A geospatial machine learning approach for solar radiation prediction in a changing climate: Integrating atmospheric and surface parameters

This study presents a Geospatial Artificial Intelligence (GeoAI) framework to enhance solar radiation predictions under climate change. Using Bayesian Regularised Artificial Neural Networks (BRANNs), the framework downscales General Circulation Model (GCM) outputs to high-resolution data over the island of Ireland, addressing GCM limitations in capturing localised solar patterns. Key atmospheric and surface variables, including cloud cover, relative humidity, pressure gradients, and topography, are incorporated to improve projection accuracy. The analysis employs two climate change scenarios, SSP2–4.5 and SSP5–8.5 (Shared Socioeconomic Pathway), for the period 2000–2050. A dedicated post-processing toolbox is developed and applied to the case study, demonstrating its capability to refine GCM simulations for both current and future conditions. The BRANN models achieved high accuracy during the baseline period (2000–2014). Projections indicate a decrease in monthly mean solar radiation of 2.01 % under SSP2–4.5 and 1.61 % under SSP5–8.5 by 2050, corresponding to mean monthly values of 9.87 × 10⁶ J.m⁻² and 9.91 × 10⁶ J.m⁻², respectively, with higher values in southeastern Ireland. The framework supports optimal solar panel site selection and provides long-term performance projections, enabling climate scientists and policymakers to make informed decisions for sustainable energy planning.