| 17 Apr 2026
Towards the development of a national drought monitoring framework for India: Reconstruction of historical droughts using CLM5
Abstract. In this study, we evaluate the Community Land Model version 5.0 (CLM5.0) as a first step toward building a National Drought Monitoring Framework – DRISHTI (Drought Risk & Impact Surveillance using Hydrometeorological, Terrestrial & Inference Models) in India. CLM5.0 was applied over India at 0.1 ° resolution (1980–2020), using India Meteorological Department (IMD) precipitation and two atmospheric forcing datasets: the Indian Monsoon Data Assimilation and Analysis (IMDAA) regional reanalysis and the fifth generation ECMWF Re-Analysis (ERA5). Model performance was assessed for soil moisture (SM), evapotranspiration (ET), and runoff under rainfed and irrigated conditions against in-situ measurements and satellite/reference datasets: Soil Moisture Active Passive (SMAP), Global Land Evaporation Amsterdam Model (GLEAM), and Global Runoff Reconstruction (GRUN).
CLM5.0 reproduced spatial and temporal hydrological variability well. Model diagnostics revealed three climate-zone biases: (1) an apparent 'Dry Bias' in humid zones (Am) characterized by low SM, runoff, and ET. This bias reflects underestimation of precipitation in IMD rather than model process deficiencies; (2) a 'Wet Bias' in semi-arid and arid zones (BSh, BWh), where the Dry Surface Layer scheme suppresses evaporation, generating excess storage and runoff; and (3) a 'Runoff Paradox' in temperate zones (Cwa, Cwb), where high surface SM coincides with low runoff. Systematic runoff underestimation in complex terrains reflects limitations in sub-grid hillslope lateral flow representation, which led to high moisture retention. Relative to ERA5-driven simulations, IMDAA reduces RMSD in SM and runoff by 10% and 2.5 %, respectively, while ERA5 improves ET estimates by 17 %. At the national scale, rainfed simulations generally outperform irrigated runs; however, irrigation leads to localized improvements - most notably over the Indo-Gangetic Plain - with a stronger influence on root-zone SM than on surface SM. Specifically, irrigation reduces errors over ~26 % of irrigated grid cells for root-zone SM compared to ~23 % for surface SM, while corresponding improvements are observed over ~33 % and ~45 % of irrigated grid cells for runoff and ET, respectively. Despite these limitations, CLM5.0 successfully reproduces historical droughts, demonstrating its utility for drought monitoring in data-scarce regions while highlighting the need for improved irrigation parameterization, targeted model calibration, and careful selection of meteorological forcing datasets.
Competing interests: At least one of the (co-)authors is a member of the editorial board of Hydrology and Earth System Sciences.
Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.
1. General Comments
The authors present an evaluation of the Community Land Model version 5 (CLM5) over India at 0.1 degree resolution for 1980–2020, as a first step towards a national drought monitoring framework (DRISHTI). The authors perform four simulations (IMDAA-Irr, IMDAA-Rfd, ERA5-Irr, ERA5-Rfd) forced with two reanalyses (IMDAA, ERA5) using common IMD precipitation, with (Irr) and without (Rfd) CLM’s default irrigation scheme. They evaluate soil moisture (surface and root zone), evapotranspiration and runoff against COSMOS, SMAP, GLEAM and GRUN, and compute SPEI/SSI/SRI-based drought indices to reconstruct historical drought events. They diagnose three climate‑zone‑dependent bias regimes and discuss implications for DRISHTI.
Overall, the manuscript presents a meaningful and timely attempt to evaluate CLM5 as a basis for drought monitoring over India. The study is potentially valuable because it examines CLM5 simulations under different meteorological forcing and irrigation configurations, and compares simulated soil moisture, evapotranspiration, and runoff against several independent datasets. However, in its current form, the manuscript contains critical weaknesses in the description of datasets, design of the performance evaluation, interpretation of inter-dataset differences, and strength of the conclusions. These issues need to be addressed before the manuscript can provide a robust basis for assessing the suitability of CLM5 for national-scale drought monitoring.
2. Specific Comments
- A key contribution of the study is not simply that CLM5 is run over India, because CLM5 and related global LSM outputs already exist. The novelty lies in the specific India-domain simulations using different combinations of meteorological forcing and irrigation settings, and in the evaluation of the resulting outputs against SMAP, COSMOS, GLEAM, and GRUN. Accordingly, the manuscript must provide much more complete and precise descriptions of all forcing, model, and evaluation datasets. At present, the descriptions are insufficient to allow readers to understand the technical basis of the assessment or to reproduce the analysis. For example, I failed to find any information of the COSMOS dataset other than their locations. Quality and reliability of ‘reference’ or ‘ground truth’ data are critical information for studies of this kind. Key information about the datasets used is insufficient needs to be substantially elaborated.
- Relevant to the previous comment, the IMD precipitation dataset requires substantially more detail. The manuscript mentions that the IMD precipitation product is based on approximately 6,995 rain gauges and inverse-distance weighting, but the number and distribution of gauges are time-dependent over the historical period and data acquisition frequency (e.g., hourly or daily) varies between gauge stations. Because the study spans 1980–2020, the authors should provide information on the temporal variation in station numbers and spatial coverage over the simulation period, and discuss how these changes may affect precipitation uncertainty and the subsequent CLM5 evaluation. The citation should be to the IMD gridded precipitation dataset documentation or relevant dataset paper, rather than to the inverse-distance weighting method. A citation to Shepard (1968) is not necessary unless the interpolation method itself is being discussed in detail; if it is retained, Shepard (1968) must also be included in the reference list.
- There is an inconsistency in the SMAP-based evaluation. The manuscript presents figures and text describing SMAP comparisons for 1980–2020, but SMAP soil moisture data are only available from 2015 onward. This must be corrected throughout the manuscript, including figure captions, methods, results, and conclusions. If the SMAP evaluation was in fact performed only for 2015–2020, the authors should state this clearly and explain how the much shorter SMAP period affects the robustness of conclusions about forcing choice, irrigation effects, and drought-monitoring suitability over the full 1980–2020 simulation period.
The treatment of SMAP soil moisture as an absolute ground truth is problematic. Microwave soil moisture products often have product-specific dynamic ranges and regionally varying biases. They are highly useful for capturing temporal wet–dry variability, but their absolute values may differ from modelled soil moisture because of retrieval assumptions, sensing depth, vegetation effects, soil texture assumptions, radio-frequency interference, and scaling. Although the manuscript reports ME/MD, RMSE/RMSD, and Spearman correlation, many interpretations and key conclusions appear to rely primarily on RMSD. This can lead to misleading conclusions when the evaluated datasets have different absolute dynamic ranges. The performance assessment should give greater emphasis to temporal correlation, anomaly correlation, rank correlation, bias-adjusted RMSE/RMSD, seasonal anomalies, and/or standardized soil moisture indices, in addition to absolute-error metrics.
- The COSMOS comparison also needs correction and clarification. COSMOS does not measure soil moisture at a fixed 5 cm depth in the same sense as an in-situ point probe. Cosmic-ray neutron sensing provides an area-integrated soil moisture estimate over a footprint of hundreds of metres and a variable effective sensing depth, commonly on the order of 10–70 cm depending on soil wetness. Therefore, comparison with CLM5 soil moisture at a single 5 cm layer is not a like-for-like evaluation. The authors should either compute a depth-weighted CLM5 soil moisture estimate that better represents the COSMOS effective measurement depth or explicitly acknowledge this mismatch as a major source of uncertainty. The spatial support mismatch between COSMOS, SMAP, and CLM5 should also be treated more conservatively.
- The authors should also justify the exclusive or primary use of SMAP for satellite soil moisture evaluation. SMAP provides a relatively short record (2015-) compared with the 1980–2020 CLM5 simulation. If the aim is to assess model performance over a multi-decadal period relevant to drought monitoring, the authors should either provide a clear rationale for using SMAP despite its short duration or extend the evaluation using longer microwave soil moisture datasets, such as ESA CCI soil moisture or other suitable multi-sensor products. This is particularly important because conclusions about the historical reconstruction of droughts should not rely heavily on a short satellite validation period.
- The statistical comparison among the four CLM5 configurations requires much more rigorous treatment. In several figures and tables, the differences among the four simulations appear small relative to the overall variability and uncertainty of the reference datasets. Conclusions such as one forcing being superior to another, or rainfed simulations generally outperforming irrigated simulations, should not be based only on small differences in domain-mean RMSD or on maps of the “best-performing” configuration at each grid cell. The authors should test whether these differences are statistically significant and practically meaningful, while accounting for spatial autocorrelation, temporal autocorrelation, multiple testing, and the different spatial support scales of CLM5, SMAP, COSMOS, GLEAM, and GRUN. I strongly recommend that the authors consider the guidance in Wilks (2016), “The stippling shows statistically significant grid points”: How research results are routinely overstated and overinterpreted, and what to do about it, Bulletin of the American Meteorological Society, 97, 2263–2273. Without such a conservative statistical treatment, the current conclusions are likely overstated.
- The interpretation of ET differences against GLEAM also needs to be reconsidered. Differences between CLM5 and GLEAM ET cannot be attributed primarily to meteorological forcing unless other major structural differences are accounted for. GLEAM and CLM5 differ in model structure, vegetation stress representation, rooting depth, soil moisture constraints, and the way evaporation components are estimated. In particular, differences in effective root-zone depth and water-stress representation may cause ET differences that are unrelated to IMDAA versus ERA5 forcing. In the extreme case of this kind, a “better performing” combination may give the right answer for wrong reasons. Therefore, statements attributing ET discrepancies mainly to forcing data are not sufficiently supported. The authors should either provide additional analysis isolating the forcing effect from structural/model-product differences or substantially soften these interpretations.
- The conclusions should be revised to better reflect the limitations of the analysis. The current manuscript makes strong statements about CLM5’s ability to capture spatiotemporal variability, reconstruct historical droughts, and identify climate-zone-dependent biases. However, these conclusions are weakened by inconsistencies in dataset periods, insufficient description of forcing and reference datasets, scale mismatches among CLM5, SMAP, COSMOS, GLEAM, and GRUN, and a lack of robust significance testing. The conclusion should explicitly distinguish between findings that are well supported by the analysis and those that remain hypotheses requiring further evaluation.