Effects of climatic control on microenvironmental responses in desert PV systems: A comparative analysis with machine learning interpretation

Microenvironmental responses to desert photovoltaic (PV) systems have been increasingly reported, yet how these responses vary across contrasting arid climate backgrounds remains insufficiently understood under comparable engineering settings. In this study, cross-climate field observations, Δ-response modeling, and SHAP (Shapley additive explanations) interpretation were combined to examine PV-induced environmental change at the Wuwei and Hami sites in northwestern China. Instead of reproducing absolute local states, the models estimated PV-induced deviations from the reference (REF) position, ΔY(position) = Y(position) − Y(REF), for the below-panel (BL) and interrow (IT) positions. The Wuwei site exhibited coupled hydrothermal regulation, with enhanced shallow soil-water retention after rainfall, whereas the Hami site exhibited predominantly thermal regulation because its moisture-retention pathway was largely inactive under extremely low precipitation, coarse soil, and strong evaporative demand. Soil-temperature perturbations were consistently predictable (best-model R 2 = 0.93–0.99), whereas air-temperature perturbations showed moderate skill (R 2 = 0.41–0.65) and relative-humidity perturbations were strongly site dependent (Hami: 0.14–0.19; Wuwei: 0.47–0.97). Grouped SHAP analysis distinguished the predictor groups associated with air-temperature, soil-thermal, and soil-moisture responses. Sparse interaction networks further elucidated the joint roles of these predictor groups in shaping model responses across climates and distinguished relatively stable thermal structures from climate-sensitive hydrothermal linkages. It is hoped that these analyses will contribute to the development of climate-adaptive, site-specific environmental monitoring strategies for desert PV plants.