TERRAIN ANALYSIS FOR LANDSLIDE HAZARD ZONATION IN TEHRI RESERVOIR RIM REGION USING GIS

Abstract

The advent of frequent natural disasters necessitates a sustained management and mitigation strategy. Landslides are the most frequent disaster in the Uttarakhand Himalaya. Construction of major infrastructures such as dams, tunnels, roads and industries complicates the impact of disaster and mitigation measures. The present study area is situated around a huge reservoir (67 km), which was developed because of the construction of the Tehri dam. Landslide hazard zonation mapping around the reservoir rim area is a prerequisite for the health of the reservoir and settlements situated in the surrounding region of the reservoir. Landslide hazard zonation mapping is practiced to facilitate the planners for mitigation strategies in the wake up of any landslide related disaster. To carry out landslide hazard zonation mapping, a number of causal parameters belonging to the geo-environment are assumed. In the present research, thirteen terrain factors namely, lithology, soil cover, land use/land cover, photo-lineament, slope, relative relief, aspect, profile curvature, topographic wetness index, stream power index, drainage buffer, road buffer and reservoir buffer are considered. Along with the terrain factors, landslide inventory map is also prepared on the basis of remote sensing data, field observations past landslide information. Using remote sensing data, important terrain factors such as land use/land cover, drainage, photo-lineaments, slope, aspect, relative relief, profile curvature, topographic wetness index and stream power index were derived. Remote sensing imageries of varying spatial, spectral and temporal resolution were also used to generate credible data of the terrain factors. Digital elevation models of varying spatial resolution were used to extract primary and secondary topographic parameters. Data management was done in the GIS platform. Several digital image processing techniques such as topographic correction, NDVI, supervised classification, band ratioing and edge detection were extensively used in the process of terrain factor extraction. Visual interpretation based on colour, tone, texture, shape, size, pattern and shadow were also performed on the remote sensing multispectral data for the delineation of important causative factors. A comprehensive landslide inventory for the study area was generated using combination of the remote sensing data, field data and historical information about the landslides. Total 195 landslide incidences of dimension varying from 25 m2 to 3000 m2 ii were covered in the point vector format. Majority of the landslide incidences were found to be belonging to rotational and talus slope failure categories. Few landslides were found to be of plane failure categories observed within the region. An attempt was made to analyse the feasibility of the causal factors considered in the present study. Landslide frequency ratio analysis and weights of evidence analysis were performed to determine the relationship between landslide incidences and the terrain factor classes. Landslide frequency ratio values were used to identify the association between the landslide incidences and terrain factors/classes. Contrast between positive weights and negative weights derived from the weights of evidence analysis was used to determine the relationship between terrain factor classes and landslide incidences. A number of methods are available for the delineation of landslide hazard zones. Here, it is made very clear that the term 'landslide hazard zonation' was adopted according to the guidelines of the Bureau of Indian Standards (1998). These guidelines do not consider temporal factors such as rainfall, seismicity, temperature variation for landslide hazard zonation mapping. So this term best resembles with the term 'landslide susceptibility zonation' mapping. In the Uttarakhand Himalaya, several methods of landslide hazard zonation mapping belonging to heuristic, semi-quantitative and quantitative approaches have been extensively applied. But a suitable method in accordance with ground physical conditions needs a comparative study of several important landslide hazard zonation methods. Accordingly, seven different landslide hazard zonation methods were used to compute five relative landslide hazard zones namely, very low hazard, low hazard, moderate hazard, high hazard and very high hazard and validated on the basis of cumulative percentage curve/cumulative frequency diagram technique along with a bar diagram technique showing frequency of landslides in each identified hazard zone. A comparison between them was carried out to find the suitable method for the Tehri reservoir rim region. Among the heuristic methods, a GIS based weighted overlay method and a modified BIS (LHEF) method was used. In case of GIS based weighted overlay method, weights/ratings of the factors/classes were awarded by considering landslide density in the factor classes. Landslide frequency ratio assumes landslide densities in the factor classes and hence was considered in awarding the ratings of the factor classes. A methodology was evolved to subjectively scale the landslide frequency ratio value and award the ratings to the factor classes. Arithmetic weighted overlay of factors/classes was performed to iii generate landslide hazard index map, which was further classified into five relative hazard zones with an accuracy of 74 %. In case of the modified BIS approach, six inherent factor along with the external factors, seismicity and rainfall were rated according to the guidelines of BIS. Slope facets were prepared on the basis of the digital elevation model. Density of the photo-lineaments in each facet was incorporated at the place of structural discontinuity. Total estimated hazard was calculated for each slope facet and the whole area (covering 126 slope facets) was classified into five landslide hazard zones using BIS guidelines. A methodology was evolved to validate this method using cumulative percentage curve which resulted in an accuracy of 62 %. Two semi - quantitative methods, namely, combined fuzzy logic and frequency ratio method and AHP method were used for the landslide hazard zonation mapping. In the first method, fuzzy membership was derived by incorporating normalized landslide frequency ratio value and fuzzy integration was performed by applying fuzzy OR operator and fuzzy gamma operator. Six different fuzzy gamma values were used to compute six landslide hazard index maps. Among the fuzzy OR integration and fuzzy gamma integration, fuzzy gamma (gamma = 0.95) was found to be most suitable for the landslide hazard zonation mapping with an accuracy of 78.2 %. A comprehensive analytical hierarchy process resulted in delineation of landslide hazard zones with an accuracy of 80%. Three different landslide inventory driven methods namely, landslide frequency ratio, weights of evidence and logistic regression method was used to delineate landslide hazard zones. In the case of landslide frequency ratio method, the normalized frequency ratio values (0-1) of the terrain factors were used for the ratings of the factors. Rated factor classes were integrated in the GIS domain using fuzzy SUM overlay. Fuzzy SUM overlay can be used not only with the fuzzy membership values but also in the case of raster data having a range of 0 to 1. This method successfully resulted in delineation of landslide hazard zones with a prediction accuracy of 72%. In the case of weights of evidence method, a landslide posterior probability map was generated using weight positive (W+), weight negative (W-) and contrast (C) values. A total of 134 landside incidences were used for the calculation of W+, W-, C and other parameters. The posterior probability map was classified into five landslide hazard zones and it gave a prediction accuracy of 82%. In the case of binary logistic regression method, correlation between factor classes and landslides were computed using binary logistic regression method and a probability iv estimate of landslide occurrence on cell by cell basis for entire study area was obtained. Probability map was further classified into five landslide hazard zones using statistical class break technique. Accuracy assessment of the model was performed using cumulative frequency diagram technique along with ROC curve technique which in turn gave accuracies of 83.5% and 82.65%. A comparison between the seven different landslide hazard zonation methods used in this study was performed on the basis of landslide density method and cumulative percentage curve method. Results of the landslide density method have indicated the consistency of the landslide hazard map produced from different models. Comparison of the prediction rate curves have reflected the different accuracy estimates calculated from the LHZ mapping method. Least accuracy was achieved in the case of heuristic models, where as peak accuracy was achieved from the multivariate logistic regression mapping method. On the basis of comparisons between landslide hazard maps computed from different methods, it was observed that quantitative methods such as weights of evidence and binary logistic regression method are most suitable techniques for the study area. Heuristic methods also have good prediction capability for landslides and can be utilized in the study area. AHP and fuzzy logic approaches have resulted in better prediction accuracy than the heuristic methods. On the basis of this analysis, the multivariate quantitative method -binary logistic regression-has been found to be the most suitable method for landslide hazard zonation mapping in the Tehri reservoir rim region.