University of Twente Student Theses


Classification of urban morphology and its relationship with air pollution using deep learning

Amouei, Morteza (2023) Classification of urban morphology and its relationship with air pollution using deep learning.

[img] PDF
Abstract:Air pollution poses a significant threat to public health and the environment. Several factors, rooted in the underlying processes in urban areas, affect air quality. This research aims to analyze, model, and develop the relationship between urban forms and PM2.5 concentration using a deep learning-based model with scene-based comprehension to capture complex interactions applied to earth observation data. The Local Climate Zones (LCZ) framework is selected for this research. Research objectives include developing an accurate deep learning model for LCZ classification, and training a suitable model to represent the impact of LCZ on PM2.5 distribution, followed by analyzing the sensitivity and feature importance of different LCZ categories. The study presents a two-stage framework that classifies local climate zones (LCZ) using three supervised convolutional neural networks models, namely the designed CNN by the author, ResNet-50, and EfficientNet models in the first stage and predicts PM2.5 concentration through the regression task of both XGBoost and LSTM models. The methodology involves data acquisition, preparation, and modeling using Sentinel-2 imagery, PM2.5 measurements, meteorological data, and traffic data. The period of the temporal data covers the hourly values between 2021 and 2022. A noteworthy aspect of this research involves citizen science data for air pollution.
Item Type:Essay (Master)
Faculty:ITC: Faculty of Geo-information Science and Earth Observation
Subject:38 earth sciences, 43 environmental science, 54 computer science, 74 (human) geography, cartography, town and country planning, demography
Programme:Geoinformation Science and Earth Observation MSc (75014)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page