Statistical approach to storm event-induced landslides susceptibility

Abstract

Abstract. For the interpretation of the storm event-induced landslide distribution for an area, deterministic methods are frequently used, while a region's landslide susceptibility is commonly predicted via a statistical approach based upon multi-temporal landslide inventories and environmental factors. In this study we try to use an event-based landslide inventory, a set of environmental variables and a triggering factor to build a susceptibility model for a region which is solved using a multivariate statistical method. Data for shallow landslides triggered by the 2002 typhoon, Toraji, in central western Taiwan, are selected for training the susceptibility model. The maximum rainfall intensity of the storm event is found to be an effective triggering factor affecting the landslide distribution and this is used in the model. The model is built for the Kuohsing region and validated using data from the neighboring Tungshih area and a subsequent storm event -the 2004 typhoon, Mindulle, which affected both the Kuohsing and the Tungshih areas. The results show that we can accurately interpret the landslide distribution in the study area and predict the occurrence of landslides in the neighboring region in a subsequent typhoon event. The advantage of this statistical method is that neither hydrological data, strength data, failure depth, nor a long-period landslide inventory is needed as input.

Integrating spatial, temporal, and size probabilities for the annual landslide hazard maps in the Shihmen watershed, Taiwan

“…The standard deviation of the elevation calculated by the elevations of the entire grid within each slope unit was used to indicate the terrain roughness. The slope roughness was simultaneously selected in this study according to the effectiveness in other researches (Lee et al, 2008;Chen et al, 2013). Similarly, the standard deviation of the slope in each slope unit was calculated to express the slope variability, which was then used to indicate the slope roughness.…”

“…The spatial probability in this study was therefore determined using the relationship between the landslide ratio and landslide susceptibility index. The landslide ratio was the ratio of the landslide sample numbers to the number of slope units for each susceptibility index interval (Lee et al, 2008). The ratios represented the landslide spatial probabilities for the slope units with different susceptibility indices.…”

“…In addition, the height differential from the crest to the toe of the slope in each slope unit was used to indicate the total slope height. The total slope height may be physically related to the magnitude of the stress and the pore-water pressure in the lower slope, and for long slopes the surface and subsurface water is more likely to be concentrated in the lower slope, which causes instability (Lee et al, 2008). NDVI values were determined by taking advantage of the absorption of red light and reflection of near-infrared light emitted by green plants.…”

“…Quantitative statistical analysis methods included bivariate analyses (Chung and Fabbri, 1993;Zêzere et al, 2007), multivariate regression (Carrara, 1983;Baeza and Corominas, 2001), logistic regression (Lee et al, 2008;Rossi et al, 2010;Nefeslioglu and Gokceoglu, 2011), and discriminant analysis (Guzzetti et al, 2006;Carrara et al, 2008).…”

Integrating Geo-Spatial Data for Regional Landslide Susceptibility Modeling in Consideration of Run-Out Signature

“…A review of the literature indicates that the continuing improvements in remote sensing and geographic information systems (GIS) have led to cooperate with statistical and data mining (or machine learning) models to evaluate the regional landslide susceptibility. In particular, GIS-based models with geo-spatial data (van Westen et al, 2008;Wang et al, 2005) and event-based landslide inventory (Guzzetti et al, 2012;Lee et al, 2008) have been emphasized and discussed in recent years. With increasing high spatial, spectral and temporal resolutions of remote sensing imageries, landslide areas can be automatically or semi-automatically detected during a single triggering event using pixel-based (e.g.…”