Accurate rainfall forecasting is essential for agriculture, water resource management, and disaster preparedness in monsoon-dominated regions. This study evaluates the performance of four global numerical weather prediction (NWP) models, ECMWF (European Centre for Medium-Range Weather Forecasts), NCEP (National Centers for Environmental Prediction), JMA (Japan Meteorological Agency), and NCMRWF (National Centre for Medium Range Weather Forecasting), and their Multi-Model Ensemble (MME) in predicting daily monsoon rainfall over the semi-arid state of Rajasthan, India, from 2020 to 2023. Forecast skill is assessed at grid and district scales using continuous (correlation, MAE, RMSE, and bias) and categorical (POD, FAR, and CSI) verification metrics for Day-1, Day-2, and Day-3 lead times. Results show a decline in forecast skill with increasing lead time across all models. For Day-1 forecasts, the MME achieves the highest correlation (0.56) with IMD (India Meteorological Department) observations and the lowest RMSE 9.36 mm day⁻¹, compared to individual models, ECMWF correlation 0.50, RMSE 10.30 mm day⁻¹, and NCMRWF (12.29 mm day⁻¹). At Day-2, correlation decreases to 0.40 for the MME, yet it continues to show the lowest RMSE (10.87 mm day⁻¹) and MAE (4.84 mm day⁻¹). By Day-3, correlations drop below 0.20 for all models; however, the MME still shows lower errors (12.70 mm day⁻¹) than most individual models. Event-based verification shows moderate skill for heavy rainfall at Day-1, with CSI values reaching 0.16 for the MME. Although NCMRWF shows higher detection but it is associated with a high false alarm rate. The inter-comparison of model forecasts reveals that the MME method can generate skilful rainfall forecasts over Rajasthan for operational use during the monsoon season and can improve accuracy, spatial consistency, and reliability.
Gurjar et al. (Wed,) studied this question.