...

History of modifications

List of datasets covered by this document

Acronyms

General definitions

Aquifer: An aquifer is a subsurface layer of permeable rock, sediment, or soil that stores and conveys water, usually referred to as groundwater. Aquifers are classified as either confined or unconfined depending on their geological setting. An unconfined aquifer is open to the atmosphere through permeable material above it, allowing the water table to rise and fall freely in response to recharge and discharge. A confined aquifer is bounded above by a layer with low or no permeability, causing the groundwater within it to be under pressure greater than atmospheric. This could result in so-called artesian conditions where groundwater flows upwards under pressure when the aquifer is penetrated by a well.

...

Uncertainty: Uncertainty measures jointly the precision and accuracy of the GWSC data, expressed as a standard deviation. The uncertainties of the GWSC product are influenced, among others, by the uncertainties of the input data streams (TWSA, SWE, RZSM, GM, and SWS), which are propagated to the final product. Spatially and temporally varying uncertainty estimates are provided alongside the GWSC data in the variable gwsc_uncertainty.

Executive summary

The Copernicus Climate Change Service (C3S) Groundwater Storage Change (GWSC) data product provides global monthly gridded estimates of groundwater storage anomalies derived from satellite gravimetry observations and complementary water storage compartment data. This document presents the quality assessment of the GWSC Climate Data Record (CDR) v1.0, produced in March 2026. The data are provided on a 0.5° × 0.5° grid and cover the period from April 2002 to September 2023, with some months missing within this timespan due to the operational characteristics of the GRACE and GRACE-FO satellite missions. For further details on the product, please refer to the GWSC Algorithm Theoretical Basis Document (ATBD) and Product User Guide and Specification (PUGS).

...

Overall, the validation results indicate that GWSC v1.0 is a robust and scientifically sound product. The global signal-to-noise ratio (mean 1.79) confirms that the deterministic groundwater signal generally exceeds residual noise. Comparison against independent in situ groundwater observations in three aquifers shows moderate to strong agreement (Pearson r = 0.63 to 0.81), capturing both long-term trends and seasonal dynamics. Comparison against global hydrological models, by contrast, performed poorly. However, this is probably attributed to known limitations in how these models represent groundwater and related processes, rather than to deficiencies in GWSC. The provided uncertainty estimates are conservative but reliably track the temporal variability of the true signal. Validation of the individual input compartments (TWSA, SWE, RZSM, GM, SWS) shows each performs well, with no evidence of instability across the record. In summary, GWSC v1.0 meets most GCOS target requirements and provides a credible estimate of global groundwater storage change, with the caveat that ground-based validation currently remains geographically limited to a few well-monitored aquifers.

Section 1 introduces the product validation methodology, covering the validated product itself (section 1.1), the validation datasets (section 1.2), the validation regions (section 1.3), and a description of the validation methods (section 1.4).

Section 2 presents the validation results. In section 2.1 the signal-to-noise ratio of the product is addressed. In section 2.2, GWSC is evaluated against in situ groundwater level observations from three well-monitored aquifers representing different hydrogeological and climatic settings: the Paris Basin (Europe), the High Plains Aquifer (North America), and the Guarani Aquifer System (South America). The in situ data were obtained from the Global Groundwater Monitoring Network (GGMN) portal and converted to groundwater level anomaly time series for comparison with GWSC estimates. Section 2.3 compares the GWSC product to groundwater storage estimates from an ensemble of global hydrological models. Section 2.4 validates the uncertainty estimates provided with the GWSC product. Section 2.5 assesses the quality of the individual input data streams, i.e. TWSA, GM, SWE, RZSM, and SWS, drawing on the respective product quality assessment reports where available, and providing a dedicated validation of the SWE input. Section 2.6 discusses the stability of the GWSC product based on the stability assessments of the individual input data streams.

The document concludes with section 3, which addresses compliance with Global Climate Observing System (GCOS) user requirements.

Product Validation Methodology

Anchor
product_validation_methodology product_validation_methodology

Validated products

Anchor
validated_products validated_products

The C3S GWSC data product provides monthly gridded groundwater storage change fields from April 2002 to September 2023. GWSC is derived from observations of the GRACE (2002–2017) and GRACE-FO (since 2018) satellite missions, combined with estimates of the remaining terrestrial water storage compartments. The primary input to the GWSC production system is the C3S Terrestrial Water Storage Anomaly (TWSA) v1.0 product, which is itself based on spherical harmonic coefficients provided by the Combination Service for Time-variable Gravity Fields (COST-G). The remaining input data streams consist of anomalies of glacier mass (GM), snow water equivalent (SWE), root zone soil moisture (RZSM), and surface water storage (SWS). Figure 1 provides a flowchart detailing the processing steps taken to produce the GWSC CDR v1.0, which is validated in this Product Quality Assessment Report (PQAR). A detailed description of the product generation is provided in the GWSC Algorithm Theoretical Basis Document (ATBD), and further information on the product's usage and limitations is given in the GWSC Product User Guide and Specification (PUGS).

...

Figure 1: Overview of the processing steps in the GWSC production system. Input Water Storage Compartment (WSC) data streams (TWSA, SWS, SWE, RZSM, GM) and their uncertainties are converted to anomalies, resampled to a common 0.5° grid under a consistent land-ocean mask, filtered with a 250 km Gaussian filter, whose width is informed by a spatial correlation analysis, and finally combined through the subtraction approach. Uncertainties are propagated throughout the chain to the final GWSC product.

Validation datasets

Anchor
validation_data_sets validation_data_sets

Two different validation datasets are used for the quality assurance of the C3S GWSC product. First, the GWSC dataset is evaluated against in situ groundwater level observations from the Global Groundwater Monitoring Network (section 1.2.1). Second, it is compared to an ensemble of global hydrological model outputs (section 1.2.2). 

 In situ groundwater observations

Anchor
twsa_data twsa_data

Ground-based groundwater level measurements are used to evaluate the GWSC product against independent observations. The data were sourced from the Global Groundwater Monitoring Network (GGMN) portal, a collaborative data sharing initiative operated byInternational Groundwater Resources Assessment Centre (IGRAC) that aggregates groundwater monitoring records from national and sub-national agencies worldwide (IGRAC, 2025). For each of the three evaluation aquifers, data were provided by the responsible national survey: the French Geological Survey (BRGM) for the Paris Basin, the United States Geological Survey (USGS) for the High Plains Aquifer, and the Geological Survey of Brazil (SGB) for the Guarani Aquifer System.

Prior to use, all records were resampled to monthly resolution and subjected to outlier removal. Wells were retained if they contained no more than 20% missing values over the full study period (April 2002 to September 2023) and no consecutive data gap exceeding six months. Data for the Guarani Aquifer System were only available from October 2011 onwards. Gaps in otherwise retained time series were filled by linear interpolation. These selection criteria reduced the initial pool of 6,070, 1,872, and 95 boreholes for the Paris Basin, the High Plains Aquifer, and the Guarani Aquifer System to final validation sets of 440, 53, and 32 wells, respectively.

Global Hydrological Model (GHM) outputs

Anchor
hydro_data hydro_data

The following GHM model outputs from the Inter-Sectoral Impact Model Intercomparison Project (ISIMIP) simulation round 3b are used to validate the GWSC product. All ISIMIP data is freely available from the ISIMIP repository.

Water GAP Global Hydrological Model

Community Water Model (CWatM)

Hanasaki 2008 (H08)

Preprocessing

ISIMIP data is already available in the format comparable to the GWSC product, i.e. monthly fields with timestamp set to the first of each month and in a 0.5° x 0.5° regular grid. Due to the unique characteristics of the spatial resolution of the TWSA dataset (see sections 1.1.2 and 1.4.5 of the TWSA PUGS), which is inherited by the GWSC product, a grid-cell-wise validation is not advisable. Consequently, validation will be done only for regionally aggregated time series. For each model the datasets are concatenated to cover the period of interest from April 2002 to September 2023.

Validation regions

Anchor
validation_regions validation_regions

For the validation of the GWSC dataset, we employ three distinct validation regions:

Aquifers 

Anchor
aquifers aquifers

The evaluation of GWSC against in situ groundwater level observations is carried out in three major aquifers: the Paris Basin (France), the High Plains Aquifer (United States), and the Guarani Aquifer System (Brazil). These aquifers were selected because they are comparatively well monitored with several groundwater wells and span a range of hydrogeological settings and climatic conditions. Location and extent of the aquifers, and groundwater well locations are shown in Figure 2.

The Paris Basin is a multi-layered sedimentary aquifer system in northern France, covering approximately 250,000 km², with a mix of unconfined and confined portions. The basin experiences a temperate oceanic climate with annual precipitation of around 650–900 mm and has been extensively used for irrigation and water supply.

...

Figure 2: Properties of the three aquifers (Paris Basin, High Plains Aquifer, Guarani Aquifer System) and the location of the groundwater wells used for validation.

River basins

We use the river basin outlines provided by the"GRDC Major River Basins of the World" (GRDC, 2020, https://www.kliwas.de/SharedDocs/ExterneLinks/GRDC/mrb_shp_zip.html, URL last accessed 19/05/2026), which incorporates data from the HydroSHEDS database (Lehner, 2013, http://www.hydrosheds.org, URL last accessed 08/10/2025). Due to the limited spatial resolution of the TWSA datasets — approximately 300 km — we restrict the analysis to basins with an area greater than 100,000 km², and then rasterise the polygons onto a grid. The resulting 121 basins are presented in Figure 3.  Appendix A provides more detailed information on these basins, including their MRBID (GRDC-issued basin ID), basin name, and the continuous ID used throughout this document.

...

Figure 3: Validation regions used for the assessment of the C3S GWSC dataset: Outlines of 121 major river basins.

Desert regions

We define four desert regions on the world’s continents, characterised byvery low mean annual precipitation of less than approximately 100 mm/year, based on climatological data from the Global Precipitation Climatology Centre (GPCC; Schneider et al., 2014).

...

• the Sahara Desert, subdivided into East and West Sahara,

• the Arabian Peninsula, and

• theGobi Desert in Central Asia.

Figure 4 shows the outlines of these regions.

...

Figure 4: Validation regions used for the assessment of the C3S GWSC dataset: Defined desert regions: the West and East Sahara, the Arabian Peninsula, and the Gobi Desert.

Description of Product Validation Methodology

Anchor
validation_method validation_method

This section provides details about the validation methods used in this document. The validation methods of the GWSC dataset are then split into five major parts:

1. Assessment of the signal-to-noise ratio of the GWSC product in section 1.4.1,
2. Validation methods to compare the GWSC dataset against observations in section 1.4.2,
3. Validation methods to compare the GWSC dataset against the GWSC output of global hydrological models in section 1.4.3,
4. Validation methods for the GWSC uncertainties in section 1.4.4, and
5. Validation methods for the SWE compartment (section 1.4.5).

Quality assessment and validation of the remaining WSCs are presented in the corresponding PQARs and summarized in section 2.5.

Signal-to-noise ratio 

Anchor
signal-to-noise signal-to-noise

This validation measure of the GWSC dataset is based on the assumption that the ratio between signal content and noise level should be maximised. To this end, the signal-to-noise ratio (S2N) is computed a) for each of the major river basins described in section 1.3.2. and b) as a global mean S2N from all grid cells containing data. The signal content is defined as the standard deviation of the full signal gwsc(t). Due to the unavailability of a noise component in the GWSC dataset, the noise is defined as the standard deviation of the residual signal gwsc_red(t) after signal decomposition following eq. (1), i.e., after the deterministic signals (trend, annual, and semiannual) have been removed. 

...

\[ S2N = \frac{std(gwsc(t))}{std(gwsc_{red}(t))}. \quad (2) \]

Comparison to observations

Anchor
comp_obs comp_obs

The evaluation of GWSC against in situ groundwater level observations is carried out following closely the methodology in Güntner et al. (2026) and is done for the aquifers described in section 1.3.1.

The in situ groundwater level time series are converted to groundwater level changes (GWLC) by removing the long-term mean level from each well record, making them conceptually comparable to the GWSC estimates. To aggregate the point-scale well observations to the spatial scale of the GWSC grid, wells are assigned to the 0.5° grid cell in which they are located. Where multiple wells fall within the same grid cell, a cell-level average is computed to avoid spatial bias from clustered boreholes. Grid-cell averages are then averaged across all grid cells within a given aquifer to obtain a single aquifer-mean GWLC time series.

...

To account for the fact that in situ groundwater levels and GWSC are expressed in different units and may differ in absolute magnitude due to the unknown specific yield of each aquifer, both time series are standardised prior to comparison by removing the respective mean and dividing by the standard deviation. Agreement between the standardised GWLC and GWSC time series is then assessed using the Pearson correlation coefficient.

Comparison to hydrological models

Anchor
comp_hyd_mod comp_hyd_mod

The comparison between the C3S GWSC dataset and the outputs of the global hydrological models, i.e. modelled GWSC, presented in section 1.2.2 is done analogous to the comparison to observations in section 1.4.2. To make the datasets comparable they are standardised prior to comparison by removing the respective mean and dividing by their standard deviation. Comparison is then done via area averages for the aquifers described in section 1.3.1.

Uncertainty over desert regions

Anchor
uncert_validation uncert_validation

As described in section 1.4.2.1 of the TWSA PQAR, the open-ocean noise level is commonly used in the GRACE data processing community to assess the relative quality of different datasets and is seen as a proxy for data uncertainty. Since the GWSC dataset does not contain data over the oceans, we use the noise level, expressed as the standard deviation of the residual signal, over desert regions as a substitute. As with the ocean, it is assumed that after removing the trend and seasonality, no significant signal remains. Although hydrological signals over deserts are generally weak, some interannual variability does occur, making the desert noise level systematically slightly larger than the open-ocean noise. Nevertheless, because this effect is consistent across all datasets, it does not affect their relative comparison.

The residual standard deviation σ(t) for each month t after removal of deterministic signals, i.e., trend, annual and semi-annual signals is defined in eq. (3) (twsa_red; see Eq. (1), for details):

\[\sigma(t) = \sqrt{\frac{1}{\sum_{i=1}^a w_i}\sum_{i}^a w_i twsa_{red}(\lambda_i, \theta_i,t)^2} \quad (3)\]  

where a represents the number of included grid points, and wᵢ is the area weight of the i-th grid cell at longitude λᵢ and latitude θᵢ. Grid points included are all grid points covered by the polygons delimiting the deserts in Figure 4. σ(t) is not included in the published dataset, but becomes available during the dataset processing.

For the C3S GWSC product, σ(t) in desert regions is dominated by propagated observational uncertainties rather than by a real hydrological signal. Accordingly, σ(t) serves as an empirical proxy for the true uncertainty, and is compared directly to the provided gwsc_uncertainty.

Validation of Snow Water Equivalent (SWE) dataset 

Anchor
method_swe method_swe

The Snow Water Equivalent (SWE) dataset used in the GWSC production system is a merged product combining satellite-based observations from the C3S with reanalysis model data to achieve global coverage.

...

Model and satellite-based SWE datasets, including the combined SWE product, are validated against ground-based in situ SWE measurements. The in situ data consist of manual gravimetric snow measurements conducted at multiple locations along a pre-defined transect, of which an averaged value was used as representative for the corresponding pixel or grid cell of the model or remote sensing dataset to be validated. In situ SWE observations for validation purposes were available for Russia (Bulygina et al., 2011), Finland (Haberkorn, 2019) and Canada (Brown et al., 2019). Russian in situ data has been made available via the All-Russia Research Institute of Hydrometeorological Information-World Data Centre (RIHMI-WDC). It contains data from routine snow surveys operated at 515 meteorological stations. For validation of the products used for GWSC production, data from 2002 to 2018 has been used. The Finnish Environment Institute (SYKE) has a network of about 160 snow survey courses (Haberkorn, 2019). From these data from 2011-2021 has been used here. For Canada, data are sourced from CanSWE v6 (Vionnet et al., 2021). The Canadian dataset spans 1979–2024. A list of all the access points for the individual snow course measurements used for this validatom is also given in Appendix B.

Figure 5 shows the spatial distribution of snow course measurements available for this validation exercise. The results of the validation activity are summarised in section 2.5.

...

Figure 5:  Location of available snow course measurements over Canada, Russia, and Finland

Validation results

Anchor
validation_results validation_results

Results of the Signal-to-Noise Ratio

Signal-to-noise ratio (S2N) is computed for each of the 121 major river basins and as a global mean from all grid cells containing data. The resulting basin-wise S2N values are shown in Figure 6. The global mean gridwise S2N is 1.79 with a standard deviation of 0.83, indicating that on average the deterministic signal content of the GWSC product exceeds the residual noise.

...

Figure 6: Signal-to-noise ratio in the 121 river basins >100,000 km².

Results of the comparison to observations

Anchor
res_obs res_obs

GWSC is evaluated against in situ groundwater level changes (GWLC) for three major aquifers spanning different climatic and hydrogeological settings: the Paris Basin (Europe), the High Plains Aquifer (North America), and the Guarani Aquifer System (South America). The aquifer characteristics and correlation results are summarised in Table 1. The standardised GWSC and GWLC time series for the three aquifers are shown in Figures 7, 8 and 9.

...

Paris Basin

Agreement between GWSC and in situ GWLC in the Paris Basin is moderate to strong, with a Pearson correlation coefficient of 0.68 (p << 5%, Table 1, Figure 7). Both datasets capture similar interannual and seasonal variation patterns. The in situ data exhibit slightly larger normalised amplitudes at the seasonal scale, while the GWSC time series shows somewhat more short-term variability, which is consistent with the influence of TWSA uncertainties at this comparatively small spatial scale. A tendency for GWSC to exceed in situ estimates prior to 2013 and fall below them thereafter may reflect a stronger negative long-term trend in the GWSC signal over the Paris Basin.

...

Figure 7: Standardised C3S GWSC and in situ GWLC time series for the Paris Basin. The dark grey period indicates the temporal gap between the GRACE and GRACE-FO missions.

High Plains Aquifer

The correlation between GWSC and in situ GWLC in the High Plains Aquifer is high (r = 0.81, p << 5%, Table 1, Figure 8), reflecting strong agreement in both long-term trends and seasonal patterns. Agreement is somewhat higher prior to 2016; after this point, in situ observations tend to record lower values during periods of overall reduced groundwater storage, which may partly reflect localised groundwater depression cones around pumping wells that are not representative of aquifer-wide conditions at the spatial scale of the GWSC grid.

...

Figure 8: Standardised C3S GWSC and in situ GWLC time series for the High Plains Aquifer. The dark grey period indicates the temporal gap between the GRACE and GRACE-FO missions.

Guarani Aquifer System

GWSC and in situ GWLC in the Guarani Aquifer System show strong agreement (r = 0.81, p << 5%, Table 1, Figure 9), with highly consistent interannual patterns over the available comparison period, which begins in October 2011. Pronounced sub-annual or seasonal storage variations are absent in both datasets, consistent with the behaviour expected of a largely confined aquifer system. Some random noise is apparent in the GWSC time series at monthly scales.

...

Figure 9: Standardized C3S GWSC and in situ GWLC time series for the Guarani Aquifer System. The dark grey period indicates the temporal gap between the GRACE and GRACE-FO missions.

Summary

All three aquifers yield moderate to high correlation between GWSC and in situ GWLC (r = 0.63–0.81), with the High Plains Aquifer and the Guarani Aquifer System showing particularly strong agreement across both unconfined and confined hydrogeological settings. Remaining discrepancies reflect the inherent differences in spatial scale and measurement principle between satellite gravimetry and borehole observations, as well as localised anthropogenic influences such as pumping effects that are not resolved at the GWSC grid scale.

Results of the comparison to global hydrological models 

Anchor
res_mod res_mod

Comparison of time series of standardised C3S GWSC with standardized GWSC from 3 GHMs in the 3 aquifers showed very poor results, going as far as complete phase shifts in the annual signal. Comparing these findings to the very good results from validation against observations presented in section 2.2 we conclude that a comparison with modelled GWSC from GHMs is not feasible. Whether this is due to the poor representation of groundwater in these models is not subject to this document. The results are presented in Figure 10 for the readers' convenience and further reference.

...

Figure 10: Standardised C3S GWSC and GWSC from the 3 GHMs in the 3 major aquifers.

Uncertainty validation

Anchor
unc_val unc_val

In Figure 11 the uncertainty provided with the C3S GWSC dataset, gwsc_uncertainty, is compared to the empirical standard deviation, σ, of the C3S GWSC dataset across the four desert regions presented in Figure 4. The provided gwsc_uncertainty exceeds σ by a factor of approximately 2.5. This is expected, as gwsc_uncertainty reflects propagated observational uncertainties from all input data streams (TWSA, SWE, RZSM, GM, and SWS), which are substantial in arid regions even where the actual groundwater signal is small. The empirical σ, by contrast, only captures the residual variability of the final GWSC product itself. The factor of 2.5 is therefore not indicative of overestimation, but rather a consequence of the conservative uncertainty propagation through the subtraction approach. Furthermore, gwsc_uncertainty captures the temporal variability of σ well, suggesting that the provided uncertainty correctly tracks the temporal dynamics of the noise.

...

Figure 11:  Time series of the C3S-provided uncertainty, gwsc_uncertainty, compared with the empirical standard deviation, σ, for the four desert regions.

Validation of input datastreams 

Anchor
val_wscs val_wscs

Snow Water Equivalent (SWE)

The individual and merged SWE datasets are validated against the ground-based in situ snow course measurements described in section 1.4.5. A more detailed validation of the baseline satellite SWE product is available in Snow Water Equivalent (SWE) version 1.0: Product Quality Assessment Report (PQAR).

SWE validation

The reference data is spatially and temporally matched with the modelled SWE grid cells. The mean of all reference observations falling within a SWE product grid cell is then compared to the corresponding co-located SWE estimate. Validation metrics include the unbiased root-mean-squared error (RMSE), bias, and correlation coefficient. Overall comparison with in situ snow course measurements shows that gap-filling satellite remotely sensed SWE with a combination of model SWE of ERA5-Land and MERRA-2 provides a good representation of SWE. The validation results of the final merged SWE product with respect to in situ observations is shown in Figure 12. The figures include annual mean Spearman correlation coefficient and RMSE for each individual SWE product (Figure 12 a and b). Additionally the individual scatter and SWE distribution of in-situ and merged product for the year 2019 are given (Figure 12 c and d).

With a relatively high spearman correlation coefficient of over 0.7 and RMSE below 70mm throughout all years, the combined SWE product used for GWSC production performs quite well for the Northern hemisphere (Figure 12 a +b). Also scatter plots for individual years show good agreement between combined SWE product and groundbased measurements (Figure 12 c). The pixel number distribution for SWE estimates (Figure 12 d) confirms the absence of significant biases and representativeness of the merged SWE product. 

Figure 12: Mean annual Spearman correlation coefficient (12 a) and mean annual RMSE (12 b) between individual SWE products and in situ snow course measurements in the Northern Hemisphere (NH) from Russia, Canada, and Finland. Shown are the C3S satellite-based SWE product (green), MERRA-2 model-based SWE (red dashes), ERA5-Land (orange), and the combined SWE product used for GWSC production, C3SHYDRO SWE (blue).

SWE uncertainties

Satellite remote sensing of SWE provides pixel-level uncertainties according to error propagation through the algorithm described in the Snow Water Equivalent (SWE) version 1.0: Algorithm Theoretical Basis Document (ATBD). No pixel-level uncertainties are provided for the model-based SWE products used for gap-filling. Pixel-level uncertainties for the model-based SWE are therefore calculated based on their RMSE with respect to in situ measurements.

Figure 13 top shows the distribution of model RMSE as a function of SWE magnitude, binned at 25 mm intervals. Each box represents the interquartile range of RMSE values within that bin, with the median shown as a horizontal line. It is assumed that the RMSE derived from model and in situ comparisons provides a reasonable estimate of overall product accuracy. The quadratic fit to the median RMSE distribution then allows the uncertainty to be estimated as a function of the absolute SWE value. A similar technique for deriving pixel-level uncertainties in remote sensing from validation results is described in Sayer et al. (2020).

...

\[ uncertainty (SWE)_{MERRA2} = 6.50+(1.7881 \cdot 10^{-1} \cdot SWE)+ {3.1099}^{-4} \cdot SWE^2 \ \quad (5)\]

Figure 13 bottom shows the dependencies of the RMSE and absolute SWE values, including the corresponding quadratic fits to those curves.

Figure 13: Boxplots of RMSE distributions binned in 25 mm intervals (top); quadratic fit (green dashed line) to the median RMSE (red line), used to derive model uncertainty as a function of SWE magnitude (bottom).

SWE validation summary

Satellite-based PMR products provide reliable SWE estimates across the Northern Hemisphere but lack coverage over mountainous regions and the Southern Hemisphere. Since the GWSC production system requires SWE with full global coverage, gaps in the satellite-based product are filled with a combination of ERA5-Land and MERRA-2 reanalysis data. Validation of the resulting merged product against in situ snow course measurements shows good overall agreement. Only spring and early winter remain challenging seasons in deriving information about snowpacks and SWE. For the underlying satellite-based SWE product and its detection capabilities of the radiometer instrument is also discussed in Snow Water Equivalent (SWE) version 1.0: Product Quality Assessment Report (PQAR). Winter months with SWE values typically above 50mm are well represented by the merged product. The combined product achieves a mean Spearman correlation coefficient of 0.7 across all validation sites, demonstrating its suitability as an input to the GWSC production system.

Terrestrial Water Storage Anomaly (TWSA) summary

Quality of the TWSA product is assessed in the Terrestrial Water Storage Anomaly v1.0: Product Quality Assessment Report (PQAR). The validation is carried out through comparison with independently processed TWSA datasets derived from the same GRACE and GRACE-FO observations, as well as with modelled TWSA from an ensemble of global hydrological models, with assessments covering noise characteristics, time series agreement, trend and amplitude consistency, uncertainty, and product stability (see section 2.6).

The C3S TWSA dataset shows strong consistency with the ensemble of other operational TWSA datasets. In only three out of 121 major river basins does the median normalised difference between the C3S dataset and the ensemble exceed the 1-sigma threshold, and in 67 basins even the third quartile remains below that threshold. For trend and annual amplitude, only three and two basins respectively fall outside the confidence interval defined by the other datasets. The S2N of the C3S TWSA is distributed evenly relative to the other datasets, with no systematic regional pattern, confirming that the C3S product lies within the quality range of other operational solutions.

Comparison with modelled TWSA from six global hydrological models shows strong agreement in basins dominated by annual variability. Larger discrepancies occur in glaciated basins with strong trends, which is expected as none of the hydrological models explicitly account for glacier mass changes. The provided twsa_uncertainty matches the magnitude and temporal variability of the empirical standard deviation over desert regions closely, indicating that the uncertainty estimates are well-characterised.

Root Zone Soil Moisture (RZSM) summary

Quality of RZSM product (and the surface soil moisture product it is based on) is assessed in the Soil Moisture v202505: Product Quality Assessment Report (PQAR). The validation is carried out using in situ measurements from the International Soil Moisture Network (ISMN) and the ERA5-Land reanalysis, with assessments covering spatial and temporal completeness, accuracy, stability, and uncertainty.

...

No additional validation for the 100-200 cm layer also included in the GWSC production system has been done.

Glacier Mass (GM) summary

Quality of the GM product is assessed in the Glacier Mass-Change Product Version WGMS-FOG-2023-09: Product Quality Assessment Report (PQAR). The report validates the product using 32 WGMS reference glaciers through a leave-one-out cross-validation approach, chosen because no fully independent reference dataset exists, i.e. the input data themselves are typically used as validation references. The validation results show a mean deviation (RMSE) of 0.55 m water equivalent (m w. e.) across all 32 glaciers, with performance varying by location: Hintereisferner in the Alps performs best with an RMSE of 0.3 m w.e. and a strong correlation (r=0.93), while Gulkana in Alaska shows a weaker correlation (r=0.45). The product is further benchmarked against three independent global assessments, Zemp et al. (2019), Hugonnet et al. (2021), and GRACE satellite gravity data (Wouters et al., 2019), all showing good agreement. While the product meets most GCOS requirements, its 0.5° spatial resolution falls short of GCOS targets.

Surface Water Storage (SWS) summary

Quality of Open-source (OS) LISFLOOD output that underpins the model-based SWS product has been assessed in Alfieri et al. (2020). The study evaluates the model against observed discharge records at 1226 river stations across 66 countries, using the Kling-Gupta Efficiency (KGE) as the primary performance metric. The calibrated model achieves median KGE scores of 0.67 in calibration and 0.61 in validation, with a median correlation of 0.80, and more than 60% of stations showing a percent bias within ±20%. Performance is strongest in data-rich regions such as Europe and North America, where station density is highest. Africa remains the most challenging region with a median KGE of only 0.02, largely attributable to positive discharge bias and data quality issues in the observed records used for calibration. Stations located downstream of reservoirs also show below-average performance, reflecting the inherent difficulty of capturing complex water release dynamics driven by competing human demands such as hydropower, irrigation, and flood protection. It should be noted that calibration performance is specific to each gauged river section and is not directly representative of neighbouring uncalibrated basins, which retain default model parameter values.

Assessment of the Product Stability 

Anchor
stability stability

Product stability of the GWSC product is assumed to be governed by and propagated from the individual input data streams. Accordingly, no stability assessment of GWSC itself has been done. A discussion of GWSC product stability based on the stability assessments of the individual input data streams follows below.

...

The stability of the subtracted compartments, i.e. SWE, RZSM, GM, and SWS, is likewise relevant to the overall stability of the GWSC product, and their importance varies regionally. In areas where one or more of these compartments exhibit large storage variations relative to groundwater, any uncharacterised drift or inconsistency in those products could become a comparably significant or even more dominant source of instability in the residual than TWSA.

Climate Change Assessment

...

For the GWSC product climate intelligence is produced on 3 levels::  

...

The classification is done by percentiles, with:

• "much below" = below 10th percentile (P10),
• "below" = between 10th and 25th percentile (P10 to Q1),
• "normale" = within interquartile range (Q1 to Q3),
• "above" = between 75th and 90th percentile (Q3 to P90),
• "much above" = above 90th percentile (P90).

Examples for the individual levels are given below.

Level-1

Figure 14 shows the global map for the Level-1 climate intelligence produced from the GWSC product. Maps like these are helpful to assess groundwater conditions on regional and global scale. Well visible, for example, are the deficits in groundwater that followed the multi-year drought in central and eastern Europe, or the impact of human groundwater overuse in northern India. In contrast, south-eastern Australia shows a strong increase in groundwater, which probably stems from the unusually wet conditions starting in 2020 until 2022 linked to three consecutive La Niña years.

Image RemovedFigure 14: Classification of the deviation of GWSC based on percentiles for the year 2022 compared to the reference period January 2003 to December 2022.

Level-2

Figure 15 shows two examples for the Level-2 climate intelligence produced from the GWSC product. Monthly plots like these are particularly helpful to assess conditions with shorter durations or to track the temporal development of conditions. Comparing the same month (September) from the beginning (2002) and the end (2023) of the GWSC dataset the differences in the above-mentioned area in Europe become apparent. As GWSC data comes in anomalies, the region shows conditions (much) above the reference period in 2002 and (much) below in 2023.

Image Removed

Figure 15: Two examples (September 2002 and September 2023) for the monthly classification of the deviation of GWSC based on percentiles compared to the reference period January 2003 to December 2022.
Plots like these are produced for each of the 225 months covered by the GWSC dataset.

Level-3

Figure 16 shows the multi-panel figure that is produced from the GWSC product for the Level-3 climate intelligence. With these plots of area averages it is possible to compare different regions of the world. In the figure it becomes apparent that only tropical South America (3) and tropical Asia (9) have somewhat stable conditions over the entire time series. All other regions have either strongly fluctuating conditions, like Australia (10) with longer dry and wet spells, or clearly visible trends, like northern Africa (5), temperate Eurasia (8) or Europe (11).

...

Output from climate intelligence activities will be added here as it becomes available.

Application(s) specific assessments

Currently, no application(s) specific assessments have been undertaken for the GWSC product version 1.0.

Compliance with user requirements concerning data quality

Anchor
gcos gcos

The product target requirements have been defined by the GCOS (WMO, 2025). Table 2 summarises the requirements and characteristics of the C3S GWSC product.

For each requirement, a threshold and a goal value, for some also a breakthrough value, are defined as listed in Table 2. The goal is the ideal level. Once this is met, further improvements in that criterion are not strictly necessary for most applications. Breakthrough is the level considered a major step towards operational provision. Threshold is the minimum acceptable requirement.

...

Acknowledgements

We would like to thank and acknowledge the modelling groups, the ISIMIP sector coordinators and the ISIMIP cross-sectoral science team for their roles in producing, coordinating, and making available the ISIMIP input data and impact model output.

References

• Alfieri, L., Lorini, V., Hirpa, F. A., Harrigan, S., Zsoter, E., Prudhomme, C., and Salamon, P., 2020. A global streamflow reanalysis for 1980–2018, J. Hydrol. X, 6, 100049, https://doi.org/10.1016/j.hydroa.2019.100049. [URL last accessed 02/06/2026].
• Brown, R. D., Fang, B., & Mudryk, L., 2019. Update of Canadian Historical Snow Survey Data and Analysis of Snow Water Equivalent Trends, 1967–2016. Atmosphere-Ocean, 57(2), 149–156. https://doi.org/10.1080/07055900.2019.1598843. [URL last accessed 02/06/2026].
• Bulygina, O., Groisman, P. Ya., Razuvaev, V., and Korshunova, N., 2011. Changes in snow cover characteristics over Northern Eurasia since 1966, Environ. Res. Lett., 6, 045204. doi.org/10.1088/1748-9326/6/4/045204. [URL last accessed 02/06/2026].
• Burek, P., Satoh, Y., Kahil, T., Tang, T., Greve, P., Smilovic, M., Guillaumot, L., Zhao, F., Wada, Y, 2020. Development of the Community Water Model (CWatM v1.04) – a high-resolution hydrological model for global and regional assessment of integrated water resources management. Geoscientific Model Development, 13, 3267-3298. https://doi.org/10.5194/gmd-13-3267-2020. [URL last accessed 02/06/2026].

• Gelaro, R., McCarty, W., Suárez, M. J., Todling, R., Molod, A., Takacs, L., Randles, C. A., Darmenov, A., Bosilovich, M. G., Reichle, R., Wargan, K., Coy, L., Cullather, R., Draper, C., Akella, S., Buchard, V., Conaty, A., da Silva, A. M., Gu, W., Kim, G.-K., Koster, R., Lucchesi, R., Merkova, D., Nielsen, J. E., Partyka, G., Pawson, S., Putman, W., Rienecker, M., Schubert, S. D., Sienkiewicz, M., and Zhao, B., 2017. The Modern-Era Retrospective Analysis for Research and Applications, version 2 (MERRA-2). J. Clim. 30, 5419–5454. doi.org/10.1175/JCLI-D-16-0758.1. [URL last accessed 02/06/2026].

• Gosling, S.N., Müller Schmied, H., Bradley, A., Burek, P., Gedney, N., Grillakis, M., Guillaumot, L., Hanasaki, N., Ito, A., Kou-Giesbrecht, S., Koutroulis, A., Nishina, K., Otta, K., Qi, W., Sahu, R.-K., Satoh, Y., Schewe, J., 2025. ISIMIP3b Simulation Data from the Global Water Sector. https://doi.org/10.48364/ISIMIP.230418.4. [URL last accessed 02/06/2026].

• GRDC, 2020: GRDC Major River Basins. Global Runoff Data Centre. 2nd, rev. ed. Koblenz: Federal Institute of Hydrology (BfG).
• Güntner, A., Sharifi, E., Haas, J., Boergens, E., Cao, F., Dahle, C., Darbeheshti, N., Dobslaw, H., Dussaillant, I., Dorigo, W., Flechtner, F., Jäggi, A., Kosmale, M., Lasser, M., Luojus, K., Meyer, U., Pasik, A., Preimersberger, W., Ruz Vargas, C., and Zemp, M., 2026. The Global Gravity-based Groundwater Product (G3P), Earth Syst. Sci. Data Discuss. [preprint], https://doi.org/10.5194/essd-2025-797, in review. [URL last accessed 02/06/2026].
• Gurney, K., Law, R., Denning, A., Rayner, P., Baker, D., Bousquet,P., Bruhwiler, L., Chen, H., Ciais, P., Fan, S., Fung, I., Gloor,M., Heimann, M., Higuchi, K., John, J., Maki, T., Maksyu-tov, S., Ken, M., Peylin, P., Prather, M., Pak, B. C., Rander-son, J., Sarmiento, J., Taguchi, S., Takahashi, T., and Yuen,C.-W., 2002. Towards robust regional estimates of CO2sources andsinks using atmospheric transport models, Nature, 415, 626–630, https://doi.org/10.1038/415626a

• Haberkorn, A., 2019. European Snow Booklet – an Inventory of Snow Measurements in Europe, EnviDat. doi.org/10.16904/envidat.59. [URL last accessed 02/06/2026].

• Hanasaki, N., Yoshikawa, S., Pokhrel, Y., Kanae, S., 2018. A global hydrological simulation to specify the sources of water used by humans. Hydrol. Earth Syst. Sci., 22, 789–817. https://doi.org/10.5194/hess-22-789-2018. [URL last accessed 02/06/2026].
• Hugonnet, R., McNabb, R., Berthier, E., Menounos, B., Nuth, C., Girod, L., 2021. Accelerated global glacier mass loss in the early twenty-first century. Nature 592, 726–731. doi: 10.1038/s41586-021-03436-z.
• IGRAC, 2025. The Global Groundwater Monitoring Network (GGMN) [dataset], https://doi.org/10.58154/6Z0Y-DA34. [URL last accessed 02/06/2026].
• Lehner, B., Grill, G., 2013. Global river hydrography and network routing: baseline data and new approaches to study the world’s large river systems. Hydrological Processes 27, 2171–2186. https://doi.org/10.1002/hyp.9740. [URL last accessed 02/06/2026].

• Müller Schmied, H., Trautmann, T., Ackermann, S., Cáceres, D., Flörke, M., Gerdener, H., Kynast, E., Peiris, T.A., Schiebener, L., Schumacher, M., Döll, P., 2024a. The global water resources and use model WaterGAP v2.2e: description and evaluation of modifications and new features. Geoscientific Model Development 17, 8817–8852. https://doi.org/10.5194/gmd-17-8817-2024. [URL last accessed 02/06/2026].

• Müller Schmied, H., Trautmann, T., Ackermann, S., Cáceres, D., Flörke, M., Gerdener, H., Kynast, E., Peiris, T.A., Schiebener, L., Schumacher, M., Döll, P., 2024b. The global water resources and use model WaterGAP v2.2e - daily water storage model output driven by gswp3-era5 and historical setup of direct human impacts. https://doi.org/10.25716/GUDE.17VN-ZP9G. [URL last accessed 02/06/2026].

• Muñoz-Sabater, Dutra, E., Agustí-Panareda, A., Albergel, C., Arduini, G., Balsamo, G., Boussetta, S., Choulga, M., Harrigan, S., Hersbach, H., Martens, B., Miralles, D. G., Piles, M., Rodríguez-Fernández, N. J., Zsoter, E., Buontempo, C., Thépaut, J.-N., 2021. ERA5-Land: A state-of-the-art global reanalysis dataset for land applications, Earth Syst. Sci. Data,13, 4349–4383. doi.org/10.5194/essd-13-4349-2021. [URL last accessed 02/06/2026].
• Sayer, A., Govaerts, Y., Kolmonen, P., Lipponen, A., Luffarelli, M., Mielonen, T., Patadia, F., Popp, T., Povey, A., Stebel, K., Wiltek, M., 2020. A review and framework for the evaluation of pixel-level uncertainty estimates in satellite aerosol remote sensing. Atmos. Meas. Tech., 13, 373–404. https://doi.org/10.5194/amt-13-373-2020. [URL last accessed 02/06/2026].
• Schneider, U., Becker, A., Finger, P., Meyer-Christoffer, A., Ziese, M., Rudolf, B., 2014. GPCC’s new land surface precipitation climatology based on quality-controlled in situ data and its role in quantifying the global water cycle. Theor Appl Climatol 115, 15–40. https://doi.org/10.1007/s00704-013-0860-x. [URL last accessed 02/06/2026].
• Vionnet, Vincent, Mortimer, Colleen, Brady, Mike, Arnal, Louise, & Brown, Ross., 2021. Canadian historical Snow Water Equivalent dataset (CanSWE, 1928-2020) (Version v1), Zenodo, http://doi.org/10.5281/zenodo.4734372
• WMO, 2025. The 2022 GCOS ECVs Requirements, 2022 edition - Updated in 2025, https://library.wmo.int/idurl/4/58111. [URL last accessed 02/06/2026].
• Wouters, B., Gardner, A. S., Moholdt, G., 2019. Global Glacier Mass Loss During the GRACE Satellite Mission (2002-2016). Front. Earth Sci. 7. doi: 10.3389/feart.2019.00096.
• Yoshida, T., Hanasaki, N., Nishina, K., Boulange, J., Okada, M., Troch, P. A., 2022. Inference of parameters for a global hydrological model by applying Approximate Bayesian Computation: Identifiability of climate-based parameters. Water Resources Research, 58, e2021WR030660. https://doi.org/10.1029/2021WR030660. [URL last accessed 02/06/2026].
• Zemp, M., Huss, M., Thibert, E., Eckert, N., McNabb, R., Huber, J., 2019. Global glacier mass changes and their contributions to sea-level rise from 1961 to 2016. Nature 568, 382. doi: 10.1038/s41586-019-1071-0.

Anchor
appendixA appendixA

Appendix A: List of river basins

Appendix B: Snow course measurements for validation of Snow Water Equivalent SWE

Candian SWE data can be accesed from: Canadian historical Snow Water Equivalent dataset (CanSWE, 1928-2025) | Zenodo

US American SWE data can be accesed from:

...

Related articles