Multi-scale landslide hazard and risk assessment: A modeling and multivariate statistical approach
Landslide susceptibility and hazard assessments in common use are applied to small regions, where high-resolution, in situ, observables are available. When approximated over larger areas, these analyses are often restricted by the absence of landslide inventories, complex or data-intensive methodologies, and surface observable data resolution and availability. As a result, few analyses have considered how landslide hazard assessment can be approximated and modeled at regional and global scales. The increasing availability of remotely sensed surface data and precipitation products presents the opportunity to explore how these small-scale investigations may be scaled up to larger areas. In this thesis, I seek to bridge the gap between site-specific and global landslide susceptibility, hazard and risk assessments by identifying potential modeling approaches and surface observable data that can be transferable over a variety of spatial scales. A preliminary satellite-based global algorithm provides the framework in which to evaluate how the landslide susceptibility and satellite derived rainfall estimates can forecast potential landslide conditions globally. An analysis of this algorithm using a newly developed global landslide inventory dataset suggests that forecasting errors are geographically variable due to improper weighting of surface observables, resolution of the current susceptibility map, and limitations in the availability of landslide inventory data. These methodological and data limitation issues can be more thoroughly assessed at the regional level, where I develop a new susceptibility map using pre-existing regional landslide inventory data and higher resolution surface observables. These analyses explore the scaling relationships and transferability of two methodologies and several surface data products from local to regional scales and indicate that simple bivariate modeling techniques can more effectively approximate susceptibility at regional scales compared to more popular and complex modeling methodologies. Data availability and resolution, particularly for geomorphologic information, landslide inventories, and rainfall extremes, remain limiting factors in affecting the accuracy of landslide susceptibility assessments at multiple spatial scales. However, this thesis provides evidence that integrating available surface observable and precipitation products can greatly enhance landslide susceptibility, hazard and risk assessments and provide the foundation for improving dynamic landslide hazard forecasting in the future.
0799: Remote sensing