Environmental indicator changes associated with solar photovoltaic plants: A data-driven multi-regional assessment

With the rapid expansion of photovoltaic (PV) construction, the potential environmental impact of PV power plants has become a crucial consideration for large scale development. This study develops a machine learning-based model, termed the PV Environmental Indicators Forecasting (EIF) model, to predict the changes in typical environmental indicators caused by PV plants. Using one-year environmental monitoring data from four representative PV plants across major climate zones and application scenarios in China, the model enables prediction of six atmospheric and soil environmental indicators for shaded and non-shaded areas of PV plants, including Atmospheric Temperature (AT), Atmospheric Humidity (AH), Deep Soil Temperature (DST), Shallow Soil Temperature (SST), Deep Soil Moisture Content (DSMC), and Shallow Soil Moisture Content (SSMC). Among the tested algorithms, the random forest model achieves the best predictive performance. Under the selected hyperparameters, the PV EIF model yields high accuracy for the six environmental indicators in the internal test dataset, with an average R 2 of 0.99 and RMSE of 0.93. External validation using field data from two additional PV plants in Hami and Wujiaqu further support the model’s predictive capability under comparable climatic and fixed-layout PV conditions, with average R 2 values of 0.90 and 0.95, respectively. This study provides an effective framework for forecasting PV-induced changes in typical environmental indicators. The proposed approach is particularly applicable to fixed-layout, ground-mounted PV systems under climatic and deployment conditions similar to those represented in the monitored datasets, and it can support environmental assessment and ecological design during PV planning.