Singh et al. (2025) Multi-scale assessment and entropy-MCDM framework for evaluating reanalysis precipitation datasets over Indian basins

Identification

• Journal: International Journal of Applied Earth Observation and Geoinformation
• Year: 2025
• Date: 2025-10-25
• Authors: Hrishikesh Singh, Mohit Prakash Mohanty
• DOI: 10.1016/j.jag.2025.104919

Research Groups

• Department of Water Resources Development and Management, Indian Institute of Technology Roorkee, Roorkee, India

Short Summary

This study systematically evaluates four reanalysis precipitation datasets (ERA5, IMDAA, MERRA-2, CFSR/CFSv2) across India using a multi-scale, multi-metric, and spatially adaptive Shannon Entropy-TOPSIS framework, finding ERA5 to be the most consistent and IMDAA strong in monsoon/mountainous regions.

Objective

• To systematically evaluate the accuracy of reanalysis precipitation datasets (ERA5, IMDAA, MERRA-2, CFSR/CFSv2) in representing rainfall patterns across India's diverse hydro-climatic conditions.
• To objectively rank these datasets using a multi-metric evaluation framework that integrates statistical and hydrological relevance at both grid and river basin scales.

Study Configuration

• Spatial Scale: Entire India (approximately 3.287 million km²), evaluated at individual grid cells and across 25 major river basins.
• Temporal Scale: 34 years (1990–2024), with evaluations performed at daily, monthly, and seasonal (Winter, Pre-Monsoon, Monsoon, Post-Monsoon) timescales.

Methodology and Data

• Models used: ERA5, Indian Monsoon Data Assimilation and Analysis (IMDAA), Modern-Era Retrospective Analysis for Research and Applications Version 2 (MERRA-2), Climate Forecast System Reanalysis / Climate Forecast System Version 2 (CFSR/CFSv2).
• Data sources:
  • Reference Data: India Meteorological Department (IMD) gridded rainfall dataset (0.25° × 0.25° resolution), derived from a dense rain-gauge network.
  • Reanalysis Data:
    • ERA5 (0.25° × 0.25° resolution, 1-hourly native)
    • CFSR/CFSv2 (0.5° × 0.5° resolution, 6-hourly native)
    • MERRA-2 (0.625° × 0.5° resolution, 1-hourly native)
    • IMDAA (0.12° × 0.12° resolution, 1-hourly native)
  • Methodology:
    • 11 statistical metrics (Correlation Coefficient, Kling–Gupta Efficiency, Nash–Sutcliffe Efficiency, Percent Bias, Probability of Detection, False Alarm Ratio, Critical Success Index, Root Mean Squared Error, Mean Absolute Error, Daily Mean, Standard Deviation).
    • Multi-Criteria Decision-Making (MCDM) framework combining Shannon Entropy (for objective weighting) and Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) for ranking.
    • Spatially adaptive framework applied independently at each grid cell and then aggregated for basin-level ranking.
    • All reanalysis datasets were spatially regridded to match the MERRA-2 grid using bilinear interpolation and aggregated to daily totals.

Main Results

• Overall Performance: ERA5 consistently demonstrated the most reliable performance across all spatial and temporal scales, securing Rank 1 in most grid cells (1067) and all 25 river basins.
• ERA5 Specifics: Exhibited high mean monthly Correlation Coefficient (~0.81), Kling–Gupta Efficiency (~0.56), Nash–Sutcliffe Efficiency (~0.53), and low Percent Bias (+6%). While reliable for long-term analyses, its distributions were ~50–60% narrower than IMD, indicating underrepresentation of moderate-to-extreme rainfall.
• IMDAA Specifics: Showed strong regional performance, particularly during the monsoon season and in mountainous regions (e.g., Western Ghats, Himalayas), with seasonal Correlation Coefficient of 0.74 and Kling–Gupta Efficiency of 0.49. Kernel Density Estimation confirmed its superior representation of precipitation extremes, with the widest inter-quartile range among reanalyses (~30% narrower than IMD). It ranked second in 17 out of 25 basins.
• MERRA-2 Specifics: Displayed mixed performance, with localized overestimations in mountainous areas and inconsistent grid-scale results. It surprisingly outperformed IMDAA to achieve Rank 2 in 7 basins (e.g., Brahmaputra, Mahanadi), suggesting more uniformly distributed spatial errors. Overall low performance with KGE of 0.09 and PBIAS of -22%. Its distributions were also ~50–60% narrower than IMD.
• CFSR/CFSv2 Specifics: Consistently underperformed across all metrics and domains, showing the lowest KGE (0.01) and significant negative bias (PBIAS: -29%). It ranked last (Rank 4) in 19 out of 25 basins, indicating limited suitability for hydrological applications in India.

Contributions

• Conducted the first systematic, multi-scale evaluation of four state-of-the-art reanalysis precipitation datasets across the entire Indian subcontinent, addressing gaps in national-scale, multi-dimensional assessments.
• Developed and implemented a novel, spatially adaptive Multi-Criteria Decision-Making framework combining Shannon Entropy for objective weighting and TOPSIS for hierarchical ranking at both grid and river basin scales, minimizing subjective bias and accounting for spatial heterogeneity.
• Provided actionable, evidence-based insights and a clear ranking of reanalysis datasets for hydrological modeling, flood risk management, and climate adaptation planning in India.
• Proposed a scalable evaluation framework transferable to other hydro-climatically diverse regions globally.

Funding

• Prime Minister’s Research Fellowship (PMRF) (grant PM-31-22-752-414) by the Ministry of Education, Government of India.
• National Supercomputing Mission (NSM) for access to the ‘PARAM Ganga’ supercomputing facility at IIT Roorkee (implemented by C-DAC, supported by the Ministry of Electronics and Information Technology (MeitY) and the Department of Science and Technology (DST), Government of India).
• Department of Water Resources Development and Management, Indian Institute of Technology Roorkee.

Citation

@article{Singh2025Multiscale,
  author = {Singh, Hrishikesh and Mohanty, Mohit Prakash},
  title = {Multi-scale assessment and entropy-MCDM framework for evaluating reanalysis precipitation datasets over Indian basins},
  journal = {International Journal of Applied Earth Observation and Geoinformation},
  year = {2025},
  doi = {10.1016/j.jag.2025.104919},
  url = {https://doi.org/10.1016/j.jag.2025.104919}
}