A novel approach by integrating physically and Machine leaning-based models for landslide susceptibility assessment

Al-Najjar, Husam Abdulrasool Hammadeh

A novel approach by integrating physically and Machine leaning-based models for landslide susceptibility assessment

Al-Najjar, Husam Abdulrasool Hammadeh

Permalink

Publication Type:: Thesis
Issue Date:: 2023

Embargoed

	Filename	Description	Size
	thesis.pdf	thesis	7.19 MB	Adobe PDF	View/Open

Copyright Clearance Process

Recently Added
In Progress
Embargoed
Open Access

This item is currently unavailable due to the publisher's embargo.

The embargo period expires on 4 Apr 2025

Full metadata record

Field	Value	Language
dc.contributor.author	Al-Najjar, Husam Abdulrasool Hammadeh
dc.date.accessioned	2023-07-12T00:09:43Z
dc.date.available	2023-07-12T00:09:43Z
dc.date.issued	2023
dc.identifier.uri	http://hdl.handle.net/10453/171451
dc.description	University of Technology Sydney. Faculty of Engineering and Information Technology.	en_US.UTF-8
dc.description.abstract	One of the most destructive natural hazards is landslide, and Landslide Susceptibility Mapping (LSM) estimates the probability of the hazard in a region. Physical and Machine Learning (ML) based models are common methods for evaluating landslide susceptibility. Physically-based models are appropriate at a local scale such as single slope, basin/ catchment and require site-specific geotechnical. Although reliable geotechnical parameters are essential for such models, lack of geotechnical data throughout a large area (e.g >100 km²) and the expensiveness remain the main obstacles in the physically-based models. In comparison, ML-based models are not explainable and not always effective, depending on the availability and characteristics of the inventory data. This thesis proposes an integrated method based on physically and ML models to address these challenges. Moreover, the proposed method will compensate the common drawback of ML methods (i.e., black box nature) using SHapley Additive exPlanations (SHAP) algorithm. The proposed method was tested at Chukha, Bhutan (area of 1,879.5 km²), a frequent landslide-prone area in the Himalayan region. As the first objective, the study develops a novel model based on Generative Adversarial Networks (GANs) algorithm to create synthetic inventory data to solve the problem of Data Imbalanced (DI) inventory, then it was compared with several ML models. Then, integrating RS time-series data, namely Synthetic Aperture Radar (SAR) time-series and Normalized Difference Vegetation Index (NDVI) time-series with other geo-environmental features, including Digital Elevation Model (DEM) derivatives was conducted in the second objective. The result showed that time-series data such as SAR and NDVI assist in more precise and certain information extraction, as each feature reveals specific characteristics of the Earth's surface. As the third objective, this research designs an integrated architecture to combine physically-based and ML models for large-sacle landslide susceptibility models by using the advantages of both models. The physical parameters of soil and rainfall analysis were utilised to generate a Factor of Safety (FOS). Consequently, the FOS result was integrated with ML and RS time-series features to improve the LSM. The fourth objective presents a novel implementation of Explainable Artificial Intelligence (XAI) for interpreting the outputs and explaining the feature's impact and contribution. The SHAP approach was utilised to understand how and what decisions ML-based models are making at different levels. This could help to build trust among the decision-makers while making geohazard predictions.	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/10453/171451/1/thesis.pdf
dc.rights	info:eu-repo/semantics/embargoedAccess
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 Husam Abdulrasool Hammadeh Al-Najjar
dc.rights	au.edu.uts.lib/cph
dc.title	A novel approach by integrating physically and Machine leaning-based models for landslide susceptibility assessment	en_US.UTF-8
dc.type	Thesis
utslib.copyright.status	embargoed	*
utslib.copyright.embargo	2025-04-04T00:00:00+1000

Abstract:

One of the most destructive natural hazards is landslide, and Landslide Susceptibility Mapping (LSM) estimates the probability of the hazard in a region. Physical and Machine Learning (ML) based models are common methods for evaluating landslide susceptibility. Physically-based models are appropriate at a local scale such as single slope, basin/ catchment and require site-specific geotechnical. Although reliable geotechnical parameters are essential for such models, lack of geotechnical data throughout a large area (e.g >100 km²) and the expensiveness remain the main obstacles in the physically-based models. In comparison, ML-based models are not explainable and not always effective, depending on the availability and characteristics of the inventory data. This thesis proposes an integrated method based on physically and ML models to address these challenges. Moreover, the proposed method will compensate the common drawback of ML methods (i.e., black box nature) using SHapley Additive exPlanations (SHAP) algorithm. The proposed method was tested at Chukha, Bhutan (area of 1,879.5 km²), a frequent landslide-prone area in the Himalayan region. As the first objective, the study develops a novel model based on Generative Adversarial Networks (GANs) algorithm to create synthetic inventory data to solve the problem of Data Imbalanced (DI) inventory, then it was compared with several ML models. Then, integrating RS time-series data, namely Synthetic Aperture Radar (SAR) time-series and Normalized Difference Vegetation Index (NDVI) time-series with other geo-environmental features, including Digital Elevation Model (DEM) derivatives was conducted in the second objective. The result showed that time-series data such as SAR and NDVI assist in more precise and certain information extraction, as each feature reveals specific characteristics of the Earth's surface. As the third objective, this research designs an integrated architecture to combine physically-based and ML models for large-sacle landslide susceptibility models by using the advantages of both models. The physical parameters of soil and rainfall analysis were utilised to generate a Factor of Safety (FOS). Consequently, the FOS result was integrated with ML and RS time-series features to improve the LSM. The fourth objective presents a novel implementation of Explainable Artificial Intelligence (XAI) for interpreting the outputs and explaining the feature's impact and contribution. The SHAP approach was utilised to understand how and what decisions ML-based models are making at different levels. This could help to build trust among the decision-makers while making geohazard predictions.

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

http://hdl.handle.net/10453/171451