| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 8.11 MB | Adobe PDF |
Authors
Advisor(s)
Abstract(s)
Air pollution is one of the most significant environmental threats to public health and ecosystems,
with substantial socio-economic repercussions. According to the European Environment Agency (EEA),
it remains the leading environmental health threat in Europe, causing over 400,000 premature deaths
annually.
This dissertation aims to characterize the spatiotemporal behavior of pollutants, focusing on PM10
in mainland Portugal between 2003 and 2022. It explores the variations in pollutant concentrations
during this period, highlighting extreme events, and the relationships between pollutants and various
atmospheric variables. The study utilizes the Copernicus Atmosphere Monitoring Service (CAMS) data
and the Portuguese Environment Agency’s QualAR monitoring network.
Machine learning methods were tested and implemented. Two different neural network model
architectures were developed and tested for each Portuguese NUTS II region– a multilayer perceptron
(MLP) and a deep learning long short-term memory (DL-LSTM) model. Data from 2003-2021 was used
for training and cross-validation, while 2022 served as an independent testing year. Finally, the models
were tested in two extreme events, the October 2017 wildfires in central Portugal, and a dust intrusion
episode in March 2022 in southern Portugal.
The findings reveal significant trends in pollutant levels and their correlation with meteorological
factors, emphasizing the influence of extreme weather events like forest fires and dust intrusions.
The models demonstrated strong predictive performance, with MLP achieving a correlation
coefficient of 0.97, and having a slight advantage over the DL-LSTM model, but both models show high
effectiveness in predicting PM10 values. This work contributes to a better understanding of air quality
dynamics in Portugal and can inform future environmental and public health strategies. We expect that
in the future, this work can serve as the basis for an operational tool that can help authorities adopt
suitable measures during air quality emergencies.
Description
Tese de mestrado, Ciências Geofísicas, 2025, Universidade de Lisboa, Faculdade de Ciências
Keywords
Qualidade do ar Poluentes Meteorologia Redes neuronais Previsão Teses de mestrado - 2025