Developing a Spatial Framework for Earthquake-Induced Building Damage Assessment Using Remote Sensing Data and Advanced Machine Learning methods

Soleimanimatin, Sahar

Developing a Spatial Framework for Earthquake-Induced Building Damage Assessment Using Remote Sensing Data and Advanced Machine Learning methods

Soleimanimatin, Sahar

Permalink

Publication Type:: Thesis
Issue Date:: 2023

Open Access

Copyright Clearance Process

Recently Added
In Progress
Open Access

This item is open access.

Adobe PDF

Download thesisAdobe PDF (3.6 MB)

View statistics

Full metadata record

Field	Value	Language
dc.contributor.author	Soleimanimatin, Sahar
dc.date.accessioned	2023-09-19T02:14:48Z
dc.date.available	2023-09-19T02:14:48Z
dc.date.issued	2023
dc.identifier.uri	http://hdl.handle.net/10453/172192
dc.description	University of Technology Sydney. Faculty of Engineering and Information Technology.	en_US.UTF-8
dc.description.abstract	Building-damage mapping using remote sensing images is crucial to support post-earthquake rescue and relief activities. Further, in the long term, knowledge regarding details of damage after an earthquake aids decision-makers in better designing buildings and infrastructure which can withstand destruction in future seismic events. Building damage mapping in real-world earthquake scenarios is limited mainly to the manual interpretation of optical satellite images, which is costly and labour-intensive but transparent and reliable. In this regard, a large body of literature proposes novel machine learning-based frameworks for automating earthquake-induced building damage mapping. These frameworks in this domain are promising yet unreliable for several reasons, including but not limited to the dependency on the availability of both pre and post-earthquake images, lack of quality and the limited number of labelled images, along with transferability and explainability of the machine learning models. Therefore, this research addresses these issues by developing a transferable and explainable machine learning-based framework for earthquake-induced building damage assessment using a single post-earthquake satellite image. The city of Palu, Indonesia and Salina Cruz, Mexico, was selected for this research because of the high number of collapsed buildings after the 2018 Indonesia and 2017 Mexico earthquakes, respectively, and the availability of the post-earthquake optical satellite images for these events. In the first step, the performance of the U-Net, multilayer perceptron (MLP), Random Forest (RF) and Support Vector Machine (SVM) was evaluated and discussed to select the best model for identifying the collapsed and non-collapsed buildings in Palu City. The result showed that i) a single post-earthquake image can be used for building damage mapping, and ii) MLP outperformed the other models with an overall accuracy of 84%. Next, to build an explainable model, the Shapley additive explanation (SHAP) was utilised to analyse the impact of each feature descriptor on the output of the proposed MLP model. The results showed that i) the spectral features contribute the most to classifying the collapsed and non-collapsed buildings in the Palu City study area, and ii) SHAP enables us to analyse the data in detail and guides us to build a more diverse dataset. Next, Salina Cruz, Mexico, was considered a geographically different study area to assess the transferability of the proposed model. The conclusions of this research can be used by the researchers and the decision makers towards the use of explainable AI for post-earthquake building-damage mapping.	en_US.UTF-8
dc.format	Thesis (PhD)
dc.language.iso	en_US	en_US.UTF-8
dc.relation	https://opus.lib.uts.edu.au/bitstream/10453/172192/1/thesis.pdf
dc.rights	info:eu-repo/semantics/openAccess
dc.rights	The author owns the copyright in this thesis including all reproduction and reuse rights for the work. The work may not be altered without the permission of the copyright owner. Attribution is essential when quoting or paraphrasing from this thesis.
dc.rights	© 2023 Sahar Soleimanimatin
dc.rights	au.edu.uts.lib/cph
dc.title	Developing a Spatial Framework for Earthquake-Induced Building Damage Assessment Using Remote Sensing Data and Advanced Machine Learning methods	en_US.UTF-8
dc.type	Thesis
utslib.copyright.status	open_access	*

Abstract:

Building-damage mapping using remote sensing images is crucial to support post-earthquake rescue and relief activities. Further, in the long term, knowledge regarding details of damage after an earthquake aids decision-makers in better designing buildings and infrastructure which can withstand destruction in future seismic events. Building damage mapping in real-world earthquake scenarios is limited mainly to the manual interpretation of optical satellite images, which is costly and labour-intensive but transparent and reliable. In this regard, a large body of literature proposes novel machine learning-based frameworks for automating earthquake-induced building damage mapping. These frameworks in this domain are promising yet unreliable for several reasons, including but not limited to the dependency on the availability of both pre and post-earthquake images, lack of quality and the limited number of labelled images, along with transferability and explainability of the machine learning models. Therefore, this research addresses these issues by developing a transferable and explainable machine learning-based framework for earthquake-induced building damage assessment using a single post-earthquake satellite image. The city of Palu, Indonesia and Salina Cruz, Mexico, was selected for this research because of the high number of collapsed buildings after the 2018 Indonesia and 2017 Mexico earthquakes, respectively, and the availability of the post-earthquake optical satellite images for these events. In the first step, the performance of the U-Net, multilayer perceptron (MLP), Random Forest (RF) and Support Vector Machine (SVM) was evaluated and discussed to select the best model for identifying the collapsed and non-collapsed buildings in Palu City. The result showed that i) a single post-earthquake image can be used for building damage mapping, and ii) MLP outperformed the other models with an overall accuracy of 84%. Next, to build an explainable model, the Shapley additive explanation (SHAP) was utilised to analyse the impact of each feature descriptor on the output of the proposed MLP model. The results showed that i) the spectral features contribute the most to classifying the collapsed and non-collapsed buildings in the Palu City study area, and ii) SHAP enables us to analyse the data in detail and guides us to build a more diverse dataset. Next, Salina Cruz, Mexico, was considered a geographically different study area to assess the transferability of the proposed model. The conclusions of this research can be used by the researchers and the decision makers towards the use of explainable AI for post-earthquake building-damage mapping.

Please use this identifier to cite or link to this item:

http://hdl.handle.net/10453/172192