Ambient Air Pollution Prediction

Accurately predicting air pollutant concentration remains challenging but essen-
tial in preventing the public from being exposed to high concentrations that lead
to several hundred thousand premature deaths across Europe yearly (European
Environment Agency, 2023).

This research focuses on enhancing hourly regional pollutant forecasts in local
urban environments using machine learning (ML). To achieve this, a data set
was collected, harmonized, and preprocessed to represent eleven major German
cities that served as the basis for answering the research question. The ability to
predict P M2.5 and N O2 at a target location with and without corresponding local
measurements at that point was evaluated for several employed ML algorithms.
Incorporating measurements at the designated sites enabled the locally implemented
ML algorithms to diminish the error in the regional forecast by 29.77% and 44.99%
for P M2.5 and N O2 , respectively. Even in the absence of measurements, a notable
reduction in error by 15.14% and 18.71% compared to the regional forecast was
evident at the target location.

Thus, the presented study shows how valuable locally employed ML algorithms
can be to enhance regional forecasts in urban environments, suggesting that these
proposed methods can help warn the citizens about estimated high concentrations
if operational.

In der Regel sind die Vorträge Teil von Lehrveranstaltungsreihen der Universität Bremen und nicht frei zugänglich. Bei Interesse wird um Rücksprache mit dem Sekretariat unter sek-ric(at)dfki.de gebeten.

zuletzt geändert am 31.03.2023