Quick estimation of critical acceleration of slopes through a single prediction equation

Abstract

The seismic slope performance is usually evaluated by estimating the permanent downslope displacement of potential sliding mass based on Newmark-type sliding block procedures. As a key input of these procedures, the critical acceleration (a_c) is typically derived as the horizontal seismic acceleration that produces the unity factor of safety in iterative pseudo-static slope stability analyses. This process is time-consuming especially when considering a number of slope configurations (e.g., in regional landslide hazard mapping). This preliminary study aims to present a single prediction equation for a_c using a neural network. The input parameters of the prediction equation include the cohesion, friction angle, and unit weight of soil, as well as the slope height and slope angle, which are common in practice. The dataset for model development is obtained from iterative pseudo-static stability analyses for thousands of generic slope configurations. The results indicate that the proposed model can reasonably predict a_c under various scenarios. A slope example is used to illustrate the application of the prediction equation in seismically-induced slope displacement analysis. Further efforts are needed to enhance the generalization capability of the a_c prediction.