Vegad et al. (2026) Reconstructing Long‐Term Reservoir Storage in India Using Hydrological Modeling and Machine Learning

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: Water Resources Research
• Year: 2026
• Date: 2026-08-01
• Authors: Urmin Vegad, Vimal Mishra
• DOI: 10.1029/2026wr044312

Research Groups

Not specified in the provided text.

Short Summary

The study reconstructs long-term daily live storage for major reservoirs in India using a hybrid approach of hydrological modeling and machine learning, finding that normalized storage has moderately declined due to increased water withdrawals.

Objective

• To reconstruct long-term daily live storage for major Indian reservoirs for the period prior to 2000 and analyze the resulting long-term trends and variability in storage dynamics.

Study Configuration

• Spatial Scale: Major reservoirs across India.
• Temporal Scale: Long-term daily resolution (including the pre-2000 period and post-2000 observation period).

Methodology and Data

• Models used: Hydrological model simulations (as baseline), Random Forest (RF), and XGBoost (XGB).
• Data sources: Reservoir storage observations (from 2000 onwards), meteorological variables, and hydrological variables.

Main Results

• Machine learning models (RF and XGB) substantially improved the accuracy of reservoir storage reconstruction compared to raw hydrological simulations.
• Despite an increase in total reservoir capacity, normalized storage shows a moderate long-term decline, indicating an increase in water withdrawals.
• Pre-monsoon storage trends are larger and exhibit more spatial variability than post-monsoon storage trends.
• There is a observed decrease in the frequency of low-storage conditions and an increase in high-storage states.
• Peak storage variability is concentrated in two periods: late July to mid-August (corresponding to peak monsoon inflows) and around January (corresponding to peak irrigation demand).

Contributions

• Fills a critical data gap by reconstructing daily reservoir storage for India prior to the year 2000.
• Demonstrates the efficacy of integrating machine learning with hydrological models to improve the accuracy of storage reconstructions.
• Provides quantitative insights into the evolving dynamics of water withdrawals and storage variability under changing hydroclimatic conditions in India.

Funding

Not specified in the provided text.

Citation

@article{Vegad2026Reconstructing,
  author = {Vegad, Urmin and Mishra, Vimal},
  title = {Reconstructing Long‐Term Reservoir Storage in India Using Hydrological Modeling and Machine Learning},
  journal = {Water Resources Research},
  year = {2026},
  doi = {10.1029/2026wr044312},
  url = {https://doi.org/10.1029/2026wr044312}
}