BUILDING ON LANDSLIDE SUSCEPTIBILITY MAPPING IN KENTUCKY: ASSESSING THE COMPLEXITY OF STATISTICS-BASED MODEL PERFORMANCE AND MAP QUALITY

Geomorphic and statistics-based landslide susceptibility modeling can involve endless choices of geomorphic variable inputs, machine-learning techniques, and model performance metrics. Much of the success and practicality of these models hinges on the quality of geomorphic data such as elevation, slope angle, curvature, plan curvature, roughness, and aspect. In addition, the quality and type of a landslide inventory (points vs. polygons, for example), which is used to harvest geomorphic data, is critical to the resulting model performance and applicability of susceptibility maps. However, as the methods of susceptibility modeling advance, particular concerns arise that include the complexity of the effects of specific model inputs on model performance and the transferability of data along political or project area boundaries. To address these concerns, we developed a two-step machine learning approach which includes decision tree algorithm bagged trees followed by multinomial logistic regression. We build on this approach with the goal of improving landslide susceptibility modeling and mapping in Kentucky in two ways. First, we added new variables yet to be considered in the two-step machine learning approach. These include topographic wetness, flow accumulation, stream order, proximity to streams, stream density, and proximity to roads. These new variables will be added altogether and iteratively to determine statistical significance. Secondly, we used a geographically distributed inventory that is perhaps more representative of landslide occurrence in the entire study area. We will test the distributed inventory with the inclusion of the additional variables to compare the statistically significant results, assess model performance, and compare landslide susceptibility maps. Finally, we will evaluate how these efforts of optimizing statistics-based landslide susceptibility modeling will influence quality landslide maps to be used by emergency managers, transportation planners, and local stakeholders in order to reduce risk.