Abstract

This research project, in collaboration with the Department for Climate and Environmental Protection (RKU), analyses data from NO2 measuring stations in the Munich city area, provided by the State Office for the Environment. The project used machine learning models to evaluate the prediction of NO2 emissions and perform an influence factor analysis by supplementing the data with traffic measurements from the mobility department and weather data from the German Weather Service. The results showed that prediction and influence factor analysis are possible, but the study also highlighted several limitations to these methods, which are thoroughly discussed. The research project achieved a prediction of NO2 with an RMSE of 13.9 by using available covariates. The study also explored use cases such as special extrapolation with these models. The influence factor analysis was conducted using Forest-Guided Clustering, mean decrease in impurity, Permutation Importance, and Partial Dependence Plots, with the latter yielding the best results. The analysis revealed that wind, temperature, ozone, and humidity significantly influenced NO2 concentrations, while precipitation had no effect.

Presenters

Leon Lukas
Machine Learning Engineer, AI-Competence-Center, City of Munich(it@m), Germany

Details

Presentation Type

Paper Presentation in a Themed Session

Theme

Ecological Realities

KEYWORDS

NO2 Emissions, Machine Learning, Influence Factor Analysis, Prediction, Munich, Climate

Digital Media

Downloads