Spatial distributions of earthquake-induced landslides and hillslope preconditioning in the northwest South Island, New Zealand

Abstract. Current models to explain regional-scale landslide events are not able to account for the possible effects of the legacy of previous earthquakes, which have triggered landslides in the past and are known to drive damage accumulation in brittle hillslope materials. This paper tests the hypothesis that spatial distributions of earthquake-induced landslides are determined by both the conditions at the time of the triggering earthquake (time-independent factors) and the legacy of past events (time-dependent factors). To explore this, we under\-take an analysis of failures triggered by the 1929 Buller and 1968 Inangahua earthquakes, in the northwest South Island of New Zealand. The spatial extents of landslides triggered by these events were in part coincident. Spatial distributions of earthquake-triggered landslides are determined by a combination of earthquake and local characteristics, which influence the dynamic response of hillslopes. To identify the influence of a legacy from past events, we first use logistic regression to control for the effects of time-independent variables. Through this analysis we find that seismic ground motion, hillslope gradient, lithology, and the effects of topographic amplification caused by ridge- and slope-scale topography exhibit a consistent influence on the spatial distribution of landslides in both earthquakes. We then assess whether variability unexplained by these variables may be attributed to the legacy of past events. Our results suggest that hillslopes in regions that experienced strong ground motions in 1929 were more likely to fail in 1968 than would be expected on the basis of time-independent factors alone. This effect is consistent with our hypothesis that unfailed hillslopes in the 1929 earthquake were weakened by damage accumulated during this earthquake and its associated aftershock sequence, which influenced the behaviour of the landscape in the 1968 earthquake. While our results are tentative, they suggest that the damage legacy of large earthquakes may persist in parts of the landscape for much longer than observed sub-decadal periods of post-seismic landslide activity and sediment evacuation. Consequently, a lack of knowledge of the damage state of hillslopes in a landscape potentially represents an important source of uncertainty when assessing landslide susceptibility. Constraining the damage history of hillslopes, through analysis of historical events, therefore provides a potential means of reducing this uncertainty.