Time Filter Assisted Deep Learning to Predict Air Pollution
Time Filter Assisted Deep Learning to Predict Air Pollution
dc.contributor.advisor | Schultz, Martin Georg | |
dc.contributor.author | Leufen, Lukas Hubert | |
dc.date.accessioned | 2023-07-13T11:08:35Z | |
dc.date.available | 2023-07-13T11:08:35Z | |
dc.date.issued | 13.07.2023 | |
dc.identifier.uri | https://hdl.handle.net/20.500.11811/10941 | |
dc.description.abstract | Exposure to ground-level ozone harms human health as well as the entire ecosystem, so accurate prediction of ozone exposure is of particular importance. Machine learning (ML), and deep learning (DL) in particular, has emerged as a powerful method with a vast variety of applications, including meteorology and Earth system sciences, making it a strong alternative to conventional methods such as chemical transport models (CTMs) or regression based solutions to forecast ground-level ozone. However, to date, classical as well as ML approaches have experienced challenges in reliably forecasting ozone pollution at the local scale. These shortcomings can be attributed to the challenges posed by inherent uncertainties about near-future weather conditions and the superposition of patterns on different time scales. In this thesis, a time series filtering approach to split up long-term and short-term variations and DL are applied to allow for accurate predictions of air pollution attributable to ground-level ozone. This is complemented by integrating large amounts of data from air quality monitoring stations distributed across Central Europe, climatological statistics on air pollutants and meteorological data from numerical weather models. The DL approach is framed by a well-defined workflow for training and validation called MLAir, which ensures the reproducibility of the findings. Results substantiate that the combination of sophisticated DL architectures and time series filtering enables accurate ozone prediction. The DL approach thereby achieves a nearly bias-free prediction and has a good performance with regard to the seasonal variability of ozone. This leads to a great improvement compared to simpler reference forecasts based on climatology and persistence, as well as to the Copernicus Atmosphere Monitoring Service (CAMS) regional multi-model ensemble forecast, which combines nine individual state-of-the-art CTMs deployed operationally by public weather services and research institutions. Averaged over a forecast horizon of four days, the prediction for the daily maximum 8-hour running average (dma8) of ozone by the CAMS regional ensemble has a root mean squared error (RMSE) of 7.6 ppb, whereas the newly developed method here achieves an RMSE of 5.1 ppb. The approach presented in this thesis thus marks an important advance in DL-based air pollution prediction, benefiting the general public through more reliable forecasts. Furthermore, this study opens up the prospect of further research opportunities towards the prediction of a range of other air pollutants or related applications in meteorology. | en |
dc.language.iso | eng | |
dc.relation.ispartofseries | Bonner Meteorologische Abhandllungen ; 96 | |
dc.rights | In Copyright | |
dc.rights.uri | http://rightsstatements.org/vocab/InC/1.0/ | |
dc.subject | Deep Learning | |
dc.subject | Ozone | |
dc.subject | Air Pollution | |
dc.subject | Forecast | |
dc.subject | Neural Networks | |
dc.subject.ddc | 004 Informatik | |
dc.subject.ddc | 500 Naturwissenschaften | |
dc.subject.ddc | 550 Geowissenschaften | |
dc.title | Time Filter Assisted Deep Learning to Predict Air Pollution | |
dc.type | Dissertation oder Habilitation | |
dc.publisher.name | Universitäts- und Landesbibliothek Bonn | |
dc.publisher.location | Bonn | |
dc.rights.accessRights | openAccess | |
dc.identifier.urn | https://nbn-resolving.org/urn:nbn:de:hbz:5-71217 | |
dc.relation.doi | https://doi.org/10.5194/gmd-14-1553-2021 | |
dc.relation.doi | https://doi.org/10.1017/eds.2022.9 | |
dc.relation.doi | https://doi.org/10.1175/AIES-D-22-0085.1 | |
ulbbn.pubtype | Erstveröffentlichung | |
ulbbnediss.affiliation.name | Rheinische Friedrich-Wilhelms-Universität Bonn | |
ulbbnediss.affiliation.location | Bonn | |
ulbbnediss.thesis.level | Dissertation | |
ulbbnediss.dissID | 7121 | |
ulbbnediss.date.accepted | 20.06.2023 | |
ulbbnediss.institute | Mathematisch-Naturwissenschaftliche Fakultät : Fachgruppe Erdwissenschaften / Institut für Geowissenschaften | |
ulbbnediss.fakultaet | Mathematisch-Naturwissenschaftliche Fakultät | |
dc.contributor.coReferee | Hense, Andreas | |
ulbbnediss.contributor.orcid | https://orcid.org/0000-0003-4154-3397 |
Dateien zu dieser Ressource
Das Dokument erscheint in:
-
E-Dissertationen (4137)