Improved flood detection and susceptibility modelling using remote sensing and geographic information system

Natural hazards such as floods, landslides, and land subsidence are destructive events which cause catastrophic damages to both human lives and properties. Accurate and easy to implement prediction models are needed to forecast these hazards and delineate the susceptible areas. Although several meth...

Full description

Saved in:
Bibliographic Details
Main Author: Shafapourtehrany, Mahyat
Format: Thesis
Language:English
Published: 2015
Online Access:http://psasir.upm.edu.my/id/eprint/50753/1/FK%202015%2083RR.pdf
http://psasir.upm.edu.my/id/eprint/50753/
Tags: Add Tag
No Tags, Be the first to tag this record!
id my.upm.eprints.50753
record_format eprints
institution Universiti Putra Malaysia
building UPM Library
collection Institutional Repository
continent Asia
country Malaysia
content_provider Universiti Putra Malaysia
content_source UPM Institutional Repository
url_provider http://psasir.upm.edu.my/
language English
description Natural hazards such as floods, landslides, and land subsidence are destructive events which cause catastrophic damages to both human lives and properties. Accurate and easy to implement prediction models are needed to forecast these hazards and delineate the susceptible areas. Although several methods and techniques have been proposed and examined by researchers to map the flood susceptible areas and to provide flood inventory maps, however optimized approaches for flood susceptibility mapping and modeling could not be encountered in the international literature. In traditional way of flood mapping, multiple field works are generally performed to map and monitor floods which is often time consuming and not economically viable. In the last few decades remote sensing based mapping has become hugely popular among the research fraternity. However, optical remote sensing (RS) data and the available classification schemes may not be appropriate for flood extent mapping. This is mainly attributed due to the severe presence of cloud cover especially during the flood seasons. For that reason, flood modelers and remote sensing scientists has to rely on the use of active remote sensing data such as space-borne radar data for flood area mapping. In this regard, a combination of optical and radar data is highly sought after in flood mapping. In disaster management flood susceptibility mapping is one of the basic steps. There are various types of methods exist in flood susceptibility mapping e.g. traditional based hydrological methods, statistical, probabilistic and data mining based approaches. Traditional hydrological methods are based on linear assumption and require extensive field work. The most popular statistical methods in flood susceptibility assessment are frequency ratio (FR), weights-of-evidence (WoE), and logistic regression (LR). Similarly, the most commonly used data mining approaches in flood susceptibility assessment are artificial neural networks (ANN), fuzzy logic and many more models. However, each of the above mentioned techniques has certain pros and cons. For example, LR is not able to assess the impact of each class of flood conditioning factor on flood occurrence. On the other hand, FR and WoE are capable of evaluating the correlation between them, but they neglect such correlation among the conditioning factors themselves. Consequently, ANN method is well known for over-training of the dataset. This study adopted several approaches to investigate and analyze flood occurrence in terms of detection, modeling and optimization of the flood conditioning factors. The current research is divided into two general aspects. The first aspect mainly explored the use of RS technology to detect the flooded areas in Kuala Terengganu, Malaysia using TerraSAR-X image. A TerraSAR-X satellite image was captured during the flood occurrence and Landsat image was captured before the flood occurrence. Both images were classified using object-based method and flooded locations were extracted by subtracting two classes of water bodies. Subsequently, confusion matrix was used to evaluate the results. The second aspect of the current research is related to the use of geographic information system (GIS) in flood susceptibility mapping. A Decision tree (DT) method was implemented for the first time in flood susceptibility mapping. The efficiency of DT to map the flood prone areas in Kelantan, Malaysia was evaluated using the well-known area under the curve (AUC) validation technique. Validationresults showed 87% and 82% for success rate and prediction rate respectively. In order to improve the prediction accuracy of the individual methods such as FR, LR, WoE, and a data-mining based support vector machine (SVM) model, the current research proposed three novel ensemble methods in GIS environment. The overall theory of the ensemble method includes combining the statistical and data-mining methods by integrating the outputs of multiple classifiers to decrease the generalization error. It started with the model development by ensembling FR and LR methods which was then tested in two study areas: Busan, South Korea and Kelantan, Malaysia. In the case study of Busan, the results of the accuracy assessment showed a success rate of 92.7% and a prediction rate of 82.3%. Similarly, the ensemble result of FR and LR models in the Kelantan achieved 90% and 83% for success rate and prediction rate respectively. Next, the second ensemble method was realized by integrating FR and SVM models and was applied for flood susceptibility mapping in Kelantan, Malaysia. The validation results showed 88.71% and 85.21% for success rate and prediction rate respectively. Next, a new ensemble method was proposed by utilizing WoE and SVM models to produce flood susceptibility map and was applied in Kuala Terengganu area, Malaysia. Validation results of the WoE-SVM ensemble model showed 96.48% (success rate) and 95.67% (prediction rate) accuracy. Another objective of this research was to implement SVM model individually and to evaluate the performance of all its kernel types in flood susceptibility mapping. The validation results for SVM using different kernel types showed that the highest achieved prediction rate (82.16%) was for SVMRBF. The last goal of the current research was to perform the optimization of the flood conditioning factors using the SVM model aided with Cohen's kappa index. The result demonstrated that the most influential factors were altitude and slope for all kernel types. Overall, this thesis proposed several new methodologies for flood area mapping and flood susceptibility assessment. The outcome of the current research may assist researchers and local government agencies in flood mitigation strategies and planning.
format Thesis
author Shafapourtehrany, Mahyat
spellingShingle Shafapourtehrany, Mahyat
Improved flood detection and susceptibility modelling using remote sensing and geographic information system
author_facet Shafapourtehrany, Mahyat
author_sort Shafapourtehrany, Mahyat
title Improved flood detection and susceptibility modelling using remote sensing and geographic information system
title_short Improved flood detection and susceptibility modelling using remote sensing and geographic information system
title_full Improved flood detection and susceptibility modelling using remote sensing and geographic information system
title_fullStr Improved flood detection and susceptibility modelling using remote sensing and geographic information system
title_full_unstemmed Improved flood detection and susceptibility modelling using remote sensing and geographic information system
title_sort improved flood detection and susceptibility modelling using remote sensing and geographic information system
publishDate 2015
url http://psasir.upm.edu.my/id/eprint/50753/1/FK%202015%2083RR.pdf
http://psasir.upm.edu.my/id/eprint/50753/
_version_ 1643834755862495232
spelling my.upm.eprints.507532017-02-08T07:34:45Z http://psasir.upm.edu.my/id/eprint/50753/ Improved flood detection and susceptibility modelling using remote sensing and geographic information system Shafapourtehrany, Mahyat Natural hazards such as floods, landslides, and land subsidence are destructive events which cause catastrophic damages to both human lives and properties. Accurate and easy to implement prediction models are needed to forecast these hazards and delineate the susceptible areas. Although several methods and techniques have been proposed and examined by researchers to map the flood susceptible areas and to provide flood inventory maps, however optimized approaches for flood susceptibility mapping and modeling could not be encountered in the international literature. In traditional way of flood mapping, multiple field works are generally performed to map and monitor floods which is often time consuming and not economically viable. In the last few decades remote sensing based mapping has become hugely popular among the research fraternity. However, optical remote sensing (RS) data and the available classification schemes may not be appropriate for flood extent mapping. This is mainly attributed due to the severe presence of cloud cover especially during the flood seasons. For that reason, flood modelers and remote sensing scientists has to rely on the use of active remote sensing data such as space-borne radar data for flood area mapping. In this regard, a combination of optical and radar data is highly sought after in flood mapping. In disaster management flood susceptibility mapping is one of the basic steps. There are various types of methods exist in flood susceptibility mapping e.g. traditional based hydrological methods, statistical, probabilistic and data mining based approaches. Traditional hydrological methods are based on linear assumption and require extensive field work. The most popular statistical methods in flood susceptibility assessment are frequency ratio (FR), weights-of-evidence (WoE), and logistic regression (LR). Similarly, the most commonly used data mining approaches in flood susceptibility assessment are artificial neural networks (ANN), fuzzy logic and many more models. However, each of the above mentioned techniques has certain pros and cons. For example, LR is not able to assess the impact of each class of flood conditioning factor on flood occurrence. On the other hand, FR and WoE are capable of evaluating the correlation between them, but they neglect such correlation among the conditioning factors themselves. Consequently, ANN method is well known for over-training of the dataset. This study adopted several approaches to investigate and analyze flood occurrence in terms of detection, modeling and optimization of the flood conditioning factors. The current research is divided into two general aspects. The first aspect mainly explored the use of RS technology to detect the flooded areas in Kuala Terengganu, Malaysia using TerraSAR-X image. A TerraSAR-X satellite image was captured during the flood occurrence and Landsat image was captured before the flood occurrence. Both images were classified using object-based method and flooded locations were extracted by subtracting two classes of water bodies. Subsequently, confusion matrix was used to evaluate the results. The second aspect of the current research is related to the use of geographic information system (GIS) in flood susceptibility mapping. A Decision tree (DT) method was implemented for the first time in flood susceptibility mapping. The efficiency of DT to map the flood prone areas in Kelantan, Malaysia was evaluated using the well-known area under the curve (AUC) validation technique. Validationresults showed 87% and 82% for success rate and prediction rate respectively. In order to improve the prediction accuracy of the individual methods such as FR, LR, WoE, and a data-mining based support vector machine (SVM) model, the current research proposed three novel ensemble methods in GIS environment. The overall theory of the ensemble method includes combining the statistical and data-mining methods by integrating the outputs of multiple classifiers to decrease the generalization error. It started with the model development by ensembling FR and LR methods which was then tested in two study areas: Busan, South Korea and Kelantan, Malaysia. In the case study of Busan, the results of the accuracy assessment showed a success rate of 92.7% and a prediction rate of 82.3%. Similarly, the ensemble result of FR and LR models in the Kelantan achieved 90% and 83% for success rate and prediction rate respectively. Next, the second ensemble method was realized by integrating FR and SVM models and was applied for flood susceptibility mapping in Kelantan, Malaysia. The validation results showed 88.71% and 85.21% for success rate and prediction rate respectively. Next, a new ensemble method was proposed by utilizing WoE and SVM models to produce flood susceptibility map and was applied in Kuala Terengganu area, Malaysia. Validation results of the WoE-SVM ensemble model showed 96.48% (success rate) and 95.67% (prediction rate) accuracy. Another objective of this research was to implement SVM model individually and to evaluate the performance of all its kernel types in flood susceptibility mapping. The validation results for SVM using different kernel types showed that the highest achieved prediction rate (82.16%) was for SVMRBF. The last goal of the current research was to perform the optimization of the flood conditioning factors using the SVM model aided with Cohen's kappa index. The result demonstrated that the most influential factors were altitude and slope for all kernel types. Overall, this thesis proposed several new methodologies for flood area mapping and flood susceptibility assessment. The outcome of the current research may assist researchers and local government agencies in flood mitigation strategies and planning. 2015-08 Thesis NonPeerReviewed application/pdf en http://psasir.upm.edu.my/id/eprint/50753/1/FK%202015%2083RR.pdf Shafapourtehrany, Mahyat (2015) Improved flood detection and susceptibility modelling using remote sensing and geographic information system. PhD thesis, Universiti Putra Malaysia.
score 13.209306