Prediction of Slope Failures Using Bivariate Statistical Based Index of Entropy Model Omar F. Althuwaynee1, Biswajeet Pradhan*1, Ahmad Rodzi Mahmud 1, Zainuddin Md Yusoff 1 1
Faculty of Engineering, Department of Civil Engineering, Geospatial Information Science Research Centre (GISRC), University Putra Malaysia, Serdang, Selangor Darul Ehsan 43400, Malaysia. *Email.
[email protected] or
[email protected] (Corresponding author) Tel. +603-8946 8466; Fax. +603-8946 8470
Abstract The main objective of this research is to evaluate the spatial prediction of potential slope failures in Kuala Lumpur and surrounding areas using an index of entropy based statistical model. Based on potential information of entropy method (IoE), subjective weights were calculated for fourteen landslide conditioning factors used in this study such as, (slope, aspect, curvature, altitude, surface roughness, lithology, distance from faults, NDVI (normalized difference vegetation index), land cover, distance from drainage, distance from road, SPI (stream power index), soil type and precipitation). A landslide inventory map of the study area was produced using previous reports and aerial photographs interpretation aided with extensive field survey and total of 220 main scarps were identified. Out of this, 153 (70%) landslide locations were used to build the IoE model, while remaining 66 (30%) landslide locations were used for validation purpose. For validation, the area under the curve (AUC) was used to quantify the predictive performance of the employed IoE model. The validation results show that the prediction accuracy of the model is 0.80 (80%) and the success rate equals to 0.81 (81%) that consider fine indicator of the reliability of bivariate model based IoE model employed in this study.
STUDY AREA Kuala Lumpur city and its vicinity areas is located in the 0 Selangor state, occupying the geographic zone of (2 0 0
0
2
covering an area of 1975 km (Fig. 1). High intensity of rainfall was considered as the main triggering factor for the initiation of landslides with precipitation amounting to 58 to 240 (mm/month). Majority of study area is covered by vegetation cover like swamp, forest, scrub and grassland [14, 15].
Keywords- Landslides, Kuala Lumpur, Bivariate model, Index of Entropy, Geographic Information Systems (GIS), Remote Sensing
INTRODUCTION In recent years, the assessment of landslide hazard and risk has become a topic of major interest for both geoscientists and engineering professionals, as well as for the community and the local administrations in many parts of the world, to predict zones prone to hazardous of slope mass movement and implementation for hazard mitigation [1-6]. In Malaysia the most catastrophic landslides or slope failures have occurred during following years (1993, 1999, 2002, 2003, 2006, 2009, and 2011), majority from June to December, causing a hundreds of fatalities and economic losses[7]. This paper evaluates landslide susceptibility assessment in Kuala Lumpur metropolitan city and its surrounding areas by using statistical bivariate based index of entropy (IoE) model [6, 8-13].
Fully supported by the UPM-RUGS project grant numbers RUGS/05-021871RU and 9344100.
Figure 1. Study area location map
DATA Remote sensing methods were used to obtain historical records of the landslides. Archived 1:5000 1:50,000 aerial photographs, SPOT 5 panchromatic satellite images were used for the visual detection of landslide scars. Also landslide reports over the past 25 years were used in the inventory mapping. A total of 220 landslides were mapped and used in this study. The main usage of landslide inventory map are: delineating the landslide scarp areas, landslide risk assessment , provide information about landslide types and its relation with the morphological and geological characteristics, and monitor the development of landscapes dominated by mass-wasting processes [16] 1. Topographic map: A topographic map with scale of 1:25000 was used in order to construct a digital elevation model from which various landslide conditioning factors were extracted. Altitude, slope angle, slope aspect, roughness and curvature, were calculated from the DEM by ArcGIS v.9.3 software. SPI (Stream power index) was computed from slope and catchment area, distance from road, and distance from drainage. 2. Geological map: A geological map at scale 1:63000 was used to produce the digital lithological map and distance from faults was also calculated using Euclidean distance method.
science experts opinion[20]. The following seven equations represent the various steps in the calculation of index of entropy model. Wi represents the weight of each parameters in the final susceptibility map. See (1) and (2).
Pc: area of the category after primary reclassification Psd: area of landslides within the given category. (Pij): probability density. H j and H (j max) are the entropy values (3) and (4) and they are written as;
Sj: number of classes. H j and H j max: Entropy values. Ij: Information coefficient.
3. Soil map: A soil map at the scale 1:100,000 was used to extract the soil properties of the study area. 4. Landsat Thematic Mapper (TM): Alandsat TM satellite image with 30x30 m pixel resolution was used to obtain the landcover and NDVI (Normalize difference vegetation index) maps. Finally a precipitation map was prepared using the past 29 years (1981 2010) of historical rainfall data.
Wj : represents the resultant weight value for the parameter. The final landslide susceptibility map was prepared by summation of weighted multiplications of the secondarily reclassified parametric maps, see (7).
INDEX OF ENTROPY MODEL A quantitative approach of landslide assessment using statistical approach compares the spatial distribution of landslides with respect to the landslide conditioning factors that are being measured [17]. Bivariate statistical analysis performs comparison of each landslide conditioning factor with the landslide inventory map individually and then assign weight on the basis of landslide density in each individual class [8, 18]. The method proposed by [8] and adopted by many authors [18, 19]. The weight of parameters are defined by entropy level that indicates the extent of disorder in the environment, then the weight will determine which factor is more significant in slope failure mechanism. Two main drawbacks of bivariate statistical analysis (BSA) are: first it assumes that all landslide inventory happened under same condition and carry the same geological
y : value of landslide susceptibility in the final map; where factor_recl2 is the value in a particular pixel after reclassifying the parametric map of the conditioning factors. Finally the susceptibility map intervals were divided into five classes, using quantile based classification approach.[11, 12, 21] RESULTS AND DISCUSSIONS The weights of entropy for each landslide conditioning factor were calculated based on the aforementioned equations. For an example, the tabulated values of the slope angle are shown in Table 1. In Table 1, values of (Pij) (probability density) of slope class 16-25, confirmed with the result obtained by other authors [14, 15] for the same study area. In the case of curvature map, concave and convex types show higher ratio, due to concave surface, the rain water is accumulated causing
infiltration and shear failure [14, 22]. In the case of altitude, the class interval of 64-126 m was found as high susceptible representing the second major percentage (21%) in the study area. Surface roughness value between 1.5 and 2.5 refers to higher weight of susceptibility. In the case of precipitation map, it showed higher susceptibility value between 2680 to 2765 mm/year. For ndvi, weight between the class interval 0.2 and 0.6 showed a high susceptibility. SPI represent the erosive power of flowing water with higher weight value between 0.7 to 1. For distance from road, the weight between 51 m and 120 m is considered to be the most effective class because of the ll phenomena. In the case of distance from drainage the result indicate that class interval between 21 to 50 m is high susceptible [15, 22]. In the case of distance from faults, the results indicate that lower weight for the classes far from the fault lines. For the soil types, the higher weight is seen in the case of Serdang_ Munchong_ Seremba_ Association.
This may be due to the sedimentation process which occur in the fine to coarse quartz sand set in a clay matrix [23]. This result confirm the findings of [14]. Similarly, in the case of lithology, phyllite, slate, shale and sandstone showed higher weights [14]. Fig 2 shows the significant contribution of each class for each landslide conditioning factor used in this study. Fig. 3 shows the landslide susceptibility zonation map for the study area. The validation of the resultant susceptibility maps were performed using success and prediction rate curves and area under the curve (AUC). The success rate and prediction rate graph indicates that the AUC values are 0.80 (80%) and 0.81 (81%) accuracy respectively for IoE model employed in this study,Fig.4.
Figure 2. Distribution of landslides within each conditioning factor according to the probability of density (pij). (a) Slope angle, (b) Slope aspect, (c) Curvature, (d) Altitude, (e) Surface roughness, (f) Precipitation, (g) NDVI, (h) Landcover, (i) SPI.
Figure 2 (continued). Distribution of landslides within each conditioning factor according to the probability of density (pij), (j) Distance from roads, (k) Distance from drains, (l) Distance from faults, (m) Soil type, (n) Lithology.
Table 1. Calculation result for slope angle weight
Spatial factor 4
Slope
class
no. of pixel
no. of slides
Pij
(Pij)
Hj
0 - 15 16 - 25 26 - 35 36 - 85
9912130 3894975 608274 25743
84 65 4 0
0.80 1.58 0.62 0.00
0.27 0.53 0.21 0.00
0.5 0.48 0.47 0
153
1.00
Hj max
2
Ij
0.27
Wj
Reclass value
0.2
3rd 4th 2nd 1st
Figure. 3. (a) Landslide susceptibility map, (b) Frequency ratio graph showing the areal coverage of each susceptibility zone in percentage.
Figure 4. The success rate and prediction rate curve of the landslide susceptibility map.
CONCLUSION This study was carried out in order to compare and verify the results presented in [14] showing the spatial relationship between Index of Entropy as a bivariate method and Evidential belief function as multivariate statistical approaches in terms of the ease, flexibility and costeffective in the application on landslide susceptibility mapping for Kuala Lumpur metropolitan city and surrounding areas, Malaysia. Success and Prediction rate of area under the curve (AUC) equal to 0.81 (81%) for model validation and prediction accuracy of 0.80 (80%), were
valued the quality of the work through robustness validation process. The results and findings obtained through this study can be used for land-use planning and slope management. However, the result is very much dependent upon the quality of the input data used in this study. We recommend for future work, to address the temporal resolution of recent slope treats by and mass movements, by effectively field prospect for effective risk mapping and management.
ACKNOWLEDGMENT This research is fully supported by the UPM-RUGS project grant numbers RUGS/05-02-1871RU and 9344100 and partially by 05-01-11-1283RU (vote number 9199892), 0503-11-1450RU (vote number 9301900) REFERENCES [1] B. Pradhan, et al., "Landslide Susceptibility Mapping by Neuro-Fuzzy Approach in a Landslide-Prone Area (Cameron Highlands, Malaysia)," Geoscience and Remote Sensing, IEEE Transactions on, vol. 48, pp. 4164-4177, 2010. [2] H. J. Oh and B. Pradhan, "Application of a neuro-fuzzy model to landslide susceptibility mapping for shallow landslides in a tropical hilly area," Computers & Geosciences, 2011. [3] B. Pradhan and S. Lee, "Landslide risk analysis using artificial neural network model focusing on different training sites," International Journal of Physical Sciences, vol. 4, pp. 1-15, 2009. [4] B. Pradhan and S. Pirasteh, "Comparison between prediction capabilities of neural network and fuzzy logic techniques for landslide susceptibility mapping," Disaster Adv, vol. 3, pp. 26-34, 2010. [5] B. Pradhan and A. M. Youssef, "Manifestation of remote sensing data and GIS on landslide hazard analysis using spatial-based statistical models," Arabian Journal of Geosciences, vol. 3, pp. 319-326, 2010. [6] B. Pradhan and S. Lee, "Delineation of landslide hazard areas on Penang Island, Malaysia, by using frequency ratio, logistic regression, and artificial neural network models," Environmental Earth Sciences, vol. 60, pp. 1037-1054, 2010. [7] B. Pradhan, "Application of an advanced fuzzy logic model for landslide susceptibility analysis," International Journal of Computational Intelligence Systems, vol. 3, pp. 370-381, 2010/09/01 2010. [8] W. P. Vlcko J, Rychlikova Z "Evaluation of regional slope stability. Mineralia Slovaca " vol. 12, pp. 275 283, 1980. [9] B. Pradhan and S. Lee, "Regional landslide susceptibility analysis using back-propagation neural network model at Cameron Highland, Malaysia," Landslides, vol. 7, pp. 13-30, 2010. [10] B. Pradhan and M. F. Buchroithner, "Comparison and validation of landslide susceptibility maps using an artificial neural network model for three test areas in Malaysia," Environmental & Engineering Geoscience, vol. 16, pp. 107-126, 2010. [11] B. Pradhan and S. Lee, "Landslide susceptibility assessment and factor effect analysis: backpropagation artificial neural networks and their comparison with frequency ratio and bivariate logistic regression modelling," Environmental Modelling and Software, vol. 25, pp. 747-759, 2010.
[12] B. Pradhan, et al., "Remote sensing and GIS-based landslide susceptibility analysis and its cross-validation in three test areas using a frequency ratio model," Photogrammetrie, Fernerkundung, Geoinformation, vol. 2010, pp. 17-32, 2010. [13] B. Pradhan, et al., "A GIS-based back-propagation neural network model and its cross-application and validation for landslide susceptibility analyses," Computers, Environment and Urban Systems, vol. 34, pp. 216-235, 2010. [14] O. F. Althuwaynee, et al., "Application of an evidential belief function model in landslide susceptibility mapping," Computers & Geosciences, vol. 44, pp. 120135, 2012. [15] S. Lee and B. Pradhan, "Landslide hazard mapping at Selangor, Malaysia using frequency ratio and logistic regression models," Landslides, vol. 4, pp. 33-41, 2007. [16] F. Guzzetti, et al., "Landslide inventory maps: New tools for an old problem," Earth-Science Reviews, vol. 112, pp. 42-66, 2012. [17] C. J. Van Westen, et al., "Landslide hazard and risk zonation why is it still so difficult?," Bulletin of Engineering Geology and the Environment, vol. 65, pp. 167-184, 2006. [18] M. Constantin, et al., "Landslide susceptibility assessment using the bivariate statistical analysis and the index of entropy in the Sibiciu Basin (Romania)," Environmental Earth Sciences, vol. 63, pp. 397-406, 2011. [19] M. Bednarik, et al., "Landslide susceptibility assessment of the Kralovanycase study," Physics and Chemistry of the Earth, Parts A/B/C, vol. 35, pp. 162-171, 2010. [20] H. Kojima, et al., "Strategy on the landslide type analysis based on the expert knowledge and the quantitative prediction model," International Archives of Photogrammetry and Remote Sensing, vol. 33, pp. 701-708, 2000. [21] D. Tien Bui, et al., "Spatial prediction of landslide hazards in Hoa Binh province (Vietnam): A comparative assessment of the efficacy of evidential belief functions and fuzzy logic models," Catena, vol. 96, pp. 28-40, 2012. [22] H.-J. Oh, et al., "Predictive landslide susceptibility mapping using spatial information in the Pechabun area of Thailand," Environmental Geology, vol. 57, pp. 641651, 2009. [23] A. Alansi, "The Effect of Development and Land Use Change on Rainfall-Runoff and Runoff-Sediment Relationships Under Humid Tropical Condition: Case Study of Bernam Watershed Malaysia," European Journal of Scientific Research, vol. 31, pp. 88-105, 2009.
