Kalman filter decision systems for debris flow hazard assessment

This paper aims to develop a model for debris flow hazard assessment, since Taiwan is a mountainous country subject to bouts of heavy rainfall during the rainy and typhoon seasons and is thus frequently subject to landslide disasters. The database used is comprised of information from actual cases that occurred in areas of Hualien in eastern Taiwan during 2007 and 2008. The Kalman filter model is utilized to assess the occurrence of debris flows from computed indexes, to verify modeling errors. Comparisons are made between two models to determine which one is better in practical applications. The efficiency of the Kalman filter decision system has been proved, showing smaller average relative error and correspondingly larger ratio of success on debris flow assessment when compared to the neural network model. The relative error is calculated between the differences in the ratio of model outputs and that of actual occurrences. Such an error index can be decreased from 4.65 to 3.39% by introducing the concept of geographic divisions to the Kalman filter model, which demonstrates its ability to forecast the occurrence of debris flows for the coming year.