Rushchak V. O. Geoecological Assessment of Landslide Development Risks Considering the Condition of the Forest Cover of the Territory. Dissertation for the degree of Doctor of Philosophy in specialty 101 “Ecology”, field of knowledge 10 “Natural Sciences”. Ivano-Frankivsk National Technical University of Oil and Gas, Ivano-Frankivsk, 2026.
The dissertation is devoted to the development and practical testing of a comprehensive methodology for the geoecological assessment of landslide development risks, taking into account the condition of forest cover. The proposed approach is based on integrated geoinformation modelling of landslide susceptibility on the Google Earth Engine cloud platform, where forest cover is distinguished as an independent structural block with quantitative parameterisation.
The relevance of the study is determined by the widespread occurrence of landslides in the Ukrainian Carpathians, where their activation is influenced by flysch geological structure, steep slopes, considerable precipitation, and anthropogenic forest transformation. In many regional models, forest cover is considered only in the binary form “forest/non-forest”, which does not fully reflect its influence on slope stability. Therefore, a comprehensive quantitative assessment of forest condition, structure, and spatial position is required.
The aim of the study is to develop, theoretically substantiate, and practically test a methodology for the geoecological assessment of landslide development risks based on an integrated geoinformation model of landslide susceptibility.
The object of the study is landslide development processes as a component of the environmental safety of the Ukrainian Carpathians. The subject of the study is geoinformation methods for assessing landslide distribution risks, taking into account quantitative characteristics of forest cover.
The study applies systemic and comparative analysis, remote sensing methods, raster and vector data analysis, buffer analysis, the Frequency Ratio method, the Random Forest Feature Importance algorithm, logistic calibration, cluster and factor analyses, and principal component analysis. The research was conducted within three study areas in the Ukrainian Carpathians, covering 236, 190, and 1,509 landslide occurrences.
The developed model includes five thematic blocks: morphometric, hydrological, forest, geological, and climatic. Six indicators were proposed for the forest block: proximity to the forest edge, forest cover density, forest type, NDVI amplitude, forest loss during 2001–2023, and the proportion of forest upslope relative to a landslide occurrence. Data from ESA WorldCover 2021, Hansen Global Forest Change, Copernicus Global Land Cover, Sentinel-2, ALOS AW3D30, MERIT Hydro, WorldPop, and JRC GHSL were used.
The scientific novelty lies in the fact that, for the first time in the Ukrainian Carpathians, the role of forest cover as a factor regulating landslide susceptibility has been scientifically substantiated and quantitatively confirmed. The statistical significance of indicators describing forest condition, structure, and spatial position was established, as well as the increased concentration of landslides in forest-edge zones between forested and non-forested areas. The modelling approach was improved by combining Frequency Ratio for factor calibration and Random Forest Feature Importance for determining factor weights. The methodology for assessing risks to the population, infrastructure, and protected areas was further developed.
It was established that forest cover within the Zakarpattia study area decreased from 71% in 1990 to 62% in 2010 and then recovered to 70% in 2020. It was found that 73% of the 1,509 landslides were located within 500 m of the forest edge. Within individual study areas, 61% and 84% of landslides were located within 100 m of areas with forest cover disturbance.
According to the modelling results, 40.2% of the primary study area was classified as having high and very high susceptibility, with 74.6% of documented landslides concentrated in these classes. Validation showed that 88.1% of the test landslides fell within the medium, high, and very high susceptibility classes. The AUC ROC values were 0.72 for model success and 0.69 for prediction, indicating satisfactory model performance.
The practical significance of the results lies in the possibility of using the developed methodology and maps to prioritise monitoring of landslide-prone areas, substantiate forest management measures, support spatial planning, adapt to climate change, and preserve biodiversity. Implementation on the Google Earth Engine platform ensures reproducibility and the potential to scale the approach to other mountainous regions.
Eleven scientific papers have been published on the dissertation topic, and the main results have been presented at international and regional scientific and practical conferences.
