INVESTIGATING LANDSLIDE SUSCEPTIBILITY MODEL VARIATION FROM DISTRIBUTED OR SUBSET INVENTORY DATA, EASTERN KENTUCKY, USA

Landslides are a costly natural hazard in the Appalachian Basin of eastern Kentucky, and therefore landslide susceptibility models are an important mitigation tool for emergency managers. Often, landslide susceptibility models are generated for specific regions with a landslide inventory from a single subset area. For example, our previous landslide susceptibility modeling in eastern Kentucky relied on leveraging a single county to represent a larger region. However, this method assumes that the geographic characteristics within a subset landslide inventory fully represent the region. To test the effects of landslide inventory sampling locations on model results, we consider two digitized inventories within the eight counties defined by the Kentucky River Area Development District in eastern Kentucky: a full county inventory (513.5-km²) and four 7.5-minute quadrangle inventories (total of 614.6-km²). Using a dual machine learning, bagged trees-logistic regression model for the full study area, we plan to input both inventories and compare the resulting susceptibility models to gauge the impact of inventory on distributions and qualitative map quality. If the model results from the distributed inventory are more representative of the full study area, then this sampling method should be the preferred input for future regional landslide susceptibility studies. Additionally, more representative model results from distributed inventory sampling should reduce differences along political boundaries such as county lines. As a result, distributed inventories could assist in decreasing disparities between preceding and forthcoming landslide susceptibility modeling in neighboring regions. Landslide susceptibility models are leveraged by emergency managers and many other local stakeholders, and nuances such as differences from inventory location sampling need to be well understood to qualify their continued use.