Identifying the causal effects of photovoltaic installations on grassland productivity using double machine learning: a case study in inner Mongolia

Driven by the global energy transition and the “dual-carbon” goals, the rapid deployment of large-scale photovoltaic (PV) installations has profoundly reshaped land surface processes. This transformation is particularly pronounced in arid and semi-arid grassland ecosystems, where the potential ecological impacts of PV construction remain both critical and controversial. However, most existing studies rely primarily on correlation analyses, which fail to accurately identify the true causal effects of PV installations on ecosystem productivity. In this study, we focus on typical grasslands in Inner Mongolia, China, integrating multi-source remote sensing datasets including MODIS net primary productivity (NPP), meteorological, topographic, and anthropogenic factors. A double machine learning (DML) approach is employed within a quasi-experimental framework to quantify the ecological causal effects of PV construction. The results reveal that the average treatment effect (ATE) of PV installation on grassland NPP is − 0.00427 (p > 0.05), indicating a statistically insignificant overall impact. Nevertheless, substantial spatial heterogeneity exists: approximately 62.1% of PV sites exhibit positive ecological effects, whereas 37.9% show negative ones. SHapley Additive exPlanations (SHAP) based analysis further identifies that the spatial variability of PV-induced ecological effects is primarily regulated by environmental factors such as distance to water bodies, mean annual temperature, potential evapotranspiration, soil moisture, and drought index. This study demonstrates the effectiveness and advantages of the DML framework in identifying causal ecological interventions. The findings provide a scientific basis for implementing “ecology-prioritized, site-specific” PV development strategies and highlight the necessity of integrating ecological baseline assessments and spatially precise management to achieve a synergy between energy security and ecosystem conservation.