Globally Scalable, QGIS-Integrated Workflow for Solar Photovoltaic System Segmentation and Capacity Estimation: A Case Study in Algeria

This study presents a deep learning-based approach for the segmentationof Solar Photovoltaic (PV) systems using a ResNet-basedDeepLabV3 architecture. The method is seamlessly integrated intothe open-source Geographic Information System QGIS environmentthrough the user-friendly Deepness plugin and allows intuitive applicationon geospatial base maps anywhere in the world. A casestudy was conducted in the Piat region of Algeria, focusing onseven large PV systems within an autonomous power grid. Aftersegmentation, a power density factor in megawatt-peak (MWp) wasused to calculate the capacity of each identified PV system. Thecalculated capacities were compared with the official values of theoperators, showing a mean relative error of 8.5 %. To further refinethe capacity estimates, we introduce a locally adjusted factor of 73.5MWp/km², which reflects the expected results for the region moreprecisely. The approach is globally applicable, easily transferable todifferent geographical contexts and accessible to users without programmingknowledge, offering significant potential for renewableenergy analysis. The dataset is publicly available for download [1]and can be used interactively [2].