Machine learning-assisted dispersion modelling based on genetic algorithm-driven ensembles: An application for road dust in Helsinki

A novel Machine Learning (ML) approach is proposed to assist the dispersion modelling of road dust in an urban area. The aim is to improve ENFUSER model's coarse particle nowcasting estimations, especially during the road dust period in springtime. We assist the dispersion model by combining road condition measurements, regionalscale air quality forecasts, meteorological data, road maintenance information, detailed traffic patterns, as well as GIS-data, for the Helsinki area, via a ML based data fusion procedure. The proposed pipeline consists of feature engineering, a set of feature selection methods, and a Genetic Algorithm search for the optimal stacking ensemble configuration, as well as robust validation in both space and time dimensions. Results suggest that our approach enhances the current ENFUSER capabilities, and thus could be applied to assist its performance under operational conditions, while also offering potential as a versatile tool for addressing a broad range of similar tasks.

Automated summary

A novel machine learning approach is proposed to enhance the dispersion modelling of road dust in urban areas. The approach combines various data sources such as road condition measurements, air quality forecasts, meteorological data, and road maintenance information through a ML based data fusion procedure. The results show that this approach improves the current model's capabilities and can be applied to a wide range of similar tasks.