Results Helmholtz Institute
Description of the working task
Tectonics, together with climate and erosion forms the topographic gradients of the Tien Shan, highly susceptible to gravitational mass movements. As a consequence, it is important to determine the dimensions and frequency of recent processes but also to consider the longer-term tectonic and climatic history. To do so we selected geomorphic indices as local relief, surface roughness, hypsometric integral, surface index and topographic position index, whose are previously used to identify recent tectonic activity, erosion rates, landslides and geoforms, and integrate them in a landslide susceptibility model.
A series of refined, integrated or new datasets are created to assess the landslide susceptibility in the Tajik-Tien Shan and NW Pamir. The datasets are subdivided in thematic groups: geology, hydro-climatic, land cover and morphological.
|Historical landslide catalogue for Tien Shan and NW Pamir
|A total 859 polygons representing landslides. The catalogue is based on the Tien Shan Geohazard Database (Havenith et al., 2015a), manual delimitation of landslides from Google Earth imagery and field work.
|Lithological units grouped by age and rock type. Source : Kufner et al., 2018., Federal State Budgetary Institution A.P. Karpinsky Russian Geological Research Institute.
|Fault traces in the area of study. It include active faults identify by CAFD and faults from FGUP VSEGEI Source: Central Asia Fault Database (CAFD) https://esdynamics.geo.uni-tuebingen.de/faults/about.php, Federal State Budgetary Institution A.P. Karpinsky Russian Geological Research Institute (FGUP VSEGEI), 2018)
|Topographic wetness index (TWI)
|Describes the potential saturation of a given site in the landscape as a function of the upslope area and the local slope. Higher TWI index values highlight flat locations with large upslope areas, which are expected to have relatively higher water availability while lower index values correspond to steep locations, which are expected to be better drained. Source: 1 arcsec SRTM data
|Distance from channel
|Identify areas of the landscape located at the same distance with respect to the drainage network, which defines the regional base-level. Source: Flow paths and river network extracted from a 1 arcsec SRTM data
|Elevation above channel
|Identify areas of the landscape located at the same elevation with respect to the drainage network, which defines the regional base-level. Source: Flow paths and river network extracted from a 1 arcsec SRTM data
|Mean precipitation for the years 1979-2013. Source: Karger, et al., 2017 https://chelsa-climate.org/
|Mean isothermality for the years 1979-2013. Source: Karger, et al., 2017.https://chelsa-climate.org/
|Normalized difference vegetation index aim to quantify vegetation by measuring the difference between the spectral response of materials on the near-infrared (which vegetation strongly reflects) and red light (which vegetation absorbs). Vegetation content is proportional to the NDVI values with a maximum of 1 for a dense vegetated area. Source: Sentinel 2 imagery taken in the months of July and August 2017.
|The terrain slope represent the maximum change in elevation over the distance between the cell and its eight neighbors. Source: 1 arcsec SRTM data
|Downslope orientation of the maximum rate of change in value from each cell to its neighbors. Source: 1 arcsec SRTM data
|Represents the difference in elevation between the highest and lowest points in a given window of observation. Source: 1 arcsec SRTM data
|The area ratio approach to calculate the surface roughness evaluates the similarities between a topographic surface within a given are and a flat surface with the same geographical extend. The ratio is close to 1 for flat areas and increases rapidly as the topographic surface becomes irregular Source: 1 arcsec SRTM data
|The hypsometric integral also known as elevation relief ratio, shows the distribution of landmass volume with respect to a basal reference plane. It is commonly used to identify the evolutionary stage of a landscape where young stages are represented by a HI > 0.6 while old stages has a HI < 0.35. Landscape with HI values between 0.35 and 0.6 are consider mature. Source: 1 arcsec SRTM data
|Surface index (SI)
|The surface index combines the elevations from the DEM with the computed maps of hypsometric integral and surface roughness. This index allows to discriminate elevated areas with low relief amplitude from areas with a more rugged topography. Positive surface index values correspond to elevated surfaces with low relief amplitude while negative values highlight rugged landscapes. Source: 1 arcsec SRTM data elevation close to the regional base level.
|Topographic position index
|Represent the difference between the elevation of a given pixel and the average elevation of its neighboring pixels within a given window of observation. Positive and negative values will highlight ridges and valleys, respectively. Source: 1 arcsec SRTM data
RESULTS: Landslide susceptibility assessment
The landslide susceptibility is assessed by the workflow presented in the Figure. The implementation of Random Forest algorithm uses the landslides scarps as training data and the datasets described above as predicting variables. The result is the probability of having a landslide at each analyzed pixel (30m x 30m) within the area of study. To evaluate the landslide susceptibility map the receiver operator curve (ROC) is used. An area under de receiver operator curve (AUC ROC) close to 1 indicates an accurate model while an AUC ROC lower than 0.5 is consider random. Uncertainties are estimated by the standard deviation from the mean landslide susceptibility for 50 iterations. High standard deviation values highlights areas with less reliable results.
The landslide susceptibility map is highly evaluated with a mean area under the receiver operator curve (AUC ROC) of 0.9 (see Figure). High landslide susceptibility areas (>80%) are located along the Zerashon river where multiple events has been reported in the past. Another interesting location is the area north to the Iskanderkul lake where landslides are mapped within a broad area distribution, ranging from mega landslides as the one that started the lake formation, until few meters rock slides. The valleys located north-west Dushanbe are particularly interesting to performed detailed studies since small size landslides are reported with strong affectation to the capital city and the surroundings. The calculation of uncertainties are represented by the spatial distribution of standard deviation (sd). It points out locations that needs to be interpreted more carefully. The landslide susceptibility and uncertainties maps are guidelines for decision and policy makers. The results supports the selection of local sites to implement monitoring systems, develop preventing planing or execute detailed studies.