Quantitative prediction and driving factors analysis for the eco-environmental impacts of photoelectric development in the Tengger Desert

With the full implementation of the "14th Five-Year Plan" and under the policy impetus of the "dual carbon" strategic goals, China is tilting its new infrastructure construction towards the new energy sector. Among new energy resources, solar energy, given its outstanding environmental friendliness and sustainability, has become a key focus for the green transformation of the energy system. Against this backdrop, the construction of large-scale photovoltaic (PV) bases serves as a core measure driving the green transformation of the energy structure. Therefore, systematically assessing their potential eco-environmental impacts holds significant practical significance. This study was performed in the largest PV industrial park located on the southern edge of the Tengger Desert, the Zhongwei PV Industrial Park in Ningxia, based on remote sensing data and meteorological data from the Moderate Resolution Imaging Spectroradiometer (MODIS) and LANDSAT satellites. It analyzed the spatiotemporal variations and interrelationships of the normalized difference vegetation index (NDVI), the fraction of photosynthetic active radiation (FPAR), the leaf area index (LAI), the gross primary productivity (GPP), the snow-free surface albedo (SSA) and land surface temperature (LST) from 2000 to 2023, using the Theil-Sen trend analysis, Mann-Kendall trend test, partial derivative analysis, and machine learning. The study also conducted a comparative analysis of data from inside and outside the PV park at different time periods, aiming to reveal the impact of PV power station construction on the ecological environment of the Tengger Desert. The results indicate that the study area generally exhibited an upward trend in vegetation and a downward trend in LST and albedo. The PV area demonstrated superior vegetation growth compared with the area outside the PV panels. Variations in the following parameters were primarily driven by climate: NDVI, FPAR, LAE, SSA and LST, with climate contributions estimated to be 0.433, 0.360, 0.354, -0.209, and -0.355, respectively. In contrast, GPP was jointly influenced by photovoltaic activities and climate (contribution: 2.356 and 0.374 respectively). Further machine learning-based analysis indicates that the time series model integrating convolutional neural networks (CNN) and long short-term memory (LSTM) can effectively predict the trend of NDVI changes. The results using the model show that the NDVI in the PV area showed a significant upward trend. This study provides new methods and ideas for the quantitative assessment and future prediction of the eco-environmental impacts of PV power stations in deserts.