the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 13 Oct 2025
An improved GRACE-derived groundwater storage anomaly (igGWSA) dataset over global land with full consideration of non-groundwater components based on current new datasets
Abstract. Accurate quantification of global groundwater storage anomaly (GWSA) is imperative for global water security and socio-economic sustainability. The Gravity Recovery and Climate Experiment (GRACE) satellite has emerged as a prevailing methodology for estimating GWSA. However, oversimplification of non-groundwater components potentially compromised its accuracy in most previous studies. Here we present an improved GRACE-derived GWSA dataset at the global scale, namely igGWSA, with full consideration of non-groundwater components including glaciers, snow, permafrost, lakes, reservoirs, surface runoff, profile soil moisture (PSM), and plant canopy water based on current new datasets. In particular, PSM was generated based on Catchment Land Surface Model and random forest algorithm. igGWSA demonstrated strong agreement with well-observed groundwater level and model-simulated GWSA in five globally recognized hotspots of groundwater depletion. Compared to igGWSA with full consideration, simplified estimation would lead to misinterpretations of groundwater storage variations in glacier-covered regions, giant lakes, and deep-soil areas, highlighting the necessity of comprehensively accounting for non-groundwater components in estimating GWSA, especially under a changing environment. igGWSA dataset is publicly available on Zenodo through https://doi.org/10.5281/zenodo.16871689 (Wang et al., 2025).
- Preprint
(2417 KB) - Metadata XML
-
Supplement
(866 KB) - BibTeX
- EndNote
Status: open (until 13 May 2026)
-
RC1: 'Comment on essd-2025-497', Anonymous Referee #1, 15 Dec 2025
reply
-
CC1: 'Reply on RC1', Zongxia Wang, 16 Jan 2026
reply
Dear reviewer:
Thank you very much for your valuable comments which help to improve the manuscript ESSD-2025-497 greatly.
We have replied the comments one by one and revised the manuscript as well as the supplements accordingly.
For your convenience, the detailed response is provided in the attached ZIP file rather than within this text box.
The attached ZIP file contains five documents:
1. Response_to_Anonymous_Referee_#1_ESSD-2025-497.docx
2. Revised_Manuscript_with_Changes_Marked_ESSD-2025-497.docx
3. Revised_Manuscript_without_Changes_Marked_ESSD-2025-497.docx
4. Revised_Supplement_with_Changes_Marked_ESSD-2025-497.docx
5. Revised_Supplement_without_Changes_Marked_ESSD-2025-497.docx
Thank you again for your valuable comments!
-
AC2: 'Reply on CC1', Suxia Liu, 28 Jan 2026
reply
Just recognized that the reply from authors to the comments raised by Reviewer 1 was wrongly submitted here as CC1. Authors have resubmitted the reply as AC1. please ignore this comment .
Citation: https://doi.org/10.5194/essd-2025-497-AC2
-
AC2: 'Reply on CC1', Suxia Liu, 28 Jan 2026
reply
-
AC1: 'Reply on RC1', Suxia Liu, 28 Jan 2026
reply
Dear reviewer:
Thank you very much for your valuable comments which help to improve the manuscript ESSD-2025-497 greatly.
We have replied the comments one by one and revised the manuscript as well as the supplements accordingly.
For your convenience, the detailed response is provided in the attached ZIP file rather than within this text box.
The attached ZIP file contains five documents:
1. Response_to_Anonymous_Referee_#1_ESSD-2025-497.docx
2. Revised_Manuscript_with_Changes_Marked_ESSD-2025-497.docx
3. Revised_Manuscript_without_Changes_Marked_ESSD-2025-497.docx
4. Revised_Supplement_with_Changes_Marked_ESSD-2025-497.docx
5. Revised_Supplement_without_Changes_Marked_ESSD-2025-497.docx
Thank you again for your valuable comments!
-
CC1: 'Reply on RC1', Zongxia Wang, 16 Jan 2026
reply
-
RC2: 'Comment on essd-2025-497', Anonymous Referee #2, 08 May 2026
reply
My personal view is that this study represents a relatively meaningful contribution to the field. Previous global GRACE-based groundwater studies have often suffered from inaccuracies due to insufficient representation of surface water and other non-groundwater storage components, making it difficult to accurately isolate groundwater anomalies. This work is among the first to address this gap at the global scale by systematically accounting for multiple surface water-related components. The methodology is sound, the validation is robust, and the resulting dataset has significant potential to advance our understanding of global groundwater dynamics and support water security assessments. However, I have several major comments. This work has a fundamental computational error.
Regarding the datasets (Section 2.1) used in this study, I would like to raise the following concern. Even though the GRID_CSR_GRACE_REC dataset from Li et al. (2021) provides a valuable reconstructed product, it cannot replace the actual observations from GRACE and GRACE-FO. The original observations remain the most accurate source of data. In my opinion, the Li et al. dataset should at most be used only to fill the 11-month gap between GRACE and GRACE-FO and possibly a few other missing months, but not as a primary or full substitute for the actual satellite observations.
I have also identified a fundamental computational error in this manuscript that must be corrected. From my reviewing experience, I have noticed that many studies make the same mistake when calculating GWSA from GRACE data—namely, subtracting runoff from TWSA. Unfortunately, the authors have made this same error. In your calculation of GWSA, you subtract SRSA (surface runoff storage anomaly) from TWSA (Equation 1). However, runoff is a flux, not a storage state variable. Both GRACE-derived TWSA and GWSA are state quantities. The correct water storage component that should be subtracted from GRACE TWSA when isolating groundwater is channel water storage anomaly—i.e., the amount of water stored within river channels. For relevant methodological guidance, I refer the authors to Coss et al. (2023), "Channel Water Storage Anomaly: A New Remotely Sensed Quantity for Global River Analysis" (Geophysical Research Letters) , and the recent Nature article "Wide-swath altimetry maps bank shapes and storage changes in global rivers" . I hope that after reading these two papers, the authors will recognize the conceptual error in their current approach and revise their methodology accordingly.
The number of in-situ groundwater level observations used by the authors to validate the igGWSA dataset remains insufficient. Why has a monthly-scale validation not been performed? Furthermore, the 2024 Nature article "Rapid groundwater decline and some cases of recovery in aquifers globally" (Jasechko et al., 2024) should provide a substantial amount of groundwater-level data at the global scale that could serve as a valuable reference for validation purposes. I recommend that the authors consult this study and consider incorporating its data to strengthen their validation efforts.
Citation: https://doi.org/10.5194/essd-2025-497-RC2 -
RC3: 'Comment on essd-2025-497', Anonymous Referee #3, 09 May 2026
reply
This manuscript tries to construct a high-precision, long-term time-series global dataset of groundwater storage anomalies (igGWSA) based on GRACE satellite gravity observations. However, I have many concerns related to the dataset selection, method, and validation.
Major Concerns
- The major concern is the selection of GRACE reconstruction dataset from Li et al. (2021). In their method, the authors first adopted the PCA decomposition method to decompose the original time series. However, this approach tends to overestimate signal amplitude and fails to retain the trend component, which is essential for GWSA analysis. When conducting GWSA estimation based on existing reconstructed products, the absence of such trend information constitutes a critical limitation. Furthermore, with a growing variety of reconstructed datasets becoming available in recent years, the conventional method is no longer adequate for practical application.
- In Lines 115–117, the authors claim that Li’s reconstructed dataset outperforms all previous products. However, Li only concluded that his results are superior to those of Vincent, rather than achieving better performance than the standard GRACE TWS solutions. At present, Li’s dataset cannot be regarded as the optimal product available.
- Various hydrological signals are subtracted to extract groundwater storage anomaly (GWSA) information. Nevertheless, each type of hydrological signal inherently contains uncertainties, which can seriously degrade the accuracy of the final GWSA products. Therefore, a comprehensive and detailed evaluation of such uncertainties is indispensable.
- Regarding the rationale for adopting PSM, the manuscript only explains that GLDAS‑Noah and ERA5 introduce large uncertainties when subtracting soil moisture at fixed depth layers. However, the justification for preferring PSM is insufficient and requires further clarification.
- Lines 176–178 raise a concern regarding the improvement of CLSM-based PSM using machine learning. In Section 3.2, the RF model takes CLSM-simulated PSM as observational input. It is questionable to train the model by regarding simulated PSM data as actual observations. Given that CLSM PSM itself suffers from considerable uncertainty, the physical justification of this processing scheme remains unclear. In addition, it is necessary to clarify why raw PSM data are not adopted directly instead.
- In Section 3.2, a machine learning approach is adopted to generate PSM data. Similar to the previous comment, it remains questionable whether machine learning can produce PSM data with lower uncertainty than the original observations. This study inclines to the view that the RF model merely performs nonlinear smoothing on the data, rather than genuinely reducing the uncertainty level.
- Lines 392–394 show that the CLSM-derived PSM yields the poorest performance among all datasets. This raises an obvious question: why was CLSM PSM initially selected as the training observational input? Would FLDAS-Noah-based PSM not serve as a more appropriate alternative?
- Lines 421–422 show that the uncertainty of igGWSA is 0.56, only 0.03 lower than the 0.59 of GWSA200. Such a marginal improvement cannot be regarded as a notable advancement and may be largely accidental and algorithm-dependent. If a different machine learning algorithm or an alternative dataset were adopted, igGWSA might no longer exhibit lower uncertainty than GWSA200. Furthermore, GWSA200 represents a straightforward and convenient processing approach, and its uncertainty differs only slightly from that of the improved igGWSA. In this case, it remains unclear how to justify to readers the necessity of adopting igGWSA, which integrates multiple datasets.
Specific Suggestions
- The result section in the abstract mentions that “igGWSA demonstrated strong agreement with well-observed groundwater level and model-simulated GWSA in five globally recognized hotspots of groundwater depletion.” This description is too general. Specific numerical values of evaluation metrics should be provided to make the conclusion more convincing.
- Section 3.2.2 only tests “ntree” values of 250, 500, 750, and 1000. The rationale for selecting these parameters should be supplemented. Meanwhile, overfitting tests are recommended to verify whether the model suffers from overfitting.
- For the high-density in-situ monitoring well data in the North China Plain, Central Valley of California, and High Plains of the United States, the well density, heterogeneity processing, outlier removal methods, and other procedures are not specified. A more detailed preprocessing workflow for validation data should be added.
- Section 4.3.2 only compares the relative uncertainties of five sets of GWSAs, without analyzing the uncertainty propagation of input datasets (e.g., glaciers, lakes, etc.). It is recommended to add the uncertainty contribution proportion of major components.
- The random forest model is applied pixel by pixel at the global scale, and its performance is found to vary significantly across climate regions. Zoned modeling is recommended to improve the overall accuracy of the random forest model.
- Section 5.3 only mentions “limited in-situ data” as a limitation, but does not address other limitations such as the accuracy variation across climate regions and incomplete quantification of human activities. It is suggested to supplement the applicable scope and usage recommendations of the dataset.
- Descriptions of legends should be added in the titles of figures and tables to improve their readability.
Citation: https://doi.org/10.5194/essd-2025-497-RC3
Data sets
An improved GRACE-derived groundwater storage anomaly (igGWSA) dataset over global land with full consideration of non-groundwater components based on current new datasets Zongxia Wang, Suxia Liu, and Xingguo Mo https://doi.org/10.5281/zenodo.16871689
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 651 | 269 | 49 | 969 | 141 | 60 | 79 |
- HTML: 651
- PDF: 269
- XML: 49
- Total: 969
- Supplement: 141
- BibTeX: 60
- EndNote: 79
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
In this study titled ‘An improved GRACE-derived groundwater storage anomaly (igGWSA) dataset over global land with full consideration of non-groundwater components based on current new datasets’, the authors consider various surface water changes that are likely to contribute to the GWSAnomaly based on GRACE data. They show that glacier melting, change in global lakes storage among others, contribute to the GWSA signal and that ignoring these contributions could lead to misinterpretation of ground water depletion or recharge trends. The overall content is well suited for this journal. As the scientific validity appears sound, my feedback mostly concerns the readability, organisation and clarification of the manuscript/content. Some remarks:
Specific comments:
LN22: “… would lead to misinterpretations of groundwater storage variations in glacier-covered regions” – …misinterpretation of GW storage? You detail some of the misinterpretations in the last sections but you could also briefly state some here.
LN26: “https://doi.org/10.5281/zenodo.16871689 (Wang et al., 2025)” – The data is provided as a .mat file. Matlab is not freely available, making the data un-accessible to many. Can the authors provide the dataset in an open format; e.g. in .nc or a multi-band .tiff?
LN133: ‘Snow water equivalent simulations from seven reanalysis products’ - which metric is used for the SWE ensemble-wightedaverage used in this study? average, median … ?
LN161-178: ‘2.2.7. Profile soil moisture (PSM) …’ – since you do not present the other datasets/variables with this much detail, this block of text could be well suited for the supplements (only keep a short summary here)
LN180: ‘meteorological variables, and vegetation index’ - You mention the meteorological variables and indices used later in the text (i.e. precipitation and ndvi) but you should also list them here.
LN181: ‘…selected as predictor variables’ - Why was evapotranspiration (which is a main component of the water cycle, thus a likely driver of water storage anomalies) not used as one of the predictors/covariates ? there are many global ET products that are freely available
LN208: Why not use the much simpler nearest neighbour. Unlike bilinear, it does not create new data (i.e. keeps the original data as-is)?
LN219,220: ‘generate ensemble simulations‘ - do you mean to ‘generate the weighted average from the ensemble simulations’?
‘… (Fig. 1, Text S5’ - In Text S5 in the supplementary document you write ‘𝜎𝑖 is the error variance of the 𝑖th dataset’ – 𝜎𝑖 is the standard deviation; change 𝜎𝑖 to 𝜎𝑖2, which is the variance
LN280: ‘To quantify the impacts of incomprehensive considerations of non-groundwater components, five kinds of non-improved GWSA were further estimated as listed below.’ - Seems arbitrary; any justifications for selecting these 5 and not any other combinations ?
LN298: ‘five globally recognized hotspots of groundwater depletion, …’ – reference/citation needed
LN304-306: ‘Accordingly, point-scale data were first converted into pixel-scale by averaging observations of wells located in the specific grid cell. Then in situ GWL and GWSA estimation at a 0.5° X0.5° resolution were upscaled to obtain basin-averaged time series.’ - not enough information for the reader to determine how this was done
LN342: ‘loess and chernozem zones worldwide’ – reference needed
LN372: ‘Given this, evaluation of interannual trends in PSM was carried out additionally’ – grammar: rephrase or remove ‘additionally’
LN380-387: section 4.2.1 - Why do you compare the era5-land-SMS289 (and other SMS) estimates to SMSimproved? Some justification needed here.
LN417: ‘Validation of igGWSA against GWDin situ and GWSAWGHM’ - Since you also compare your igGWSA with the GWSA_WGHM, can you also provide other metrics for a more exhaustive comparison, e.g. bias …
LN419-…: ‘section 4.3.2: Uncertainty analysis’ - recall the metric used to quantify the uncertainty here. Is it GTCH, similar to how uncertainties in PSM are quantified?
LN440: ‘This pattern was found in Region 5, 9, and 12’ - igGWSA in region 12 does not appear to show a decreasing trend
LN447: ‘Therefore, absence of glaciers would inevitably ‘ - not clear …rephrase
LN456-458: interesting observations. Can you expound a bit on this.’ - Interesting. Can you elaborate a bit on this?
Chapter 4: What about permafrost? The authors seem to have left out presenting results on effects of ignoring permafrost when estimating the GWSA
LN522: ‘Mann-Whitney U test…’ - Reference needed or provide a bit more details in the supplements