy is ± (0.226 - 0.0028 × air temperature) °C.

Relative humidity (RH) is needed to estimate atmospheric moisture content, evaluate sublimation processes, and calculate longwave radiation. In the Rotronics setup , relative humidity (RH; in %) is measured alongside the PT100 sensor inside the aspirated radiation shield using a HC2A-S3 (or HC2) HygroClip, which has an accuracy of ± 0.8 %.

The OTT Lufft WS401 hygrometer features a capacitive humidity sensor designed for humidity measurement, operating over a range of 0 %–100 % relative humidity with an accuracy of ± 2 % in the 10 %–90 % range at 20 °C. It is temperature-compensated and housed in an aspirated enclosure.

The Vaisala HMP155E sensor for relative humidity measurement, offering an accuracy of up to ± 1 %. It operates over the 0 %–100 % range. With an operating temperature range of -80 to +60 °C, the sensor provides fast response times, temperature compensation, and resistance to contamination.

Relative humidity is measured with respect to water, meaning calibration is performed above the freezing point. Hygrometers are typically recalibrated when possible, but in practice this occurs every 1 to 4 years. Calibration is conducted in a closed chamber at room temperature under controlled relative humidity conditions at levels of 10 %, 35 %, and 80 %. Alternatively, the instruments may be sent to the manufacturer for recalibration.

For temperatures below freezing, relative humidity is recalculated relative to ice in post-processing. To distinguish between the two relative humidities in the data products, the prior humidity (adjusted below freezing) is called “relative humidity with respect to water or ice”, whereas the latter is simply referred to as “relative humidity”. The conversion of relative humidity relative to ice is after . See Table  for further information.

Liquid precipitation is monitored at stations using the OTT Lufft WS401 tipping-bucket rain gauge (in mm). Rainfall collects in a small, seesaw-like bucket that tips once a set volume (typically 0.1 or 0.2 mm) is reached, triggering a magnetic switch to record one “tip.” The total rainfall is calculated by counting these tips over time. This design performs well in most weather conditions, though regular maintenance is needed to prevent clogging or debris buildup. It is not optimal for measuring solid precipitation (e.g., snowfall) since the instrument lacks heating, which can lead to snow accumulation and delayed melting within the gauge. Nevertheless, it is adopted here as a rain gauge in a large range of conditions experienced across Greenland (Table ).

Wind measurements are essential for interpreting turbulent heat fluxes, drift snow processes, and station exposure. We use anemometers from the manufacturer Young. The wind speed and direction (in ms-1 and degrees, respectively) measurement height, like the other measurements, has a reduced measurement height if a winter snow layer is present (Table ). An AC sine wave voltage signal is produced by the rotation of the four-bladed propeller, and the pulse count converts to wind speed using a multiplier. According to the manufacturer, the sensor can measure wind speeds between 0 and 100 ms-1, with an accuracy of ± 0.3 ms-1 or 1 % if the measured value is higher than 30 ms-1. Wind direction is measured through changes in the vane angle by a potentiometer housed in a sealed chamber on the instrument. The output voltage is directly proportional to vane angle wind direction and is measured between 0 and 360° with an accuracy of ± 3°. When possible, every three years the sensor is replaced and tested for drift and functionality with an “anemometer drive”, rotating the propeller shaft at a known rate. The instrument's orientation is logged and reset to “geographic north” during each maintenance visit to keep wind direction data accurate within ± 15° (although much larger station rotations have been encountered).

Radiometers measure the incoming and outgoing shortwave and longwave radiation needed to compute the surface energy balance. The Kipp & Zonen CNR1 and CNR4 are net radiometers (in Wm-2), designed to measure the balance between shortwave and longwave radiation. The CNR1 is an instrument with two pyranometers and two pyrgeometers, suitable for meteorological and environmental research. The CNR4 is an advanced model offering improved accuracy, including lower thermal offset, and better longwave response. These radiometers are targeted for recalibration every three years (Table ), however in a few cases, recalibration happens every 4–5 years.

The pyranometers, housed within hemispherical glass domes to minimize water droplet adhesion, record upward and downward shortwave irradiance between 0.3 and 2.5 µm. The manufacturer specifies a sensor uncertainty of 10 %, though practical assessments in Antarctica suggest an approximate 5 % uncertainty for daily totals .

The pyrgeometers measure upward and downward longwave irradiation with an estimated field uncertainty of 10 % for the CNR1 and 5 % for the CNR4. These values account for instrumental and environmental factors, including calibration accuracy and thermal offsets. Both pyrgeometers use silicon windows, sensitive to infrared wavelengths between 4.5 and 42 µm.

The radiometer data is stored in the data logger as voltage (µV) due to the unique calibration coefficients assigned to each radiometer, while all logger programs on the AWSs remain standardized for operational efficiency. During post-processing, raw sensor readings (SRraw) are converted into physical measurements (SRm) using the equation: 1SRm=SRrawCSR, where CSR (unit: µV(Wm-2)-1) is the sensor specific calibration coefficion provided by the manufactorer. SRm represents either the downward or upward shortwave irradiance. Similar to shortwave radiation, longwave radiation readings are stored in the data logger as voltage (LRraw) and later converted into physical units (LRm) during post-processing using the formula: 2LRm=LRrawCLR+5.67×10-8⋅(Trad+T0)4, where CLR (unit: µV(Wm-2)-1) is the sensor calibration coefficient. Trad represents the sensor temperature recorded within the radiometer casing (in °C), and T0 = 273.15 K.

Surface height changes from snow accumulation or ice ablation are measured using the Campbell SR50A sonic ranger. The SR50A is an ultrasonic depth sensor (sonic ranger) designed for measurement of height changes, e.g. snow accumulation (Table ). It operates by emitting ultrasonic pulses toward the surface and measuring the time delay of the reflected signal. The SR50A is durable, weather-resistant, and suitable for use in harsh environmental conditions, making it ideal for AWSs on the Greenland ice sheet. On both station designs, the sensor boom height (in m) is measured by a sonic ranger mounted approximately 0.1 m below the boom itself. For the tripod design, a SR50A is also mounted on a stake assembly drilled into the ice recording surface height changes.

Boom or stake height Hm (in m) is derived from raw sensor data (Hraw), corrected for air temperature during post-processing: 3Hm=Hraw⋅Tair+T0T0, where T0 = 273.15 K. After temperature correction, the manufacturer-reported uncertainty for the SR50A sonic ranger (Campbell Scientific) is ± 1 cm or ± 0.4 % of the measured distance. An uncertainty assessment for sonic ranger readings, based on wintertime accumulation-free data from SCO_U, found standard deviations of 1.7 and 0.6 cm after spike removal, corresponding to uncertainties of 0.7 % and 0.6 % of the measured distance, respectively .

Surface lowering in the ablation zone can also be measured using a pressure transducer assembly (PTA). The Ørum & Jensen NT 1700 is a robust pressure transducer designed for accurate measurement of water pressure and level in environmental and industrial applications (Table ). It features a piezoresistive sensor element, housed in a durable stainless steel casing, and is suitable for long-term deployment. The NT 1700 offers reliable performance, stable output, and compatibility with standard data logging systems. The tripod AWSs are equipped with a pressure transducer assembly (PTA) that measures surface height changes caused by ice ablation. Originally developed in Greenland in 2001 by and later refined under PROMICE , the PTA consists of a hose filled with a 50/50 antifreeze-water mixture and a pressure transducer at its base. The hose is typically drilled up to 14 m into the ice, and the transducer registers the pressure from the vertical liquid column above it. A schematic showing how to construct the PTA system is provided in the appendix. Similar to the radiometer, each PTA has a unique calibration coefficient, which is why measurements (Hraw) are stored as voltage in the data logger and is converted into physical units as Hm (in m): 4Hm=CPTA⋅ρwρaf⋅Hraw, where CPTA is the calibration coefficient. The constants ρw and ρaf are the densities of water and the antifreeze/water solution, respectively. See Appendix Fig.  for further details.

Subsurface temperature is used to track firn and ice thermal structure, freeze–thaw cycles, and englacial processes. We have two types of thermistor strings (temperature-dependent resistor) that measure subsurface temperatures (in °C). We have a digital and an analogue type: the analogue type is designed and constructed internally, while the digital type is made by Geoprecision (Table , see Appendix for further details).

For sites in the ablation area, the subsurface ice temperature is measured using a 10 m thermistor string with 8 thermistors unevenly spaced. The string records temperatures at depths that may vary due to surface ablation and accumulation as seasons change through the year.

For sites in the accumulation area, the subsurface firn temperature is similarly measured using a 10 m thermistor string (digital type), but with 11  thermistors unevenly spaced to capture the higher temperature variability in the near-surface snow/firn. When installed, the top thermistor is placed at surface level, with a spacing of 50 cm to the next down to 3 m depth, then a spacing of 1 to 4 m depth, followed by 2 m spacings down to the bottom thermistor at 10 m depth. The thermistor string is inserted into a standard 32 mm (outer diameter) polypropylene pipe (PP), commonly used for sewer water and consisting of 6 × 2 m pieces, allowing for a 2 m extension above the snow surface. The PP pipe system is sealed at the bottom and a custom-made cap allowing passage of the cable is placed at the top end, providing a waterproof system with the thermistor string extended in a relatively narrow air-filled pipe. The PP pipe is usually reinforced with a stake for structural support and a thin bamboo pole is attached as an extension at the top to help locating the system if the snowfall exceeds the height of the PP pipe before next visit.

Tilt and orientation measurements (in degrees) help correct radiation and wind data for boom inclination or rotation. We use two types of inclinometers. HL Planartechnik GmbH manufactures a series of tilt sensors designed for measuring angular displacement. The NS-25/E2 inclinometer, for example, consists of a flexible circuit that can be integrated into different measurement systems and provides an analog voltage output proportional to tilt. The Planar inclinometer measures the tilt (in degrees) both across (left-right) and along (up-down) the sensor boom, which is interpreted as tilt-to-east and tilt-to-north when the sensor boom is oriented north-south. The inclinometer's voltage readings (Tiltraw) are converted into Tiltm in degrees using the following equation: 5Tiltm=21.1⋅|Tiltraw|-10.4⋅|Tiltraw|2+3.6⋅|Tiltraw|3-0.49⋅|Tiltraw|4, where all constants were determined in-house using. The constants were determined using a separate inclinometer by deriving a best-fit polynomial relationship between the measured voltage and the corresponding measured tilt.

The Rion compass was chosen to replace the HL Planar tiltmeter in our AWS systems. It uses magnetic field sensors to determine azimuthal orientation, providing accurate and reliable heading (tilt) data for applications that require precise directional alignment, such as measuring downward shortwave irradiance and wind direction (see Table ).

Since their inception, all AWSs have been equipped with a single frequency GPS that records site position and position metrics hourly. The same technology has been applied to GC-Net stations starting in 2020 for the SWC site and in 2021 onward for an increasing number of the GEUS carry-forward GC-Net sites. The GPS antenna and the receiver, which is part of the Iridium 9602-LP modem, are housed inside the data logger enclosure. The receiver is a NEO-6Q model, operating at 1575.42 MHz (L1), with 16 channels and C/A code. Its accuracy is reported to be within 2.5 m (Table ). The manufacturer states the the vertical accuracy is typically 1.5–2 times worse than the horizontal. Implicit in the single frequency measurements is the use of the EGM96 geoid to obtain orthometric height a.k.a. elevation above mean sea level. See also AWS position in the post-processing section for more information.

In the AWS configuration, the GPS receiver is activated for one minute before each Iridium transmission attempting to acquire position. The coordinates with the lowest horizontal dilution of precision is saved to memory.

The single frequency GPS can produce relatively noisy data and suffer from occasional data gaps. For the users' convenience, we distinguish between these direct GPS measurements, called gps_lat, gps_lon, gps_alt, and our best estimate of the station position at all time step, simply called lat, lon, alt derived in post-processing (see Sect. ).

CR1000X is a rugged, low-power data logger from Campbell Scientific, ideal for long-term monitoring in harsh environments. It offers faster processing, more memory, and improved analog precision compared to the CR1000, along with USB, RS-232, and Ethernet connectivity. These upgrades enhance data acquisition efficiency, reliability, and flexibility.

The NAL Research 9602-LP modem is a low-power, compact device designed for reliable long-range communication. It uses Iridium satellite connectivity, providing global coverage for data transmission. The modem supports Iridium Short Burst Data (SBD), which is a communication protocol designed for sending small amounts of data. It is optimized for low-bandwidth applications that need to transmit short bursts of data, such as sensor readings, GPS locations, or status updates. Iridium SBD enables reliable communication in areas without cellular coverage.

We use two types of batteries for AWS power:

Lead-acid batteries are known for their ability to supply high surge currents, cost-effectiveness, and robustness in harsh environments (Table ). Although their energy density is lower compared to more advanced battery chemistries, they offer consistent performance over a wide temperature range. In a controlled freezer test, we evaluated lead-acid battery performance and found that while charging below even -40 °C is feasible, only a limited amount of energy is stored; nevertheless, they remain a reliable power source under extreme low temperatures.

For the reasons above, lead-acid batteries were chosen for high-elevation/far-North GC-NET sites at risk of temperatures below -40 °C, despite their lower energy density compared to the NiMH batteries.

RS PRO solar panels are high-efficiency photovoltaic modules designed to convert sunlight into electrical energy. These panels are made of either monocrystalline or polycrystalline silicon. RS PRO solar panels are known for their robust construction, suitable for harsh environments.

The GC-NET mast configuration is a two-boom system, designed for accumulation area deployment. The two booms on the mast allow for vertical profiling of the near-surface boundary layer air temperature, wind speed & direction and humidity. The historical GC-NET mast configuration, originally deployed at 18 sites starting in 1995 as described in and , consisted of a 4” diameter aluminum tube with a 1/4” wall thickness, providing a very stiff, but relatively heavy mast. Extending the mast thus required a tripod crane.

To enable a more light-weight system that would not require a crane, it was decided to investigate alternative mast solutions. The first light-weight alternative, deployed in 2021, was based on a 48 mm diameter titanium tube, fitted with two titanium booms. These tubes turned out to be too flexible for stable mounting of instruments, despite being anchored with 3 mm braided stainless-steel wires and reinforced with wooden poles inside. The second version, which has subsequently been deployed at all accumulation area sites, re-introduced the original outer mast diameter of 100 mm and aluminium, based on a combined analysis of weight, rigidity and usability of titanium, steel, aluminium and carbon fibre, respectively. To meet the weight requirements, the wall thickness of the 100 mm aluminium mast was reduced to 3 mm. Apart from reducing the weight of the mast itself, the number of the booms carrying the instruments were reduced from 5 to 2, with the two booms positioned orthogonally, 1250 mm apart on the mast. Additionally, the choice of a more light-weight construction has enabled a field team of four to raise a fully instrumented mast without the use of a tripod crane and winch, further reducing the total weight to be carried on the airplane.

However, the light-weight construction has also required a positioning of instruments, solar panel and logger box that could potentially reduce the quality of the measurements, as they are more closely spaced than previously, increasing the risk of shading and turbulence. Similarly, the orthogonal position of the two booms provides different conditions for the two levels of wind measurements in terms of down-wind turbulence from the measurement frame.

Despite these potential issues, introducing a more light-weight mast structure has been deemed necessary in order to provide capacity for additional personnel (generally 4–6 people), regular transport of battery boxes, tools and replacement sensors, and equipment.

Figure  illustrates the schematic of an accumulation area AWS, and Fig.  gives and in-situ impression from the field. Liquid precipitation is measured using a OTT Lufft WS401 sensor (1a), while humidity and temperature are recorded by a Vaisala sensor (1b). Visual- and infra light is measured by a CNR4 radiometer (2), which is equipped with with an integrated inclinometer/compass (2). Wind speed and direction are recorded by two anemometers (4), and snow height is measured by two sonic rangers (5). Power is supplied by a south-facing solar panel (6), connected to a battery box (9). Data acquisition and positioning are handled by a GPS and a logger box (7). Structural stability is ensured by an anti-torque rod (8) and a thermistor string (10) is drilled into the firn measuring temperatature at 11 different levels.

The standard accumulation area mast structure consists of 6 aluminium tube pieces as shown in Figs.  and :

A lower part with a plastic cup in the bottom, length 230 cm.

A mast and an insert are fastened using 3 rivets (5 mm) separated by 120°. The accumulation area mast design makes it possible to always bring down the instruments for routine maintenance and rotation, without the use of a crane.

The AWS tripod is constructed using 32 mm (1.25 in.) and 44 mm (1.75 in.) diameter aluminum tubes, reinforced with 3 mm braided stainless-steel wires to form a stable tetrahedral structure (Figs.  and ). Most sensors are mounted on a 1.7 m long horizontal boom positioned 2.7 m above the surface (Fig. ). To enhance stability, a battery box weighing approximately 50 kg is suspended beneath the mast on wire ropes with shackles, lowering the centre of gravity of the AWS installation. The tripod design allows it to be folded for transport in small helicopters and tilted for sensor replacements.

The sensor housing with thermometer and hygrometer is located approximately 2.6 m above the ice surface (i.e. as high as possible underneath the sensor boom). The measurement height varies when a winter snow cover is present (Fig. , item 1). The inclinometer is mounted on the sensor boom (Fig. , item 2) and aligned with the radiometer to allow for tilt correction of the CNR1 shortwave radiation measurements. In contrast to the previous setup described in , the inclinometer and compass are now integrated into the CNR4 radiometer for improved alignment. The radiometer is installed as an extension of the boom and faces south, as shown in Figs.  and . In the latest AWS designs, the compass also provides orientation data to record any rotation of the boom or tripod. The wind sensor is mounted on the opposite side of the boom from the radiometer (Fig. , item 4), measuring approximately 40 cm above the boom. The sonic ranger is mounted on the boom directly under the wind sensor with a distance to the ice surface of approximately 2.6 m (Fig. , item 5), while the SR50A mounted on a stake is drilled into the ice and melts out with the ablating ice surface during summer (Fig. , item 12). As the tripod rests freely on the ice surface, it moves down as the ice melts, meaning the sonic ranger measurements on the AWS do not capture ice melt, only snowfall and snowmelt. The separate sonic ranger on the stake (8 m), constructed from 40 mm carbon fibre tubing and typically drilled 6–7 m into the ice, does record any sort of accumulation and ablation (Fig. ). In addition to sensor-related uncertainties, occasional complications arise when a stake assembly melts out and falls over. The data logger enclosure also includes the Iridium modem and GPS receiver. A single-frequency GPS receiver is used to measure the position and elevation of each station to determine e.g. ice flow velocity (Fig. , item 7). The Iridium antenna is mounted on the boom (Fig. , item 3) to ensure optimal satellite reception. The PTA bladder box (Fig. , items 8 and 10) is mounted on the mast approximately 1.5 m above the ice surface, with any spare or melted-out hose resting on the surface and the remaining hose drilled into the ice, measuring ice melt. Figure  illustrates the free-standing AWS tripod, which moves downward as the ice surface ablates, while the hose itself melts out of the ice. This process reduces the hydrostatic pressure exerted by the liquid column over the transducer, allowing direct calculation of ice ablation. The power system includes rechargeable batteries connected to a relatively small solar panel without a charge controller (Fig. , items 6 and 9). The solar panel is mounted on the mast, facing south, and positioned well above the ice surface to prevent winter snow accumulation from covering the panel. The thermistor string (Fig. , item 11) is drilled 10 m into the ice.

All stations on bedrock uses the tripod setup without a pressure transducer assembly, a thermistor string, and a stake assembly. AWS KAN_B additionally includes a rain gauge of type Geonor T200B for precipitation measurement. KAN_B is the only station equipped with a Geonor T200B, whose rain-gauge setup differs from the other stations and is not part of the standard configuration.

There are multiple types of L0 data collected from the AWS data loggers, while the formats can vary depending on local installations and logger programs.

Raw data: Recorded every 10 min and retrieved from the data logger during maintenance visits. This can either be retrieved directly from the memory card or downloaded from the data logger.

The pipeline supports all formats with raw data, when available, in favour of transmissions. Transmissions cover the period since the latest visit and serve as a fallback in case of missing or corrupted raw data. New transmission data is processed in near-real-time every hour, with a latency of approximately five minutes between transmission and production-ready data.

AWS data are processed by the pypromice Python package (https://github.com/GEUS-Glaciology-and-Climate/pypromice, last access: 28 May 2026) (version 1.5.1), a peer-reviewed suite of algorithms in a standardised workflow for transforming original AWS data (hereafter referred to as L0) to a processed, finalised, user-ready L3 data product (Fig. ). The pypromice package is available via pypi and conda-forge for easy deployment and contains full-coverage unit testing to ensure continuous integration and compatibility across versions and updates. Pypi is the official repository for Python packages, while conda-forge is a community-maintained channel that provides conda-compatible packages across platforms.

This section provides an overview of the derived and corrected variables included in the dataset. It outlines the calculations used to generate these variables and presents them in a structured dataset variables table. Methods for deriving new variables or correcting existing ones are described, ensuring transparency and reproducibility of the data processing steps. In the available netcdf files, the long variable names and a dedicated attribute indicate whether a variable is a direct measurement or calculated in post-processing. This metadata is also summarized in a CSV file “variable.csv” distributed along with the data .

The specific humidity q (in kgkg-1) is calculated from the relative humidity RH (with respect to water or ice, depending on the temperature) using the following equations: 6q=RH100⋅qsat, where 7qsat=εesatp-(1-ε)esat.

In these equations, ε=0.622 is the ratio of the specific gas constants for dry air and water vapor, p is the air pressure (in Pa), and esat is the saturation water vapor pressure (in Pa) over either ice (for below freezing) or water (for above freezing), as calculated following .

The surface temperature Ts (in °C) is calculated using Stefan–Boltzmann law by using the measured downward and upward longwave irradiance (LRin and LRout, respectively) with the following equation: 8Ts=LRout-(1-ε)LRinε⋅5.67×10-80.25-T0, where the ice sheet surface emissivity is assumed to be ε=0.97 and T0 = 273.15 K.

With key surface meteorological values known, such as temperature from longwave radiation, saturated humidity, and zero wind, turbulent heat flux gradients can be calculated without a second sensor boom. The sensible heat flux (SHF) and latent heat flux (LHF), expressed in (Wm-2), are estimated from vertical gradients in wind speed, potential temperature, and specific humidity between the instrumented boom height and the surface, following the method described by and . Based on Monin–Obukhov similarity theory, SHF and LHF are approximated as: 9SHF=ρCpκ2uln⁡zuz0-ψuT-Tsln⁡zTz0,T-ψT,10LHF=ρLs/vκ2uln⁡zuz0-ψuq-qsln⁡zqz0,q-ψq, where ρ denotes the air density, and Cp = 1005 JK-1kg-1 is the specific heat capacity of air at constant pressure. The latent heat of sublimation and evaporation are Ls = 2.83 × 10⁶ Jkg-1 and Lv = 2.50 × 10⁶ Jkg-1, respectively. The von Kármán constant is κ=0.4. Positive fluxes contribute energy to the surface, whereas negative fluxes withdraw energy from it.

To estimate turbulent heat fluxes, we require measurements of the following variables at given heights: wind speed (zu), temperature (zT), and specific humidity (zq). Additionally, we need the surface roughness lengths for momentum (z0), heat (z0,T), and moisture (z0,q). A constant value of z0 = 0.001 m is used, while z0,T=z0,q is calculated based on the formulation for rough surfaces by . Atmospheric stability corrections are applied using the functions ψu,T,q from for stable conditions and from for unstable conditions. Surface temperature (Ts) is derived from longwave radiation (see Eq. 8), and the surface specific humidity is assumed to be at saturation, i.e., qs=qsat.

Several sources of uncertainty affect the calculation of sensible (SHF) and latent heat fluxes (LHF). The aerodynamic surface roughness length z0 varies with surface type and over time . Assuming a constant value of z0 = 0.001 m may overestimate surface roughness in snowy conditions and thus lead to overestimations of both turbulent fluxes.

showed that one- and two-level methods underestimate downward latent heat flux under extreme stability. found similar biases in sensible heat flux, with one-level methods offering longer records. highlighted the use of unrealistically high surface roughness lengths (z0) to match surface energy balance closure with melt-driven ablation rates during intense heat flux events. Turbulent heat flux estimates over ice and snow are uncertain due to assumptions of surface homogeneity, stable polar boundary layers limiting turbulence, surface variability, scarce measurements, and sensitivity to temperature errors. Thus, these estimates require cautious interpretation.

Tilt correction of solar radiation follows the method outlined by , which is also described by . Downward shortwave radiation (SRin) is composed of both diffuse and direct beam components, but only the direct beam component requires correction for surface tilt. For a horizontal radiation sensor, the direct beam component, equivalent to SRin, is reduced by the diffuse fraction (fdif). For a tilted sensor, SRin is derived from the measured radiation (SRin,m) using a correction factor C, as follows: 11SRin,cor=SRin,mC1-fdif+Cfdif, with 12C=cos⁡(SZA)⋅(sin⁡(d)sin⁡(lat)cos⁡(ϕsensor)-sin⁡(d)cos⁡(lat)sin⁡(θsensor)cos⁡(ϕsensor)+cos⁡(d)cos⁡(lat)cos⁡(θsensor)cos⁡(w)+cos⁡(d)sin⁡(lat)sin⁡(θsensor)cos⁡(ϕsensor)cos⁡(w)+cos⁡(d)sin⁡((θsensor)sin⁡(ϕsensor)sin⁡(w))-1, where SZA is the solar zenith angle, d is the solar declination (the angle between the Sun and the Earth's equatorial plane), ω is the hour angle (the angular distance between the Sun's current position and solar noon), lat is the site's latitude in radians, and θsensor and ϕsensor represent the radiometer's tilt angle and azimuth orientation, respectively. The procedures for calculating d (solar declination), ω (hour angle), and SZA (solar zenith angle) are found in . Table  presents the average bias or correction applied to incoming solar radiation based on Eq. (). For most AWS stations, the standard deviation shows that the average correction is small, typically less than 15 Wm-2, although a few stations exhibit a broader range of correction values. We estimate the diffuse fraction (fdif) to range from 0.2 under clear-sky conditions to 1.0 during overcast skies, assuming a linear relationship with cloud cover fraction, as described by .

To approximate the cloud cover fraction, we rely on the relationship between near-surface air temperature (Tair) and downward longwave radiation (LRin), following . Specifically, we compute the theoretical clear-sky downward longwave radiation flux using the formula proposed by : 13LRclear=5.31×10-14⋅(Tair+T0)6, where LRclear is the clear-sky longwave radiation flux (in Wm-2) and Tair is the near-surface air temperature (in Kelvin). This allows us to estimate the cloud cover fraction by comparing observed longwave radiation to the clear-sky baseline.

Theoretical downward longwave radiation under overcast conditions is estimated by assuming black-body emission from a cloud base at the near-surface air temperature. This is calculated using the Stefan–Boltzmann law: 14LRovercast=5.67×10-8⋅(Tair+T0)4, where LRovercast is the overcast longwave radiation flux (in Wm-2), Tair is the near-surface air temperature (in °C), and T0 is the conversion offset to Kelvin (273.15 K).

The cloud cover fraction (cc), constrained within the range [0,1], is then estimated by linearly scaling the observed longwave radiation between clear-sky and overcast conditions: 15cc=LRin-LRclearLRovercast-LRclear=fdif-0.20.8.

The cloud cover estimation is only valid over ice and snow surfaces, and therefore is not computed for stations installed on bedrock.

Surface broadband solar reflectivity in the 0.3–2.5 µm wavelength range, commonly referred to as albedo (unitless), is derived from 10 min tilt-corrected measurements of downward and upward solar irradiance. Hourly albedo values are computed when the solar zenith angle is under 70° (i.e., when the sun is more than 20° above the horizon), ensuring optimal measurement reliability for the pyranometer. Daily mean albedo values are then calculated from the valid hourly data. Shadows cast by AWS components, such as the mast or sensor arms, together with surface contrast with AWS infrastructure (e.g., solar panel, battery box, legs, enclosure), and the presence of features such as melt ponds beneath the station can reduce observed albedo values by up to 0.03 on average, depending on surface type and snow surface height . This bias source is variable with snow surface height, effectively zero when snow thickness exceeds 1.5 m. compared ablation area AWS albedo measurements with unmanned aerial vehicle (UAV)-derived and satellite-based albedo products, finding increasing discrepancies during the late melt season due to spatial inhomogeneity and limited representativeness of point measurements. While reported a 5 % uncertainty for pyranometer-based albedo measurements, the instrument manufacturer (Kipp & Zonen) suggests a more conservative estimate of 10 %, adopted here for the calculated albedo values.

The pressure transducer assembly (PTA; Fig. ) is sensitive to fluctuations in atmospheric pressure, which can influence the measured signal HM. To correct for this effect, the contribution of air pressure is removed using the following expression: 16HL=HM+PC-PAgρl, where PA (in hPa) represents the ambient air pressure, while PC (in hPa) is the reference pressure specified by the manufacturer during sensor calibration. The gravitational acceleration is assumed to be g = 9.82 ms-2, and the density of the antifreeze mixture used in the system is ρl = 1090 kgm-3, for a temperature of 0∘C. Cumulative variations in the corrected liquid level HL directly correspond to ice surface ablation. validated PTA-derived ablation measurements against manual hose-based readings and sonic ranger data, and found the PTA measurements to be accurate within ± 0.04 m.

Following , we correct liquid precipitation measurements for undercatch using wind speed (U). We apply the undercatch correction factor (k) for an unshielded Hellmann-type gauge under liquid-only precipitation conditions, using the catch-efficiency relation from : 17k=100100-4.37U+0.35U2, where U is wind speed (in ms-1) at measurement height. derived a well-tested wind-based correction for unshielded Hellmann-type gauges using extensive World Meteorological Organization intercomparison data, showing that wind speed is the primary driver of undercatch. Because our gauge type matches theirs, and the method performs robustly across varied climates, their relationship is an appropriate choice. Some uncertainty remains, as wind alone cannot explain all undercatch variability and the correction was originally derived from daily mean wind speeds up to 6.5 ms-1, but is assumed to be applicable to hourly wind speed data. For wind speeds exceeding 6.5 ms-1, an extrapolated correction is applied . As with all automated precipitation measurements, considerable uncertainty persists in the corrected values, as wind speed alone does not fully account for the observed undercatch.

Moreover, only rainfall is considered in this undercatch correction by excluding measurement periods where air temperature is below -2 °C. However, this does not eliminate the affect of delayed snow melt errors, when snow accumulates in the gauge and is only registered as precipitation as it melts into the tipping bucket. Such instances can occur during short atmospheric warm spells within otherwise sustained below-freezing conditions during winter. As a result, corrected rainfall should be used and interpreted with caution. In addition, no corrections are applied for evaporation, wetting, or trace precipitation.

The datasets are distributed with a comprehensive set of metadata, following the Climate and Forecast (CF, ) conventions and the Attribute Convention for Data Discovery (ACDD). In addition, specific attributes are included to capture station- and site-levels details relevant for interpretation, reuse and reproducibility. For example, the specific attribute site_type is added to describe the environment type of the installation site (e.g., ablation, accumulation, or tundra).

The data variables are CF-compliant according to CF-1.7 and use an updated naming convention relative to our earlier products. The L3 hourly datasets contain a full set of data variables from our processing pipeline and are summarised in Table . Many variables are measured at both the upper and lower boom in cases where stations or sites follow the accumulation area two-boom station design. In addition, instantaneous measurements are provided for key variables (air temperature, air pressure, relative humidity, wind speed, and wind direction), whereby instantaneous measurements are recorded at the top of each hour. For stations or sites with the ablation area one-boom station design, variables are assigned as upper boom measurements, with no corresponding lower boom values provided.

The datasets are provided in both NetCDF and CSV formats. The data itself is unchanged between these two versions, however, the NetCDF format includes metadata and variable attributes which better inform about the collection and quality of the data. We therefore recommend users adopt the NetCDF format where possible.

An upper and lower threshold is adopted to filter out erroneous measurements (Table ). These thresholds are informed by the instrument upper and lower measurement capabilities, commonly documented by the instrument manufacturers (see Appendix). Measurements exceeding these limits are flagged as outliers and removed from subsequent analysis. These thresholds are designed to reflect realistic environmental conditions and are adapted to local site characteristics. Figure  illustrates an example of how these thresholds are applied to a time series, highlighting the removal of values that fall outside the accepted range.

The rate of change (ROC) between consecutive measurements is used to detect anomalous values in air temperature, air pressure, relative humidity, and subsurface temperature. Initial testing showed that a fixed ROC threshold is insufficient: a high threshold fails to capture outliers, while a low threshold removes valid observations during periods of naturally high variability.

To address this, we compute both forward and backward ROC for each variable and derive a dynamic threshold based on local variability. For each time step, the 95th percentile of the ROC is calculated within a 1 week rolling window (i.e. ± 3.5 d), separately for forward and backward differences: ROC95%fwd(t),   ROC95%bwd(t).

These percentiles represent the typical variability at that time, taking into account seasonal and site-specific conditions.

A value is flagged as a potential outlier if both its forward and backward ROC exceed a variable-specific multiple of the corresponding threshold: ROCfwd(t)>f⋅ROC95%fwd(t) AND ROCbwd(t)>f⋅ROC95%bwd(t), where ROCfwd(t) and ROCbwd(t) denote the forward and backward ROC values at time step t. The factor f is variable-specific, e.g. f=2.2 for air temperature and f=3.5 for relative humidity (Table ). Figure  shows the ROC filter applied to NAE air temperature data. In the first part of the series, the 10 min data appear outlier-free, although forward- and backward-looking ROC values often exceed their thresholds (Fig. a–c, left of the gray line). In the second half, the hourly data contain clear outliers and exhibit several threshold crossings (Fig. a–c, right of the gray line). Thanks to adaptive thresholding and reassessment of flagged samples via linear interpolation, the ROC filter flags only a few early values while correctly identifying the obvious outliers later in the record (Fig. d).

When a time step is adjacent to a data gap, the AND condition above is relaxed to an OR, since only one-sided information is available. At the final time step of the time series (e.g. for incoming transmissions), only the forward condition is applied, with thresholds computed from preceding data.

To avoid removing valid rapid variations, all flagged values are re-evaluated using linear interpolation between neighboring valid measurements. If the observed value lies within a variable-specific tolerance of the interpolated estimate, the flag is removed.

To detect sensor or data logging malfunctions that result in unchanging measurements, a persistence-based filter is applied as part of the quality control. This filter is designed to identify and flag periods where values remain constant over time, a typical symptom of readout failures or stuck sensors. Persistence filtering targets a known behavior of some logger programs: when a sensor readout fails, the system may fall back to returning the last successfully measured value. If the issue persists, the output becomes artificially constant for hours or days. Figure  shows an example of persistent relative humidity readings from station CP1 in January 2022. The red line shows the values before the persistence filter and black line shows after.

At times, manual intervention is required when it is known that the recorded data does not represent the actual conditions at the station. This can occur in situations such as sensor malfunction, the sensor being covered by snow, frost, or rime, or the station becoming tilted or moved during maintenance. Data collected during these periods can either be flagged and removed from the dataset or adjusted using a predefined set of supported operators.

Manual quality control is implemented as an asynchronous and collaborative process based on a public GitHub repository https://github.com/GEUS-Glaciology-and-Climate/PROMICE-AWS-data-issues (last access: 12 November 2025) that allows both internal and external users to contribute by either raising a data issue here, or proposing their own adjustments to the dataset. This is reviewed by a member of the PROMICE | GC-NET AWS data team. Flagging and adjustment rules defined in this repository are integrated into the pypromice production pipeline, where they are applied to the data products on an hourly basis.

Using the daily data product, AWS data coverage was assessed using a “success rate”, defined as the ratio of days with valid daily averages for all variables required to estimate the surface energy budget (air pressure, air temperature, humidity, wind speed, and downward and upward shortwave and longwave radiation) to the total days since AWS installation. Performance for these critical variables by site and measurement period is shown in Fig. . The historical GC-NET dataset lacks full surface energy balance coverage due to differing instrumentation from all accumulation area stations .