Abstract:

Arid regions are key areas for photovoltaic （PV） development， yet their ecosystems are fragile and highly sensitive to external disturbances. This study investigates the spatiotemporal dynamics of PV expansion and its ecological impacts on vegetation in the Aksu River Basin， Xinjiang， China. Using multi-temporal Landsat 8 imagery from 2015， 2020， and 2024， combined with SRTM DEM data， a multi-dimensional feature set integrating spectral， texture， and topographic variables was constructed. Object-oriented classification based on SNIC superpixel segmentation and Random Forest algorithm was applied for accurate PV facility identification， achieving an overall accuracy of 92.6%. Based on NDVI annual maxima， vegetation changes within PV areas and their 100 m， 500 m， and 1000 m buffer zones were extracted. Global Moran's I and LISA statistics were employed to assess spatial clustering and ecological responses. The results show： （1） From 2015 to 2024， PV area expanded more than 12-fold， with a "belt-patch" spatial pattern concentrated along transportation routes and low-slope sandy lands； （2） NDVI increased across all buffer zones， especially in the 500 m and 1000 m zones， forming a spatial gradient of “central disturbance-peripheral restoration”； （3） NDVI differences in 2024 were significantly higher than in 2020， indicating a clear time-lagged ecological effect； （4） Spatial autocorrelation analysis revealed strong positive clustering of NDVI， with High-High （HH） clusters expanding from local patches to regional scales， suggesting a potential “positive ecological spillover effect” of PV development； （5） The object-based approach integrating multi-source features demonstrated high accuracy and robustness in PV mapping. This study provides technical references and scientific support for ecological assessment and sustainable planning of PV deployment in arid regions.

Key words: PV detection, NDVI, arid regions, SNIC segmentation, random forest, Moran's I, LISA, ecological response