日本地球惑星科学連合2023年大会

講演情報

[E] 口頭発表

セッション記号 H (地球人間圏科学) » H-DS 防災地球科学

[H-DS05] 地すべりおよび関連現象

2023年5月26日(金) 09:00 〜 10:15 106 (幕張メッセ国際会議場)

コンビーナ:王 功輝(京都大学防災研究所)、今泉 文寿(静岡大学農学部)、齋藤 仁(名古屋大学 大学院環境学研究科)、千木良 雅弘(公益財団法人 深田地質研究所)、座長:今泉 文寿(静岡大学農学部)、Zheng-yi Feng(National Chung Hsing University)

09:00 〜 09:15

[HDS05-01] Mapping intense-rainfall-induced landslide hazards for Puerto Rico, USA, using empirical and physics-based approaches

★Invited Papers

*William Schulz2、Kenneth Stephen Hughes1、Rex Baum2、Mark Reid2、Dianne Brien2、Colin Cronkite-Ratcliff2、Matthew Tello2、Jonathan Perkins2 (1.Univ. Puerto Rico-Mayaguez、2.U.S. Geological Survey)

キーワード:landslide, physics-based modeling, empirical modeling, Puerto Rico

Landslides present significant hazards in the tropical, mountainous U.S. territory of Puerto Rico, primarily an 8,870 km2 island in the Caribbean Sea that is home to ~3.3 million people. For example, Hurricane Maria triggered more than 70,000 landslides during 2017 that caused casualties and widespread damage. Following Maria, we developed landslide susceptibility maps to help reduce future landslide hazards.

To help guide hazard mapping, we mapped point locations of Maria-induced landslides and boundaries and runout of 5,720 landslides using aerial imagery and lidar topography. We studied 123 landslides in the field and tested materials in the lab. Results show that intense-rainfall-induced landslides in Puerto Rico are generally translational, less than 1 m thick and 10 m across, form in saprolite, and mobilize as debris flows that may travel more than 1 km when channelized with other slides. Hence, Puerto Ricans face 2 primary types of landslide hazard: ground failure at landslide sources, and impact or inundation from landslide runout. We developed maps in multiple ways to depict these hazards, with benefits and limitations of each.

Many runout modeling approaches, including ours, require landslide location and volume estimates. We advanced the one-dimensional (1D) TRIGRS pore pressure and slope-stability model to quasi-three-dimensional (3D) to evaluate pixel regions as potential landslides, rather than individual pixels, and applied the model over ~600 km2. Lab data provided initial material property parameters, full saturation and slope-parallel flow was assumed, and soil depth was estimated using a nonlinear area and slope dependent model. Compared to 1D, the quasi-3D analysis more clearly delineated susceptible areas and provided landslide sources for runout modeling, although validation using the Maria inventory required conservative definition of potential instability that resulted in large regions of relatively high susceptibility. Using those results, we modeled runout in two ways. For landslides unlikely to reach channels, we mapped runout by topographic routing of individual sources projected downslope within an angle of reach of 20°, conservatively estimated from observations. To map channelized debris flows, we used empirical landslide volume-inundation area relations along with empirical debris-flow growth factors. Considering high observational data and computational needs, we followed these approaches for areas where landslide risk is particularly great.

Empirically derived landslide hazard maps can account for more factors related to landslide initiation than current physics-based modeling approaches, are computationally simpler, and use more readily obtainable data. Using 75% of the Maria landslide point inventory and the frequency-ratio approach, we found increasing correlation between landslide location and proximity to roads and streams, mean annual precipitation, geologic terrane, land cover, topographic curvature and slope, satellite-based soil moisture estimates, and soil class. These analysis results were used to create a Puerto Rico-wide map depicting relative susceptibility to landslide occurrence during intense rainfall. Validation using 25% of the inventory produced a receiver operating characteristic area-under-curve value of 0.87, indicating high map accuracy.

Landslide hazard maps should depict where landslides occur and travel downslope, but physics-based landslide initiation modeling most appropriate for producing runout modeling input data cannot yet account for all factors contributing to landslide occurrence. By contrast, with ample observational data, empirically derived landslide initiation modeling can better reveal potential landslide locations, but not readily provide discrete source areas necessary for runout modeling. Future hazards from rainfall-induced landslides in Puerto Rico may be best estimated by using the empirically derived and physics-based maps in combination.