LANDSLIDE SUSCEPTIBILITY MODELLING UNDER ENVIRONMENTAL CHANGES: A CASE STUDY OF CAMERON HIGHLANDS, MALAYSIA

Modeling landslide susceptibility usually does not include multi temporal factors, e.g. rainfall, especially for medium scale. Landslide occurrences in Cameron Highlands, in particular, and in Peninsular Malaysia, in general, tend to increase during the peak times of monsoonal rainfall. Due to the lack of high spatial resolution of rainfall data, Normalized Different Vegetation Index (NDVI), soil wetness, and LST (Land Surface Temperature) were selected as replacement of multi temporal rainfall data. This research investigated their roles in landslide susceptibility modeling.
In doing so, four Landsat 7 Enhanced Multi Temporal Plus (ETM+) images acquired during two peak times of rainy and dry seasons were used to derive multi temporal NDVI, soil wetness, and LST. Topographic, geology, and soil maps were used to derive ‘static’ factors namely slope, slope aspect, curvature, elevation, road network, river/lake, lithology, soil geology lineament maps. Landslide map was used to derive weighting system based on spatial relationship between landslide occurrences and landslide factor using bivariate statistical method. A non-statistical weighting system was also used for comparison purpose. Different scenarios of data processing were applied to allow evaluation on the roles of multi temporal factors in landslide susceptibility modeling in terms of the accuracy of the landslide susceptibility maps (LSMs), the appropriate weighting system of the models, the applicability of the model, the ability to confirm the relation between landslide occurrences and rainfall.
The results show that the average accuracy of LSMs produced by the developed models with inclusion of multi temporal factors is 49.1% on the overall. Addition of LST tends to improve the accuracy of LSMs. NDVI can be a suitable replacement for rainfall data since it can explain the relation between landslides occurrences and rainfall cycle. Statistical-based weighting system produced more accurate LSMs than non-statistical-based one and is applicable for landslide susceptibility modeling elsewhere. Significant causative factors were proven to produce more accurate LSMs.