the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 20 Jan 2026
Enhancing hydrological representation of the Brahmaputra basin through terrestrial water storage and surface soil moisture Data Assimilation
Abstract. Understanding the dynamics of terrestrial water storage (TWS), and its components such as surface soil moisture (SSM) and groundwater, is important for the Brahmaputra River Basin, where water resources are expected to experience increasing demand and are highly vulnerable to extreme hydrological events and climate change. However, water storage dynamics are complex and difficult to capture by state-of-the-art large-scale hydrological models. In this study, we implement a multi-variate daily TWS and daily SSM sequential Data Assimilation (DA) with the aim of improving model-derived water storage dynamics. In our methodology, we propose a model space covariance localization approach that is compared with three other approaches used in the previous literature. The results show that this new approach is the only one to effectively mitigate cross-variable influences along the vertical water storage profile, which have been reported as one of the main challenges of multi-variate land DA. A validation of the multi-variate DA estimates (for the period 2004–2015) indicates that more realistic decadal trend and inter-annual variability are introduced into the groundwater estimates, increasing the correlation coefficients with the Standardized Precipitation Evapotranspiration Index and observed groundwater levels by +0.24 to +0.54 correlation points. With respect to SSM, DA induces a general phase shift, especially around mountain areas. Improved land water storage estimates reveal a land water decline of 70.9 GT per decade for the period 2004–2015 in the Brahmaputra River basin, which constitutes approximately half of the TWS decline in that period, with the other half caused by glacier retreat (67.5 GT per decade).
- Preprint
(6804 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 03 Mar 2026)
- RC1: 'Comment on egusphere-2025-5625', Anonymous Referee #1, 15 Feb 2026 reply
-
RC2: 'Comment on egusphere-2025-5625', Anonymous Referee #2, 20 Feb 2026
reply
“Enhancing hydrological representation of the Brahmaputra basin through terrestrial water storage and surface soil moisture Data Assimilation” [egusphere-2025-5625]
General comment:
This manuscript presents a methodologically innovative contribution to large-scale hydrological data assimilation by integrating GRACE/GRACE-FO Terrestrial Water Storage (TWS) anomalies into the W3RA land surface model using an Ensemble Kalman Filter (EnKF) framework with a novel model space mixed localization scheme. The study is applied to the Brahmaputra basin, a hydrologically complex and data-scarce region, and demonstrates improvements in both TWS and streamflow simulations at daily temporal resolution. The multi-variable validation approach, encompassing TWS, streamflow, and surface soil moisture (SSM), strengthens the credibility of the results.
The paper addresses a relevant scientific problem and makes genuine contributions to the field. However, several major concerns must be addressed before the manuscript can be considered for publication in the Hydrology and Earth System Sciences journal. I believe it needs “major revisions”. Some suggestions for revisions are as follows.
Major comments:
- The authors employ an ensemble of N=30 members in the EnKF framework. Given that the W3RA model state vector encompasses a large number of variables across the Brahmaputra basin, this ensemble size appears insufficient to adequately sample the model's uncertainty space and may lead to significant sampling errors and filter degeneracy. The authors provide no explicit justification for this choice, nor do they present a sensitivity analysis demonstrating that N=30 yields stable and converged results. It is strongly recommended that the authors either:
- Provide a formal justification based on the effective degrees of freedom of the system, or;
- Include a sensitivity analysis showing key metrics (e.g., NSE, RMSE, ensemble spread) as a function of ensemble size (e.g., N = 20, 30, 50, 100).
- The Brahmaputra basin contains significant glaciated areas, and GRACE/GRACE-FO TWS signals include a long-term trend associated with glacial mass loss. The authors appear to assimilate the full TWS signal and subsequently subtract the glacial component a posteriori. This approach raises important questions:
- Does the assimilation of the glacial trend introduce systematic biases into the model state variables (soil moisture, groundwater) that are not physically meaningful?
- How does this a posteriori correction compare to an approach where the glacial signal is removed prior to assimilation?
The authors should provide a more explicit discussion of the implications of this methodological choice and, if possible, a quantitative comparison between both approaches.
- While the assimilation improves TWS estimates, the manuscript reports a non-trivial reduction in Nash-Sutcliffe Efficiency (NSE) for streamflow at certain gauging stations. This result is concerning and warrants a more thorough investigation. Specifically:
- What are the dominant mechanisms driving this degradation? Possible causes include cross-variable error covariances, inadequate localization, or structural limitations of the W3RA baseflow parameterization.
- Is the W3RA model's baseflow representation adequate for the Brahmaputra's complex glacial and snowmelt-driven hydrological regime?
- Have the authors considered recalibrating the model after assimilation to mitigate this effect?
A dedicated subsection analyzing the sources of streamflow degradation would substantially strengthen the manuscript.
- The validation of surface soil moisture relies on WaterGAP model outputs as a reference dataset rather than on independent observational data. This introduces a circular dependency, as WaterGAP is itself a model subject to its own structural and parametric uncertainties. The authors should:
- Quantify the uncertainty associated with WaterGAP SSM estimates and discuss how this uncertainty propagates into the validation conclusions.
- Consider supplementing the validation with observational datasets such as ESA CCI Soil Moisture, SMAP, or in-situ measurements where available.
- The proposed model space mixed localization scheme is a central methodological contribution of this work. However, the manuscript does not present a systematic sensitivity analysis of the localization radii applied to the different model state variables. Given that localization is known to critically influence EnKF performance, the authors should demonstrate:
- How sensitive are the results to the chosen localization radii?
- What criteria were used to select the final parameter values?
- Were alternative localization configurations tested?
- The manuscript reports quantitative estimates of land water loss in the Brahmaputra basin. However, no uncertainty bounds (e.g., confidence intervals derived from the ensemble spread, or sensitivity to model assumptions) are provided for these estimates. Given the policy relevance of these figures, particularly in the context of climate change and transboundary water management, the authors must include a rigorous uncertainty analysis accompanying all reported water loss values.
Minor comments:
- Table 1: The notation used to describe model state variables is inconsistent with the notation employed in the equations. A unified nomenclature should be adopted throughout the manuscript.
- Equation 6: There appears to be a typographical error in the formulation of the observation operator. The authors should verify the consistency of this equation with the surrounding text and correct it accordingly.
- Definition of SSM: The manuscript uses the term "surface soil moisture" inconsistently, at times referring to the top soil layer and at others to a vertically integrated quantity. A precise and consistent definition should be provided at first use.
- Asynchronous Assimilation: The authors propose a hypothesis regarding the effects of asynchronous assimilation on streamflow performance. This hypothesis, while plausible, remains speculative and is not supported by direct evidence within the manuscript. It should either be tested explicitly or clearly framed as a working hypothesis requiring future investigation.
- Relegation of Results to Appendices: Several results that appear central to the evaluation of the proposed localization scheme are presented in appendices. The authors should consider whether these results merit inclusion in the main body of the manuscript to improve readability and scientific transparency.
- Model Recalibration: Given the known structural limitations of W3RA in glacially influenced basins, the authors should discuss whether a recalibration of the model prior to assimilation was considered and, if not, justify this decision.
- GRACE Spatial Resolution: The coarse spatial resolution of GRACE/GRACE-FO (~300 km) relative to the spatial heterogeneity of the Brahmaputra basin may limit the physical interpretability of the assimilated signal at sub-basin scales. The authors should explicitly discuss this limitation and its potential impact on the results.
Data Availability:
The Zenodo DOI provided for data availability appears to be reserved but not yet publicly accessible. The authors should confirm that all data and code necessary to reproduce the results will be made available upon publication, in accordance with the journal's open science policy.
Conclusion:
This manuscript makes a genuine and timely contribution to the field of hydrological data assimilation in data-scarce, high-mountain basins. The proposed localization scheme is innovative, the validation framework is multi-variable, and the application to the Brahmaputra basin is scientifically relevant. Nevertheless, the major concerns outlined above, particularly regarding ensemble size justification, glacial signal treatment, streamflow degradation, and uncertainty quantification, must be thoroughly addressed before the manuscript can be recommended for publication.
-
RC3: 'Comment on egusphere-2025-5625', Bramha Dutt Vishwakarma, 24 Feb 2026
reply
Summary: The manuscript improves assimilation of satellite-derived TWS and SSM products into a hydrological model. They explore various data assimilation strategies, including univariate and multi-variate, to obtain better representation of water storage dynamics over the Brahmaputra River basin. They claim to improve TWS, groundwater, and soil moisture representation in the model. The manuscript is well written and addresses a significant challenge/problem in the field.
Major comments:
1. From a methodological development point of view, the manuscript tries to add information from TWS and SSM in a multivariate DA setting., which is an advance over usual two step DA schemes. The authors also discuss a potential challenge in their approach, which is deciding the weightage given to one variable relative to the other. They discuss tunning approaches and propose observation space mixed localization as a solution. The idea seems interesting. However, see the next point.
2. They fix a smaller region of influence for SSM while a larger for TWS. This appears to be driven by data resolution. However, it would be nice to have some discussion and possibly analysis based on spatial wavelengths of processes responsible for changes in SSM or TWS. Groundwater has a larger wavelength than SSM but the current space localization scheme have these wavelengths at least an order of magnitude higher and I believe it could be two order of magnitudes for the time scale at which the assimilation is done, i.e. daily. Around line 270, the arguments presented are again appearing to be motivated by data resolution rather than from process spatial variability.
3. Equation 4, suggests independence between TWS and SSM, while it is known to be strongly correlated with the TWS budget equation defining causality. Would it be better to not have off diagonals as zero. This issue also haunts the framework, where TWS and SSM are ingested while information from SSM is a subset of TWS, although at different spatial resolution. A philosophical as well as analytical explanation would help a lot.
4. The validation could be improved a lot. SPEI-12 is very smooth. With simulations and DA done at daily scale, SPEI-3 itself is good enough. Majority of drought studies use either SPEI-6 or 3. Secondly, the well observations are compared in term of their correlation in Figure 7. The GW storage from MV-DA appears to do poorly in comparison to OL in c and f subplots (RMSE). This correlation plot also suggests that groundwater spatial-variability is not very large (in line with point 2 above).
5. SSM relation with TWS is tricky in irrigated regions. For example, in rainy season, a change in TWS is positively correlated with SSM but when groundwater irrigation is done, SSM is anti-correlated to TWS change. This has a significant impact on water budget also, which can lead to phase shifts tin ET (for example, see Goswami et al., 2025). How is the assimilation approach able to deal with such scenarios?
Other comments:
- Section 2.1: here authors acknowledge groundwater irrigation and thus point 5 above can be discussed.
- Daily GRACE products are generated using a KF, which introduces a strong autocorrelation in time. How would this impact the quality of assimilation?
- Line 170: Well data should also be checked for errors (repeating values, negative values, large gaps, see Kuruva et al., 2025). In South-east Asia, the dataset quality check is necessary sometimes. If that has been done, please mention.
- Since the covariance localization approach uses damped-spatial correlation based on distance b/w observations and model, do we expect observations to be at a higher spatial resolution than model? If yes, then it must be clarified how a coarse resolution GRACE product was used? Same values for all grids in a sub-basin?
- L 265: Smooth distant, whould be replaced by smooth distance
- Line 299 and 303: use comma, not full stop (33.808 -->33,808)
- Result section was slightly harder to follow. Maybe improve the writing aspect.
Best,
Bramha
References:
Kuruva, S. K., Suryawanshi, M. R., Shakya, A., Va, C., Shaw, B., Sukumaran, V., ... & Vishwakarma, B. D. (2025). Quality controlled, reliable groundwater level data with corresponding specific yield over India. Scientific Data, 12(1), 1609.
Goswami, S., Rajendra Ternikar, C., Kandala, R., Pillai, N. S., Kumar Yadav, V., Abhishek, ... & Dutt Vishwakarma, B. (2024). Water budget-based evapotranspiration product captures natural and human-caused variability. Environmental Research Letters, 19(9), 094034.
Citation: https://doi.org/10.5194/egusphere-2025-5625-RC3
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 183 | 121 | 17 | 321 | 18 | 14 |
- HTML: 183
- PDF: 121
- XML: 17
- Total: 321
- BibTeX: 18
- EndNote: 14
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Retegui-Schietekatte et al. address the problem of observation localization in multivariate ensemble data assimilation. The application here is integration of both GRACE TWSA maps and satellite SSM maps in hydrological modelling for the Brahmaputra basin.
The Brahmaputra basin (BB here) represents a hydrologically complex region with significant long-term changes due to warming climate in the Himalayas, above-average sea level rise with increased river blocking and salinization, and socio-economic processes such as land use change, population pressure and water diversion, but also with frequent large-scale flooding caused by cyclone landfall. Since the hydrological ground infrastructure is sparse, improved understanding of hydrological processes with the aid of satellite data is welcome. The manuscript addresses thus an important topic with broad relevance.
GRACE data assimilation has been often evaluated with sparse surface soil moisture data, and only few papers address joint assimilation. As most frameworks nowadays seem to rely on some localization technique, a systematic study such as provided here is welcome. The proposed model space mixed localization appears new to me in this context. The results are interesting and support the new idea. This would certainly warrant publication. The authors may want to elaborate somewhat more on the reason for the need for localization, is it due to spurious model ensemble correlations and if yes why, is it due to the observation error characteristics or is it both?
However, the manuscript in present form suffers from two major drawbacks, in my view: (1) The hydrological model is not well motivated and described, and it is thus not clear at all what the baseline for the data assimilation is. (2) The evaluation of results appears not very thorough. Here are more details:
The hydrological model that is used in this study is neither mentioned in the abstract nor in the introduction, which raises a red flag. The BB hydrology is really complex and the choice of a model should be motivated by the study purpose; are we interested in river discharge forecasting, in flood potential monitoring on timescales of days of weeks, in agricultural forecast, in water management at which level, water resources, or in climate change assessment on decadal timescales? The authors need to clarify this and explain why their model is fit for purpose, and how it compares to other global or regional models that have been used to describe the BB hydrology. The authors mention some limitations of W3RA, e.g. the model has not been calibrated over the BB region, they use ERA5/ERA5 land forcing but there seems no bias-correction. Even if the authors must rely on this model due to the assimilation framework, they should compare the open-loop total water storage, surface soil moisture and river discharge to the results from some published global models such as GLDAS, WaterGAP, PCR-GLOBWB in the BB. Since TWSA and SSM are assimilated, the authors should also explain the W3RA parameterization of processes relevant to these observables such as crops, water demand and anthropogenic water withdrawal, groundwater, surface water storage beyond river storage (lakes, floodplanes, reservoirs), and soils (which soil maps are used). How is evapotranspiration computed in W3RA? There will be severe limitations in the W3RA representation of these processes, as it is for other models, but whether assimilation is useful depends on our understanding of such error sources. Put in other words, if the baseline model is poor then the assimilation will almost always successed in „bringing the model closer to reality“ (page 25).
Validation in the BB is challenging, as the authors are aware. This raises the question why the study did not focus on Australia where the model was calibrated and the monitoring infrastructure appears much more complete, or some other region like the US. Comparison against the SPEI appears doubtful to me – W3RA also uses precipitation and radiation input and computes evapotranspiration and I wonder why then the simple SPEI could be considered as a „truth“ for groundwater change? Many studies have shown that accumulated precipitation (minus evapotranspiration) correponds well to TWSA but this does not prove anything about groundwater, in a region where surface water dynamics plays a BIG role. In this reviewer’s opinion, the authors need to validate groundwater change against quality-controlled well data with documented specific yield information (it wasn’t clear to this reviewer whether this was avaliable for the wells in Fig. 7). Or they validate against specific groundwater 3D modelling that has been calibrated against wells. If this is not possible, this kind of valiation cannot be performed. And the situation is similar for SSM, the authors need to work with in-situ data from the representative depth range, potentially with models informed by in-situ data. WaterGAP as a conceptual model has known limitations in soil moisture representation, and even disregarding this, comparing soil moisture across different models with differing layers, model physics, data integrated is highly problematic. If there is no ground data, no validation is possible. The authors could still compare their results to measured streamflow. Of course the evaluation should be tailored to the purpose of the study, i.e. looking at either short or long timescales. Last, I noticed the authors compared river water storage from the assimilation to gauge observations, this requires an explanation. While stage-discharge (rating curve) relations are typically derived empirically from simultaneous measurements with ADCPs and gauges, it is less clear how the routed model simulation can be mapped to observable water level change as this would require river cross-section assumptions.