Forecasting heavy precipitation in Portugal: AROME numerical prediction model performance

Abstract High-quality precipitation forecasts are essential to ensure public safety and minimise economic losses during extreme precipitation events (EPEs). However, precipitation forecasting remains challenging due to the complexity of its driving mechanisms and model limitations in resolving short temporal scales and subgrid processes. This study presents the first assessment of the precipitation forecast skill of the Application of Research to Operations at Mesoscale (AROME) model in Portugal. AROME precipitation was compared with data from 105 operational automatic weather stations for 2022–2023. Forecast skill scores derived from contingency tables (Probability of Detection, Heidke Skill Score, Symmetric Extreme Dependency Score, and Frequency Bias) indicate a decline in model performance with increasing precipitation thresholds. However, the AROME model shows improved performance for longer accumulation periods and larger spatial neighbourhoods, reflected by the Fraction Skill Score, with better performance in northern Portugal. In an attempt to identify skilful predictors of these events, convective conditions were then analysed during two illustrative EPEs in the Douro (Northern Portugal) and Alentejo (Southern Portugal) wine regions, owing to the potential damage these events may have on viticulture. Thunderstorm diagnostic parameters (K-index, Total-Totals index, Potential instability, and Bulk-Shear) were derived from the AROME model for different forecast ranges. These indices showed a good agreement with the observed lightning discharges, highlighting the model's skill in predicting convective activity. The similarity between forecasts for different steps suggests that such events can be timely forecast by the aforementioned indices. Their operational assessment may enhance early warning and enable the implementation of suitable risk reduction measures.