Neural Networks for Spatial and Temporal Data Classification ─ A Case Study for Landslide Susceptibility Analysis

The amount of data and factor of typhoon hazard is big. There are many high mountains and short rivers in Taiwan, so it is easy to landslide and collapse when typhoons occur. A prediction model for finding the landslide susceptibility area can reduce the loss of collapse event. Factors causing collapse (ex: rainfall, wind speed, terrain, and geology etc.) have their own spatial and temporal correlation and they are different by each typhoon hazard and region. Therefore, the subject of the research is to establish different landslide susceptibility analysis model with spatial and temporal data in a different basin. In this research, using geographic facet analyzing the landslide data in 2005~2014 solves the problem of integrity in past data and the imbalanced landslide data. Using time delay factor in multiple times data as training data of double layer Elman network and spatial attribute shows the spatial and temporal correlation in data. Find the key factor and draw landslide susceptibility map by the classification model. The result of classification is 96% of the real landslide data is in the high and mid susceptibility region. Neural network for spatial and temporal data classification with geographic facet can increase the accuracy and effective in hazard warning.