Predictability and Skill of 10-m Wind Speed Forecasts Over Argentina on Seasonal Timescales

Abstract

In recent years, rapid growth in wind power generation has driven a strong demand for skilful wind speed forecasts, especially in regions with high-quality wind resources, such as Argentina. To date, most surface wind speed forecast analyses and applications have been focused on micrometeorological and synoptic timescales. This paper presents the first assessment of the predictability and skill of 10-m height wind speed forecasts over Argentina during the austral summer. The analysis focuses on the three-month mean for December–January–February (DJF), using one-month lead-time predictions from two seasonal forecast systems: the European Centre for Medium-Range Weather Forecasts system 5 (SEAS5) and the Climate Forecast System version 2 (CFSv2). Predictability was assessed through the decomposition of model variance into signal and noise components, and the calculation of potential predictability and perfect-model skill. The forecast skill of the ensemble mean was evaluated using several verification indices. Predictability and deterministic forecast skill are higher for CFSv2. Significant predictability is found in most of Argentina for CFSv2. Both models exhibit poor deterministic forecast skill. A more detailed analysis was conducted in three regions with a high density of wind farms, evaluating both deterministic and probabilistic forecasts. Three-year-out cross-validation was used to build the probabilistic forecasts for three categories based on the terciles of the distribution, and for two simplified categories of events above and below the median. The models demonstrate higher reliability and discrimination ability for predicting median-based events compared to tercile-based categories.