AI-Based Spatiotemporal Analysis of Solar and Wind Energy Potential Using Satellite and Ground Sensor Data

Abstract

Renewable energy planning requires accurate estimates of solar and wind resources over space and time. This study develops a spatiotemporal framework that uses satellite and reanalysis data to assess solar irradiance and wind speed. We use NASA’s Power and PVOUT satellite data, ECMWF’s ERA5 reanalysis, and ground observations as inputs. These data are processed to capture both spatial variability (via convolutional neural networks) and temporal trends (via LSTM and graph neural networks) in renewable potential. Our approach is tested on a case region using Google Earth Engine and public datasets. Results show that combining spatial satellite information with temporal weather data improves prediction accuracy (e.g. solar PV output and wind power) over purely numerical models. For example, we observe that the windiest sites align with historical patterns, and solar irradiance maps match known sunny regions. The methodology also reveals solar-wind complementarity: in some areas wind peaks during seasons of weaker sun, helping balance supply. Key contributions of this study include integration of high-resolution satellite and ERA5 data, hybrid AI models capturing spatiotemporal dependencies, and case-study results demonstrating enhanced resource mapping for policy planning. The proposed framework can aid grid operators and planners in identifying high-potential solar and wind sites under varying conditions.