the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 08 May 2026
Assessing Terrestrial Water Storage Change Since the 1980s
Abstract. Water availability for societies and ecosystems depends upon Terrestrial Water Storage (TWS), yet global, spatially resolved measurements are largely unavailable before the advent of the Gravity Recovery and Climate Experiment (GRACE) gravimetric measurements in 2002. By exploiting a larger set of model and observations-based datasets than previously considered, along with statistical and machine learning techniques, we advance understanding of TWS changes since the 1980s, including accounting for human water management (HWM).
A decline in TWS during 2002–2019 is identified for three global hydrologic models with HWM and bias-corrected precipitation forcing (-0.91 to -0.06 mm yr-1) with only one showing larger decreases than observed by GRACE observations (-0.80 mm y-1). We further identify a longer-term decline in TWS during 1980–2019 in these models, linked with regional precipitation decreases and the net effects of HWM through TWS drawdowns over northern India, southwest U.S. and northeastern China, yet the amplitude of the global land trends remains poorly quantified, ranging from -0.72 to +0.04 mm y-1. Statistical / Machine Learning (ML) reconstructions are found to match GRACE variability but their fidelity in the pre-GRACE/FO period remains unknown.
A stronger decline in TWS since 1980 in the European Centre for Medium-range Weather Forecasts 5th generation reanalysis (ERA5) enhanced land component (ERA5-Land) is linked to an artificial drop in precipitation around 2000–2002 in ERA5 that is most pronounced over equatorial central Africa, northeastern China and the northern Argentina / La Plata region. Our findings urge caution in inferring changes in hydroclimate variables from ERA5-Land and other reanalyses due to inhomogeneities in the assimilated observational data. Continued emphasis on bias corrections to hydrometeorological data and better modeling of HWM are crucial to improving all retrospective analyses of changes in land surface hydrology and terrestrial water stores.
- Preprint
(2334 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on egusphere-2026-2539', Anonymous Referee #1, 31 May 2026
The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2026/egusphere-2026-2539/egusphere-2026-2539-RC1-supplement.pdfCitation: https://doi.org/
10.5194/egusphere-2026-2539-RC1 -
AC1: 'Reply on RC1', Franklin Robertson, 04 Jun 2026
The authors thank the reviewer for providing a variety of suggestions that will lead to the improvement of our paper. The reviewer statement that the analysis is interesting and that the main conclusions appear reasonable is an important assessment.
Two of the most substantive critiques of our paper involve (1) the claim that the choice of title is too broad and (2) the uneven use of the Satellite LASER Ranging (SLR) data. Regarding the title, our survey of four different methodologies for determining terrestrial water storage (TWS) included GRACE determinations, global hydrologic models, a state-of-the-art land reanalysis, and statistical methods. The breadth of these approaches led us to view this work as an “assessment” of current abilities. We acknowledge that our discussion in the paper is more heavily slanted toward the ERA5 land reanalysis. In large part this is because we were able to use precipitation observations / analyses, the ISIMIP3a model experiment results, and records of satellite sensor availability to more deeply assess its performance. However, we did intercompare all of the four methodologies in terms of variability and trends, pointing out the abilities and limitations of each. We suspect that our conclusions section needs to state more clearly that support for a continued downward trend in TWS, however varied in amplitude among the estimates, does constitute an assessment of current estimates by available methods. We will reconsider the title and, pending comments from other reviewers, try to formulate a clearer and more informative version.
In terms of the SLR data, you are correct that the SLR trend should be included in Table 3. Regarding its inclusion in Table 2, the SLR mascons or “footprints” are quite large, covering multiple river basins, which makes them difficult to compare to the gridded data in Figure 2. Furthermore , the SLR data does not extend back to 1980. In any case we could provide a supporting SLR figure in an Appendix to the paper. These changes will be made in a subsequent revision to the present paper.
Further criticisms involve formatting of the manuscript, many acronyms and other elements of style. Revisions to our paper will insure we adhere to HESS style. We will also streamline the presentation for readability. Please note that we did provide a list of abbreviations as noted on page 2.
We’re confident that a subsequent revision, pending comments from other reviewers, will satisfy the concerns raised in your evaluation.
Citation: https://doi.org/10.5194/egusphere-2026-2539-AC1
-
AC1: 'Reply on RC1', Franklin Robertson, 04 Jun 2026
-
RC2: 'Comment on egusphere-2026-2539', Anonymous Referee #2, 11 Jun 2026
This paper evaluates many methods for extending the evaluation of terrestrial water storage to decades earlier than the GRACE/FO mission so longer term trends and variability could be studied. A comprehensive set of methods is used, including remotely sensed, modeled, and statistically estimated TWS. With so many methods the paper is lengthy, and the dozens of acronyms can be a hindrance to digesting everything. The conclusions are supported by the analysis, and this is a meaningful step forward in understanding the evolution of water storage in recent decades.
Comments:
- Lines 120-122, conclusions in general don’t belong in the introduction.
- Line 138, remove the URL from the text – it is correctly included in the “code and data availability” section later, and should not be in the body of the text. This also applies to lines 161, 164, 166, and 258, and perhaps in other locations.
- Line 168, it would be helpful to also include a (~300 km) equivalent to describe the GRACE resolution.
- Line 267, how were glaciated grid points determined?
- Section 3.1.1 and Table 3, trends should be tested for statistical significance, and only significant trends should be considered real trends.
- Following on from comment 5, Figure 1 shows the trends in TWS. These trends over different periods could be illustrated as the annual TWS/dt (as is shown later in Fig 4), which is more meaningful as the TWS flux is part of the water budget. In that way the well-documented ERA5 issues in 2000-2002 could be easily removed to show (and calculate) the trend without that effect.
- Line 209, to be consistent with other details provided for other models, consider noting that H08 also has some HWM representation (per Table 1).
- Figure 1, could the models with HWM be shown in one set of colors or line types to help with interpretation?
- Liune 302, the caption for Fig 1, please refer to “Parentheses Are (Are Not) for References and Clarification (Saving Space)”
- Lines 304-306, the role of ENSO variability in the TWS signal is considered. This is also discussed in several other parts of this paper. A straightforward way to assess this would be to work with TWS (or TWS/dt) residuals, after a regression with ENSO has removed that effect. That could be done here, or at least discussed as something to gain more insight into driving mechanisms for evident trends. Similarly, at line 349 the role of AMO is discussed, and the effect of that could similarly be isolated.
- Lines 326-328, and 389-391, these repeat lines 275-276 and could be removed.
- Figure 2, shade or stipple the areas with statistically significant trends.
- Figure 8, It is helpful that the caption highlights that the color bars are very different. However, different scales makes intercomparison between the columns basically impossible. While one column may look uninteresting with identical color bars, that would seem to be the point. I highly suggest using the same color bar for the entire figure.
Typos:
- Line 113, “In particular, [a] number..”
- Line 167, “product[s]”
- Line 190, “sets of forcing data [sets]
- Line 357 (Table [23])
- Line 476, “et[ ]al.”
Citation: https://doi.org/10.5194/egusphere-2026-2539-RC2 -
AC2: 'Reply on RC2', Franklin Robertson, 14 Jun 2026
The authors acknowledge the in-depth evaluation that the reviewer has provided for our paper and that overall, the “conclusions are supported by the analysis, and this is a meaningful step forward in understanding the evolution of water storage in recent decades”. In addition, the reviewer cites a number of specific changes and corrections that will lead to an improvement in the value and the readability of the work. Our intent is to accommodate as many as is possible in a revised version of the paper.
The most significant of the points raised concerns the need for measures of statistical significance for global trends and regional patterns. Our omission of these stems from our aim to uncover intrinsic reasons for the different trend estimates. For example, the change in the ERA5 water cycle near the turn of the 21st Century is likely due in large part to changes in the assimilated data stream. So, while statistical tests may quantify an associated trend, we thought the greater value is in understanding the origin of that behavior. Nevertheless, a concrete numerical value is always of value and we will work to provide the results of significance tests for the trends.
A second suggestion was to try to quantify how much interannual variability (e.g. ENSO) might be important in inducing trends. We believe that a full treatment of this is beyond the scope of our paper since it would likely involve not only simple regressions but also numerical experiments with various SST forcing configurations. Here we are focusing on different methods of arriving at estimates of terrestrial water storage change. That said, our discussion may not adequately emphasize that large interannual signals that are not rectified over the short period of years we are studying may result in significant trend components. We will be doing some regression diagnostics to explore this to address the reviewer’s concerns. These may or may not be included in the revision, but we will certainly tighten the discussion on this subject
Other issues raised having to do with style, readability and figure presentation in the paper will be addressed and corrections incorporated.
Citation: https://doi.org/10.5194/egusphere-2026-2539-AC2
-
RC3: 'Comment on egusphere-2026-2539', Anonymous Referee #3, 13 Jun 2026
This manuscript evaluates changes in terrestrial water storage (TWS) since 1980 using a broad suite of data products, including ERA5-Land, GRACE/GRACE-FO mascon solutions, satellite laser ranging (SLR)-based gravimetry, ISIMIP3a global hydrological models (GHMs) and land-surface models (LSMs), and three statistical or machine-learning GRACE-like reconstructions. Its main contribution is a comparative assessment of long-term TWS trends and variability before and during the GRACE era, with particular attention to the apparent artificial decline in ERA5 precipitation around 2000-2002 and its effects on ERA5-Land TWS trends. The study is timely and potentially valuable. However, several methodological assumptions, uncertainty treatments, and reproducibility details require substantial strengthening before the conclusions can be considered robust.
Major Comments
(1) The study is primarily diagnostic and comparative; it does not introduce a new reconstruction method, hydrological model, or formal data-assimilation framework. The authors should more clearly define the novelty of the work relative to prior studies on ERA5 hydrological discontinuities, GRACE-like reconstructions, and the tendency of GHMs to underestimate TWS trends.
(2) The manuscript relies largely on visual correspondence and similarity in trend patterns. The authors should add a formal homogeneity or changepoint analysis for precipitation, runoff, evapotranspiration, and TWS tendencies around 2000-2002. This analysis should ideally include regional statistics for central Africa, northeastern China, La Plata and northern Argentina, and other regions emphasized in the manuscript. The authors should also quantify how much of the ERA5-Land TWS trend can be attributed to the precipitation discontinuity, as opposed to evapotranspiration, runoff, soil-column storage constraints, spin-up effects, or streamflow-related processes.
(3) ERA5-Land, GHMs, LSMs, GRACE/GRACE-FO, SLR, and machine-learning reconstructions do not represent identical water-storage components. ERA5-Land lacks explicit groundwater storage and human water management, whereas some GHMs include groundwater, surface water, reservoirs, irrigation withdrawals, and return flows. LSMs generally emphasize shallow soil moisture, snow, and canopy water. GRACE/GRACE-FO observes total mass change after geophysical corrections. Although the manuscript acknowledges some of these differences, it does not provide a rigorous component-level harmonization or sensitivity analysis. The authors should add a table specifying exactly which storage components are included in each TWS estimate used in the analysis and should discuss how component mismatches may affect global trends, regional trends, and inter-product correlations.
(4) Tables 3 and 4 report trends, standard deviations, and correlations, but they do not provide confidence intervals, effective degrees of freedom, autocorrelation treatment, spatial covariance estimates, GRACE mascon uncertainties, model-spread uncertainty, or sensitivity to endpoint selection. The authors should estimate trend uncertainties using methods appropriate for autocorrelated monthly data. They should also test alternative analysis periods, such as 2003-2016, 2003-2022 where available, and 1980-2014 for products that end earlier. What’s more, Table 3 shows that the correlation coefficients of the trend patterns between all physical models and GRACE/FO are below 0.40, an important issue that is not discussed in sufficient depth.
(5) GPCC is treated as the reference precipitation product, but its limitations require more careful discussion. Using GPCC as a bias-correction anchor is defensible; however, gauge coverage is sparse and temporally heterogeneous in several key regions, including central Africa, parts of South America, high-latitude regions, and politically restricted areas.
(6) SLR, as currently used, does not provide an independent validation of long-term trends. The manuscript states that SLR trends were calibrated to match GRACE/GRACE-FO over a common period. This calibration limits the independence of SLR for evaluating pre-GRACE trends. The authors should clarify exactly how SLR is used and what role it plays in the interpretation.
(7) The manuscript lists the data sources, but the analytical workflow is not specified in sufficient detail. The authors should add equations or precise definitions for anomaly calculation, area-weighted averaging, linear trend estimation, TWS tendency, masking, smoothing, spatial filtering, and the treatment of missing GRACE months.
(8) The high correlations between GRACE/GRACE-FO and the Li21 and Yin23 reconstructions are expected because these products were trained on GRACE/GRACE-FO, and Li21 imposes the GRACE trend pattern. Although the manuscript acknowledges this point, the evaluation should be framed more explicitly as an in-sample consistency check rather than as independent validation.
(9) The conclusions are somewhat stronger than the evidence supports. The statement that the TWS decline observed since 2002 extends back over four decades is broadly consistent with several products, but the global trend estimates range from near zero to strongly negative, and the regional patterns differ substantially among products.
(10) Section 4 (Discussion and Conclusions) is substantial, but it is recommended that the key conclusions be extracted into a separate '5 Conclusions' subsection.
(11) The reference list includes duplicated “References” headings and inserted editorial text, such as “Here is your reference list reformatted…” These artifacts should be removed.
Minor Comments
(1) The corresponding author is listed as “Roberton” in the correspondence line, whereas the author list gives “Robertson.” This discrepancy should be corrected.
(2) Units are written inconsistently as “mm yr-1,” “mm y-1,” and “mm yr -1.” The authors should adopt a single journal-compliant format throughout.
(3) The acronym list gives “HCTESSEL,” whereas the main text uses “CHTESSEL.” The correct acronym should be verified and used consistently.
(4) Table 4 states “Same as in Table 2,” but it appears to refer to the statistics reported in Table 3. This cross-reference should be corrected.
(5) Several references are duplicated or inconsistently formatted, including Van Beek et al. and multiple “References” headings. The Scanlon et al. (2018) entry is not formatted consistently with the rest of the reference list.
(6) Several author names and citations appear inconsistent or misspelled, including “Cleeland/Clelland,” “Muños-Sabater/Muñoz-Sabater,” and possibly “Dinz/Diniz.” These should be checked carefully against the original publications.
(7) In the footnote of Table 4, '1019' should be '2019'.
(8) In the footnote of Table 1, should be '4 soil layers, 0-10, 10-40, 40-100,100-200 cm”. “0.50.5 deg res” should be “0.5 x 0.5 deg res”？
(9) In line 389, “105 sq km” should be “105 km2”, the entire manuscript should maintain consistency.
(10) In line 505, “mm yr-1” should be removed.
(11) The font size in the figure captions may need to be standardized.
Citation: https://doi.org/10.5194/egusphere-2026-2539-RC3
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 221 | 89 | 17 | 327 | 25 | 24 |
- HTML: 221
- PDF: 89
- XML: 17
- Total: 327
- BibTeX: 25
- EndNote: 24
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1