Detailed photovoltaic mapping and multidimensional analysis based on deep learning: a case study in Jiangsu Province, China

Solar photovoltaics (PV) constitute a key component of renewable energy, and accurately mapping their spatial distribution and assessing their environmental benefits are crucial for energy planning and sustainable development. Focusing on Jiangsu Province, China, this study employs deep learning techniques to develop a semantic segmentation network named PV-AssNet for the precise extraction of PV distributions from high-resolution remote sensing imagery. Building on this, a systematic analysis is conducted to investigate their spatial patterns, land use preferences, and carbon reduction benefits. The results show that PV-AssNet enhances PV identification performance, achieving an overall mapping accuracy of over 98% and outperforming other comparative models. The total PV area in Jiangsu Province in 2020 was 142.01 km 2 , exhibiting significant spatial heterogeneity, with high-density clusters mainly concentrated in cities of Yancheng, Suqian, and Huai'an. Water surfaces (primarily reservoirs and ponds) were the dominant land use type for PV deployment, accounting for 45.71% of the total area, followed by cropland and construction land. The carbon payback period for PV systems was approximately 2.19 years, achieving considerable net carbon emission reduction benefits over its lifecycle, with the pace of carbon benefit realization showing minimal variation across different land types. The findings provide a scientific basis for PV resource planning and territorial spatial policy formulation in Jiangsu Province, as well as a reference for coordinating regional renewable energy layout and land use optimization.