The large-scale development of photovoltaic (PV) panel arrays in Northwestern China has proposed an created an urgent demand for high-precision spatial information extraction and management. In this study, Sentinel-2 MSI imagery in 2024 was used to establish a dataset of PV panel arrays. By integrating spectral and spatial features, a multi-scale segmentation approach was applied, and three classification methods—nearest neighbor, sample-based object-oriented, and BP neural network—were comprehensively evaluated. Accuracy assessment based on confusion matrices with independent validation samples identified the BP artificial neural network as the optimal model, resulting in a 10 m spatial resolution dataset of photovoltaic panel arrays in Northwestern China, with an overall accuracy (OA) of 91.6% and a Kappa coefficient of 0.89, demonstrating high precision and reliability. Temporal comparison and typical-area spatial distribution analysis using the “10 m National Ground-Mounted PV Dataset of 2020” and historical Wayback imagery, further confirmed the dataset’s reliability and advantages in temporal updating and spatial identification accuracy. The results indicate that by 2024, the total area covered by PV panel arrays in Northwestern China reached 1,676.13 km 2, with the largest areas in Xinjiang (605.80 km2), Qinghai (484.27 km2), Gansu (272.98 km2), Ningxia (174.65 km2), and Shaanxi (138.43 km2 ). These installations are mainly concentrated along the periphery of the Taklamakan Desert, the northern slopes of the Tianshan Mountains, Haixi and Hainan Prefectures in Qinghai, the Hexi Corridor in Gansu, Yinchuan and Wuzhong in Ningxia, and Yulin in Shaanxi. This dataset provides a reliable basis for energy planning, ecological assessment, and sustainable development under China’s “dual-carbon” strategy.
WEI et al. (Sun,) studied this question.