Effect of scale and mapping unit on landslide susceptibility mapping of Mandakini River Basin, Uttarakhand, India

Most of the research studies on landslide susceptibility mapping (LSM) in the Himalayan region consider small subsets of several geomorphological features such as river basins, mountainous areas except for only a few studies that have prepared susceptibility mapping for a whole basin. The selection of study areas in most studies is random and depends on a variety of parameters such as the number of landslide occurrences, availability of preparatory causal factors or conditioning factors, availability of computational resources. However, the selected study area and mapping unit may not represent landslide occurrences in the whole basin or landslide susceptibility of the region. This work analyses and problematizes the effect of scale (smaller to the larger, i.e., basin-wise to micro-watershed) in preparation of LSM based on various landslide conditioning factors. The study has been carried out in the Mandakini river basin of Garhwal Himalaya, Uttarakhand. To evaluate the effect of geographical and computational scale, we have divided the river basin into ten sub-basins and further into 88 micro-watersheds. The study considers an inventory of 229 landslides (consisting of approximately 6000 pixels) and seven predisposing factors of landslide occurrence, i.e., aspect, elevation, geology, slope, soil, the buffer of streams, and buffer of thrusts. To compare the effect of scale, we have used two widely known techniques, namely Fisher discriminant analysis and logistic regression. The results of this study are validated on more than 230 landslide locations provided by the Geological Survey of India using the Heidke skill score, which is preferred for the assessment of rare events in predictive analysis. We conclude that logistic regression performs better for the whole basin and sub-basins. However, discriminant analysis performs better for micro-watersheds. Furthermore, based on the results we can conclude that if LSM is required at a sub-basin or micro-watershed scale, all categorical data must be prepared at a finer scale.

Automated summary

Most of the research studies on landslide susceptibility mapping in the Himalayan region focus on small subsets of geomorphological features and the selected study area may not accurately represent landslide occurrences in the whole basin. This study analyzes the effect of scale in landslide susceptibility mapping and compares two widely known techniques. The results show that logistic regression performs better for larger scales, while discriminant analysis performs better for smaller scales.