Deb et al. (2026) Multi-model assessment of seasonal rainfall and temperature forecast skill for operational drought prediction in India

Identification

• Journal: Climate Dynamics
• Year: 2026
• Date: 2026-01-01
• Authors: Pranab Deb, Saurabh Verma, L. Karthikeyan
• DOI: 10.1007/s00382-025-08024-9

Research Groups

• Centre of Studies in Resources Engineering, IIT Bombay, Powai, Mumbai, India
• Centre for Climate Studies, IIT Bombay, Powai, Mumbai, India

Short Summary

This study comprehensively evaluates the seasonal forecast skill of precipitation and 2 meter temperature from four Global Climate Models (CCSM4, GEOSS2S, CFSv2, and SEAS5) for operational drought prediction across India at 2 and 3-month lead times. It reveals that SEAS5 generally offers the most robust temperature and drought detection skill, while CFSv2 performs well for monsoon precipitation, though all models face challenges in complex and dry regions and exhibit skill deterioration with increased lead time.

Objective

• Evaluate the deterministic skill of seasonal precipitation and 2 meter temperature forecasts from multiple dynamical models at 2 and 3-month lead times over India.
• Assess model performance in predicting lower tercile precipitation and upper tercile 2 meter temperature events (which typically correspond to water stress conditions) across seasons and spatial scales, with emphasis on regional variability.
• Examine the effectiveness of seasonal forecasts in detecting meteorological droughts.

Study Configuration

• Spatial Scale: India, divided into six homogeneous rainfall regions (Northwest, Central Northeast, Northeast, West Central, South Peninsular, and Hilly Regions). Data resolution: 0.25° x 0.25°.
• Temporal Scale: 42 years (March 1982 to December 2023). Forecast lead times: 2 and 3 months. Seasons analyzed: Monsoon (June-September), Post-monsoon (October-December), Winter (January-February), Pre-monsoon (March-May).

Methodology and Data

• Models used:
  • Community Climate System Model version 4 (CCSM4)
  • Global Earth Observing System version 2 - Sub-seasonal to Seasonal (GEOSS2S)
  • Climate Forecast System version 2 (CFSv2)
  • Seasonal Forecasting System version 5 (SEAS5)
• Data sources:
  • Reference precipitation: India Meteorological Department (IMD) daily gridded dataset (0.25° resolution).
  • Reference 2 meter temperature: European Centre for Medium-Range Weather Forecasts (ECMWF) reanalysis version 5 (ERA5) (0.25° resolution).
  • Forecast products: North American Multi-Model Ensemble (NMME) models (CCSM4, GEOSS2S, CFSv2) from IRI/LDEO Climate Data Library (1° x 1° resolution). SEAS5 from Copernicus Climate Data Store (36 kilometer grid point resolution).
  • All forecast datasets were bilinearly interpolated to 0.25° x 0.25° to match reference data.
  • Evaluation metrics: Deterministic (Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Anomaly Correlation Coefficient (ACC)) and Categorical (Heidke Skill Score (HSS), Probability of Detection (POD), False Alarm Ratio (FAR), Equitable Threat Score (ETS)). Meteorological drought defined by Standardized Precipitation Index (SPI1) values ≤ -0.8.

Main Results

• Overall Model Performance: SEAS5 generally exhibited the most robust performance for 2 meter temperature and drought detection across seasons. CFSv2 showed good precipitation and drought skill during the monsoon. CCSM4 offered consistent temperature forecasts and moderate precipitation skill. GEOSS2S showed stable temperature skill and effective winter drought detection.
• Precipitation Forecasts: All models demonstrated notable HSS skill for lower tercile precipitation during the monsoon, especially over West Central, Central Northeast, and Northeast regions. CFSv2 and SEAS5 performed well. Skill generally declined at 3-month lead times.
• 2 meter Temperature Forecasts: SEAS5 consistently outperformed other models across all seasons, showing the most widespread and consistent skill. CFSv2 also performed well during the monsoon and post-monsoon seasons. Temperature forecast skill remained relatively stable with increasing lead time.
• Drought Forecasts (SPI1 ≤ -0.8): CFSv2 was the most accurate for monsoon drought, particularly over West Central and Central Northeast regions. SEAS5 showed strong pre-monsoon drought detection skill, especially in Northeast India, and performed well during the post-monsoon season. GEOSS2S and CFSv2 notably performed better over West Central, Northwest, and Central Northeast regions during winter.
• Limitations: All models faced challenges over climatologically dry and topographically complex regions (Northwest, Hilly Regions, and Western Ghats). Forecast skill deteriorated with increasing lead time, particularly for precipitation and drought, while temperature forecasts remained more stable. High False Alarm Ratios were observed in the Hilly Regions and Western Ghats for precipitation.

Contributions

• Provides a comprehensive and systematic skill assessment of seasonal precipitation and 2 meter temperature forecasts from four operational climate models over India.
• Evaluates the effectiveness of these models in detecting meteorological droughts at 2 and 3-month lead times, with a focus on region-specific variability across India's homogeneous rainfall regions.
• Integrates both hindcast and forecast datasets, carefully processed for consistency, to ensure the operational utility of the findings for India's early warning systems.
• Identifies specific strengths and limitations of individual models across different seasons and regions, offering valuable insights for enhancing sub-seasonal drought forecasting and sustainable water resource management in India.

Funding

• Ministry of Earth Sciences, Government of India (Project ID: MOES/16/04/2022-RDESS/AI-ML-04)
• Department of Science and Technology through Geospatial Information Science and Engineering Hub, IIT Bombay (Project Code: RD/0123-GISIR00-006)

Citation

@article{Deb2026Multimodel,
  author = {Deb, Pranab and Verma, Saurabh and Karthikeyan, L.},
  title = {Multi-model assessment of seasonal rainfall and temperature forecast skill for operational drought prediction in India},
  journal = {Climate Dynamics},
  year = {2026},
  doi = {10.1007/s00382-025-08024-9},
  url = {https://doi.org/10.1007/s00382-025-08024-9}
}