Skip to content
Open access

Mapping the probability of landslide processes in the Sunzha river basin using GIS modelling and machine learning methods

Sep 2026 · Sustainable Development of Mountain Territories · 0 citations

Abstract

Introduction. Landslides, rockfalls and taluses are among the most frequent natural hazards of the foothill and mountain belt of the North Caucasus. No basin-wide landslide susceptibility assessment has so far been made for the Sunzha River basin. The aim of the work. To produce a landslide probability map for the basin and to measure how the introduction of the seismic factor changes the resulting spatial pattern. Methods and materials. The computation was carried out on the Google Earth Engine platform using nine morphometric, lithological, biogenic and anthropogenic predictors: slope, aspect, elevation, terrain ruggedness index, surface curvature, NDVI and NDWI, distance to roads and lithological class. The model was trained with a random forest algorithm on an inventory of documented landslide occurrences. Seismicity was introduced at a separate, second step – as a trigger coefficient ranging from 0.85 to 1.15 applied to the completed susceptibility map. Results. High and very high probability classes cover 4531 km2 (37.6 % of the classified area) in the baseline version and 4100 km2 (34.0 %) once seismicity is taken into account. The two maps agree over 90.0 % of the area with a Cohen kappa of 0.863; no pixel shifted by more than one class. Discussion. The correction raises the calculated probability in the highly seismic south of the basin and lowers it in the lowland north, leaving the overall elevation-and-aspect pattern untouched. Feeding A10 directly into the training set, by contrast, cuts the territory along the isolines of seismic zoning and destroys the monotonic link between seismicity and probability. Conclusions. A two-stage scheme is preferable for factors defined at a scale much coarser than that of the modelled process. The calculated maximum is confined not to the high mountains but to the low-mountain broadleaf-forest landscapes of the Black Mountains. Conclusions from the article. Accounting for seismicity shifts the class boundaries but does not rewrite the map: the spatial pattern is still governed by relief, lithology and the state of the vegetation cover. Recommendations for practical application and further research. The map is suitable for spatial planning in mountain districts, for the design of linear structures and for justifying a geodynamic monitoring network. Further work involves extension of the landslide inventory and substantiation of the trigger coefficient limits from regional data on seismogenic displacements.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.