Geospatial assessment of rooftop solar photovoltaic potential using multi-source remote sensing data

Abstract

Rooftop solar photovoltaics (PV) play increasing role in the global sustainable energy transition. This raises the challenge of accurate and high-resolution geospatial assessment of PV technical potential in policymaking and power system planning. To address the challenge, we propose a general framework that combines multi-resource satellite images and deep learning models to provide estimates of rooftop PV power generation. We apply deep learning based inversion model to estimate hourly solar radiation based on geostationary satellite images, and automatic segmentation model to extract building footprint from high-resolution satellite images. The framework enables precise survey of available rooftop resources and detailed simulation of power generation on an hourly basis with a spatial resolution of 100 m. The case study in Jiangsu Province demonstrates that the framework is applicable for large areas and scalable between precise locations and arbitrary regions across multiple temporal scales. Our estimates show that rooftop resources across the province have a potential installed capacity of 245.17 GW, corresponding to an annual power generation of 290.66 TWh. This highlights the huge space for carbon emissions reduction through developing rooftop PVs.

Keywords

• Rooftop photovoltaics
• Building footprints
• Remote sensing
• Deep learning
• Solar energy