Rising Ozone Exposure and Future Public Health Risks in Germany and Neighboring Regions Revealed by Automated Machine Learning (AutoML)

Abstract Ground-level ozone remains a major global health and environmental challenge, while climate change is expected to exacerbate ozone pollution. However, chemical transport models (CTMs), such as the Community Multiscale Air Quality Modeling System (CMAQ), still suffer from systematic biases and limited predictive capabilities, highlighting the need for more accurate future projection models. This study aims to develop an improved ozone projection framework by integrating Automated Machine Learning (AutoML) with CMAQ simulations that project future ozone concentrations in Germany and its neighboring regions under climate change scenarios. Specifically, a fast and lightweight AutoML-enabled framework was developed to optimize machine learning models based on CMAQ simulations, meteorological data, land use, and socioeconomic factors. The model is trained using data from 2018 and validated against ground-level ozone observations from 2019, showing substantial improvements over CMAQ simulations, with the average residual improved from −23.54 μg/m3 to 1.59 μg/m3. According to projections for 2030 under the Shared Socioeconomic Pathways (SSPs), summer ozone levels in Germany are expected to exceed the 2030 European Union (EU) air quality targets, particularly under high-emission scenarios. Health risk assessments also suggest an increase (4% of overall exposure in SSP585 than in SSP126) in ozone-related health impacts. Our findings highlight a simple yet effective approach for improving hybrid air quality modeling using AutoML. They also provide an early warning for Germany and neighboring regions about the potential exceedance of the 2030 EU standard. These results further underscore the associated public health risks and offer insights for future climate and air quality policy development.