A data-driven approach to forecasting ground-level ozone concentration

The ability to forecast the concentration of air pollutants in an urban region is crucial for decision-makers wishing reduce impact pollution on public health through active measures (e.g. temporary traffic closures). In this study, we present a machine learning approach applied forecasts day-ahead maximum value ozone several geographical locations southern Switzerland. Due low density measurement stations and complex orography use-case terrain, adopted feature selection methods instead explicitly restricting relevant features neighborhood prediction sites, as common spatio-temporal forecasting methods. We then used Shapley values assess explainability learned models terms importance interactions relation predictions. Our analysis suggests that trained effectively explanatory cross-dependencies among atmospheric variables. Finally, show how weighting observations helps increase accuracy specific ranges ozone’s daily peak values.