APPLICATIONS OF STATISTICAL AND HEURISTIC METHODS FOR LANDSLIDE SUSCEPTIBILITY ASSESSMENTS
A case study in Wadas Lintang Sub District, Wonosobo Regency, Central Java Province, Indonesia
Thesis submitted to the Graduated School, Faculty of Geography, Gadjah Mada University and International Institute for Geo-information Science and Earth Observation in partial fulfilment of the requirements for the degree of Master of Science in Joint Education Program between Gadjah Mada University (GMU)-Yogyakarta-Indonesia and International Institute for Geo-information Science and Earth Observation (ITC)- Enschede-The Netherlands,on Geo-Information for Spatial Planning and Risk Management
by: Bonaventura Firman Dwi Wahono (22624-AES) 08/276588/PMU/05636
SUPERVISOR : Dr. Danang Sri Hadmoko M.Sc.(GMU) Dr. C.J. van Westen (ITC)
GADJAH MADA UNIVERSITY INTERNATIONAL INSTITUTE FOR GEO-INFORMATION AND EARTH OBSERVATION 2010 APPLICATIONS OF STATISTICAL AND HEURISTIC METHODS FOR LANDSLIDE SUSCEPTIBILITY ASSESSMENTS
A case study in Wadas Lintang Sub District, Wonosobo Regency, Central Java Province, Indonesia
By: Bonaventura Firman Dwi Wahono (22624-AES) 08/276588/PMU/05636
Has been approved in Yogyakarta ...February 2010 By Team of Supervisors:
Chairman:
. .
External Examiner:
. . . .
Supervisor I: Supervisor II
Dr. Danang Sri Hadmoko, M.Sc Dr. C.J. van Westen
Certified by: Program Director of Geo-Information for Spatial Planning and Risk Management Graduate School Faculty of Geography, Gadjah Mada University
Dr. H.A Sudibyakto, M.S.
DEDICATED TO
MY LOVELY FAMILY
I certify that although I may have conferred with others in preparing for this assignment, and drawn upon a range of sources cited in this work, the content of this thesis report is my original work.
Signed …………………….
DISCLAIMER This document describes work undertaken as part of a programme of study at the Double Degree International Program of Geo-information for Spatial Planning and Risk Management, a Joint Education Program of Institute for Geo-information Science and Earth Observation, the Netherlands and Gadjah Mada University, Indonesia. All views and opinions expressed therein remain the sole responsibility of the author, and do not necessarily represent those of the institutes.
ABSTRACT
One of the basic activities in landslide mitigation is to generate a landslide susceptibility map which is the main objective of this research. Using appropriate method becomes the key to determine the areas prone to landslide. There are three applied methods for assessing the landslide susceptibility in Wadas Lintang sub district such as bivariate statistical analysis, multivariate statistical analysis and combination between bivariate and pair wise comparison method. The result showed that there are 71 landslide sites which are categorized into two types of landslide; rotational slides and translational slides. From the rainfall analysis, temporal probability of landslides can be formulated as Y = 0.0334X – 4.0127, where x is daily rainfall. Based on laboratory investigation, the soil textures in research area were dominated with a high percentage of clay fractions. The investigation also showed that the soil of research area can be categorized as organic clay and very plastic silt which is harmful for sliding in rainy season. The simple formulas were used to extract magnitude of landslide which is ranging from 1.848 to 2.565. Building the susceptibility map for rotational slides and translational slides should be separated because of their different characteristics. The comparison between three methods for assessing landslide susceptibility proofs that multivariate statistical analysis gives the best estimation of the location of future rotational slides as well as translational slides. The susceptible areas to landslide are determined by a combination of several factors which depend on the landslide types. Rotational slides occur on steep slopes from 15 – 30%, in areas covered by shifting cultivation and shrubs and underlain by grained sandstone, breccia and andesite. Most landslides of translational slides occur on landuse types of settlement and paddy field and in geological unit such as breccia, basalt and andesite. Translational slides dominantly occurred in flat - gentle slope from 0 to 8%. Based on the best resulted susceptibility map, the high susceptible area of rotational slides comprise 36.50% of the whole research area in which Samogede, Ngalian, Trimulyo, Lancar and Gumelar become the most susceptible villages. The high susceptible area prone to translational slides was estimated equal to 14.35% of research area at which Kaligowong, Tirip, Sumber Rejo, Wadas Lintang, and Plunjaran appear to be the most susceptible villages.
Key words: Landslide, susceptibility, statistical, bivariate, multivariate, pair-wise .
ii ACKNOWLEDGEMENT
This research had been supported by contributions of many people and institutions. Supports had come in many different ways and each input helped to bring this research to be completed. Herewith I would like to thank all who have helped me in this effort and particularly mention of person and institution whose contribution has been important. First I thank to Jesus Christ whom supported me by his grace and blessing. I would like to express my high gratitude to my GMU supervisor, Dr. Danang Sri Hadmoko, M.Sc, for his valuable guidance, encouragements and discussions during this susceptible period. This research would not been carried out and written without his attentive support. I would also like to give my sincere gratitude to Dr. C.J van Westen , my ITC supervisor, for his remarks, improvements and comments during writing proposal until the final submission. I also appreciate to all lecturer and staff members both in GMU and ITC, for their support and guidance, especially to Dr. Sudibyakto, Dr. Junun Sartohadi, Dr. Aris Marfai, Drs. Robert Voskuil, Dr. Michiel Damen, Prof.Dr.Victor Jetten, Dr. David G. Rossiter, Drs. Nanette C. Kingma, mbak Tuti, mbak Indri, mbak Win and mas Wawan. I appreciate also to Mr. Mauro Rossi (CNR IRPI) and Saibal Gosh (PhD student of ITC) for giving the scripts and guidance of multivariate statistical analysis. I wish to convey my acknowledgement for BAPPENAS, Netherlands Fellowship Programme- NESO (NEC) and local government of South Sumatra Province for giving me scholarship and financial support to continue my post graduate programme. Thanks also to my previous bosses, Dr. Doddy Supriadi, ibu Sri Hastuti, bapak Zulfikhar and mbak Neneng, for giving the opportunity to continue my study. This research had also been done by permission given by local governments of Wonosobo regency and Wadas Lintang sub district. Thank for the kindness of all the chiefs in each authority’s level. Thanks to all of my classmates for making a friendly atmosphere and sharing all their scientific abilities. This condition kept me to stay in Jogja. Thanks to Emba Tampang Allo for his critical discussion and to Fetty Febrianti for her word’s editing in my thesis. I owe a lot to my brother’s family, Peyek and mbak Tita, for giving the facilities and comfortable condition during I stay in Jogja and to my little Ale for her strengthen smiles. My sincere gratitude is also given to my parents and my parents in law for their supporting and keeping my family during I continue my study and for my siblings and brothers, Sesilia, Sari, Chepy and Bram who gave a comfortable condition to my family. The last is special and deeply thank to my lovely family, my beloved wife, Agrestiwa Perbawani and my children, Pascal Rangga Prasetya and Felicia Maharani. This research is mainly dedicated for you.
Jogjakarta, January 2010 Bonaventura Firman D. Wahono
iii
TABLE OF CONTENTS
DISCLAIMER...... i ABSTRACT ...... ii ACKNOWLEDGEMENT...... iii TABLE OF CONTENTS ...... iv LIST OF TABLE...... vi LIST OF FIGURES...... vii CHAPTER 1...... INTRODUCTION...... 1 1.1 Background...... 1 1.2 Problem Statement...... 2 1.3 Objectives...... 3 1.4 Research Questions...... 3 1.5 Recent studies...... 4 CHAPTER 2...... LITERATURE REVIEW...... 6 2.1. Types of Landslides...... 6 2.1.1. Falls ...... 6 2.1.2. Flows ...... 6 2.1.3. Slides ...... 7 2.1.4. Topples ...... 7 2.1.5. Spread...... 8 2.2. Landslide’s influencing factors ...... 8 2.2.1. Landslide triggering factors...... 9 2.2.2. Landslide preparatory factors...... 9 2.2.2.1. Slope Gradient...... 9 2.2.2.2. Landuse...... 10 2.2.2.3. Lithology ...... 10 2.2.2.4. Geological structure...... 10 2.2.2.5. Distance to road...... 11 2.2.2.6. Distance to river...... 11 2.3. Soil Properties...... 11 2.4. Remote Sensing and GIS in mapping Landslide Inventory...... 12 2.5. Landslide Susceptibility ...... 12 2.5.1. Heuristic approach...... 13 2.5.2. Statistical approach...... 13 2.5.3. Physically-based modeling approach ...... 13 2.5.4. Probabilistic approach ...... 13 2.6. Magnitude of landslides...... 14 CHAPTER 3...... METHODOLOGY ...... 15 3.1. Raw materials, data requirement and equipment...... 15 3.2. Data preparation ...... 16 3.3. Methods ...... 17 3.3.1. Landslide Inventory...... 17 3.3.2. Rainfall Frequency Analysis ...... 19 3.3.3. Soil properties analysis...... 19 3.3.4. Determining landslide influencing factors of research area ...... 20 3.3.5. Landslide Density...... 20
iv 3.3.6. Landslide susceptibility analysis...... 20 3.3.6.1. Bivariate statistical analysis ...... 20 3.3.6.2. Multivariate statistical analysis...... 21 3.3.6.3. Spatial Multi Criteria Evaluation...... 22 3.3.6.4. Improved method...... 22 3.3.7. Verifying landslide susceptibility maps ...... 23 3.3.8. Classifying the landslide susceptibility map ...... 25 3.3.9. Extracting Landslide Magnitude ...... 25 CHAPTER 4...... RESEARCH AREA ...... 27 4.1. Geographic position...... 27 4.2. Rainfall activity ...... 27 4.3. Geological and geomorphological condition of research area ...... 28 4.4. Landuse of Wadas Lintang sub district ...... 30 4.5. Reported landslides of Wonosobo regency ...... 30 CHAPTER 5...... LANDSLIDE CHARACTERISTICS ...... 32 5.1. Soil Properties Analysis...... 32 5.2. Generating the landslide inventory map...... 33 5.2.1 Recognizing landslides by using aerial photos...... 33 5.2.2 Recognizing landslides by using SPOT Imagery...... 34 5.2.3 Field observation as a final mapping process...... 35 5.3. Landslide Profile...... 39 5.4. Volumetric Analysis...... 40 5.5. Slope morphology analysis...... 41 5.6. Magnitude of landslides...... 42 5.7. The probability of occurrence rainfall triggering landslides ...... 42 CHAPTER 6...... LANDSLIDE SUSCEPTIBILITY ASSESSMENT...... 45 6.1 Selecting the landslide influencing factors of research area...... 45 6.2 Bivariate statistical analysis...... 48 6.1.1 Bivariate statistical analysis for rotational slides ...... 48 6.1.2 Bivariate statistical analysis for translational slides...... 49 6.1.3 Bivariate statistical analysis for mixed types ...... 50 6.3 Multivariate statistical analysis by using logistic regression...... 52 6.2.1 Multivariate statistical analysis for rotational slide...... 52 6.2.2 Multivariate statistical analysis for translational slides...... 54 6.2.3 Multivariate statistical analysis for mixed types ...... 56 6.4 Improved Method (BSA and pair-wise)...... 59 6.4.1 Improved Method for Rotational Slides...... 60 6.4.2 Improved Method for Translational slides ...... 62 6.4.3 Improved Method for mixed types...... 64 6.5 Verifying landslide susceptibility index map...... 66 6.6 Defining the best method for assessing the landslide susceptibility ...... 68 6.7 Determining the classification of susceptibility maps...... 69 6.8 Comparison of spatial predictions...... 73 6.9 Defining the areas prone to landslide in Wadas Lintang Sub District...... 74 CHAPTER 7...... CONCLUSIONS, LIMITATIONS AND RECOMENDATIONS...... 77 7.1 Conclusions ...... 77 7.2 Limitations...... 78 7.3 Recommendations ...... 79 REFERENCES...... 80
v
LIST OF TABLE
Table 1-1. The Disaster Event of Indonesia in 2008,...... 1 Table 2-1. Velocity of Several Types of Movement...... 6 Table 2-2.Causal Factors of landslides ...... 9 Table 3-1. Data required, row materials and preparation processes...... 15 Table 3-2. Equipment used in the field observation...... 16 Table 3-3. Scale for Pair wise Comparison...... 22 Table 4-1 Geology formation of research area...... 28 Table 4-2 the composition of slope classes in the research area...... 30 Table 4-3 .The composition of landuse in the research area...... 30 Table 5-1 Relationship between the limits and engineering properties ...... 33 Table 5-2 Landslide magnitudes in the research area ...... 42 Table 5-3 Return Period (TR) of Annual Rainfall in Wadas Lintang...... 43 Table 6-1 The weight values of each class parameter induced rotational slides...... 46 Table 6-2 The weight values of each class parameter induced translational slides ...... 47 Table 6-3 The weight values of each class parameter induced mixed types of landslides...... 50 Table 6-4 Summary of training datasets for rotational slides in MSA...... 52 Table 6-5 Classification of dataset for rotational slides in MSA ...... 52 Table 6-6 The coefficients of of dataset for rotational slides in MSA...... 53 Table 6-7 Summary of training datasets for translational slides in MSA ...... 54 Table 6-8 Classification table of dataset for Translational slides in MSA...... 55 Table 6-9. The coefficients of of dataset for Translational slides in MSA ...... 55 Table 6-10 Summary of training datasets for both types in MSA...... 56 Table 6-11 Classification table of dataset for mixed types in MSA...... 57 Table 6-12 The coefficients of dataset for mixed types in MSA ...... 57 Table 6-13. The weight value for each group and parameter for rotational slides by using pair wise comparison – SMCE ...... 60 Table 6-14 . Final weight values of rotational slides by using Pair-wise comparison...... 61 Table 6-15 The weight value for each group and parameter for translational slides by using pair wise comparison – SMCE...... 62 Table 6-16 Final weight values of translational slides by using Pair-wise comparison method...... 63 Table 6-17 The weight value for each group and parameter for translational slides by using pair wise comparison – SMCE...... 64 Table 6-18 Final weight values of mixed types by using Pair-wise comparison method ...... 65 Table 6-19 the matrix showing the areas classified as the same class of rotational susceptibility (in percentage of study area)...... 73 Table 6-20 the matrix showing the areas classified as the same class of translational slides susceptibility (in percentage of study area)...... 73 Table 6-21 the matrix showing the areas classified as the same class of mix landslide susceptibility (in percentage of study area) ...... 74 Table 6-22 Susceptible classes of rotational slide of each village in the research area ...... 75 Table 6-23 Susceptible classes of translational slides of each village in the research area ...... 76
vi LIST OF FIGURES Figure 1-1. The latest landslide susceptibility map including research area...... 2 Figure 2-1. Fall Type...... 6 Figure 2-2. Flow type...... 7 Figure 2-3. a) Rotational slide, b) Translational Slide ...... 7 Figure 2-4.Topples type ...... 8 Figure 2-5.Spread type...... 8 Figure 3-1. Flow chart of building landslide inventory map...... 18 Figure 3-2. Flowchart of Landslide Susceptibility Analyses ...... 23 Figure 4-1 : Administrative Map of Wonosobo Regency (Source Local Government) ...... 27 Figure 4-2 Average Monthly Rainfall of Wadas Lintang sub district 1980 - 2008 ...... 28 Figure 4-3 Litology map of research area ...... 29 Figure 4-4 Topographic visualization of research area...... 29 Figure 4-5 Landslide occurrences in Wonosobo Regency, 2000 – 2008...... 31 Figure 5-1 Soil texture distribution of landslide sites ...... 32 Figure 5-2 Casagrande Plasticity Chart of soil samples...... 33 Figure 5-3 Temporal Landslide investigation 1(a) aerial photo 1994,(b) SPOT 2006, (c) Field Observation 2009 ...... 34 Figure 5-4 Temporal landslide investigation 2 (a) SPOT 2006, (b) Field Observation, 2009 ...... 34 Figure 5-5 Misinterpretation Landslide investigation (a) SPOT 2006, (b) Field Observation...... 35 Figure 5-6 Field observation (a) direct measurement, (b) taking soil sample of landslide sites...... 35 Figure 5-7 Unreported landslides (a) in the mix garden (b) forest, (c) shifting cultivation area ..... 36 Figure 5-8 The damages caused by translational slides in the settlement area. (a) wall break, (b) floor cracks...... 36 Figure 5-9 The boundaries of translational slides (a)land drops down,(b) land breaks away...... 36 Figure 5-10 Geomorphological Distribution of Landslide in Wadas Lintang ...... 37 Figure 5-12 Landslide profile of Dadap Gede...... 39 Figure 5-13 Volumetric Analysis process...... 40 Figure 5-14 Transect lines of Dadap Gede’s Landslide...... 40 Figure 5-15 Slope condition before and after sliding at Dadap Gede...... 41 Figure 5-16 Relationship between Left Probability and amount of Rainfall ...... 44 Figure 6-1 Success rate of rotational susceptibility index based on sensitivity analysis ...... 45 Figure 6-2 Landslide occurrences triggered by road expanded...... 46 Figure 6-3 Success rate of translational slides susceptibility index based on sensitivity analysis... 47 Figure 6-4 Landslide occurrences triggered by river bank erosion in toe part...... 48 Figure 6-5 Landslide Susceptibility index map of Rotational Slides by using BSA...... 49 Figure 6-6 Landslide Susceptibility index map of Translational slides by using BSA...... 50 Figure 6-7 Landslide Susceptibility index map of mixed types by using BSA...... 51 Figure 6-8 Susceptibility index map of Rotational Slides by using MSA ...... 54 Figure 6-9 Susceptibility index map of translational slides by using MSA...... 56 Figure 6-10 Susceptibility index map of mixed types by using MSA ...... 58 Figure 6-11 Susceptibility index map of rotational slides by using Improved Method ...... 62 Figure 6-12 Susceptibility index map of translational slides by using Improved Method...... 64 Figure 6-13 Susceptibility index map of mixed types by using Improved Method ...... 66 Figure 6-14 Comparing the prediction rate of 3 methods for mixed types ...... 67 Figure 6-15 Comparing the prediction rate of 3 methods for rotational slides ...... 67 Figure 6-16 Comparing the prediction rate of 3 methods for translational slides...... 68 Figure 6-17 Class breaks of rotational susceptibility map ...... 69 Figure 6-18 Final susceptibility map for rotational slides...... 70 Figure 6-19 Class breaks of translational slides susceptibility map...... 70 Figure 6-20 Final susceptibility map for translational slides ...... 71 Figure 6-21 Class breaks of susceptibility map for mixed landslides...... 72 Figure 6-22 Landslide susceptibility map for mixed types ...... 72
vii
CHAPTER 1. INTRODUCTION
1.1 Background Landslides are regarded as natural degradation processes as stated (Van Westen,1994a) produced by natural and human activities. Natural factors such as high rainfall, earthquake, and volcanic eruption can trigger landslide occurrences. This condition becomes worse when human activity also contributes to landslide occurrences. Landslides are among of the major hazardous events in Indonesia. It caused a large number of casualties and damages. Badan Nasional Penanggulangan Bencana Indonesia / National Bereau of Disaster Management in Indonesia (BNPB) (2008) revealed the data of disaster occurrences in 2008 as presented in Table 1: Table 1-1. The Disaster Event of Indonesia in 2008,
Number of The Number of No Disaster Types Occurrences Affected and Casualties Evacuated People 1 Flood 19 7 587.190 68 2 Typhoon 56 2.564 3 3 Landslide 39 1.599 73 4 Flood and 22 15.915 54 5 Tidal Wave 8 3.911 - 6 Earthquake 8 24.002 12 7 Fire 7 2.392 - 8 Technological 3 - 30 9 Forest Fire 1 - 3 10 Volcanic Eruption 1 9.708 - 11 Social Disturbance 1 - 2 Total 343 647.281 245 Source BNPB, 2008 Although most parts of Indonesia have more experience related to landslides, the preparedness and mitigation activities are not running well in many regions. The lack of required data, and hazard experts, limited budget and the lack of awareness of local government are some of the reasons why the mitigation and preparedness activities are far from adequate. In other hand, those activities are absolutely needed to reduce the effect of hazard. One of the activities in landslide mitigation is to generate landslide susceptibility maps. Landslide susceptibility can be defined as the relative degree of instability of the terrain. This map presents the probability of future landslide occurrences. It can be made by correlating the environmental factors, which influence the landslide occurrences with past distribution of slope failures (Brabb,1984). To build a landslide susceptibility map we should consider the published methods or we can modify these to get a better-expected result. In spite of that, the chosen mapping method depends on data availability, financial budget and time for observation, while detailed level of required data counts on working scale and the proposed application of the mapping result (Soeters and Van Westen ,1996 ).
1 In Indonesia, many local institutions tried to apply the known landslide susceptibility assessment methods in their regions. Most of them only implemented the heuristic method to the area prone to landslide and the problem of this method is in the subjectivity of defining parameters, the weighted and scoring values. These given values were based on either individual opinion or taken values from another location. For example, Wuryanta et al (2004) developed the landslide susceptibility map in Kulon Progo, Kebumen and Purworejo. They used the weighted value based on their opinions and they did not use the landslide inventory map to build a landslide susceptibility map. Therefore the resulted landslide susceptibility maps have many weaknesses and uncertainties. To assess the landslide susceptibility, the statistical analysis requires a landslide occurrence map that will be combined with environmental factors and the assumption that landslides will occur under the similar condition of each factor like the occurrences in the past. (Van Westen et al., 2005 ). In this approach landslide analysis requires a large number of input variables which several of them need a high cost and time consuming to collect (Van Westen ,1994 ). The combination and comparison between statistical and heuristic method will become the major discussion in this research. The author chooses Wadas Lintang District where located in the southern part of Wonosobo Regency as the study area. This location was selected because a number of landslide occurrences are highly recorded. As a part of landslide mitigation activities, building the landslide susceptibility map is an important thing to generate the spatial planning for this district.
1.2 Problem Statement The intensive human activities in utilizing land are responsible for reducing the qualities of environmental condition. The decreasing becomes the causal factor for triggering several hazard events, such as landslides. Landslide mitigation processes are absolutely needed for hilly and mountainous area. Wonosobo is one of the landslide susceptible areas in the Central Java Province where landslide occurrences almost occur during the rainy season. Unfortunately the availability of landslide susceptibility map just is in Provincial level with scale 1 : 250.000 just showing the mass movement activities and not presenting the hazard zonation. This map was published by Bakosurtanal (2005). The latest landslide susceptibility map is shown in the Figure 1.
Research Area
Figure 1-1. The latest landslide susceptibility map including research area.
2 The local authorities need a more detailed landslide susceptibility map to support the better environmental management and for reducing the victims and damages caused by future landslide occurrences. Actually there are some data related to landslides in Wonosobo Regency during 2000 – 2008 in tabular format. However the lack of detailed spatial data of landslide occurrences and its distribution is still a main problem. Next, we need the landslide susceptibility map yet in the more detailed scale than the provincial one. The other problem is how to improve susceptibility assessment method related to landslides. This improvement is expected to solve the problem related to the weakness of both methods; heuristic and statistical method. Using multivariate statistical analysis (MSA) completely depends on the accuracy of data. MSA may get the result objectively, but expert’s knowledge is entirely needed in the estimation of future landslides. In the end of this research, the author will compare the combined method with statistical methods. This research is hoped not only to produce a map, but also to contribute the suitable approaches to landslide susceptibility mapping. Anyway, limiting the landslide causal factors may overcome as a question to be answered in this research. This can be solved by field observation which aims to identify the physical characteristics of landslide causal factor.
1.3 Objectives The main objective of this study is to assess and evaluate the landslide susceptibility in Wadas Lintang Sub District using statistical analysis and pair-wise comparison method in the medium scale. The specific objectives are: 1. To inventorize the distribution of landslide occurrences in Wadas Lintang Sub District. 2. To define the appropriate method for assessing the landslide susceptibility in the research area. 3. To determine and to map the areas of landslide susceptibility based on the best method.
1.4 Research Questions The research questions are formulated as follow: 1. How to develop the map of landslide occurrences in Wadas Lintang Sub District? 1.1. What are the characteristic of soil on the landslide site? 1.2. How to extract the data of the past landslide occurrences either using satellite imagery or aerial photos? 1.3. How to collect, measure, and map the recent landslide occurrences through field work? 1.4. How to differentiate the landslides according to type and date of occurrences? 1.5. How to express the landslide magnitude in the research area? 1.6. What is the probability of occurrence rainfall triggering landslide? 2. What is the most appropriate method to assess landslide susceptibility in the research area? 2.1. Can the combination between bivariate statistical and pair-wise comparison method predict the future landslide more accurately than that by using either bivariate or multivariate statistical method?
3 3. How to determine the prone area of landslide in Wadas Lintang Sub District? 3.1. What are the landslide influencing factors of Wadas Lintang Sub District? 3.2. Which areas in parts of Wadas Lintang Sub Districts are susceptible to landslide occurrences?
1.5 Recent studies Magliulo et al (2008) built the landslide susceptibility map by using bivariate statistical method in southern Italy. They extracted the landslide susceptibility map by applying both information value and landslide detachment zone. Then the results had been compared with geomorphological field survey. The authors used orto-photo to recognize landslide sites as the based to generate landslide inventory map. As the preparatory factors, they choose lithology, landuse, slope gradient and aspect. The final result showed that both methods had good coherence with geomorphological map. Firdaini (2008) generated landslide hazard assessment to define and evaluate land capability in Purworejo regency. She used some factors to build landslide hazard map such as slope, geology, soil, and landuse. The score values for each class parameter were given by her opinion based on the rank of them. The rank was arranged based on general pattern of landslide prone area in Indonesia. All factors were assumed in the same weight. Three classes of susceptibility were given by using equal interval method. Landslide inventory map was used to determine landuse priority and not used for verifying the landslide hazard map. The susceptibility method can be included as heuristic approach Marhaento’s research (2006) tried to generate landslide hazard map by using heuristic and bivariate statistical analysis. Loano sub district in Purworejo Regency was chosen as the study area. Heuristic method was used to determine the landslide preparatory factors such as slope gradient, soil type, soil depth, land-use, lithology and distance to road. He extracted the landslide sites by using multi temporal ASTER images. The accuracy test chosen was success rate. Wuryanta et al (2004) had done the research about landslide identification and solving activities in Kulon Progo, Purworejo and Kebumen regencies in Indonesia. Some factors were used to generate the susceptibility map such as landform, slope, geology, soil and landuse. The weight values were given based on their opinion. The final weight values were extracted by adding all weight values of each factor. The susceptibility classes were divided with equal interval method where maximum values minus minimum value divided with number of classes. They also identified the landslide by using Landsat 7 ETM which has 30 m spatial resolution, but the results were not used for verification. The identification of landslide site was used to combine with susceptibility map for producing the final susceptibility map. The susceptibility method is so subjective and can be included as heuristic approach. Castellanos and van Westen’s research (2003) in Cuba gave details how to construct the landslide hazard map by means of developed heuristic method. They developed analytical hierarchy process, a part of the decision support system, as the heuristic model. Pair-wise and ranking methods were chosen to extract the weighting value. The causative factors chosen were topographic, geological, hydrological, geomorphology and tectonic factors. They gave the initial weights for each class parameters and the parameter levels based on both their analysis of three existing landslide area and their knowledge. The procedure to build the hazard map can be described as selecting the
4 parameters, designing the hierarchical tree, assigning the weighting value, and generating the final map.
Compared with those researches, this study is significantly different. Two different maps of landslide susceptibility had been generated based on the types of landslides. The different can also be seen in the improved method which author used combination between pair-wise comparison - analytical hierarchy process (included as heuristic method) and bivariate statistical analysis. The BSA was only applied to extract the initial weighting value for each class parameter. Pair-wise comparison method was used to give the weighting value of parameter and then combining with the level of initial value of each class parameter to produce the final weighting value for each class parameter. Pair-wise comparison method which based on analytical hierarchy process was applied by means of Spatial Multi Criteria Evaluation (SMCE). In this research, the method for assessing susceptibility is not only using the improved method, but also bivariate statistical analysis and logistic regression (multivariate statistical analysis). The results of the three methods had been compared and the best results are used as final landslide susceptibility maps. Another difference is in testing the accuracy of landslide susceptibility map. The author selected prediction rate method where there were two scenarios of landslide inventory map; for generating and for verifying landslide susceptibility map.
5 CHAPTER 2. LITERATURE REVIEW
2.1. Types of Landslides United States Geological Survey defined landslide as a type of mass movement caused either by natural occurrence or human activity or combination of both. Based on the velocity of the movement, landslides can be classified into several types such as fall, flow and slide (Abbot,2004)as described in table 2-1, later Soetoers and van Westen gave new types; topple and spread. Table 2-1. Velocity of Several Types of Movement Type of Less than 1 cm a 1 mm/day to 1 1 to 5 km/hour Generally greater Movement year km/hour than 4 km/hour slowest ------ fastest flow creep/debris ------debris flow------ earth flow mudflow (water debris avalanche saturated debris) (debris) rock avalanche (rock) slide ------debris slide------ ------rock slide (bedrock)------ fall rock fall (bedrock) debris fall (debris) ------landslides------ Source: Abbot, 2004
2.1.1. Falls These types occur in very steep slope when elevated masses are separated along joints, bedding or the weaknesses. The detached masses go downward by falling, bounding and rolling which can be seen in figure 2-1
Figure 2-1. Fall Type
2.1.2. Flows Flows are the type of mass movement with a fluid motion. The speeds of the movements vary from slow to rapid. The ranges of material are differently gradated from massive boulders to sand, clay, snow and ice. The differences between slide and flow are based on the water contents and slip surface. Falls are strongly influenced by gravity, mechanical weathering, and the presence of interstitial water. There are several sub types of flows such as earth-flow, mudflow, debris-flow, and debris avalanche. The illustration of flows is presented in figure 2-2
6
Figure 2-2. Flow type
2.1.3. Slides Slides can be defined as movements of masses on failure surfaces or on the zone of shear strain. Typically of failure surface are either 1) curved in concave upward sense (rotational slides) and 2) nearly planar (translational slides). • Rotational slides These slides occur along a curved surface. In the crown parts, the displaced masses moves downward vertically and the top surface of displaced masses angles back toward on the scarp. Movement is more or less rotational on axis parallel to the slope (Cruden and Varnes,1996). To identify the centre of rotation, we can determine by piercing a cross section with the point of compass and then swinging the compass in an arc to draw the failure surface (Abbot,2004). This slides is illustrated in Figure 2-3a • Translational slides Translational slides are types of slides where the sliding can extend downward and outward along a broadly planar surface, and slide out over the original surface (Cruden and Varnes,1996) as depicted in Figure 2-3b. This type typically takes place along structural features, such as a bedding plane or the interface between resistant bedrock and weaker overlying material. (Abbot,2004) They are found globally in all types of environments and conditions. Translational slides generally fail along geologic discontinuities such as faults, joints, bedding surfaces, or the contact between rock and soil. These slides typically reflect a weak layer or existing structural discontinuity (Cruden and Varnes,1996). If the overlying material moves as a single, little-deformed mass, it is called a block slide.
a) b)
Figure 2-3. a) Rotational slide, b) Translational Slide
2.1.4. Topples Toppling is privileged by the presence of a steeply inclined join set with the strike aligned by more or less parallel to the slope face which can be seen in Figure 2.4. Topples are similar to falls except that the initial movement involves a forward rotation of the mass.
7
Figure 2-4.Topples type
2.1.5. Spread The distinctive type of landslide is a spread. This type generally takes places on very gentle slopes or flat terrain. The upper layer is broken and moves outward on the underlying layer. Figure 2-5 shows the illustration of this type. The surface’s boundaries of mass movement can be difficult and diffuse to recognize (Soeters and Van Westen ,1996 ). The failure is usually triggered by ground motion, such as that experienced during an earthquake, but can also be artificially induced.
Figure 2-5.Spread type
2.2. Landslide’s influencing factors There are many factors that should be considered to analyze landslide hazard (Varnes,1984). Soeters & Westen (1996) divided those factors into 5 groups of factors as described follow - Geomorphology factors such as data of terrain unit, geomorphological sub unit, types of landslide. - Topography factors such as data of digital terrain model, slope direction and length, concavities - Engineering geology factors such as data of lithology, material sequences, structure of geology, and seismic acceleration. - Landuse factors such as data of infrastructure (recent and older) and landuse map (recent and older) - Hydrology factors such as data of drainage, catchments area, rainfall, temperature, evaporation and water table map. As mentioned by Soeters & Westen (1996), it may not be necessary to include all parameters; because it depends which ones are relevant for the study area. It also provides to conduct landslide susceptibility assessment by using few parameters. Cruden and Varnes (1996) separated landslide causal factors into two group; preparatory and triggering factors. They also classified the causal factors of landslide including preparatory and triggering factors into geological, morphological, physical and human induced causes as given in the table 2-2.
8 Table 2-2.Causal Factors of landslides Geological Factors Morphological factors • Tectonic or volcanic uplift • Weak materials • Glacial rebound • Sensitive materials • Fluvial erosion of slope toe • Sheared materials • Wave erosion of slope toe • Jointed or fissured materials • Glacial erosion of slope toe • Adversely oriented mas discontinuity • Erosion of lateral margins • Aversely oriented structural • Sub terrain erosion discontinuity • Deposition loading slope or its • Contrast in permeability crest • Weathered materials • Vegetation loss
Physical factors Human Induced Factors • Intense rainfall • Excavation of slope and toe • Rapid snow melt • Loading of slope or its crest • Prolonged exceptional precipitation • Draw down of reservoir • Rapid draw down of floods and tides • Deforestation • Earthquake • Irrigation • Volcanic eruption • Mining • Freeze and thaw weathering • Artificial Vibration • Shrink and swell weathering • Water leakage from utilities Source : Cruden and Varnes, 1996
2.2.1. Landslide triggering factors Triggering factors are the events that change the condition of slope from generally stable to actively unstable state. The movement is initiated by triggering factors, such as earthquake, volcano eruption and relatively high rainfall. Rainfall is the most significant and common triggering factor for causing landslide occurrences in Indonesia. Generally, the infiltration process of rainfall influences the results in changing soil suction, positive pore pressure, ground water table, as well as increasing the weight of soil unit, diminishing shear strength of rock and soil (Huang and Lin, 2002)
2.2.2. Landslide preparatory factors It is different with preparatory factors which cause the changes from susceptible slope to movement without initiating it. Preparatory factors have a tendency to place the slope in generally stable state such as slope gradient, landuse, drainage pattern, lithology, geological structure, etc . The relationship between landslide occurrences and landslide preparatory factors in the research area will describe as below.
2.2.2.1.Slope Gradient The slope gradient is the most important factor to build landslide susceptibility mapping. This factor is generated from Digital Elevation Model. Based on USDA, the classes of slope gradient are extracted in percentage classified into 7 classes; 0 – 3, 3 – 8, 8 – 15, 15 – 30, 30 – 45, 45 – 65, and higher than 65%. Gentle slopes are predicted to give the low value for shallow landslide based on commonly lower shear stresses
9 correlated with low gradient. At slope from 10 to 45 %, landslides may have high probability. The remained slope angle more than 45% will give a low probability relatively because the existing of resistant lithologic units. The assumption that these units are not covered by weathered units which are susceptible for land-sliding.
2.2.2.2.Landuse Landuse is one of the key factors responsible for the occurrence of landslides. The comparison between aerial photo 1994 and SPOT Imagery 2006 shows the drastically changes of forest area. Many forest areas were converted into agricultural areas and it was followed with increasing the number of landslide occurrences. Theoretically, barren land and shifting cultivation are more prone to landslides than other landuses. It could be happened because there is no deep root which can hold the soil. Contrarily, forest areas tend to decrease the landslide occurrences due to the natural anchorage provided by the tree roots. Landuse data is extracted from SPOT satellite imagery taken in 2006. The interpretation is done visually and then supported by supervised classification. Validating and verifying are done by checking most of research area. The classes of landuse are based on the landuse classification published by Mallingreu,J.P (1977) who researched about landuse in Indonesia, and the classes are categorized into reservoir / water body, settlement, forest, mixed garden, paddy field, shifting cultivation, filed crop, barren land, and shrub.
2.2.2.3.Lithology Lithology data was extracted from geological map. This data had been prepared for common geological functions without take into account the special needs of the susceptibility evaluation procedure. So, the geological units were regrouped based on lithological attributes rather than their stratigraphic content and age (Suzen and Doyuran,2004). Lithology is one of the main factors influencing the type and the intensity of the morphodynamic processes, including landsides. Thus, many researchers involved lithology as a factor for susceptibility mapping (e.g. Dai et al. (2001);; van Westen et al.(2003); Ayalew and Yamagishi (2005); Ayenew and Barbieri (2005); Ermini et al. (2005). Based on geological map scale 1 : 100.000 published by Geological Research and Development Centre of Indonesia, various rock formations in the study area have been grouped into seven classes to prepare the lithology data layer. The five classes correspond to (a) Schits and Phyllite, (b) Penosogan Formation, (c) Halang Formation Breccia, (d) Waturanda Formation (e) Tuff Member Waturanda Formation, (f) Totogan Formation, (g) Peniron formation. For area utilized for reservoir was defined as water body. The description of contents per each formation is discussed in chapter of research area.
2.2.2.4.Geological structure Saha et al, (2005) used the geological structure as a landslide causal factor. It means that landslide occurrences rise with proximity to geological structures. In this term, the lineament was presented as geological structure. The distance to lineament map signifies the presence of joint-fracture affecting the shear strength (Suzen and Doyuran,2004). The distance to lineament was worked in meter scale, and it tends to evaluate the local situation. Structurally, the research area is relatively complex. Based on visual interpretation from SPOT imagery taken in 2006 and geological map, we can identify that there is a major fault that pass through the area, named Kalibawang Fault . In addition to fault,
10 there are several lineaments in this area. Field observation shows that some of the lineaments are taken place on the cuesta area. Euclidean distance processing, one of the distance function, has been done to determine a buffer zone to signify the influenced area on the landslide occurrences.
2.2.2.5.Distance to road Beside the natural environment, human induced factor may raise the probability of landslide occurrences. Cutting the toe of steep slope and filling along the road are the common human activities on the hilly areas which increase the susceptible area to landslide. It is convinced when the author found that many landslide sites are near by cutting road area. The road network map was taken from topographic map scale 1 : 25.000. The best way to contain the effect of road factor in landslide study is by making a buffer on the upslope part.
2.2.2.6.Distance to river In order to evaluate the influence of drainage pattern on landslide occurrence, a spatial analysis is performed. Many landslide sites took place near by the river, especially river with steep cliff. Water flow scraps the edges of river causing the broken of upper part. The areas near by the river are more susceptible for sliding. It is supported by Suzen and Doyuran’s statement that corrected distance to river network is a sign of the possibility of river bank erosion in toe part of landslide. The river network map was taken from topographic map scale 1 : 25.000. Similar with the process in the geological structure and distance to road, the map of river distance was identified by buffering the river network. This is measured to be a reason of the terrain modification caused by undercutting of the slopes in the research area.
2.3. Soil Properties Soil properties in term of geotechnical engineering support to influence landslide occurrences. It could be happened when the soil moisture content exceeds the liquid limit in the field. This event generates the dangerous and prospectively devastating results as a soil become visible to be stable and then when distressed can unexpectedly break away (Handy and Spangler,2007). The soil loses its shear strength and become changed into a quick churning, and flow which cross anything in its path. Similar with that statement, Denisov in Handy and Spangler, (2007) declared that if the exceeding of saturated soil moisture content pass over the liquid limit, the soil become harmful to collapse and density under its own weight (if it ever becomes saturated). Increase in soil moisture content decreases soil strength by diminishing capillary cohesion. Liquid and plastic limits are significantly controlled by the clay content and clay mineralogy. Lambe and Whitman (1951) gave the description that the liquid limit is the water content where the soil has smaller shear strength. It gets to flow closely to a path of standard width when vibrated in a specific manner. The plastic limit is the water content where the soil starts to crumble and the soil can be rolled into thread of specified size. Based on theirs definition, the equation of plasticity index is formulated below: PI = LL – PL Where PI = Plasticity Index . LL = Liquid Limit PL = Plastic Limit
11 2.4. Remote Sensing and GIS in mapping Landslide Inventory. Many researches evaluate the characteristics of landslides on the basis of geomorphological analysis fulfilled through aerial photograph interpretation and field observation (Carrara et al, 2003). Recently, some improvements have been developed for recognizing the landslide occurrences. For instant ; combination between satellite imagery and digital elevation model can be used to detect landslide sites (Barlow et al.,2006). Usually, landslide occurs in mountainous areas where are less accessible. Instead of doing the direct mapping, using the remote sensing method will be more reliable. By visual recognition from remote sensing data using visual aspects such size, contrast, and morphological expression, the identification of landslide occurrence can be conducted (Cassale et al., 1993 ). This process will show the scarp, disrupted vegetation cover and deviation in soil moisture related to past landslide occurrences (Marhaento,2006 ). The advantages of GIS technology such as geo-statistical analysis and database processing provide necessary tools for supporting landslide susceptibility map. The statistical method has been implemented by several researchers and it used to determine the relationship between slope failure and causal factor to generate landslide susceptibility map (Zhou et al.,2003).
2.5. Landslide Susceptibility The terminology used in this report concerning landslide hazards and associated concepts reflects the following definitions, based on Varnes (1984) as follow: • Landslide susceptibility refers to the likelihood of a landslide occurring in an area due to local terrain conditions. Susceptibility does not consider the probability of occurrence, which depends also on the recurrence of triggering factors. • Landslide hazard refers to the potential for occurrence of a damaging landslide within a given area. Landslide hazard includes spatial probability, size probability and temporal probability. The landslide susceptibility assessment provides important information for conducting the landslide hazard assessment. Spiker and Gori (2000) stated that landslide inventory and landslide susceptibility maps are critically needed in landslide prone regions of the nation. Landslide susceptibility map proposes to determine the areas that would be affected by landslide in the future either by natural or human induced factors. These maps must be sufficiently detailed to support mitigation action at the local level. Generally, several methods pertaining to landslide susceptibility mapping can be divided into three methods: qualitative, semi quantitative and quantitative (Lee and Jones,2004). The qualitative method is a subject oriented method and well applied on the large areas and the quantitative method is object oriented method which looks for relationship between environmental factors and previous landslide occurrences. There are some approaches which can be classified into either qualitative or quantitative method. Some approaches from those methods have been elaborated by researchers, such as heuristic approach by Ruff and Czurda (2007), statistical approach by Carara et al (1991), deterministic by Soeters and Westen (1996) and probabilistic approach by Guzzeti et al (2005).
12 2.5.1. Heuristic approach Heuristic approach is based on opinion of geomorphologic experts. Landslide inventory map is accompanied with environmental factors to be main input for determining landslide hazard zonation, and then the experts define the weighting value for each factor. Heuristic approach takes into account a hierarchical level and different method for determining weigh factors. Next, the hierarchical heuristic model becomes a part of decision support system (DSS) which aims for spatial decisions (Castellanos and Van Westen,2003). Generally, this approach can be divided into two analyses, direct mapping analysis and qualitative map combination. In the direct mapping analysis, the geomorphologist determine the susceptibility in the field directly which is based on individual experience. In the later analysis, the experts use their knowledge to determine the weighting value for each class parameter in each parameter. The main problem of this approach is in determining the exactly weighting value because this approach is mainly subject oriented method.
2.5.2. Statistical approach In term of landslide susceptibility assessment, the statistical approach has been expanded to produce higher degree of objectivity of landslide hazard assessment (Van Westen,1993). The key of this method is landslide inventory map when the past landslide occurrences are absolutely needed to forecast the future landslide’s areas. The combinations of causal factors are statistically determined, and based on similar existing condition, quantitative estimation of landslide occurrences are generated for currently free areas of landslide (Soeters and Van Westen,1996). There are two different analyses in the statistical approach; bivariate statistical analysis and multivariate statistical analysis. Both of these analyses will be deeply explain in chapter 3.
2.5.3. Physically-based modeling approach Landslide hazard assessment is determined by using slope stability model. In the contrary with other approaches, this approach quantitatively produces the stability index by calculating the safety factors. There are some weaknesses of this approach. The complicated calculation of safety factor, detailed required datasets, and the measurement of parameter’s spatial distribution relatively become the obstacles of this approach (Van Westen et al.,2005). Some experts gave some limitations for this approach. Van Westen (1993) stated that It can be implemented only at large scale and over small areas, because some input data such as ground water level, soil profile and geotechnical description are not sufficient in regional and medium scale. The resulted values of safety factors are only used to test different condition of slip surfaces and ground water depth, so they should not be used as absolute values. This approach is also acceptable only when geomorphologic and geological conditions over whole research area are moderately homogenous and the existing landslide types are simple. (Soeters and Van Westen,1996).
2.5.4. Probabilistic approach This approach is based on some following methods such as interpreting remote sensing data, field observation, interviews, and historical analysis of landslides in study area (Soeters and Van Westen,1996). Because the resulted spatial distribution of landslide occurrences presents the location of each type of landslides, it is considered as a straightforward form of landslide hazard zonation. The weakness of this approach is
13 that the result doesn’t present estimation of temporal changes in landslide distribution. It causes landslides that occurred before image taken may be unpredictable. Therefore the generating of landslide activity map should be based on multi temporal interpretation of aerial photographs (Van Westen,1993). Even though this approach doesn’t explore the correlation between landslides and causal factors, it can be used to describe the density of landslide qualitatively.
2.6. Magnitude of landslides In addition to the strength of mind of location and date of occurrences (Varnes,1984) landslide susceptibility assessment also involves the magnitude of potential phenomena (Guzzetti et al.,2005) The landslide magnitude was mostly articulated in term of landslide area which is generally easy to be measured quantitatively. In a few cases, the magnitude of landslide was expressed in term of volume of soil displaced which is much more difficult to assess (Marques,2008). The volume of soil displaced was predicted by reflecting on the horizontal area lost. Stark and Hovius (2001) stated that the reasonable proxy for estimating landslide magnitude is landslide sizes such volume and area which can be obtained from landslide observation. Recently, Malamud et al (2004) proposed a formula of landslide magnitude based on the total number of landslide events. So, three parameters which purpose to estimating magnitude of landslide were used in this research. They are total effected area, number of occurrences, and volume of the landslide.
14 CHAPTER 3. METHODOLOGY The main objective of this research is to produce landslide susceptibility maps by using the combination between bivariate statistical analysis and spatial multi criteria evaluation. The maps are expected to estimate the landslide occurrences in the future. The other objectives are to invent the distribution of landslide occurrences as a base to generate a landslide susceptibility map. At the end of this research, the author will validate the improved method by comparing the results with pure bivariate statistical analysis and multivariate statistical analysis and the best results become the selected susceptibility map of research area. The following part will briefly describe the used materials and equipment, and the methods chosen to solve the objectives.
3.1. Raw materials, data requirement and equipment The basic materials which are used and considered in this research are listed as follow • Aerial photos from the year 1993 at scale 1 : 20.000 which contains of 7 photos (just some parts of research area), source Perhutani Unit I • Aerial photos from the year 1994, date taken April 23rd, 1994, at scale 1 : 50.000 which contains of 7 photos (whole research area), source : Bakosurtanal. • SPOT Imagery, resolution 15 m, at the year 2006, date taken 6rd September 2006, source : ITC, the Netherland • Landuse map from the year 2000 at scale 1 : 25.000, source : Bakosurtanal • Topographic map from the year 2000, sheet 1408-143, Wadas Lintang, at scale 1 : 25.000, source : Bakosurtanal • Geological map from the year 1992 At scale 1 : 100.000, source Geological Research and Development Centre, Bandung • Daily rainfall of research area from year 1980 to 2008, source : Public Work Service of Wonosobo Regency and Badan Meteorologi dan Geofisika, Semarang. The spatial data availability is the most important thing to make the environmental modeling in Geographic Information Science. It helps the decision maker to manage the implementation (Skidmore,2002). Not all data related to landslide parameters can be directly extracted. Some of them must be prepared as is shown in the table 3.1. Table 3-1. Data required, row materials and preparation processes No Parameters Row Materials Preparation processes
Creating slope gradient map from 1 Contour Map (Topographic Slope gradient DEM / elevation map then reclassify Map,) based on their class 2 Detailed by using landform map and Litology Geological Map RS Image 3 Geological Detailed by using landform map and Geological Map Structure RS Image Interpreting satellite image to achieve 4 landuse information (visual and Landuse SPOT Image 2006 supervised interpretation) and then validated by field observation 6 Make a buffer from road feature in Road network Topographic Map the digital base map.
15 7 Make a buffer from river feature in River network Topographic Map the digital base map. Delineating the locations of landslide 8 RS Images, Reported events, Landslide sites occurrences and verifying all of and Field Observation samples in the field observation 9 Extracting data from a station Rainfall Data Rainfall reports (4 station) completed by data from the others 10 Soil Texture and Soil Sample of Landslide Laboratory Investigation Plasticity Index Sites
Some tools are used to measure the characteristics of landslide occurrences in the field work. They can be seen in the table 3-2 below: Table 3-2. Equipment used in the field observation No Name Function Pictures
Laser range Finder / Measure the distance 1 Laser Ace between two points,
Global Position Determining the 2 System (GPS) geographic position
3 Compass (Suunto) Measuring the azimuth
Clinometers 4 Measuring slope angle (Suunto)
6 Soil Trowel Taking a soil sample
Source : Field Work
3.2. Data preparation All of the row data can’t be directly used to put into the processes, so it had been prepared firstly. In this sub section, the author explains the preparation processes. a. Aerial photos Aerial photos were taken in TIFF or JPEG format, and to put these data to spatial format, geometric corrections were done. Geometric corrections are based on the Ground Control Points which exactly known in the field. The author did this process by using ARC GIS ver 9.3 and ARC View ver 3.3. After the process, the
16 result had been converted into ILWIS format, and then combining the image with DEM data by means of stereo pair process. The combination was shown as anaglyph map. The aim of this procedure is to analyze the landslide distribution and to generate landform map. b. SPOT Imagery. Similar with the processes did in aerial photo, this image was corrected geometrically. To give a good visual appearance, the bands of this image were arranged. Based on that appearance, we identified the landslide sites and classified landuse in the research area. There was a research in Hongkong providing that SPOT imagery could be used to identify landslide sites. They applied automated change detection using SPOT multi-spectral satellite images with 20 m spatial resolution (Nichol and Wong,2005). Although there was only a SPOT image, recognizing landslides still could be implemented by followed extensive field check. Despite the visual interpretation, determining landuse had used supervised classification by taking some locations as sample plots of specific landuse. These sample plots were utilized as the bases for identifying all parts of research area. The basic concept of supervised classification is classifying by using same value of the pixels in the sample plot to these in the whole area. c. Digital Elevation Model (DEM) In this research, DEM was used to build the slope gradient map. DEM was not directly obtained from existing map except SRTM data, so it should be generated from some raw data, for instant height point and contour data. In this case, the available data is a contour map taken from topographic map at scale 1: 25,000 with interval range 12.5 m. DEM is constructed by using Topo to Raster Process in the ARC GIS Ver 9.3, and then the slope gradient was generated in percent rise by using spatial analysis tool. The other purposes of generating DEM are to support the identification of landslide sites and to generate landform map. Extracting DEM can also be done by using other spatial software such as ILWIS, MapInfo, etc. d. Road and river networks The maps of both river and road network were taken from topographic map with scale 1 : 25,000. The whole of research area was separated into some classes based on the distances between locations and river or road network. The distances can be classified into 6 classes; 0 – 20 m, 20 – 40 m, 40 – 60m, 60 – 80m, 80 – 100m and more than 100 m e. Landform map. Some map sources (e.g. geological map) has a less detail scale. To make it same to the others, detailing processes had been done. Landform map could be used a basic to recognize more detail layer. This map was produced by identifying either anaglyph or stereo pair map.
3.3. Methods There are several methods to solve the objectives of this research which are described in the sub section below.
3.3.1. Landslide Inventory Landslide inventory map was built by using interpretation of both aerial photo (taken in 1993, 1994) and SPOT Imagery (taken in 2006) and verified by field works. Voskuil (2008) gave some indicators to recognize landslide both by RS Imagery and field check as follow:
17 • Semi-circular back scars and cracks. • Irregular/hummocky ground beneath the scars (local) • Irregular/hummocky slopes • Pounding of water in slides material, anomalies in drainage • Vegetation anomalies • Presence of man-made structures
Before field work had been done, the author built landslide inventory map by using remote sensing imageries (explained in data preparation section). The result had validated by field observation, where misinterpretations and inactive landslide sites were eliminated. To verify the realistic susceptibility maps, landslide occurrences from 2007 to 2009 were separated differently to test the landslide susceptibility map. The field work also proposed to add landslide sites which were not recognized in interpreting RS Imagery. Furthermore of these proposals, field observation was done to easily collect and recognize the types and dates of landslide occurrences. Interviewing the local communities was perceived helpfully, especially to identify the locations and boundaries of landslide sites. The other helpful material was the report of landslide occurrences from 2000 to 2009 presented by local government of Wonosobo Regency. This report showed the dates, villages and damages of landslide events. Figure 3-1 presented the procedure to generated landslide inventory map.
Aerial photos 1994 Contour map SPOT imagery 2006
Extracting DEM
Combining Combinin g DEM data
Anaglyph map Stereo pair map
Interpreting Interpreting
Landslide sites 1994 Lan dslide sites 2006
Field observation Reported landslide
Landslide inventory map
Past landslide Separated Recent landslide inventory map inventory map in 2006
Figure 3-1. Flow chart of building landslide inventory map
18 3.3.2. Rainfall Frequency Analysis This research took a center of attention on rainfall as the triggering factor of landslide occurrences. The reason is that rainfall generally causes of landslides in the subtropical and monsoonal climatic regions covered by hilly or mountainous topography. The author collected as much rainfall data in the study area. The aim of the rainfall frequency analysis is to estimate the temporal probability of landslide occurrences. By analyzing it, we can know the relationship between the temporal probability of occurrences and its return period. Gumbel Extreme Value Distribution was chosen as the method to analyze the rainfall data in the research area. The benefits of this method are that it can applied even with a small number of data sets and it can be used with extreme values (Parodi,2005). The rainfall data of research area was collected from surrounding rainfall station. The maximum value of monthly rainfall per year was extracted by summarizing the daily rainfall data. Then maximum annual rainfall values were sorted and ranked from the lowest to the highest. The lowest rank 1 was given to the lowest value of rainfall data and the highest rank N was allocated to the highest data value. The daily rainfall data before landslide event is also considered to be analyzed. It is computed over the return periods before the landslide initiated. After that process, the other step should be followed that described below. • The rainfall data per year was sorted by ascending and then we give the rank value started from smallest to the biggest value. • The left sided probability for each observation was calculated by means of equation below : R PL = N+1 Where PL : left sided probability (probability that a certain rainfall amount is lower than the one considered) R : is the rank of a given amount / value of rainfall N : number of observed years • After all of left probability value is got, the return period (TR) was determined by using equation below. 1 1 TR = = PR 1 - PL Where PR : right sided probability (probability that a certain rainfall is higher than the one under consideration)
3.3.3. Soil properties analysis Soil data should be included as a factor to build susceptibility map. The lack of the data and the time consuming to build semi-detail soil map became the reasons why it was not put into susceptibility map. The other approach is by knowing the tendency of soil in the landslide sites. In this research, the soil properties data that had been extracted is texture and plasticity index. The soil samples were collected from landslide sites by means of field observation and then brought to laboratory for the investigation.
19 3.3.4. Determining landslide influencing factors of research area In this research, the causal factors of landslides are determined by field observation, general literature inputs and data availability. Some factors had been chosen such as slope gradient, landuse, lithology, geological structure, distance to river, and distance to road. The reasons why chose them were explain in chapter 2. Exactly, soil properties data was needed to be put in the causal factor, but since there was no valid soil data, so the samples of soil in each landslide site were only taken to know the tendency of soil properties in the research area. To determine the landslide influencing factors, sensitivity analysis was done by eliminating each factor in generating susceptibility index. The factors which are not significant were removed in constructing the final susceptibility index map.
3.3.5. Landslide Density Landslide density can be defined as the area of landslide, divided by the total area. It can be implemented to calculate landslide density within the parameter class and landslide density within the entire area ( van Westen, 1993 and Yin and Yang, 1988). The formulas of landslide density are presented below ;
NpixXi ∑ NpixXi DensClas = DensMap = NpixNi ∑ NpixNi where, area is represented as a number of pixels, Densclas = the landslide density within the parameter class, Densmap = the landslide density within the entire map, Npix(Xi) = number of pixels, which contain landslides, in a certain parameter class, Npix(Ni) = total number of pixels in a certain parameter class Some times landslide density is also presenting in point unit. It means that landslide density is measured by dividing the number of occurrences with the entire area. The unit is presented as occurrences per square kilometer.
3.3.6. Landslide susceptibility analysis There are three methods which were compared in generating landslide susceptibility maps; the improved method (combination of BSA and pair-wise), bivariate statistical analysis and multivariate statistical analyses. The description of these methods are described in the sub section below
3.3.6.1.Bivariate statistical analysis In this technique, the contribution of each parameter class of causal factors was assessed independently with landslide distribution. It proposed to determine the density of landslides in different parameter classes, relative to the landslide density over the entire research area. Using these density values derived from existing landslide distribution, the landslide susceptibility was predicted for absences landslide areas, by addition of the weights of individual parameter classes. Van Westen (1993) had proposed some procedures to perform the bivariate statistical analysis (BSA) for landslide hazard zonation studies. • Categorizing each parameter map into some relevant class parameters. • Overlaying and analyzing the combination of the parameter classes with landslide inventory. • Calculating the weight values for parameter classes based on cross tabulated data • Assignment of the weight values to the parameter classes and integration of the parameter maps • In the end processes, the resulting value is classified a few hazard classes .
20 In BSA, author used Information Value Method (Yin and Yan,1988) to build landslide susceptibility map which uses the following formula as below :
Where Wi is the weight given to a certain parameter class i; DensClas is the density of the landslides in the class i; DensMap is the density of the landslides in the whole study area; NpixXi is the number of pixels falling within the polygons representing landslides occurring within the class i; NpixNi is the number of pixels within the class i; ΣNpixXi is the total number of pixels falling within the polygons representing landslides occurring in the whole study area; ΣNpixNi is the total number of pixels of the whole study area map. In addition to Information Value Method, there are some methods in term of BSA which developed by some researchers such as Landslide Susceptibility Analysis (Brabb,(1984); van Westen, (1993)), Weights of Evidence model (Lee and Choi,2004).
3.3.6.2.Multivariate statistical analysis In Multivariate statistical analysis (MSA), the combinations of all causal factors gave the weighting values which control the landslide occurrences. These values signify the contribution of each combination of causal factors to the degree of landslide susceptibility within defined land unit. Moreover, the relations among the causal factors are also met within the analyses. The general property of these analyses is their nature of being based on the presence or absence of landslide occurrences in the defined land units and there are two main approaches for multivariate statistical analysis (Van Westen,1993): • Statistical analysis of point data obtained from checklist of causal factors associated with individual landslide occurrences. • Statistical analysis performed on terrain units covering the whole study area. In comparison with bivariate statistical analysis, multivariate statistical analysis is rather time and budget consuming, in term of both collecting and processing data. We have to take samples for relevant factor either on unique condition area or terrain units and then determine either the presence or absence of the landslide occurrences. The resulted data will become the subjects in multivariate statistical procedure. The procedure to generated Multivariate Statistical Analysis can be described below : (Van Westen,1993): • Defining the landslide causal factors as the parameter maps and then overlaying all of the parameter maps to produce the land unit map. From the later process, a large matrix is resulted. • Crossing the land unit map with the landslide inventory map to extract the stable and unstable areas. • The coefficients approximated by the multivariate analysis are used to calculate the landslide susceptibility of the entire study area. • Slicing the landslide susceptibility values into a few susceptibility classes. Application of multivariate statistical analysis in this research is as the comparing method with selected method. The method of Multivariate Statistical Analysis used in this study is Logistic Regression which doesn’t required normally distributed input variables. Logistic Regression Method is a mathematical modeling approach generating
21 to explain the relationship among several independent variables and dependent variable (Hosmer and Lemeshow,2000). The relationship above can be expressed as formula below: Pr (event) = 1 / (1 + e –z) Where Pr (event) is the estimated probability of landslide occurrence which varies from 0 to 1 and Z is the linear combination:
Z = B0 + B1X1 + B 2X2 ………+ B nXn Where Bi (i=0, 1, 2 . . ., n) is the coefficient estimated from the sample data, n is the number of independent variables (i.e. landslide-related physical parameters), and Xi (i=1, 2,. . ., n) is the independent variable . The other methods in Multivariate Statistical Analysis are Multiple regression and discriminant analysis which applied by some researchers recently (Ayalew and Yamagishi, (2005); Yesilnacar and Topal, 2005). Both of these methods required the normally distributed input variables (Suzen, 2003)
3.3.6.3.Spatial Multi Criteria Evaluation Spatial multi criteria evaluation (SMCE) is a model in ILWIS Software which is categorized as a heuristic method generated in more transparent way. Implementing SMCE helps users to apply multi criteria evaluation in analyzing spatially (ITC, 2001). The model is built by making criteria tree, where the causal parameter maps is grouped, standardized and weighted. The landslide causal parameters are weighted by means of direct, pair wise and rank ordering comparison and the output is a composite index map (Castellanos and Van Westen,2007). In this research, the weighting method used is pair-wise comparison method This method assumes that the users comparably evaluate the difference of magnitude among all unique pairs of factors qualitatively. Pair-wise comparison method was established by Saaty (1980) in the context of the analytical hierarchy process (AHP). In this process, the weights are defined by standardizing the eigenvector correlated with the highest eigenvalue of the ratio matrix. The AHP consists of three main steps ; 1)generating the pair wise comparison matrix, 2)computing the weights of the criterion, and 3) estimating the consistency ratio (Malczewski,1999). In the development comparison matrix the method employs an underlying scale with values from 0 to 1 to rate the relative preferences for two criteria which can be seen in the table 3-3 below : Table 3-3. Scale for Pair wise Comparison Intensity Definition 1 Equal Importance 2 Equal to moderate importance 3 Moderate importance 4 Moderate to strong importance 5 Strong Importance 6 Strong to very strong importance 7 Very strong importance 8 Very to extremely strong importance 9 Extremely importance Source : Saaty 1980
3.3.6.4.Improved method This research used the combination between bivariate statistical analysis and pair-wise comparison. Firstly, to know the scored value for each class parameter, the author calculated the density of landslides by using partly steps in the bivariate statistical analysis. The second process is grouping the causal factors into 4 general factors such as human induced, geological, hydrological and geomorphologic factors. Next, the levels of weight values were used to standardize input value by means of pair wise comparison resulting values from 0 to 1. After this process the steps in spatial multi-criteria evaluation were followed again by means of pair
22 wise comparison method. The difference of this improved method was located on giving the weighting value of each parameter. The weighting value of this method was given by calculation process of analytical hierarchy process. The values were extracted based on the level of influences. Expert opinion which depends on observed physical characteristic of landslide sites determined the levels of the influencing factors. The levels of influence for each parameter can also be investigated from the weight values in BSA and the B value in MSA. The procedures to generate the landslide susceptibility maps are presented in the figure 3-2.
3.3.7. Verifying landslide susceptibility maps Accuracy test is absolutely needed to verify the performance of susceptibility maps. prediction rate, the chosen method, was applied to know how well the model performs for future landslide occurrences. The distribution of multi temporal landslide event is definitely required in this method. We need at least two types of landslide inventory maps. The first one is to build landslide susceptibility map and the other ones are to test the accuracy of landslide susceptibility map. It is the difference with success rate method which only need a landslide inventory map. Success rate is made just for prove how good the estimation of model when it was built (Chang-Jo et al, 1995). In both prediction and success rate, the percentage of landslide occurrences was calculated. The calculated result was plotted in the graph where the percentage of the map in the X-axis and the percentage of the landslide occurrences in the Y- axis (Van Westen et al, 2009). The curve reveals the percentage of area of landslide occurrences which is estimated by percentage of predicted area. The better result is if the less percentage of map area can predict the higher percentage of landslide area. Figure 3-2. Flowchart of Landslide Susceptibility Analyses Defining the landslide parameters Overlay Past l andslide inv map Slope gradient Geological structure Litology Landuse Presence and absence Road distance location River distance
Sampling MSA process
Sampled data
Logistic Regression Intercept and index value Analysis in SPSS
Combining Recent landslide inventory map Combination values
Calculating Landslide suscept index map probability
Landslide Prediction rate Reclassify suscept map 3
Predicted rate A graph 3
23
Defining the landslide parameters Satellite imagery Slope Gradient interpretation and aerial photo Geological Structure Litology Landuse Road Distance Past landslide inv map Calculate landslide density River Distance for each class parameter
Verified
Field Landslide density for observation each class parameter
Combined No with pair-wise
Yes
BSA Geology Hydrology Geomorphology Human Induced
Standardization and Report of recent Weighting pair wise analysis landslide occurrences
Weighted value for each Landslide suscet index parameter map 2
Field IMPROVED METHOD observation Pair wise analysis Reclassify for group parameter Landslide Recent landslide Weighted Value for each suscept map 2 inventory map group parameter
Pair wise analysis Predi ction rate for goal 2
Landslide suscept index map 1 Prediction rate graph 2 Reclassify
Landslide susceptibility Predicted rate Pr edicted rate Comparing and map 1 1 graph 1 analyzing
Conclusions and A recommendation
24 3.3.8. Classifying the landslide susceptibility map After deciding the best susceptibility index map, it is important to divide the map into a few susceptibility classes. There are four classification system to realize the necessity ; quantiles, natural breaks, equal interval and standard deviation (Ayalew and Yamagishi,2005). Ayalew and Yamagishi found some characteristics of those systems related to calculate the probability. Quantile system tends to place widely different value in the same class. For the data value with big fluctuation, the natural breaks give the better result; unfortunately this system is not suitable for the case which depicts the probability map. Next, Equal intervals, the common system in Indonesia, emphasizes one class relative to the others. It becomes unhelpful because it does not reflect the real susceptibility. The last method is by means of standard deviation. This system generates the class break supported by mean value which the number of classes are determined by the system. Finally, the standard deviation method was chosen as the classification method for landslide susceptibility maps. The mean value plus or minus a half of standard deviation will produce the first class boundary. The next boundary is generated by adding the standard deviation with the previous boundary.
3.3.9. Extracting Landslide Magnitude Because of limited data availability, the calculation of landslide magnitude only used a few available data. Malamud et al (2004) proposed some equations to estimate the magnitude of landslide based on total area of landslides, number of events, and total volume of landslide. Malamud et al fit a three-parameter inverse-gamma distribution; p : (parameter primarily controlling power-law decay for medium and large values) which is equal to 1.40 , a : (parameter primarily controlling location of maximum probability) which is equal to 1.28x10 -3km 2, and s : (parameter primarily controlling exponential rollover for small values) which is equal to -1.32x10 -4km 2 The first magnitude equation is based on number of landslide associate with the event and the formula is mL = log NLT (1)
Where mL is magnitude of landslide NLT is total number of landslides
Magnitude of landslide also can be estimated by using area of landslide which is formulated below : mL = log ALT + 2.51 (2) 2 Where ALT is total area of landslides in km
Equation 2 is extracted from combination between equation (1) and (3) below; a N + s -3 ALT = NLT ĀLT = LT − = NLT x 3,07 x 10 (3) p 1
2 Where ĀLT is average of total landslide area in km p is equal to 1.40 a is equal to 1.28x10 -3km 2, ѕ is equal to -1.32x10 -4km 2
25 The other proposed equation to measure landslide magnitude is based on landslide volume. This formula requires the average volume of landslide. Although volume analysis is not done in this research, there is an assumption to estimate the average volume which is showed in equation below:
= = −6 .1 1222 VLT V L N LT 30.7 x10 N LT (4)
3 Where VLT is total volume of landslides in km 3 V L is average volume of landslides in km
By combining equation 1 and 4, landslide magnitude is obtained based on volume landslide which is used equation below :
mL = 0,89 log VLT + 4.58 (5)
From those equations, the number of events becomes the key to generate magnitude of landslide. Using three-parameter inverse-gamma distribution, Malamud et al gave the estimation of area and volume of landslide in relation to number of events. The Malamud’s method becomes the main formula to extract magnitude of landslide in this research.
26 CHAPTER 4. RESEARCH AREA
4.1. Geographic position Wonosobo Regency is located in the middle part of Central Java Province. Geographically, Wonosobo Regency is situated at 109º45’ - 110º04’ East and 07º20’ - 07º48’ South. The total area of Wonosobo Regency is ± 100.400 Ha divided into 15 sub districts which are Wadas Lintang, Kaliwiro, Kalibawang, Kepil, Sapuran, Kalikajar, Selomerto, Leksono, Sukoharjo, Wonosobo, Kertek, Mojotengah, Watumalang, Garung and Kejajar. Administrative map of Wonosobo Regency is also presented in the Figure 4-1. The geographic condition is varying from flat to mountainous terrain of which the lowest altitude is 270 msl and the highest altitude is 2,250 msl. In regard to the number of landslide occurrences (see figure 4.4.), Wadas Lintang Sub District has been selected for the research area.
Figure 4-1 : Administrative Map of Wonosobo Regency (Source Local Government)
Boundaries of research area administratively shall be as follow: - Northern part is adjacent to Kaliwiro Sub District. - Eastern part is adjacent to Purworejo Regency. - Southern part is adjacent to Purworejo Regency. - Western part is adjacent to Kebumen Regency.
4.2. Rainfall activity To build the average monthly rainfall, rainfall data was extracted from 4 stations (Wadas Lintang, Limbangan, Sapuran and Kaliwiro stations ) surrounding the research
27 area where the Wadas Lintang station is located in the centre of study area. Next, the main data was extracted from this station supported by the others. The cart in figure 4.2 shows the average monthly rainfall taken from 4 surrounding stations.
600 502 481 488 490 500 430 385 400
300 244 191 200 146 Rainfall (mm)
100 59 40 40 0
ry y e ly st r r r ay u e ua M Ju b rua April Jun m March Aug Jan eb October F eptembe ese S NovemberD
Figure 4-2 Average Monthly Rainfall of Wadas Lintang sub district 1980 - 2008
It is noticed that the peak amount of rainfall occurred in January, March, November and December, and the lowest amount of rainfall occurred in August and September.
4.3. Geological and geomorphological condition of research area Based on geological map scale 1: 100.000 sheet Kebumen, it was known that Wadas Lintang Sub District consists of 7 formations. The formations are presented in table 4-1 Table 4-1 Geology formation of research area CODE FORMATION LITOLOGY MATRIX Amphibol,mica,glauchane schists, Km Schits and Phyllite phylite sandstone,claystone,tuffs,marls,cal influenced by Tmp Penosogan Formation carenite turbidity current tufasandstone, Halang Formation Tmpb Breccia: andesite,basalt,limestone intercalationsandst Breccia one,basalticlava grained sandstone, breccia: andesit Tmw Waturanda Formation sandstone, tuff basaltic Tuff Member glass crystal tuff, sandstone, Tmwt Waturanda For tuffaceous marls Breccia : Tomt Totogan Formation claystone,sandstone,limestone, scaly clay basalts tuffaceous Polymict breccia : Tpp Peniron formation sandstone,tuff andesit,claystone, limestone intercalations WB Water body Source : Geological map of Kebumen
28 Spatially, the distribution of litology formation is presented in Figure 4-3. Despite the litology, the geological map also show that there are some geological structures located on research area. Two fault were crossing the study area; Kalianget Fault which cross from southern to the northern part and Segayam Fault crossing the northern west part. Some lineaments also appear in some parts. It can be recognized by using satellite imagery and also depicted in the geological map.
Figure 4-3 Litology map of research area Topographic condition of research area is dominated by hilly area covering 31.18% from total area. Hilly area are widely spreads on the research area. The eastern part is covered by mountainous area with steep slope class 45 – 65%. There are only a few parts of wadas lintang sub district which have a very steep slope. The flat and undulating areas took place near by the reservoir and Medono River. The visualization of research area is depicted visual in figure 4-4 and the composition of slope classes is presented in table 4-2.
Figure 4-4 Topographic visualization of research area
29 Figure 4-4 was made by combining hill-shade visualization and digital elevation model which is showed by means of ArcScene 9.3. Hill-shade was generated based on digital elevation model and then were put into ArcScene 9.3. Its base height was set by using the base height of DEM and the appearance becomes 3D visualization. Table 4-2 the composition of slope classes in the research area. Description Slope Gradient Classes Hectares Percentage Flat 0 - 3% 2518.17 19.20 Undulating 3 - 8% 1773.16 13.52 Moderately sloping 8 - 15% 3474.62 26.50 Hilly 15 - 30% 4088.74 31.18 Moderately steep 30 - 45% 1025.48 7.82 Steep 45 - 65% 211.36 1.61 Very steep > 65% 22.69 0.17 Total 13114.22 100.00 Source : Data Analysis 2009 based on USDA Classification
4.4. Landuse of Wadas Lintang sub district Almost of research area is exploited as plantation, agriculture and settlement areas. The planted pine forest is lying on the hilly and mountainous areas. The agriculture areas take place from the hilly area to flat area such as field crop, paddy filed and shifting cultivation. Mixed garden dominated by fruit and albizzia trees is widely spread from flat to steep areas and the barren land and shrub come to pass on the steep and very steep area. In the southern part, the wadas lintang reservoir is placing. The composition of landuse can be seen in table 4-3.
Table 4-3 .The composition of landuse in the research area Number Landuse Hectares Percentage 1 Barren Land 46.68 0.36 2 Field Crop 685.15 5.22 3 Forest 2351.61 17.93 4 Mixed Garden 3505.49 26.74 5 Paddy Field 1741.46 13.28 6 Reservoir 1344.94 10.26 7 River 28.07 0.21 8 Settlement 1816.27 13.85 9 Shifting Cultivation 1094.38 8.34 10 Shrub 500.17 3.81 Total 13114.22 100.00 Source: Data Analysis 2009
4.5. Reported landslides of Wonosobo regency This sub section explains the reason why the author chooses Wadas Lintang Sub District as the study area. Landslide events used are based on the report of local authority of Wonosobo Regency. Based on figure 4-5, it was known that Wadas Lintang sub district had the highest occurrences of landslide events. The number of landslide occurrences is the highest one among another sub districts. Two crossing major faults are assumed as the causal factor.
30 60 53 50 50
40 37
30 27 24 22 20 18 17
Occurrences 20 14 11 8 10 7 7 1 0
Figure 4-5 Landslide occurrences in Wonosobo Regency, 2000 – 2008 Source: Local Government of Wonosobo Regency
31 CHAPTER 5. LANDSLIDE CHARACTERISTICS The contents of this chapter are the explanations about the activities taken to produce the landslide inventory map. The other activities that had been done were analyzing the rainfall as well as soil properties and landslide magnitude in the research area. One selected landslide sites was also used to depict the landslide profile and to estimate the volume of soil displaced and slope morphology analysis. All of data resulted by those activities were collected, verified and done by direct field observation.
5.1. Soil Properties Analysis The researcher collected 60 soil samples taken from landslide sites. They were brought and investigated in laboratory of Agricultural Faculty of Gadjah Mada University. All of those can be used to extract the soil texture content, unfortunately just 58 samples could be investigated to produce liquid and plastic limit data. The results of soil content had been plotted to soil triangle by putting percentages of the sand and clay contents. It proposed to determine the texture classes. Figure 5-1 depicts the distribution of texture classes.
100
90
80
70
60 clay 50 silty % clay clay sandy 40 clay silty clay loam clay loam 30 sandy clay loam Figure 5-1 Soil texture 20 loam silt loam distribution of landslide 10 sandy loam loamy sites silt sand sand 0 0 10 20 30 40 50 60 70 80 90 100 % sand
By avoiding the homogeneity of research area, Figure 5-1 shows that landslide sites were dominated by clay texture. Almost of clay fraction percentage is more than 30%. Soils hold a large amount of clay fraction has fairly different physical characteristics from one made up mostly of sand or silt fraction. The higher the percentage of clay fraction, the lower the permeability and the higher the probability of sliding. Generally, clay minerals are quite resistant in dry conditions, but rapidly lose their strength in wet conditions.
Extracted liquid and plastic limits were processed to produce the plasticity index which was formulated as liquid limit minus plastic limit. Next procedure is plotting Liquid Limit and Plasticity index into Casagrande Plasticity Chart as shown in figure 5-2.
32 100
90
80 70 Very Plactic Clay 60
50 40 Plasticity Index 30
20
10 Organic Clay / Plactic Silt 0 Very Plastic Silt 0 10 20 30 40 50 60 70 80 90 100 Liquid Limit
Figure 5-2 Casagrande Plasticity Chart of soil samples
Some general relationship between the limits and engineering properties (Lambe and Whitman,1951) is given in the table below:
Table 5-1 Relationship between the limits and engineering properties Characteristic Comparing soils at equal Comparing soils at equal Liquid limit with plasticity plasticity index with Liquid limit index increasing increasing Permeability decrease increase Rate of volume change decrease - Shear Strength increase decrease Source : Lambe and Whitman, 1951
Figure 5-2 shows that many samples have value of liquid limit more than 50 and based on table 5-1, the higher liquid limit corresponds to lower shear strength. As we know that the soil with low shear strength have a high possibility to landslide. When high rainfall event occurs, the exceeding of saturated soil moisture content passes over the liquid limit, and it triggers the soil to become harmful for collapsing.
5.2. Generating the landslide inventory map Landslide inventory map is the most important thing that must be clearly done to generate landslide susceptibility map by means of statistical analysis. Several sources had been extracted to identify landslide sites by using aerial photo year 1993 scale 1 : 50.000, aerial photo year 1994 scale 1 : 20.000 as well as SPOT Image year 2006. Reported landslide was also used to look for the location of landslide occurrences. The final result was produced by combining those tentative results validated by means of field observation.
5.2.1 Recognizing landslides by using aerial photos Identifying landslides by using aerial photos was done by means of stereo-pair image as well as anaglyph map. Based on stereo-pair and anaglyph map, the sites of
33 landslides were visually identified. The significant indicators showing the occurrences were irregular slope and vegetation anomalies. It was too difficult to recognize landslide by using aerial photo year 1993 because of its small scale. Unfortunately the aerial photo year 1994 which had a higher scale just covered a middle part of research area and there only was found 3 sites of landslide which have relatively large sizes. It was difficult to decide that aerial photo’s interpretations were really landslide sites. There was no convincing evidence presenting the indication of landslide. Those areas were covered by pine forest and settlement and the interviewing with local inhabitants showed that they did not really sure about landslide occurrences in those sites. So, the researcher decided not to use the resulted interpretations to avoid the mistakes of landslide’s identification.
Figure 5-3 Temporal Landslide investigation 1(a) aerial photo 1994,(b) SPOT 2006, (c) Field Observation 2009 5.2.2 Recognizing landslides by using SPOT Imagery Beside aerial photo’s interpretation, identifications of landslide occurrences by means of GIS and Remote Sensing were also done by using satellite imagery. The available data of satellite imagery was only SPOT Image year 2006. Similar with the procedure in using aerial photos, the prepared process was building stereo pair map by combining SPOT image and digital elevation model. We recognized 14 landslides where were distributed at 10 villages. Vegetation anomalies were identified by considering the absolute contrast of tone and they were combined with the irregular slope. The lack of detail satellite image caused that only the large areas of landslide could be identified. Verification was done by field work, which showed that there were 12 actual landslides from 14 interpreted landslides. One of actual landslides is presented in Figure 5-4.
(a) (b)
Figure 5-4 Temporal landslide investigation 2 (a) SPOT 2006, (b) Field Observation, 2009
One of the problems in landslide identification by using image interpretation in Indonesia is that many steep slopes have experienced to landslide at least once. To
34 determine the recent ones it must be verified by field investigations and interviewing with local communities. Field check convinced that there were 2 misinterpreted landslides. The first one happened in the pine forest area where there was slashing activity in the image’s date taken. The second one showed that it was a landuse conversion from pine forest into shifting cultivation in the steep slope area as depicted in Figure 5-5. (a) (b)
Figure 5-5 Misinterpretation Landslide investigation (a) SPOT 2006, (b) Field Observation
5.2.3 Field observation as a final mapping process One of the functions of field observation is to verify and measure the landslide sites. The reported landslides were in tabular data and did not give the spatial distribution, so all of reported landslide must be directly verified and measured in the field. The polygon shapes of landslides had been presented in spatial data as required in generating landslide susceptibility map by means of statistical analysis. The field observation of the researcher is depicted in Figure 5-6.
(a) (b)
Figure 5-6 Field observation (a) direct measurement, (b) taking soil sample of landslide sites Not only interpreted and reported landslides but also the unreported landslides were identified and measured (see Figure 5-7). The data of unreported landslide was investigated by interviewing with the local communities and the chiefs of sub villages who really know the location of landslides in their areas. It was proposed to collect landslide sites not just at the settlement areas and to identify the landslide sites as many as possible. The other benefit of interviewing activity is to easily recognize the dates and the boundaries of landslide occurrences. The boundaries became obscure when the fast plant growth covered the location and when the rebuilding activities hide the scarp of landslides.
35
Figure 5-7 Unreported landslides (a) in the mix garden (b) forest, (c) shifting cultivation area From field work activities, it was known that there are 71 sites of landslides. They were differed into 51 rotational slides and 20 translational slides. The pictures of rotational slides had been given in previous figure and the pictures of translational slides seen in the figure below. (a) (b)
Figure 5-8 The damages caused by translational slides in the settlement area. (a) wall break, (b) floor cracks
The boundaries of this type could be recognized in the ground as presented in figure 5-9 (a) (b)
Scarp
0.5 m
0. 4 m Rupture zone Rupture zone
Figure 5-9 The boundaries of translational slides (a)land drops down,(b) land breaks away
Based on the analysis of geomorphological condition, the distribution of landslide was presented in Figure 5-10. This figure shows that most of landslide occurrences were taken place in slope range 15 – 30%. It could be happened because this class (15-30%) dominates the research area which covers about 31.18% of total area. The areas with slope angle more than 30% just cover less than 10% of total area. The class 15-30% is steep enough for occurrences rotational slides. The extensive human activities
36 triggering landslide are limited until this slope class. The slope class >65% is relatively resistant to slide because ground conditions are mostly covered by bed rock.
50.00 45.00 40.00 35.00 30.00 25.00 20.00 15.00 Area of Landslide Area 10.00 5.00 0.00 0 - 3 % 3 - 8 % 8 - 15 % 15 - 30 % 30 - 45 % 45 - 65 % > 65 % Slope
Figure 5-10 Geomorphological Distribution of Landslide in Wadas Lintang The author also separated landslides based on the sources; they were separated into 38 reported landslides, 21 unreported landslides and 12 interpretation’s sites. Local authorities stated that there were 50 events of landslide in Wadas Lintang Sub Districts. After the field observation, not all of those could be classified as landslide, for instant; bridge collapse because of its scraped footprint, the fence failure because of its weak materials. The non landslides categorized were eliminated from the list of landslide occurrences. Finally, to construct and verify the susceptibility map, landslide inventory maps was separated based on types of landslides and periods of events.
The illustration of landslide distribution in Wadas Lintang Sub District was depicted in Figure 5-11 and the detail attribute data of landslide occurrences can be seen in appendix 1.
37
Landslide in Tirip
Landslide in Penerusan Landslide in Wadas Lintang
ctures Translational
Landslide in Kalidadap Landslide distribution map and pi map distribution Landslide 11 - 5 Figure Figure Landslide in Kumejing Landslide in Kalidadap Landslide in Kaligowong
38 5.3. Landslide Profile Generating landslide profile had been done by detailed measuring and depicting at landslide site. The limited time became main problem to make landslide profile for each occurrence. The reactive landslide of Dadap Gede (a part of Wadas Lintang sub District) was chosen as an example of constructing landslide profile. This site was selected because of its fresh occurrence and clear performance. The size of this site is 0.577 ha which is covered by cassava and banana trees. The recent slope morphology of Dadap Gede’s site is ranging from 0 to more than 65%. By generating DEM, it was known that the slope morphology before sliding was dominated by class > 45%. Local inhabitants stated that the sliding happened after high rainfall occurrences in 2008. The spring which flows crossing the sites could be one of the causal factors. The water is coming from seepage in the upper area which is planted with paddy field. The other processes are generating transect line, calculating the removal deposit and geomorphological analysis. The basic principle is by knowing the height of surface before and after the landslide events. It recognizes the area and volume of surface that had been change by removal or addition of surface material. The condition after the events was directly measured by using geodetic instruments such compass, clinometers, and Laser Range Finder and the condition before the events was assumed similar with contour map extracted from topographic map. The software used was ArcGIS ver 9.3, 3Dimension and Cut/Fill Volumetric Analyses. Landslide profile of Dadap Gede was illustrated in Figure 5-12
Crown
Body
Slide direction
Figure 5-12 Landslide profile of Dadap Gede
39 5.4. Volumetric Analysis Next, estimating the removal soil was done by means of volumetric analysis. Two DEM data (before and after sliding) were combined to produce volume estimation map. The process to produce volumetric estimation map was shown in Figure 5-13. DEM before 0 40 80
Meters
DEM after
Volumetric Analysis
Volumetric Estimation Map
Figure 5-13 Volumetric Analysis process
Supporting the volumetric estimation, some transect lines were built to show the surface conditions in Dadap Gede’s landslide as presented in Figure 5-14.
Surface before sliding Surface after sliding
(m)
(m)
(m)
Distance (m) Figure 5-14 Transect lines of Dadap Gede’s Landslide
Distance (m) 40 The Figure 5-14 shows that a middle part of area crossed by transects B had a loss of surface material more than 6 meter height. Transect B, C and D showed there was some surface material added to the south parts. The result of volumetric analysis stated that the material cut was estimated equal to 2,246.13 m3 located in 0.4824 ha and the material filled was predicted equal to 118.8 m 3 located in 0.0946 Ha. It means that just a small amount of soil loss was moving to filled area. Most of soil loss disappeared because of human activities and erosion as depicted in figure 5-11. The result of volumetric analysis was verified by the formula proposed by Simonett in Malamud et al, (2004). The formula is presented below:
= 1.368 VL 0.024 AL
3 Where VL is volume of landslide in km 2 AL is area of landslide in km Calculation shows that 1.368 VL = 0.024 * 0.00577 = 0.0000020773 km 3 = 2,077.3 m 3
In this case, using volumetric analysis or landslide volume equation gives a relatively similar result, about 8 percent different (equal to 168.8 m3)
5.5. Slope morphology analysis Analysis of morphology also had been done to estimate the topographical changes at the landslide site. It could be done by generating slope maps which extracted from DEM data. The slope maps depicting the topographical condition before and after the sliding were presented in Figure 5-15.
0 40 80
Meters
Before sliding After sliding
Figure 5-15 Slope condition before and after sliding at Dadap Gede
Topographical condition of Dadap Gede was significantly changing after the landslide event. As presented in figure 5.11, the ground surface changes from relatively regular pattern onto irregular pattern. In the upper part, the slope class changed from 30 – 40% class into 4 different classes; 0 – 3%, 3 – 8%, 8- 15%, 15-30%, and 30-45%. The very steep area (>65%) in the middle part became less steep areas and the moderately sloping area (8-15%) in the lower part became moderately steep (30-45%) and steep (45-65%) areas. In short, the landslide events can change the topographical condition on the certain areas from steeper onto less steep.
41 5.6. Magnitude of landslides Landslide magnitude of research area was extracted by means of area affected, number of events and volume of soil displaced. The used formulas were explained in chapter 3. The area affected and a number of events were obtained from landslide inventory map, and volume of soil displaced was estimated by using Malamud’s equation which predicts the volume by means of number of events. The research area is about 131 km2 where there are 71 landslide sites during 2000- 2009 with density 0.542 occurrence per square kilometer. Based on the area size, the landslide density of the research area is equal to 0.0086 Table 5-2 Landslide magnitudes in the research area Sources of Basic Equations Magnitudes Basic values values
1 mL = log NLT 1.8512 NLT = 71 Landslide inventory
-4 -6 1.1222 2 mL = 0,89 log VLT + 4.58 1.8573 VLT =8.726.10 VLT =7.30x10 N LT
-3 3 mL = log ALT + 2.51 1.8483 ALT =0.21797 ALT = NLT x 3,07 x 10
4 mL = log ALT + 2.51 2.5649 ALT =1.13486 Landslide Inventory
Where mL is magnitude of landslide, NLT is total number of landslides, ALT is total area of landslides in 2, 3 km and V LT is total volume of landslides in km
Based on Table 5-2, the magnitudes presented in equation 1,2, and 3 vary in the range mL = 1.848 – 1.8573. For these examples, using the total area affected, total number of events, and total volume soil displaced gives similar magnitudes, less than 0.01 different. When using landslide area based on inventory data (equation 4), the magnitude resulted is quite different with equation 1,2, and 3, less than 0.72 different. The dissimilar values were caused by the differences value between predicted landslide area and actual landslide area. The actual landslide area is significantly larger than predicted one. So far, we can estimate the magnitude of landslide in the range 1.8483 – 2.5649. Actually, the magnitude of landslide based on its volume also can be generated if there is detail volume data for each site.
5.7. The probability of occurrence rainfall triggering landslides It is quite important to know the relationship between rainfall data and yearly landslide events. The first step was building the rainfall return periods. The Gumbel extreme value distribution method was used to extract the temporal probability shown the coherence between the rainfall probability occurrences and its return period. This method only considers the maximum value of daily rainfall in a year. The rainfall data used came from Wadas Lintang station primarily because this station was taken place in the centre of research area. The empty data of this station had been completed from surrounding stations. Calculating process which uses Gumbel equation produces the rainfall return period in the Table 5-3 below
42 Table 5-3 Return Period (TR) of Annual Rainfall in Wadas Lintang Hydrology Data Sorted Rank Left Prob Right Prob TR y Years Max Daily (mm) 1-LP 1/(RP) -ln(-ln(LP)) 1980 110 85 1 0.03 0.97 1.03 -1.22 1981 147 85 2 0.07 0.93 1.07 -1.00 1982 144 90 3 0.10 0.90 1.11 -0.83 1983 102 97 4 0.13 0.87 1.15 -0.70 1984 140 102 5 0.17 0.83 1.20 -0.58 1985 106 105 6 0.20 0.80 1.25 -0.48 1986 220 106 7 0.23 0.77 1.30 -0.38 1987 125 110 8 0.27 0.73 1.36 -0.28 1988 97 123 9 0.30 0.70 1.43 -0.19 1989 136 125 10 0.33 0.67 1.50 -0.09 1990 170 126 11 0.37 0.63 1.58 0.00 1991 142 128 12 0.40 0.60 1.67 0.09 1992 128 128 13 0.43 0.57 1.76 0.18 1993 151 129 14 0.47 0.53 1.88 0.27 1994 128 134 15 0.50 0.50 2.00 0.37 1995 150 136 16 0.53 0.47 2.14 0.46 1996 123 140 17 0.57 0.43 2.31 0.57 1997 158 142 18 0.60 0.40 2.50 0.67 1998 167 144 19 0.63 0.37 2.73 0.78 1999 129 147 20 0.67 0.33 3.00 0.90 2000 178 150 21 0.70 0.30 3.33 1.03 2001 134 151 22 0.73 0.27 3.75 1.17 2002 162 158 23 0.77 0.23 4.29 1.33 2003 210 162 24 0.80 0.20 5.00 1.50 2004 126 167 25 0.83 0.17 6.00 1.70 2005 90 170 26 0.87 0.13 7.50 1.94 2006 85 178 27 0.90 0.10 10.00 2.25 2007 105 210 28 0.93 0.07 15.00 2.67 2008 85 220 29 0.97 0.03 30.00 3.38 Source: Rainfall data analysis
Described in the Table 5-3 above, the highest amount of daily rainfall per each year was 220 mm occurred in 1986 and the lowest ones were 85 mm occurred both in 2006 and in 2008. The maximum rainfall in the year 2007 was 105 mm, and it was assumed that the rainfall value induced landslides in year 2007. The return period of this rainfall was 1.25 years, and it means that in the next 1.25 year from year 2007 the same peak will happen and induce the landslide occurrences.
Gumbel method also can be used to estimate the rainfall data in the future. From those we can build an estimating equation of rainfall data as painted in the Figure 5-16 below.
43 4.00
y = 0.0334x - 4.0127 3.00
2.00
1.00
0.00 Y = -LN(-LN(LP)) 0 50 100 150 200 250 Rainfall (mm) -1.00
-2.00 Figure 5-16 Relationship between Left Probability and amount of Rainfall
Related to landslide events, It can be assumed that rainfall in 2003 (210mm) triggered the largest area of landslides in 2003 (73.58 Ha). So, return period of rainfall can be used as the temporal probability of landslides .
44 CHAPTER 6. LANDSLIDE SUSCEPTIBILITY ASSESSMENT The main goal of this research is to build landslide susceptibility map by means of statistical analysis and pair-wise comparison method. The first step that should be done was selecting the landslide influencing factors which were put in landslide susceptibility assessment. Based on field observation, there were two types of mass movement in research area, so the landslide susceptibility map was decided to separate, but the author also builds the susceptibility map by merging all of landslide types. Three analyses had been done to compare the final results which were described in this chapter.
6.1 Selecting the landslide influencing factors of research area The estimating of influencing factors was also conducted by using sensitivity analysis. Each factor was eliminated in the calculating landslide susceptibility index and furthermore verification is done by means of success rate. The sensitivity analysis had been done for both rotational slides and translational slides. After the process, the result of sensitivity analysis for rotational slides is presented in Figure 6-1.
Includes all factors Without river Without litology Without lineament Without landuse
Without slope
Without road
Figure 6-1 Success rate of rotational susceptibility index based on sensitivity analysis
It is quite difficult to see the differences of the sensitivity analysis result. The solution of this problem is by calculating the area under the curve. The area under the curve which includes all factors is equal to 92.89%. By eliminating river, the area under the curve is 92.05%, followed by curve without road factor; 91.49%, without lineament 91.24%, without slope 90.89%, without lithology 90.59% and area of curve without landuse 80.06%. From those calculations, landuse factor becomes the most influence factor and the lowest one is distance to river. The influences of each factor also had been determined by assessing the range of weight value per each factor. Weight values were extracted by calculating the landslide densities of each class parameter and the range value is the summary of absolute minimum and maximum values of weight values. Table 6-1 illustrates the weight values of each class parameter for rotational slides.
45 Table 6-1 The weight values of each class parameter induced rotational slides Weighting Values (Rotational Type) Theme Class Denclass Densmap Weight Theme Class Denclass Densmap Weight Landuse River Baren Land 0.121601 0.008226 2.6934 0-20 0.004508 0.00823 -0.6019 Field Crop 0.015347 0.008226 0.6236 20-40 0.004899 0.00823 -0.5188 Forest 0.000127 0.008226 -4.1709 40-60 0.005569 0.00823 -0.3906 Mixed Garden 0.002851 0.008226 -1.0596 60-80 0.006604 0.00823 -0.2201 Paddy Field 0.000126 0.008226 -4.1788 80-100 0.007318 0.00823 -0.1174 River 0.000001 0.008226 -9.0151 >100 0.010453 0.00823 0.2391 Settlement 0.001166 0.008226 -1.9537 Range 0.8410 Shifting Cultivation 0.031265 0.008226 1.3352 Road Shrub 0.067538 0.008226 2.1054 0-20 0.009129 0.008226 0.1042 Range 11.7085 20-40 0.008621 0.008226 0.0469 40-60 0.008644 0.008226 0.0496 Lineament 60-80 0.008441 0.008226 0.0258 0-20 0.004184 0.00823 -0.6765 80-100 0.008476 0.008226 0.0299 20-40 0.009027 0.00823 0.0924 >100 0.008008 0.008226 -0.0269 40-60 0.016993 0.00823 0.725 Range 0.1311 60-80 0.02427 0.00823 1.0815 Slope 80-100 0.030143 0.00823 1.2982 0-3 0.001029 0.008228 -2.079 >100 0.00774 0.00823 -0.0614 3- 8 0.003097 0.008228 -0.9771 Range 1.9747 8 -15 0.005548 0.008228 -0.3941 Litology 15-30 0.010271 0.007383 0.3301 Km 0.000001 0.008226 -9.0151 30-45 0.02076 0.008228 0.9255 Tmp 0.004061 0.008226 -0.7059 45-65 0.032591 0.008228 1.3765 Tmpb 0.009249 0.008226 0.1172 >65 0.033693 0.008228 1.4097 Tmw 0.020056 0.008226 0.8912 Range 3.4887 Tmwt 0.000001 0.008226 -9.0151 Tompt 0.000001 0.008226 -9.0151 Tpp 0.005543 0.008226 -0.3948
Range 9.9063
Based on both Table 6-1 and Figure 6-1, the most influence factor is landuse, followed by lithology, slope, lineament and river. The road factor has the lowest value of significant factors. However since there were many researches which emphasized the influence of road network, the road distance is still used in building susceptibility map. Many landslide occurrence located near the road indicate the influence of this factor to landslide occurrences. Figure 6-2 shows the landslide occurred near to the road in research area.
Figure 6-2 Landslide occurrences triggered by road expanded
46 The sensitivity analysis was also executed for translational slides, and the results were presented in table and Figure 6-3.
Includes all factors Without river Without litology Without lineament Without landuse Without slope
Without road
Figure 6-3 Success rate of translational slides susceptibility index based on sensitivity analysis Including all factors produces area under the curve which is equal to 97.07% of total area. The area under the curve by eliminating river is equal to 96.64%, followed by curve without lineament ; 96.47%, without road factor; 96.41%, without slope 95.89%, without lithology 94.44% and area of curve without landuse 93.85%. From those calculations, landuse factor becomes the most influence factor and the lowest one is distance to river. Similar with the process applied in rotational slides, the influences of each factor could be seen in weight value table. Table 6-2 presents the weight values of each class parameter for translational slides. Table 6-2 The weight values of each class parameter induced translational slides Weighting Values (Translational slides) Theme Class Denclass Densmap Weight Theme Class Denclass Densmap Weight Landuse River Baren Land 0.000001 0.000309 -5.7333 0-20 0.000088 0.000309 -1.256 Field Crop 0.000001 0.000309 -5.7333 20-40 0.000275 0.000309 -0.1166 Forest 0.000001 0.000309 -5.7333 40-60 0.000305 0.000309 -0.013 Mixed Garden 0.000029 0.000309 -2.3660 60-80 0.000256 0.000309 -0.1882 Paddy Field 0.000121 0.000309 -0.9376 80-100 0.000189 0.000309 -0.4916 River 0.000001 0.000309 -5.7333 >100 0.000391 0.000309 0.2354 Settlement 0.001832 0.000309 1.7798 Range 1.4914 Shifting Cultivation 0.000001 0.000309 -5.7333 Road Shrub 0.000001 0.000309 -5.7333 0-20 0.000728 0.000309 0.857 20-40 0.000973 0.000309 1.147 Range 7.5131 40-60 0.001012 0.000309 1.1863 Lineament 60-80 0.000761 0.000277 1.0106 0-20 0.00251 0.000309 2.0947 80-100 0.000464 0.000309 0.4065 20-40 0.001675 0.000309 1.6902 >100 0.000078 0.000309 -1.3766 40-60 0.001301 0.000309 1.4375 Range 2.5629 60-80 0.000691 0.000309 0.8048 Slope 80-100 0.000809 0.000309 0.9625 0-3 0.000543 0.000309 0.5638 >100 0.000249 0.000309 -0.2159 3-8 0.000477 0.000309 0.4342 Range 2.3106 8-15 0.000392 0.000309 0.2379 Litology 15-30 0.000193 0.000277 -0.3613 Km 0.000001 0.000309 -5.7333 30-45 0.000001 0.000309 -5.7333 Tmp 0.000486 0.000309 0.4529 45-65 0.000001 0.000309 -5.7333 Tmpb 0.002043 0.000309 1.8888 >65 0.000001 0.000309 -5.7333 Tmw 0.000001 0.000309 -5.7333 Range 6.2971 Tmwt 0.000001 0.000309 -5.7333 Tompt 0.000318 0.000309 0.0287 Tpp 0.000001 0.000309 -5.7333 Wb Range 7.6221
47 Based on Table 6-2 and Figure 6-2, the most influencing triggering factors of translational slides are landuse, litology and slope, followed by lineament, road and river. All of the factors show the significant influences causing landslides. Although river factor has the lowest influence, it was still kept as the parameter. Some evidences showing the landslide occurrences in the river bank are presented in Figure 6-4. Hence, causal factors elimination is not necessary in this case.
Stream
Figure 6-4 Landslide occurrences triggered by river bank erosion in toe part
6.2 Bivariate statistical analysis. Landslide Susceptibility index map was built by summarizing the weight values in the same land unit by using the formulation as follow: The Susceptibility Index = Total Weight Value of all factors used. The description for each type of landslide is presented in sub section below.
6.1.1 Bivariate statistical analysis for rotational slides According to table 6.1, some analyses had been extracted. The results indicate some landuse types such as barren land, field crop, shifting cultivation and shrub have positive correlation with the rotational slide occurrences. Reservoir and river classes have the lowest weighting value, and it is acceptable because there was no landslide in that classes. In addition, the surrounding areas of settlement also have a positive relation. In distance to lineament parameter, we see that all of class parameters had a positive correlation with rotational landslides except 0 – 20 m and > 100 m classes. Theoretically, the nearer distance to lineament has the higher susceptibility of the area. The other parameter, lithology, indicates that only Halang Formation (Tmpb) and Waturanda Formation (Tmw) had a positive coherence with rotational slides. It means that the material deposits of both formation have induced landslide in the research area. Halang Formation consists of Brescia and Waturanda Formation consists of grained sandstone and brescia. From the distance to river parameter, only more than 100 m class has the positive correlation. Actually, in some cases, the cliff of river is susceptible for rotational slides. The acceptable result of distance to road parameter can be seen in table above, which all of class parameters have a positive correlation except > 100 m class. Based on slope gradient, rotational slide has a positive correlation with some classes; 15 – 30%, 30-45%, 45-65% and > 65%. Theoretically, the areas which has slope > 65% relatively resistance to slides. In this research the areas with > 65% slope were affected by surrounding slope class areas. Landslide Susceptibility index map is presented in Figure 6-5.
48
Figure 6-5 Landslide Susceptibility index map of Rotational Slides by using BSA
6.1.2 Bivariate statistical analysis for translational slides Some analyses had been extracted according to the result shown in table 6.2. Based on landuse weight, only settlement areas had a positive correlation prone to translational slides. It happens because these mass movements were recognized in settlement areas according to local authority’s report. In distance to lineament factor, all of classes present the positive coherence with translational slides except class > 100 m. Based on litology parameter, the table shows that Penosogan (Tmp), Halang (Tmpb) and Totogan (Tompt) formation had a positive correlation with translational slides. From the distance to river weight, there are two classes with positive correlation; 40 – 60 m, and > 100 m classes. The acceptable result of distance to road parameter can be seen in table above, which all of class parameters have a positive correlation except > 100 m class. Translational slides occurred more dominant in the gentle slope, where classes 0 – 3 % and 8 – 15% become the most susceptible area. Furthermore, a landslide Susceptibility index map was generated as shown in figure 6-6.
49
Figure 6-6 Landslide Susceptibility index map of Translational slides by using BSA
6.1.3 Bivariate statistical analysis for mixed types In this scenario, the author didn’t separate the types of mass movement. The weighting values extracted from BSA are presented in the Table 6-3 below Table 6-3 The weight values of each class parameter induced mixed types of landslides Theme Class Denclass Densmap Weight Theme Class Denclass Densmap Weight Landuse River Baren Land 0.121601 0.008535 2.6566 0-20 0.004595 0.008539 -0.6197 Field Crop 0.015347 0.008535 0.5867 20-40 0.005175 0.008539 -0.5008 Forest 0.000127 0.008535 -4.2077 40-60 0.005874 0.008539 -0.3741 Mixed Garden 0.002879 0.008535 -1.0867 60-80 0.006860 0.008539 -0.2189 Paddy Field 0.000247 0.008535 -3.5425 80-100 0.007507 0.008539 -0.1288 River 0.000001 0.008535 -9.0519 >100 0.010845 0.008539 0.2391 Settlement 0.002999 0.008535 -1.0459 Range 0.8588 Shifting Cultivation 0.031265 0.008535 1.2983 Road Shrub 0.067538 0.008535 2.0685 0-20 0.009857 0.008535 0.1440 20-40 0.009594 0.008535 0.1170 Range 11.7085 40-60 0.009656 0.008535 0.1234 Lineament 60-80 0.009209 0.008535 0.0760 0-20 0.006694 0.008539 -0.2434 80-100 0.008941 0.008535 0.0465 20-40 0.010702 0.008539 0.2258 >100 0.008086 0.008535 -0.0540 40-60 0.018294 0.008539 0.7619 Range 0.1980 60-80 0.024961 0.008539 1.0727 80-100 0.030951 0.008539 1.2878 >100 0.007989 0.008539 -0.0666 Range 1.5312
50