A Geostatistical Method For Predicting The Spatial Variability Of Rainfall In Landslide Hazard Assessment: A Case Study On The Southern Fleurieu Peninsula, South Australia

Susan. J. Greene

This paper presents four methods for predicting the spatial variability of rainfall: the theissen polygon, inverse distance squared, isohyetal and kriging. These techniques were applied to an area on the southern Fleurieu Peninsula for use in landslide hazard assessment.

The theissen polygon and isohyetal methods recorded the largest prediction errors due to their dependence on having a large number of rainfall stations not available in the area. The inverse distance squared technique was more successful but was primarily restricted by the difficulties involved in incorporating elevation into the model.

The most successful technique was the multivariate geostatistical algorithm: kriging with an external drift (KED). This technique was able to account for spatially dependent rainfall values and elevation. Predictive monthly rainfall plots were calculated between 1997-2000 using the technique, based on data from the past 74 years. Rainfall values were largest in areas of high relief (Mount Lofty Ranges) and lower in the valley system (Inman Valley). High rainfall variability was shown to have a significant impact when predicting the likelihood of rainfall-triggered landslides in the area.