Tien Shan Geohazards Database: Landslide susceptibility analysis

Abstract This paper is the second part of a new geohazards analysis applied to a large part of the Tien Shan, Central Asia, focused on landslide susceptibility computations that are based on recently compiled geographic, geological and geomorphological data. The core data are a digital elevation model, an updated earthquake catalogue, an active fault map as well as a new landslide inventory. The most recently added digital data are a new simplified geological map, an annual precipitation map, as well as river and road network maps that were produced for the Kyrgyz and Tajik parts of the Tien Shan. On the basis of these records we determine landslide densities with respect to morphological (M), geological (G), river distance (R), precipitation (P), earthquake (E) and fault (F) distance factors. Correlations were also established between scarp locations and the slope angle, distance to rivers, curvature. These correlations show that scarps tend to be located on steeper slopes, farther from rivers and on more convex terrain than the entire landslides. On the basis of the landslide density values computed for each class of the aforementioned factors, two landslide susceptibility maps are created according to the Landslide Factor analysis: the first one considers correlations between the landslide occurrences and the first four factors (MGRP); the second one is based on the first map (MGRP) combined with the seismo-tectonic influence (+ E + F) on landslide distributions. From the comparison of these two maps with actual landslide distributions we infer that the distances to rivers as well as to faults and past earthquakes most strongly constrain the susceptibility of slopes to landslides. We highlight several zones where the landslide susceptibilities computed for the MGRP + E + F factors fit better the observed concentration of landslides than those computed for the MGRP factors alone. For a few zones, both maps produce high landslide susceptibilities that do not well reflect the observed low sub-regional landslide activity; for some cases, we consider that some influencing factors must not have been well taken into consideration, for others we show that we simply had missed landslide detections. At the scale of the mountain range, the computed landslide susceptibility maps fit the observed landslide distributions relatively well, but these maps only represent the spatial component of landslide hazards. Temporal aspects are not considered by this analysis.