Bridging climate modeling and Artificial Intelligence for enhanced renewable energy forecasting and siting

The accelerating transition toward electrification and decarbonization increases the need for accurate, scalable forecasting tools to guide renewable energy deployment. Most existing approaches treat climate prediction and spatial suitability analysis as independent tasks, constraining the integration of short-term forecasts into site selection frameworks. This study addresses this gap by investigating how Artificial Intelligence, specifically Gated Recurrent Units, can enhance renewable energy site planning through improved short-term climatic forecasting embedded into spatial decision-making frameworks. The primary contribution lies in directly integrating Machine Learning-based climate predictions with physics-based energy production models within a unified multi-criteria framework applicable to both photovoltaic and wind technologies. The proposed models predict key climatic variables (solar irradiance, temperature, wind speed, and direction) over a 3-day horizon. Performance is evaluated by comparing Machine Learning forecasts and a regional climate model against weather-station observations at two Italian sites. Results demonstrate substantial improvements: for photovoltaic systems, prediction errors decrease by 11%–60% with consistently higher correlations across seasons; for wind energy, errors decrease by 13%–27%. By coupling data-driven climate forecasting with physics-based energy models, the framework enhances the accuracy, robustness, and spatial relevance of renewable energy assessments, providing more reliable support for site planning and grid integration decisions. • Machine learning optimizes site selection for solar and wind projects. • Integration of geospatial, meteorological, and environmental data. • Neural networks outperform the MM5 model in forecasting accuracy. • Up to 60% reduction in renewable energy forecasting error.