Satellite imagery to select a sample of rooftops for a PV installation project in Jeddah, Saudi Arabia

Abstract A region-based convolutional neural network image segmentation approach (Mask R-CNN) was applied to identification of flat rooftops from satellite imagery in the city Jeddah Saudi Arabia. The model trained on a small sample (202) digitized 0.5 m resolution (covering 0.21 km 2 ) and then an independent area 4.5 away. precision recall were 0.98 0.96 respectively terms identifying area. spatially stratified drawn those identified by median roof not significantly different as whole. results, although at scale, demonstrate effectiveness this for selecting buildings with appropriate solar photovoltaic (PV) installation, context closely spaced flat-roofed buildings, without requiring cadastral mapping or LIDAR datasets.