Accuracy assessment of Antecedent Precipitation Model (AP-Model) for landslide early warning system

Abstract

Geotechnical Engineering Research and Development Center (GERD), Faculty of Civil Engineering, Kasetsart University has developed a Critical Rainfall Threshold by studying various landslide location and collecting rainfall data from events to create the relationship between rainfall accumulated in 3 days and rainfall on the day of landslide incident. The threshold uses Antecedent Precipitation Model (AP-Model) to analyze landslide susceptibility areas. The AP-Model creates a map for landslide early warning system. The model is analyzed by the use of predictive rainfall dataset of the Weather Research and Forecasting Model (WRF) by Hydro Informatics Institute to calculate a cumulative rainfall of 3 days and compare it with the Critical Rainfall Threshold. Thelimitation of the rainfall dataset usage affects the model because the predictive rainfall dataset of WRF model is only 69 percent accurate. This limitation may reduce accuracy of AP-model for landslide early warning system. Therefore, the purpose of this research is to evaluate the accuracy of the AP-Model for landslide early warning system by ROC method and comparing statistical data of landslide with simulate landslide susceptibility areas of the AP-model during 2014 to 2019. This result shows that area under curve (AUC) of landslide probability 20-50% and more than 50% is 0.736 and 0.639 that means good and medium respectively.