Carbon reduction assessment of PV-assisted mine land reclamation using deep learning from remote sensing images

Abstract Coal mining in high groundwater table regions (HGTRs) of central-east China has caused severe land degradation and greenhouse gas emissions, necessitating sustainable energy solutions. Photovoltaic (PV) technology offers significant potential for carbon reduction in these areas, yet existing studies face limitations in PV mapping (e.g., reliance on single-sensor data and coarse spatial resolutions) and lifecycle carbon assessments (e.g., incomplete stage-specific emission analysis). To address these gaps, this study proposes a multi-scale two-stage deep learning framework integrating YOLOv7 for preliminary PV localization and DeepLabv3 + for refined segmentation, combined with a lifecycle assessment (LCA) to quantify carbon reduction benefits for assessing carbon reduction of PV in high groundwater table regions (HGTRPVs) during land reclamation. Experiments across five HGTR provinces demonstrate high accuracy (98.11% localization, 94.74% segmentation) and identify 1,491.85 km 2 of PV-deployable areas. Results reveal that over 45% ofche lifecycle emissions originate from PV manufacturing, The annual PV potential ranges from 6193.74 GWh to 24,118.06 GWh, with carbon payback periods ranging from 7.05 to 8.29 years. This framework provides actionable insights for optimizing PV deployment in ecologically sensitive mining regions, supporting China’s carbon neutrality goals.