ИСТИНА

Systematizing the set of exogenic processes and their spatial distribution is a critical component of studying sediment budgets in small catchments. This is particularly important in mountainous regions, where significant elevation gradients and variations in slope steepness create high spatial heterogeneity of exogenic processes. Standard methods for mapping predominant exogenic processes include fieldworks, aerial surveys, and visual interpretation. However, manually generated results can be used as training data to automate the mapping of these processes. While the accuracy of this approach may be limited, automatically generated maps can provide a basis for subsequent refinement. An experiment to map predominant exogenic processes was conducted for two small catchments in the central sector of the Greater Caucasus Mountains: the Lake Donguz-Orun catchment and the Djankuat Stream catchment. Both are located on the northern slope of the Main Divide Range near the Russia-Georgia border. They are characterized by similar elevation ranges but differ in overall geomorphic characteristics. For the Lake Donguz-Orun catchment (A = 13 km²), a map of predominant exogenic processes was created (approach: each point corresponds only to the most active process). Key process groups include glacial erosion, fluvial erosion, various types of creep (fine-grained and coarse debris), rockfalls, sheetwash, gully erosion, and stream erosion. A Random Forest model was trained on 80% of the catchment area, with model accuracy evaluated on a 20% test subset and independently validated via 5-fold cross-validation. Predictors included elevation, TPI (radius = 100 m), slope, topographic wetness index, and Haralick texture metrics derived from RGB orthomosaic channels. Accuracy estimates ranged from 80% to 84%, exceeding baseline (random guessing) by 7–7.5 times. The trained model was applied to the Djankuat Stream catchment (A = 10 km²), located 20 km away. Model results indicate the most widespread processes are fine-grained material creep (3.1 km²), coarse debris creep (1.5 km²), rockfalls (1.6 km²), glacial and fluvioglacial erosion (1.3 km²), gully erosion (0.9 km²), and stable surfaces takes 1.3 km². Future work will validate the semi-automated mapping approach against manually generated reference maps for the second catchment. This work was supported by the grant of the The Government of the Russian Federation (Agreement №075-15-2024-614 date 13.06.2024).