Estimating the impact of PM2.5 on solar power with machine learning: Evidence from South Korea

This study examines the effect of air pollution on solar photovoltaic (PV) power generation in South Korea, utilizing hourly provincial data from 2017 to 2023. We apply a double/debiased machine learning (DDML) framework to estimate the impact of PM2.5 on solar PV output, addressing potential endogeneity by using wind direction as an instrumental variable. Our results reveal that increased PM2.5 concentrations significantly reduce solar PV generation, with the DDML model showing a larger marginal impact compared to conventional ordinary least squares (OLS) and instrumental variable (IV) methods. Specifically, a 10 % increase in PM2.5 leads to a 4.4 % decline in solar PV output, far exceeding the 0.4 % reduction estimated with OLS and the 3.2 % decline with IV. These findings highlight the limitations of OLS and IV methods in capturing the complex, potentially non-linear relationship between air pollution and PV performance. To address this critical methodological gap, the DDML approach leverages the flexibility of machine learning techniques to offer more robust causal estimates, mitigating biases from functional form misspecification and high-dimensional confounding. Our results indicate that a 1μg/m3 increase in hourly average PM2.5 leads to substantial economic losses in the solar sector, estimated at 0.53 million USD per year. These findings highlight the significant economic benefits of policies aimed at reducing air pollution, as cleaner air can enhance the efficiency of renewable energy systems and support the transition to a sustainable energy future.