Probabilistic Modeling of Shallow Landslides at Regional Scales

Abstract

Mountainous areas are challenging to manage and maintain access due to their remoteness and steep topography. Shifting hydrologic regimes from changing climate are projected to intensify these challenges. Of particular concern are the effects and uncertainties from climate change on hillslope stability that may lead to increased landslides, which adds sediment to streams, elevates flooding, and impacts downstream natural and built resources. This dissertation aimed to improve mapping landslide hazard by integrating process-based and data-driven statistical models. To achieve this, we organized the dissertation into four chapters that begins with motivation and background (Chapter 1) and a climate change vulnerability assessment to access over a large regional area (Chapter 2). Chapter 3 describes a new probabilistic model of shallow landsliding based on a physical model that is coupled with a macro-scale hydrologic model and a soil evolution model explicitly addressing spatial and temporal uncertainty. This physical model is integrated with a statistical model relating observed landslides with local site factors predisposing a hillslope to fail to produce regional-scale landslide hazards from initiation, transportation, and deposition processes (Chapter 4). Concerns about hillslope stability were identified during one of the largest climate change adaptation efforts undertaken on federal lands. This effort included a transportation vulnerability assessment conducted with research scientists and federal land managers of two national parks and two national forests in north-central Washington, USA. During this assessment documented in Chapter 2, one of the top four infrastructure sensitivities recognized was increased damage associated with landslides from projected higher winter soil moisture caused by changes in seasonal precipitation and snow accumulation. Numerous strategies were identified to increase resistance and resilience of the transportation system to this impact pathway, including information needs such as “site-specific stability analysis based on soil and geologic information” and “identification of areas sensitive to high landslide frequency.” This dissertation takes on these information priorities by developing regional landslide models and demonstrates the models in one of the four jurisdictions: North Cascades National Park Complex (NOCA), Washington. Chapter 3 of the dissertation describes our development of a hydro-climatological approach to modeling of regional shallow landslide initiation that integrates spatial and temporal dimensions of parameter uncertainty. The physically-based model estimates annual probability of landslide initiation by solving the infinite slope stability equation coupled to steady-state topographic flow routing using a Monte Carlo approach. The uncertainty of soil depth often ignored in landslide hazard modeling is address by a soil development model, and subsurface flow recharge is obtained from the Variable Infiltration Capacity macro-scale hydrologic model. Thus, the model design allows for use of future hydrologic projections to estimate changing landslide probability as climate and landscape evolve. The model is available as a component in Landlab, an open-source, Python-based landscape earth systems modeling environment. It is designed to be easily reproduced and applied in various locations utilizing HydroShare cyberinfrastructure; therefore, it can be implemented in the other three federal jurisdictions and elsewhere. To better understand landslide transport and deposition impacts, we develop empirically-based probability hazard maps from a statistically-derived susceptibility index explained in Chapter 4 of this dissertation. This empirical model integrates the influence of seven site attributes on observed landslides, inventoried by NOCA park personnel, using a frequency ratio approach. The attributes assessed included: elevation, slope, curvature, aspect, land use-land cover, lithology, and topographic wetness index. The physically-based and empirically-based models were then combined to produce an integrated probabilistic map of landslide hazard for initiation, transport, and deposition processes. Thus, these maps identify locations of high and low probability of landslide impacts within the NOCA that can be used by land managers in their design, planning, and maintenance. Improved tools such as these with incorporated uncertainty can be used to reduce system vulnerabilities and lead to adaptations that allow continue use of natural areas with reduced risks.