PREDICTIVE MAPPING OF LANDSLIDE RISK: MULTIVARIABLE ANALYSIS OF GEOLOGICAL, VEGETATIVE, METEOROLOGICAL, AND HUMAN-RELATED FACTORS

While the mechanistic factors controlling landslide processes are increasingly understood, most studies have necessarily concentrated on single hillslopes or small catchments. Few studies have statistically examined spatial variability of influential factors, which requires a large number of observations of landslide locations. This has limited our ability to predict landslide risk across large regions and develop geographically extensive rather than site intensive hazard planning. We present a GIS-based modeling technique for mapping landslide susceptibility across a 2200 km2 region of coastal southern California. Using scanned and georeferenced 1:24,000 USGS Dibblee geological maps, we randomly selected 1000 landslides of Holocene origin across the study region for analysis of factors underlying landslide distribution. We developed and validated logistic regression models of landslide probability to analyze the relative predictive power of 17 risk variables describing geological, vegetative, meteorological, and human-related site conditions. The modeling indicated that susceptibility of landslides is controlled by a combination of factors. Application of the model in the GIS predicted numerous areas at considerable risk of landslide occurrence, many of which threaten human populations and transportation infrastructure. To date, geomorphological and other environmental data underlying landslide susceptibility have been largely descriptive and/or site-specific, making it difficult for engineers and policy makers to quantitatively incorporate landscape characteristics into mitigating the effects of geological hazards on larger scales. Thus, as the number of landslide disasters continues to grow, effective quantitative analyses and geospatial datasets for predicting locations most at risk are increasingly needed.