Vulnerability and pathways of global renewable energy transition under climate exposure

Under intensifying climate exposures, assessing the renewable energy transition (RET) vulnerability and identifying the effective pathways for RET are increasingly critical. However, current research on RET vulnerability (RETV) neglects climate exposure factors and lacks an interpretable assessment framework. Further, the causal pathways to RET under different climate exposures remain underexplored. To address these gaps, this study is the first to integrate the SHAP-enhanced machine learning method with panel data fuzzy-set Qualitative Comparative Analysis (fsQCA). Applying the proposed assessment framework to 16 vulnerability factors across 94 countries from 2010 to 2022, this paper finds that 1) Adaptability factors exert the greatest influence, followed by sensitivity and then exposure factors. Economic development, government efficiency, energy dependence, and government revenue emerge as the most influential variables, each contributing over 10 % to RET variation globally. Over the study period, 86 of 94 countries exhibit reductions in RETV, leading to an approximate 7.26 % decline in global RETV. 2) The fsQCA results uncover 4 pathways leading to high-RET performance and 7 with low RET performance, confirming economic development and government efficiency as core sufficient conditions. Notably, countries with lower climate exposure require less stringent conditions to achieve high RET than their high-exposure counterparts. 3) Temporal analysis reveals that the pathways leading to low-RET display strong path dependence, but this mitigates over time. In contrast, high-RET pathways grow increasingly robust, particularly in high-exposure countries, while remaining relatively stable in low-exposure settings. These trends, alongside a global decline in the RET vulnerability, point to a growing global momentum toward RET. Methodologically, this study contributes a data-driven, interpretable weighting framework for RET vulnerability assessment and demonstrates the utility of integrating SHAP-enhanced machine learning and fsQCA in capturing complex, time-sensitive causal mechanisms, offering broad applicability to other domains concerned with vulnerability and transition dynamics.