A MULTIVARIATE ANALYSIS OF LANDSLIDE SUSCEPTIBILITY ON AN OVERGRAZED RANGELAND, SANTA CRUZ ISLAND, CALIFORNIA

Severe mass wasting occurred on Santa Cruz Island (SCI) California as a result of El Niño driven storms during the 1997-98 winter. Slope failures included both shallow soil slips and deep seated slides, resulting in severe damage to the National Park Service facility at Scorpion Canyon. There was a marked contrast between extensive slope failure on eastern, sheep grazed SCI and the small number of failures on the western portion of the island, which was cleared of sheep in the mid 1980s. The purpose of this research was to determine the relative significance of land use and several environmental factors in predicting landslide susceptibility on SCI using multivariate analysis.

The main factors investigated include slope angle, slope aspect, slope curvature (plan and profile), vegetation, geology, elevation, and land use. Slope angle, slope aspect, slope curvature and elevation were extracted from a 10m DEM that was interpolated from USGS DLG hypsography. Vegetation was obtained from a rectified Normalized Difference Vegetation Index (NDVI) of Landsat 5 TM imagery acquired 10/97. Two sets of orthorectified air photographs, a 1:24,000 set shot 11/97 and a 1:12,000 set shot in 8/98, ‘book end’ the 1997-98 El Niño event and were used to precisely map the occurrence of slope failures over the entire island. A landslide census from the orthophotos show that 1922 discrete soil slips occurred as a result of the El Niño storms.

Logistic regression is a statistical method for fitting a nominal response variable (presence or absence of a landslide) to a linear model of independent predictor variables. From the regression results, the predictor variables can be ranked respectively in order of their significance. To test model accuracy, modeled slide occurrence was compared to actual slide occurrence. The model predicts actual landslide occurrence at 86% accuracy and landslide non-occurrence at 94% accuracy for SCI. Land use has the most significant control on landsliding on SCI followed by slope angle, NDVI, slope aspect, slope plan curvature, and geologic substrate. Elevation and profile curvature were insignificant predictors. Apparently a 15-20 year period of vegetative recovery in response to land use change is enough to effectively stabilize slopes in an overgrazed rangeland like SCI.