Machine learning exposure model predictions for ground-level ozone during wildfire events: results for a major wildfire in California

The aim of this paper is to develop prediction models for ground-level ozone that results from major wildfire events. We evaluate the predictive accuracy of 12 machine-learning methods. We used ground level ozone monitoring as the dependent variable and a variety of emissions model estimates, satellite products, meteorological parameters, and land use variables as predictors. The gradient boosting model supplied the best results with a cross-validation R2 of 0.653. Our results demonstrate that machine learning methods can produce models that predict ground level ozone well, even in highly dynamic wildfire events. The results can supply important information for public health protection during wildfires and for studying the health effects of the ozone exposures resulting partly from wildfires.