Loading...
Optimization of landslide susceptibility modeling: a Puerto Rico case study
Tello, Matthew
Tello, Matthew
Citations
Altmetric:
Editor
Date
Date Issued
2020
Date Submitted
Collections
Research Projects
Organizational Units
Journal Issue
Embargo Expires
Abstract
Landslide susceptibility modeling is a time-consuming and challenging endeavor. In most cases, verified field data is limited, and projections of subsurface conditions based on borings or field data are highly uncertain. In order to accurately forecast landslide events, several aspects of the geology, climate, hydrogeology, land use, soil mechanical properties and spatial variation of soils need to be considered. Additionally, modeling landslide susceptibility often requires the integration of several software packages for modeling, map building, and analysis. Once developed, however, accurate susceptibility maps help to improve public safety by helping to direct response efforts and the implementation of mitigation and zoning for development. To address the aforementioned, this thesis presents a framework for calibration and optimization of soil depth and landslide susceptibility modeling using statistical methods. It presents three data processing tools that can be used to reduce processing time and improve the accuracy of landslide susceptibility maps. It also identifies the correlation of landslide susceptibility with various land use and land cover factors, which should be considered in future model development.The developed framework is applied to the U.S. territory of Puerto Rico, where heavy rains from Hurricane Maria caused over 71,000 landslides and debris flows in 2017, impacting the majority of the island. The magnitude of this event both emphasizes the necessity for improving landslide susceptibility modeling to prepare for future events and provides a detailed dataset for such research.Data processing for soil depth and landslide susceptibility maps were completed using Regolith and TRIGRS software packages. Simulated soil depth data sets were compared with the observed data set using goodness of fit statistical summaries typical of hydrogeological model calibration. Landslide susceptibility maps were analyzed using receiver operating characteristics to identify the fit of each model type with mapped landslide polygons. Of the eight models tested, the non-linear slope and area dependent sediment transport model (NASD) was the best fit model for the Puerto Rico case study. The results also indicate that the statistical methods used in this study are in agreement with each other, providing a high level of confidence in the results. Additionally, the model type itself and its optimized model input parameters are found to be appropriate for the geological and geographical conditions considered.The results from this thesis provide insight into the importance of integrating a soil depth distribution map into the modeling process for shallow rainfall-induced landslide modeling. The developed methodology shows improvement over previous susceptibility maps, and should be considered for future studies when considering shallow rainfall-induced landslide susceptibility modeling. Finally, the framework and tools presented in this thesis can be used as a guide for implementation of batch processing and automated processing to drastically reduce the time component of calibration and optimization.
Associated Publications
Rights
Copyright of the original work is retained by the author.