Predictive Landslide Susceptibility Mapping in the Eastern Nepalese Himalaya

다국어 초록 (Multilingual Abstract)

The objective of the present study is to construct landslide susceptibility maps in a landslide-prone area, Panchthar district, eastern Nepal, by means of bivariate and multivariate analyses using geographic information system (GIS) techniques as a basic analysis tool. GIS is used for the data management and manipulation. The DEM data are collected from the survey department of Nepal government, and aerial photo interpretation is used for the depiction of lineaments. The locations of 111 landslides that occurred in the study area are identified from field survey. Six pre-existing methods (frequency ratio, class variable analysis and area density methods as bivariate analyses, and logistic regression, artificial neural networks and decision tree as multivariate analyses) are utilized to produce the respective susceptibility maps. The three bivariate-derived methods are relatively simple and similar to each other in their applications, whereas the multivariate-derived methods are somewhat complicated in their utilization since each has to use different software for analysis. A total of ten landslide-controlling factors (slope, aspect, curvature, distance from drainage, distance from lineament, stream power index, topographic wetness index, slope-length, geology and landuse) are implemented to produce final landslide susceptibility maps using individual methods, which are compared for their ability to predict landslide probability based on actual landslide events. The accuracies of the landslide susceptibility maps produced by individual methods are 81.9% for frequency ratio, 83.4% for class variable analysis, 79.0% for area density method, 81.6% for logistic regression, 78.3% for artificial neural networks, and 95.9% for decision tree method, indicating that the decision tree method is an incomparably better tool than the others.

국문 초록 (Abstract)

본 연구의 목적은 기존에 개발되어 있는 다양한 단변량 분석(bivariate analysis)와 다변량 분석(multivariate analysis)을 네팔 동부의 Panchthar지역에 적용하여 산사태 취약성도를 구축하는데 있다. GIS는 자료 관리 및 처리 하는데 이용하였다. 네팔 정부의 survey department로부터 얻은 DEM 자료를 사용하였으며 항공사진 분석을 통해 선구조를 추출하였다. 연구지역에서 발생한 100개의 산사태 위치는 필드조사에 의해 획득다. 본 연구에서 이용된 단변량 분석방법은 frequency ratio, class variable analysis, area density methods 등이며 다변량 분석은 logistic regression, artificial neural networks and decision tree등이다. . 단변량 분석방법들은 그 활용법이 비교적 용이하고 각각의 분석방법이 서로 유사한데 비해 다변량 분석 방법은 각각의 분석 방법에 활용되는 소프트웨어가 다르기 때문에 그 활용 방법이 다소 복잡하다. 산사태발생과 관련이 있는 10개의 인자(slope, aspect, curvature, distance from drainage, distance from lineament, stream power index, topographic wetness index, slope-length, geology and landuse )를 이용하여 각 방법을 이용해 최종 산사태 취약성도를 작성하였으며 그 결과를 비교하였다. 빈도비(frequency ratio), 통계적 분석(class variable analysis), 지역밀도 방법(area density method), 로지스틱 회귀분석(logistic regression), 인공신경망(neural network), 의사결정나무(decision tree) 분석 방법에 의해 작성한 산사태 취약성도의 정확도는 81.9%, 83.4%, 79.0%, 81.6%, 78.3%, 95.9%로 각각 나타났다. 의사결정나무가 분석방법 중 가장 높은 예측결과를 나타냈다.

목차 (Table of Contents)

• CHAPTER 1. INTRODUCTION 1
• 1.1 In general 1
• 1.2 Research objective and methodology 3
• 1.3 Outline of the dissertation structure 6
• CHAPTER 2. THE STUDY AREA 8
• 2.1 Over view of the study area 8
• 2.1.1 Background 8
• 2.1.2 Location and accessibility 9
• 2.1.3 Topography and drainage 10
• 2.1.4 Climate 11
• 2.1.5 Socio-economic condition 12
• 2.2 Geology of the study area 12
• 2.2.1 Higher Himalayan Crystalline 13
• 2.2.2 Lesser Himalayan Rocks 14
• I) Taplejung Group 15
• a) Bharapa Phyllite 15
• b) Phyme Khola Member 16
• c) Phidim Member 17
• d) Jorpati Augen Gneiss 17
• 2.3 Geological Structures 20
• 2.3.1 Main central thrust (MCT) 20
• 2.3.2 Taplejung Window 20
• 2.3.3 Small scale folds 22
• 2.4 Landslides in the study area 24
• CHAPTER 3. LITERATURE REVIEW 29
• 3.1 GIS modeling method 29
• 3.2 Bivariate analysis 30
• 3.2.1 Frequency ratio modeling method 30
• 3.2.2 Class variable analysis method 31
• 3.2.3 Area density method 31
• 3.3 Multivariate analysis 31
• 3.3.1 Logistic regression method 31
• 3.3.2 Artificial neural networks method 32
• 3.3.3 Decision tree method 35
• CHAPTER 4. GEOSPATIAL DATABASE AND MAP PREPARATION 37
• 4.1 Data influence landslide occurrences 37
• 4.1.1 Geology 37
• 4.1.2 Geomorphology 38
• 4.1.3 Hydrology 39
• 4.1.4 Vegetation 40
• 4.1.5 Seismicity and volcanic activity 40
• 4.1.6 Anthropogenic 41
• 4.2 Landslide Mapping 41
• 4.2.1 Landslide inventory mapping 41
• 4.2.2 Landslide susceptibility mapping 43
• 4.3 Data availability and map preparations 44
• 4.3.1 Causative factors and their map preparations 46
• CHAPTER 5. APPLIED METHODOLOGY 62
• 5.1 Bivariate analysis 62
• 5.1.1 Frequency ratio 62
• 5.1.2 Class variable analysis method 63
• 5.1.3 Area density method 64
• 5.2 Multivariate analysis 68
• 5.2.1 Logistic regression analysis 68
• 5.2.2 Artificial neural networks method 70
• 5.2.3 Decision tree method 78
• CHAPTER 6. LANDSLIDE SUSCEPTIBILITY MAPPING RESULTS 82
• 6.1 Landslide susceptibility based on bivariate analysis 82
• 6.1.1 Frequency ratio 82
• 6.1.2 Class variable analysis method 88
• 6.1.3 Area density method 93
• 6.2 Landslide susceptibility based on multivariate analysis 95
• 6.2.1 Logistic regression method 95
• 6.2.2 Artificial neural networks 98
• 6.2.3 Decision tree method 101
• 6.3 Verification and Comparisons 103
• CHAPTER 7. CONCLUSION AND DISCUSSION 108
• REFERENCES 112
• ABSTRACT (KOREAN) 133
• ACKNOWLEDGEMENT 134