Modeling risk of landslide initiation and runout in the Colorado Front Range under current and future climates

Abstract

Precipitation-induced landslides pose risks to humans through property damage, disruption of infrastructure, injury, and loss of life. Due to the spatial and temporal heterogeneity of soil moisture and landscape characteristics that impact slope stability and potential impacts of climate change on landslide location, quantifying landslide risk to humans is difficult as uncertainties are not represented in available datasets. Recent developments have improved our ability to probabilistically model landslide initiation, thus allowing for the incorporation of spatial and temporal uncertainty in the prediction of the onset of hillslope failures. The ability to incorporate uncertainty in landslide models is particularly valuable for considering how climate change, which could impact vegetation cover and associated root cohesion, might alter the vulnerability of people and infrastructure to landslides. The aim of this analysis is to probabilistically forecast landslide susceptibility under climate change by incorporating changes in the type and distribution of vegetation while accounting for uncertainties in key properties. Using Landlab, a Python-based toolkit for landscape modeling, we perform Monte Carlo simulations with an infinite slope stability model to make spatially explicit calculations of the probability of landslide initiation. The soil moisture input to the landslide model is from the Equilibrium Moisture from Topography, Vegetation, and Soil (EMT+VS) model, which downscales coarse-resolution soil moisture by incorporating the dependence of soil moisture on topographic, vegetative, and soil characteristics. We evaluate model sensitivity and identify that vegetation, which impacts cohesion and soil depth, has a large impact on the model. We evaluate model performance by simulating landslide susceptibility over a 1333 km2 area of the Colorado Front Range as there is a large inventory of more than 1300 landslides from an extreme precipitation event in 2013. One anticipated effect of climate change in the Colorado Front Range is a reduction in the survivability of trees, which we incorporate through applying reductions to vegetative cohesion and vegetation cover. For the 2013 event, the model predicts 79.6% of the mapped landslides and 5.8% of the rest of the study area as being unstable. A deterministic model using mean values from the probability model and assuming FS ≤ 1 is unstable captures only 42% of observed landslides, supporting the use of the probabilistic model. The probabilities are low (P(F) < 0.2) for the majority of predicted failures with a concentration at higher (P(F) > 0.8) values, with the latter having higher slopes and lower vegetation. 66% of nodes with P(F) > 0 occur on south facing slopes where trees are less abundant. After incorporating climate change, we see an increase in the areas susceptible to landslides and a shift to more instability on north-facing slopes. Our study suggests that vegetation changes due to climate change could result in major shifts in the people and infrastructure susceptible to landslides in the Colorado Front Range. In conjunction with landslide initiation, determining landslide runout is important to fully analyze landslide risk. Landslide runout modeling for large areas is difficult due to limited information and the complexity of landslides. The difficulties of physically modeling landslides on large spatial scales have led to the development of empirical methods based on topographic attributes. While empirical models are limited in that they require calibration in new areas and thus can only be applied to areas with landslide inventories, they provide a way to model landslide runout at large spatial scales and identify areas for further, potentially more physically-based, analyses. We investigate whether topographic controls can be used to predict landslide termination. We develop a landslide runout model and apply it to a 10-m elevation grid. Our model routes landslides downslope with d8 flow direction method and uses a critical slope, defined as a minimum slope a landslide must encounter to end, and slope persistence, defined as the distance the landslide must travel under the critical slope, to represent landslide stopping locations. We apply our model to see if it can replicate landslide runout in the Colorado Front Range due to a large landslide inventory from a 2013 precipitation event that induced approximately 1300 mapped landslides. The calibrated model has a critical slope of 3° and a slope persistence of 20 m and predicts landslide distance in both the calibration and evaluation areas with a Nash-Sutcliffe (NS) value of 0.69 and 0.58, respectively. We compare our calibrated model to an angle of reach approach, an approach that has been applied previously for landslide runout mapping which determines the slope between the start and end of a landslide, and determine that the best NS value of 0.14 occurs at an angel of 20°. Our results show that within our study area, topographic controls provide plausible initial estimates of runout endpoints and an improvement over similarly simplistic methods such as the angle of reach. The potential of using critical slope combined with slope persistence to capture topographic controls to predict runout endpoints is a promising opportunity for landslide hazard mapping at large spatial extents.