Interpretable machine learning method empowers dynamic life cycle impact assessment: A case study on the carcinogenic impact of coal power generation

Abstract

Life cycle impact assessment (LCIA) is a crucial tool for sustainable development, cleaner production, and policymaking globally. However, traditional static LCIA methods rely on fixed characterization factors, making it difficult to capture the dynamic changes in environmental impacts over time and space. This study uses an interpretable machine learning method to develop dynamic LCIA for assessing the spatiotemporal carcinogenic impact of coal power generation in China. The results show that the accuracy of the dynamic life cycle carcinogenic assessment (LCCA) outperforms the traditional LCCA. The Pearson correlation coefficient between the dynamic LCCA and cancer cases is 0.676, while that of the traditional LCCA is 0.556. The disease burden caused by pollutants released from coal power generation is spatiotemporal quantified based on dynamic LCCA, and results show that mercury pollutant emissions caused a cumulative disease burden of 661,062 DALYs from 2007 to 2016. Furthermore, the dynamic sensitivity analysis reveals the nonlinear response of disease burden to pollutant emissions. The sensitivity of disease burden to different pollutant emission levels is various, and the response of disease burden is more significant when the pollutant emission level is higher. This study supports the advancement of dynamic LCIA and sustainable environmental health.