Hazards analysis and prediction from remote sensing and GIS using spatial data mining and knowledge discovery: a case study for landslide hazard zonation

摘要

Due to the particular geographical location and geological condition, Taiwan suffers from many natural hazards which often cause series property damages and life losses. To reduce the damages and casualty, an effective real-time system for hazard prediction and mitigation is necessary. In this study, a case study for Landslide Hazard Zonation (LHZ) is tested in accordance with Spatial Data Mining and Knowledge Discovery (SDMKD) from database. Many different kinds of geospatial data, such as the terrain elevation, land cover types, the distance to roads and rivers, geology maps, NDVI, and monitoring rainfall data etc., are collected into the database for SDMKD. In order to guarantee the data quality, the spatial data cleaning is essential to remove the noises, errors, outliers, and inconsistency hiding in the input spatial data sets. In this paper, the Kriging interpolation is used to calibrate the QPESUMS rainfall data to the rainfall observations from rain gauge stations to remove the data inconsistency. After the data cleaning, the artificial neural networks (ANNs) is applied to generate the LHZ map throughout the test area. The experiment results show that the accuracy of LHZ is about 92.3percent with the ANNs analysis, and the landslides induced by heavy-rainfall can be mapped efficiently from remotely sensed images and geospatial data using SDMKD technologies.