Performance Evaluation of Global Numerical Weather Prediction Models over Tunisia: A 2025 Annual Assessment Using RMSE, MAE, and Pearson Correlation

Accurate numerical weather prediction (NWP) is critical for effective meteorological services, particularly in the Mediterranean region, where complex terrain and climatic gradients pose significant forecasting challenges. This study presents a comprehensive annual evaluation of four operational global NWP models — ARPEGE (Météo-France), GFS (NCEP/NOAA), IFS (ECMWF), and ICON (DWD) — over Tunisia for the full year 2025 (January–December). Model outputs are systematically compared against ground-truth SYNOP observations from 30 meteorological stations distributed across three climatic zones (North, Centre, and South). Three meteorological variables are assessed: maximum temperature (Tmax), minimum temperature (Tmin), and daily precipitation accumulation. Performance is quantified using the Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Pearson correlation coefficient (R). Results demonstrate that ICON consistently achieves the lowest RMSE for temperature forecasts across all regions, with values as low as 1.22°C for Tmax in the Centre zone and 2.18°C for Tmin in the South. IFS emerges as the superior model for precipitation, particularly over the North (MAE: 4.89 mm, R: 0.62) and South (MAE: 2.95 mm). A performance Weighted Ensemble Forecasting (WEF) strategy is proposed, assigning dominant weights to the best-performing model per parameter and region. These findings provide actionable guidance for operational meteorological vigilance in Tunisia, with direct implications for heat-wave, frost, and convective precipitation warnings.