Landslide assessment in a remote mountain region
: a case study from the Toktogul region of Kyrgyzstan, central Asia

  • Namphon Khampilang

    Student thesis: Doctoral Thesis


    Slope instability is a significant natural hazard in the Tien Shan mountain range, some landslide studies were carried out in small areas in the Tien Shan Mountain but no landslide susceptibility mapping has been carried out for the region. This thesis describes the creation of a digital landslide inventory and the use of a Geographical Information System (GIS) to create the first landslide susceptibility models for the area.

    This research has resulted in the landslide inventory of the Toktogul region. This was accomplished through a combination of SPOT image interpretation with the validation of Google Earth image and field mapping. 2,776 landslides were mapped with the area of approximately 202 km2 in total. This area is about 1.6% of the total study area (12,280 km2). The landslide frequency-area distribution using inverse gamma distribution was carried out using all landslides.

    A quantitative landslide susceptibility assessment was conducted. Parameter maps including slope, elevation, aspect, lithology, land cover, distance from faults and distance from drainage, normalised difference vegetation index and stream power index were constructed and compiled into a database with the landslide inventory. The bivariate (frequency ratio and weight of evidence) and multivariate (logistic regression) statistical analysis were used to establish landslide susceptibility maps. The susceptibility maps were validated using landslide inventory. Success rate curve and cumulative area under the curve were created. All susceptibility maps resulted in very similar prediction rate curves and cumulative area under the curves. Logistic regression gave the best result in cumulative area followed by frequency ratio and weight of evidence (79.51%, 77.86%, and 78.21%, respectively). The combining of landslide susceptibility using logistic regression method was performed. The result shows a good predictive accuracy (78.72% cumulative area). Based on the validation, the susceptibility map derived using logistic regression was determined to be the best.
    Date of AwardJun 2015
    Original languageEnglish
    Awarding Institution
    • University of Portsmouth
    SupervisorMalcolm Whitworth (Supervisor) & Nick Koor (Supervisor)

    Cite this