Hillslope memory and spatial and temporal distributions of earthquake-induced landslides

Large earthquakes commonly trigger widespread and destructive landsliding. However, current approaches to modeling regional-scale landslide activity do not account for the temporal evolution of progressive failure in brittle hillslope materials. Progressive failure allows hillslopes to possess a memory of previous earthquakes, which has the potential to influence landslide activity in future earthquakes. The original contribution of this thesis is to address the influence of hillslope memory on spatial and temporal patterns of earthquake-triggered landslide activity, through a combination of landslide inventory analysis and numerical modeling. An understanding of spatial distributions of earthquake-triggered landslides is first established, through analysis of inventories of landslides triggered by five large (M_w > 6.7) earthquakes. The results show how current landscape conditions at the time of earthquakes influence hillslope failure probability. By identifying factors exhibiting a common influence on landslides triggered by all five earthquakes, general spatial models of landslide probability are developed, which are transferrable between different earthquakes and regions. Analysis of model performance for landslide distributions triggered by two sequential earthquakes is then used to establish where this spatial approach breaks down. Errors in the landslide distribution predicted for the second earthquake suggest that the legacy of damage to hillslope materials accrued from the first earthquake is an important control on landslide occurrence. Given the infrequent recurrence of large earthquakes and limited temporal coverage of landslide data, a new modelling approach is developed to understand how hillslope memory influences long-term patterns of earthquake-triggered landslide activity. The model integrates the site-scale evolution of hillslope progressive failure into modeling regional-scale earthquake-triggered landslide activity, in response to sequences of earthquakes. The model results suggest that the sensitivity of landscapes to landslide-triggering increases following large earthquakes, due to damage accumulated in hillslopes that do not reach the point of failure, and decays as these hillslopes fail in response to subsequent, lower-magnitude events. Prolonged elevated levels of rainfall-triggered landslide activity observed following large earthquakes appear to reflect this result. Using the model outputs, a methodology is proposed for predicting temporal variability in landslide activity using records of seismic data. The model results also suggest that, when hillslopes undergo progressive failure, relationships between seismic forcing and landslides are influenced by the magnitude-frequency distribution of earthquakes. As a result, current approaches that use these relationships to predict levels of long-term landslide hazard and erosion rates, but do not account for regional differences in earthquake distributions, may suffer from systematic under- or over-prediction. These significant implications for predicting the geomorphological and human impact of landslides highlight the need for detailed multi-temporal datasets recording the evolution of landslide activity following major earthquakes, in order to quantitatively investigate the influence of hillslope memory in real landscape settings