Scholarlink

<p>Short-range rainfall forecasting is critical for water management and disaster risk reduction in semi-arid monsoon regions. This study evaluates the performance of four global numerical weather prediction models, the European Centre for Medium-Range Weather Forecasts (ECMWF), National Centers for Environmental Prediction (NCEP), Japan Meteorological Agency (JMA), and National Centre for Medium Range Weather Forecasting (NCMRWF), and their multi-model ensemble (MME) for daily rainfall prediction over Rajasthan, India, during 2020–2023. In addition, rainfall observations for 2000–2023 were analysed to provide climatological context regarding recent rainfall variability over the region. Model performance is evaluated using continuous (MAE, RMSE, Bias, and correlation) and categorical (POD, FAR, and CSI) metrics, as well as spatial analysis. Forecast skill decreases with lead time across all models, with RMSE increasing from 5.846 to 7.409&#xa0;mm (Day 1) to 7.528–9.474&#xa0;mm (Day 5), and correlation declining from 0.537 to 0.440 (Day 1) to 0.175–0.110 (Day 5). The MME shows consistent improvement, reducing RMSE to 5.846&#xa0;mm (Day 1) and 7.528&#xa0;mm (Day 5), corresponding to a 15–40% reduction compared to individual models, while maintaining near-neutral bias (0.200–0.655&#xa0;mm). Despite this, all models show limited ability to predict heavy rainfall events, with CSI values ranging from 0.05 to 0.15 on Day 1 and dropping below 0.02 at longer lead times, while FAR increases from 0.30 to 0.60 to 0.95–1.00. Spatial patterns indicate higher errors (10–25&#xa0;mm) over south-eastern Rajasthan and lower errors (3–8&#xa0;mm) in western regions. The MME captures overall spatial variability but tends to smooth localised extremes and slightly overestimates rainfall intensity. The results highlight the advantage of ensemble-based forecasting for improving reliability, while emphasising the need for better representation of extreme rainfall and reduced uncertainty at longer lead times. The results provide useful guidance for improving operational monsoon forecasting and strengthening early warning systems for rainfall-related hazards in semi-arid regions.</p> Graphical Abstract <p></p> <p>The graphical abstract illustrates the workflow and key findings of the study on short-range monsoon rainfall prediction over Rajasthan using a multi-model ensemble (MME) approach. It presents the integration of deterministic forecasts from global models (ECMWF, NCEP, JMA, and NCMRWF) with IMD observations, followed by verification using continuous, categorical, and advanced skill metrics. The results highlight a consistent decline in forecast skill with increasing lead time, limited ability to predict heavy rainfall events, and strong spatial variability across the region. The MME demonstrates improved performance over individual models by reducing forecast errors and enhancing reliability. The study emphasizes the operational relevance of ensemble forecasting for agriculture, disaster preparedness, and water resource management in semi-arid regions.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Performance Evaluation of Numerical Weather Prediction Models and Multi-model Ensemble Forecasts for Short-range Monsoon Rainfall over Rajasthan, India

  • 期刊论文

  • 发表日期：

    2026年7月6日

查看全文
  • Suman Gurjar,
  • N. K. Goel,
  • Manohar Arora,
  • M. K. Goel

摘要

Short-range rainfall forecasting is critical for water management and disaster risk reduction in semi-arid monsoon regions. This study evaluates the performance of four global numerical weather prediction models, the European Centre for Medium-Range Weather Forecasts (ECMWF), National Centers for Environmental Prediction (NCEP), Japan Meteorological Agency (JMA), and National Centre for Medium Range Weather Forecasting (NCMRWF), and their multi-model ensemble (MME) for daily rainfall prediction over Rajasthan, India, during 2020–2023. In addition, rainfall observations for 2000–2023 were analysed to provide climatological context regarding recent rainfall variability over the region. Model performance is evaluated using continuous (MAE, RMSE, Bias, and correlation) and categorical (POD, FAR, and CSI) metrics, as well as spatial analysis. Forecast skill decreases with lead time across all models, with RMSE increasing from 5.846 to 7.409 mm (Day 1) to 7.528–9.474 mm (Day 5), and correlation declining from 0.537 to 0.440 (Day 1) to 0.175–0.110 (Day 5). The MME shows consistent improvement, reducing RMSE to 5.846 mm (Day 1) and 7.528 mm (Day 5), corresponding to a 15–40% reduction compared to individual models, while maintaining near-neutral bias (0.200–0.655 mm). Despite this, all models show limited ability to predict heavy rainfall events, with CSI values ranging from 0.05 to 0.15 on Day 1 and dropping below 0.02 at longer lead times, while FAR increases from 0.30 to 0.60 to 0.95–1.00. Spatial patterns indicate higher errors (10–25 mm) over south-eastern Rajasthan and lower errors (3–8 mm) in western regions. The MME captures overall spatial variability but tends to smooth localised extremes and slightly overestimates rainfall intensity. The results highlight the advantage of ensemble-based forecasting for improving reliability, while emphasising the need for better representation of extreme rainfall and reduced uncertainty at longer lead times. The results provide useful guidance for improving operational monsoon forecasting and strengthening early warning systems for rainfall-related hazards in semi-arid regions.

Graphical Abstract

The graphical abstract illustrates the workflow and key findings of the study on short-range monsoon rainfall prediction over Rajasthan using a multi-model ensemble (MME) approach. It presents the integration of deterministic forecasts from global models (ECMWF, NCEP, JMA, and NCMRWF) with IMD observations, followed by verification using continuous, categorical, and advanced skill metrics. The results highlight a consistent decline in forecast skill with increasing lead time, limited ability to predict heavy rainfall events, and strong spatial variability across the region. The MME demonstrates improved performance over individual models by reducing forecast errors and enhancing reliability. The study emphasizes the operational relevance of ensemble forecasting for agriculture, disaster preparedness, and water resource management in semi-arid regions.