Authors' Names

Gina BelairFollow

Presentation Type

Poster Presentation

Category

STEM (science, technology, engineering, mathematics)

Abstract/Artist Statement

Landslides have the potential to cause substantial loss of life and damage to infrastructure. To minimize the damage incurred, landslide hazard assessments are conducted to determine areas of potential mass movements. The majority of advancements in landslide hazard assessment are implemented in a narrow set of circumstances, and their transferability to other regions is largely untested. This study follows the methodology of Kritikos et al. (2015) who investigated the transferability of co-seismic landslide relationships from well-studied regions to regions without accurate landslide locations. The results from Kritikos et al. (2015) provide evidence that statistical relationships formed from training data in multiple regions can accurately predict the occurrence of co-seismic landslides in regions that were not used in training. This study investigates whether these results can be generalized to shallow landslides that were not caused by seismic activity. We focus on three diverse regions in Utah, Washington, and the Apennine Mountains of Italy. Logistic regression (LR) and frequency ratio (FR) methods were used to determine relationships between landslide occurrence and DEM (digital elevation model)-derived contributing factors. Models will be created using 70% of the data for each individual region, each pairwise combination of regions, and the combination of all three regions. Relative operating characteristic (ROC) curves will be used to assess the efficacy of the models [3]. The ROC curve is defined as the true-positive proportion versus the false positive proportion [3]. For landslide hazard assessments, areas under this curve (AUROC) from 1-0.9 are considered excellent, 0.8-0.9 are considered very good, and 0.7-0.8 are good [1]. Preliminary studies using the contributing factors of elevation, slope, aspect, curvature, and topographic position index derived from a 30-meter DEM show promising results. Validation AUROCs were found to be greater than 0.7 for LR and FR in the Utah region, Washington region, Italian region, and the combined Utah & Washington region. If models created using data from multiple regions can be used to predict landslide occurrence in dissimilar regions, it would allow hazard scientists to conduct studies in places without accurate landslide catalogs. Only using factors derived from a 30-m resolution DEM, which is globally available, also increases the importance of this study. Many communities with high landslide risk do not have resources to employ methods that require high levels of expertise and high-resolution data. Although this study has the potential to oversimplify a very complicated system, it can be used as a first-order approach in regions that lack data and resources. [1] Abedi Gheshlaghi, H., & Feizizadeh, B. (2017). An integrated approach of analytical network process and fuzzy based spatial decision making systems applied to landslide risk mapping. Journal of African Earth Sciences, 133, 15–24. https://doi.org/10.1016/j.jafrearsci.2017.05.007 [2] Kritikos, T., Robinson, T. R., & Davie, T. R. H. (2015). Regional coseismic landslide hazard assessment without historical landslide inventories: A new approach. Journal of Geophysical Research: Earth Surface, 120, 711–729. https://doi.org/10.1002/2014JF003224 [3] Swets, J. A. (1988). Measuring the Accuracy of Diagnostic Systems. Science, 240(4857), 1285–1293.

Mentor Name

Rebecca Bendick

Personal Statement

The USGS estimates that anywhere between 25-50 annual deaths occur due to landslides in the United states alone, with the death toll in the thousands worldwide [4]. Although the causes and mechanics of landslides are relatively well understood, substantial risk still exists due to the continued development in landslide-prone areas [1,3]. Due to this development, there is an increased demand for landslide hazard and risk assessments. One of the major difficulties in landslide research is the proliferation of methods available, both for landslide identification and for hazard assessment. Visual identification is the most common method of identification, but requires a high level of expertise. Some level of experience with geomorphology is expected of someone trying to complete a landslide hazard assessment, but visual identification is a skill that is not uniform across the hazard assessment field. This leads to a high degree of variability in landslide inventories, which increases the possible bias in the hazard assessment methods dependent on them. The accumulation of these inventories are often small-scale and site specific. Many countries do not have a national database of events, and if they do they contain inconsistent and incomplete data. In the United States a national database was only released by the USGS in October 2019 [2]. Having a better understanding of where landslides occur in regions that have a complete and accurate catalog can help to better inform policies and mitigation efforts in places without the same data. The results of my work will allow scientists to use globally available data and the location of landslides in other regions, to conduct hazard assessments in unstudied regions. The small number of explanatory variables and the simplicity of the statistical methods make this approach easy to implement and easy to understand. [1] Aleotti, P., & Chowdhury, R. (1999). Landslide hazard assessment: Summary review and new perspectives. Bulletin of Engineering Geology and the Environment, 58(1), 21–44. https://doi.org/10.1007/s100640050066 [2] Jones, E. S., Mirus, B. B., Schmitt, R. G., Baum, R. L., Burns, W. J., Crawford, M., et al. (2019). Summary Metadata – Landslide Inventories across the United States. U.S. Geological Survey Data Release. https://doi.org/https://doi.org/10.5066/P9E2A37P. [3] Schuster, R. L., & Highland, L. M. (2001). Socioeconomic and Environmental Impacts of Landslides in the Western Hemisphere. U.S. Geological Survey Open-File Report. Retrieved from http://pubs.usgs.gov/of/2001/ofr-01-0276/ [4] USGS. (2019). How many deaths result from landslides each year? Retrieved from https://www.usgs.gov/faqs/how-many-deaths-result-landslides-each-year?qt-news_science_products=0#qt-news_science_products

gradcontalk.mp4 (16142 kB)

Share

COinS
 

Investigating the Transferability of Landslide Hazard Assessments

Landslides have the potential to cause substantial loss of life and damage to infrastructure. To minimize the damage incurred, landslide hazard assessments are conducted to determine areas of potential mass movements. The majority of advancements in landslide hazard assessment are implemented in a narrow set of circumstances, and their transferability to other regions is largely untested. This study follows the methodology of Kritikos et al. (2015) who investigated the transferability of co-seismic landslide relationships from well-studied regions to regions without accurate landslide locations. The results from Kritikos et al. (2015) provide evidence that statistical relationships formed from training data in multiple regions can accurately predict the occurrence of co-seismic landslides in regions that were not used in training. This study investigates whether these results can be generalized to shallow landslides that were not caused by seismic activity. We focus on three diverse regions in Utah, Washington, and the Apennine Mountains of Italy. Logistic regression (LR) and frequency ratio (FR) methods were used to determine relationships between landslide occurrence and DEM (digital elevation model)-derived contributing factors. Models will be created using 70% of the data for each individual region, each pairwise combination of regions, and the combination of all three regions. Relative operating characteristic (ROC) curves will be used to assess the efficacy of the models [3]. The ROC curve is defined as the true-positive proportion versus the false positive proportion [3]. For landslide hazard assessments, areas under this curve (AUROC) from 1-0.9 are considered excellent, 0.8-0.9 are considered very good, and 0.7-0.8 are good [1]. Preliminary studies using the contributing factors of elevation, slope, aspect, curvature, and topographic position index derived from a 30-meter DEM show promising results. Validation AUROCs were found to be greater than 0.7 for LR and FR in the Utah region, Washington region, Italian region, and the combined Utah & Washington region. If models created using data from multiple regions can be used to predict landslide occurrence in dissimilar regions, it would allow hazard scientists to conduct studies in places without accurate landslide catalogs. Only using factors derived from a 30-m resolution DEM, which is globally available, also increases the importance of this study. Many communities with high landslide risk do not have resources to employ methods that require high levels of expertise and high-resolution data. Although this study has the potential to oversimplify a very complicated system, it can be used as a first-order approach in regions that lack data and resources. [1] Abedi Gheshlaghi, H., & Feizizadeh, B. (2017). An integrated approach of analytical network process and fuzzy based spatial decision making systems applied to landslide risk mapping. Journal of African Earth Sciences, 133, 15–24. https://doi.org/10.1016/j.jafrearsci.2017.05.007 [2] Kritikos, T., Robinson, T. R., & Davie, T. R. H. (2015). Regional coseismic landslide hazard assessment without historical landslide inventories: A new approach. Journal of Geophysical Research: Earth Surface, 120, 711–729. https://doi.org/10.1002/2014JF003224 [3] Swets, J. A. (1988). Measuring the Accuracy of Diagnostic Systems. Science, 240(4857), 1285–1293.