Debris Flow Fan Recognition Study Based On Multi-spectral Image And DEM

As a typical geological disaster, debris-flow brings a great deal of harm to human life and property. Debris Flow Fan is not only an important symbol of the historical debris flow development, but also the metric for the size and hazard scope of debris-flow. Therefore, it’s very important for study on the geomorphologic processes of debris-flow, the assessment of debris-flow disaster, and the choice of the area for monitoring and warning to identify debris flow fan accurately. The Bailong River Basin in China is selected as studying area where debris flow occurs frequently. Combing with field survey and draw, through visual interpretation on SPOT images, the distribution range partial debris flow fans and non-debris flow are obtained as the sample area. The vegetation index, soil, topography, terrain features and many other indexes are extracted by using multi-spectral remote sensing images (ASTER) and DEM. Under the variance analysis and cluster analysis, each feature is evaluated. According to physical meaning of different remote sensing indicators, the most relevant features are selected as input, and then based on the object-oriented and pixel-based methods; the debris flow fan is identified respectively by means of the Support Vector Machine and Neural Net. The conclusions are shown as the following:1. The band ratio of the fused image with SPOT images and ASTER can effectively highlight the mineral composition of the rock soil, which can be further used for the classification of Debris Flow Fans; especially the indicators of iron oxide and vegetation identify it significantly.2. The band ratio can effectively identify the Debris Flow Fans with the principal components obtained by principal component analysis. For instance, the first and the third principal components reflect geological conditions, the second component reflects cover of vegetation, and the fourth and fifth ones reflect cover of soil.3. With the vegetation, soil, topography and terrain indexes selected as input, the debris flow fan is identified by using supervision classification. The remote sensing indexes and terrain indexes can be effectively combined and the result is close to the actual situation.4. The identification accuracy by using supervised classification is higher than that based on object-oriented approach.