A DYNAMIC APPROACH TO MODELING SHALLOW LANDSLIDES SUSCEPTIBILITY THROUGH RANDOM FOREST IN GREER’S FERRY LAKE, AR

This study examines the effectiveness of Random Forest (RF), a machine learning algorithm in creating a landslide susceptibility map. Greer’s Ferry Lake Arkansas is the focal area for this investigation. The objectives are to (1) create a landslide susceptibility map through (RF), and (2) examine the relationships between the explanatory variables and landslide susceptibility, with the top three variables that effect susceptibility to landslides being: precipitation, slope, and lithology. Nine explanatory variables were (slope, aspect, flow direction, distance from roads, distance from faults, precipitation, soil type, lithology, and land-use land cover) due to their potential relationship to landslides. ArcGIS Pro ran the (RF) analysis, using a weighted variable approach to create a susceptibility model that accurately demonstrates areas where there will be a high likelihood of future landslide occurrences. Analysis of the variable importance (VI) table for landslide susceptibility showed a highly positive statistical output with R² = 0 .932, indicating a relationship between the landslide triggering factors and landslide susceptibility. Variables that showed the highest relationship with landslide susceptibility were slope (26%), precipitation (19%), and land use land cover (10%), contrary to our initial expectation of lithology playing a significant role for landslide susceptibility. Based on comparisons to previous landslides inventories and field studies for Greer’s Ferry Lake AR, random forest demonstrated to be an effective method for both determining what factors contribute to landslide initiation, representing areas susceptibility to landslides through a landslide susceptibility map.