Development of a hybrid machine learning model for rockfall source and hazard assessment using laser scanning data and GIS

In mountainous and hilly areas such as Malaysia, rockfalls phenomena is a significant and ongoing threat to people and their properties in addition to infrastructure and transportation lines located within steep terrain. This is because such incidence can cause serious injuries and fatalities as...

Full description

Saved in:

Bibliographic Details
Main Author:	Fanos, Ali Mutar
Format:	Thesis
Language:	English
Published:	2019
Subjects:	Landslide hazard analysis - Risk management - Malaysia Rockslides - Remote sensing - Malaysia Geographic information system
Online Access:	http://psasir.upm.edu.my/id/eprint/84166/1/FK%202019%20122%20-%20ir.pdf
Tags:	Add Tag No Tags, Be the first to tag this record!

id	my-upm-ir.84166
record_format	uketd_dc
institution	Universiti Putra Malaysia
collection	PSAS Institutional Repository
language	English
advisor	Mansor, Shattri
topic	Landslide hazard analysis - Risk management - Malaysia Rockslides - Remote sensing - Malaysia Geographic information system
spellingShingle	Landslide hazard analysis - Risk management - Malaysia Rockslides - Remote sensing - Malaysia Geographic information system Fanos, Ali Mutar Development of a hybrid machine learning model for rockfall source and hazard assessment using laser scanning data and GIS
description	In mountainous and hilly areas such as Malaysia, rockfalls phenomena is a significant and ongoing threat to people and their properties in addition to infrastructure and transportation lines located within steep terrain. This is because such incidence can cause serious injuries and fatalities as well as severe damage to buildings and infrastructure. Therefore, proper and accurate assessment of rockfall sources and hazard is required in order to map and thus understand the characteristics of rockfall catastrophe. The identification of probable rockfall starting regions, the calculation of the rockfall trajectories in complex three-dimensional terrain, and rockfall hazard assessment are three major components of the rockfall research and issues. Although the numerous significant attempts to propose models that can accurately identify potential rockfall source areas, one major problem remain unsolved. This issue is when the focus area contains other types of landslides that have nearly similar geomorphometric characteristics such as rockfall and shallow landslides. Therefore, this research adopted various methods to investigate, analyze and assess rockfall in terms of sources identification, trajectories modeling and their characteristics, and consequently rockfall hazard. This is based on highresolution Light Detection and Ranging (LiDAR) techniques both airborne and terrestrial (ALS and TLS). Different machine learning algorithms (Artificial Neural Network [ANN], K Nearest Neighbor [KNN] and Support Vector Machine [SVM]) were tested individually and with various ensemble models (bagging, voting, and boosting) to detect the probability of the landslide and rockfall occurrences. Consequently, a novel hybrid model is developed to identify potential rockfall sources in the presence of shallow landslides. This is based on an integration of Gaussian Mixture Model (GMM) and an ensemble Artificial Neural Network (Bagged ANN -BANN) for automatic detection of potential rockfall sources at Kinta Valley area, Malaysia. Moreover, a developed 3D rockfall model is employed to derive rockfall trajectories and their characteristics in three different areas within Kinta Valley namely (Gunung Lang, Gua Tambun, and Gunung Rapat) with various scenarios. In addition, a proposed spatial model in combination with fuzzy analytical hierarchy process (fuzzy-AHP) is executed within the geographic information system (GIS) environment to extract rockfall hazard. Mitigation measures are suggested based on the modelling results. Overall, the proposed hybrid model was found to be an efficient method for identifying potential rockfall source areas in the presence of other landslides types with relatively high prediction accuracy and a good generalization performance. GMM could reproduce the slope angle distribution in an accurate way with a coefficient of determination close to 1. The obtained slope thresholds through GMM were (23° to 58°) for landslide and (> 58°) for rockfall. The results of Ant Colony Optimization show that best subset of conditioning factors contains 12 factors of 17 for rockfall with an accuracy of (86%) and 14 factors of 17 for shallow landslide with an accuracy of (82%). The proposed BANN model achieved the best training accuracies of (95%) and best prediction accuracies of (92%) based on testing data compared to other employed methods. This indicates that the model can be generalized and replicated in different regions and the proposed method can be applied in various landslides studies. The result of Fuzzy-AHP revealed the rockfall hazard is highly affected by kinetic energy, frequency, bouncing height, and impact location with weights of (0.48, 0.30, 0.12, and 0.10), respectively. In addition, the proposed spatial model effectively delineates areas at risk of rockfalls. The suggested barriers could effectively reduce the degree of rockfalls hazard. In summary, the proposed methods provide a comprehensive understanding of rockfall hazards that can assist authorities to develop proper management and protection of urban areas and transportation corridors.
format	Thesis
qualification_level	Doctorate
author	Fanos, Ali Mutar
author_facet	Fanos, Ali Mutar
author_sort	Fanos, Ali Mutar
title	Development of a hybrid machine learning model for rockfall source and hazard assessment using laser scanning data and GIS
title_short	Development of a hybrid machine learning model for rockfall source and hazard assessment using laser scanning data and GIS
title_full	Development of a hybrid machine learning model for rockfall source and hazard assessment using laser scanning data and GIS
title_fullStr	Development of a hybrid machine learning model for rockfall source and hazard assessment using laser scanning data and GIS
title_full_unstemmed	Development of a hybrid machine learning model for rockfall source and hazard assessment using laser scanning data and GIS
title_sort	development of a hybrid machine learning model for rockfall source and hazard assessment using laser scanning data and gis
granting_institution	Universiti Putra Malaysia
publishDate	2019
url	http://psasir.upm.edu.my/id/eprint/84166/1/FK%202019%20122%20-%20ir.pdf
_version_	1747813439879774208
spelling	my-upm-ir.841662022-01-04T03:43:26Z Development of a hybrid machine learning model for rockfall source and hazard assessment using laser scanning data and GIS 2019-08 Fanos, Ali Mutar In mountainous and hilly areas such as Malaysia, rockfalls phenomena is a significant and ongoing threat to people and their properties in addition to infrastructure and transportation lines located within steep terrain. This is because such incidence can cause serious injuries and fatalities as well as severe damage to buildings and infrastructure. Therefore, proper and accurate assessment of rockfall sources and hazard is required in order to map and thus understand the characteristics of rockfall catastrophe. The identification of probable rockfall starting regions, the calculation of the rockfall trajectories in complex three-dimensional terrain, and rockfall hazard assessment are three major components of the rockfall research and issues. Although the numerous significant attempts to propose models that can accurately identify potential rockfall source areas, one major problem remain unsolved. This issue is when the focus area contains other types of landslides that have nearly similar geomorphometric characteristics such as rockfall and shallow landslides. Therefore, this research adopted various methods to investigate, analyze and assess rockfall in terms of sources identification, trajectories modeling and their characteristics, and consequently rockfall hazard. This is based on highresolution Light Detection and Ranging (LiDAR) techniques both airborne and terrestrial (ALS and TLS). Different machine learning algorithms (Artificial Neural Network [ANN], K Nearest Neighbor [KNN] and Support Vector Machine [SVM]) were tested individually and with various ensemble models (bagging, voting, and boosting) to detect the probability of the landslide and rockfall occurrences. Consequently, a novel hybrid model is developed to identify potential rockfall sources in the presence of shallow landslides. This is based on an integration of Gaussian Mixture Model (GMM) and an ensemble Artificial Neural Network (Bagged ANN -BANN) for automatic detection of potential rockfall sources at Kinta Valley area, Malaysia. Moreover, a developed 3D rockfall model is employed to derive rockfall trajectories and their characteristics in three different areas within Kinta Valley namely (Gunung Lang, Gua Tambun, and Gunung Rapat) with various scenarios. In addition, a proposed spatial model in combination with fuzzy analytical hierarchy process (fuzzy-AHP) is executed within the geographic information system (GIS) environment to extract rockfall hazard. Mitigation measures are suggested based on the modelling results. Overall, the proposed hybrid model was found to be an efficient method for identifying potential rockfall source areas in the presence of other landslides types with relatively high prediction accuracy and a good generalization performance. GMM could reproduce the slope angle distribution in an accurate way with a coefficient of determination close to 1. The obtained slope thresholds through GMM were (23° to 58°) for landslide and (> 58°) for rockfall. The results of Ant Colony Optimization show that best subset of conditioning factors contains 12 factors of 17 for rockfall with an accuracy of (86%) and 14 factors of 17 for shallow landslide with an accuracy of (82%). The proposed BANN model achieved the best training accuracies of (95%) and best prediction accuracies of (92%) based on testing data compared to other employed methods. This indicates that the model can be generalized and replicated in different regions and the proposed method can be applied in various landslides studies. The result of Fuzzy-AHP revealed the rockfall hazard is highly affected by kinetic energy, frequency, bouncing height, and impact location with weights of (0.48, 0.30, 0.12, and 0.10), respectively. In addition, the proposed spatial model effectively delineates areas at risk of rockfalls. The suggested barriers could effectively reduce the degree of rockfalls hazard. In summary, the proposed methods provide a comprehensive understanding of rockfall hazards that can assist authorities to develop proper management and protection of urban areas and transportation corridors. Landslide hazard analysis - Risk management - Malaysia Rockslides - Remote sensing - Malaysia Geographic information system 2019-08 Thesis http://psasir.upm.edu.my/id/eprint/84166/ http://psasir.upm.edu.my/id/eprint/84166/1/FK%202019%20122%20-%20ir.pdf text en public doctoral Universiti Putra Malaysia Landslide hazard analysis - Risk management - Malaysia Rockslides - Remote sensing - Malaysia Geographic information system Mansor, Shattri

Development of a hybrid machine learning model for rockfall source and hazard assessment using laser scanning data and GIS

Similar Items