Soil Depth Models in Assessments of Shallow Landslide Potential

While spatially distributed information about soil depth is a critical input in landslide susceptibility models, it is time consuming to measure. We coded nine published steady state soil-depth models (four empirical, five process-based) to provide a framework for evaluating various depth models as a step within a landslide susceptibility assessment. We validated the code by applying the models to three study sites of diverse geomorphology with some existing soil depth data and landslide inventories including a steep, highly dissected sandstone upland in the Oregon Coast Range (OCR), steep-walled canyons in crystalline rock of the Colorado Front Range (CFR), and steep coastal bluffs in glacial deposits along the Puget Sound, Washington (WCB). We used modeled soil depths as input in Transient Rainfall Infiltration and Grid-Based Regional Slope Stability Analysis to compute the factor of safety, a measure of slope failure potential where a value less than one is indicative of slope failure. Then we compared the resulting factor of safety values to mapped landslide locations. Models depending on the following variables most accurately predicted areas of slope failure at each site: slope and depth, slope and contributing area in OCR; constant depth, slope and curvature in CFR; constant depth, slope and depth in WCB. Although all models could predict average depths based on existing soil depth data, each had strengths and weaknesses. The slope-dependent empirical models consistently overpredicted depths on lower slopes and convex areas, such as ridges, and underpredicted on steeper slopes. Thus these models are poorly suited for areas where drainage is well developed as in the OCR and better suited for smooth, sloping terrain like the WCB. A slope and area-dependent model based on the wetness index performed well on steep slopes and was one of the more successful models in all three study areas. Nonlinear area- and slope-dependent and nonlinear slope- and depth-dependent models also performed well due to their ability to distinguish between different terrain features. Nevertheless, the area-dependent models tend to underpredict on ridges and overpredict in swales and the depth-dependent model tends to overpredict on low slopes.