Landslide Occurrence Decision Tree Model for Western Washington Featuring Precipitation

Abstract

Landslides pose threats not only to infrastructure around the world but also to local communities. One particularly susceptible area is the Northern Cascades Mountains in the state of Washington in the United States. This study aimed to produce a J48 decision tree model using WEKA software that accurately predicts landslides in this area by validating the predictions against those historically documented landslides in this region. The unique aspect of this study was the use of precipitation in three of the nine independent variables used to generate the classification decision tree model. Besides precipitation, other independent variables included slope, elevation, land cover type, land cover change, soil series type, and bedrock type. A historical landslide event occurrence layer posed as the dependent variable classifier in the model. First, testing subsets of instances allowed an optimal model to be created with the highest kappa statistic of 0.84 and a classification accuracy of 92%. Next, this model identified the most influential factors causing landslides by using ‘information gain’ statistics in WEKA. Average precipitation, elevation, and soil type were determined as being most influential, while surprisingly land cover and land cover change were the least. Lastly, the model’s false positive instances indicated areas of potential future landslides throughout the region. Because an accurate model was created, decision tree classification models were verified as an effective way for landslides to be predicted.