How subsidy withdrawal reshapes photovoltaic deployment: Spatial heterogeneity and causal evidence from China

The rapid expansion of photovoltaic (PV) power is central to global low-carbon energy transitions, yet many PV markets are entering a post-subsidy phase as policy priorities shift toward market-based deployment and long-term efficiency. However, how subsidy withdrawal reshapes the spatial configuration of PV deployment, its drivers, and regional heterogeneity remains poorly understood, constraining energy planning in mature PV markets. This study proposes an integrated analytical framework combining spatial autocorrelation, machine-learning-based driver attribution, and causal inference to assess the impacts of China's 2018 “531 Policy” on PV deployment in the Yangtze River Delta (YRD). Results show that subsidy withdrawal coincided with marked spatial reorganization of PV deployment from 2014 to 2023, characterized by intensified spatial patterning and polarization. XGBoost–SHAP analysis reveals a shift from a multi-factorial regime dominated by infrastructure, urbanization, and economic activity to a resource-oriented regime governed by solar radiation, with reduced socioeconomic influence and weaker nonlinear interactions. Counterfactual analysis using Bayesian structural time-series models indicates that the aggregate policy effect is modest and statistically uncertain, with substantial temporal variation. PV deployment showed short-term moderation during 2018–2020 followed by a partial rebound, with no evidence of a strong long-run regional suppressive effect. Spatial causal machine learning further reveals pronounced city-level heterogeneity and localized spillovers along economic and transport linkages. Overall, post-subsidy PV development is increasingly shaped by resource suitability, spatial structure, and system-level interactions, highlighting the need for spatially differentiated and adaptive policy strategies rather than uniform expectations of policy suppression in mature PV markets.