| Summary: | 碩士 === 國立中興大學 === 土木工程學系所 === 104 === The study investigates and analyzes the landslides in Central (the upsteam of the Wu River watershed) and Southern (Cing-Shui and Ai-Liao River watershed) Taiwan. The rainfall staions data and SPOT satellite images of typhoons MINDULLE (2004) and MORAKOT (2009) were collected to interpret the landslides and calculate the three types rainfall indices, which are rainfall intensity, estimated rainfall and the number of short duration rainfall lead to landslide disasters. We use Ordinary Kriging, Simple Kringing and Co-kriging to estimate the distributions of these rainfall indices; by comparing and the RMES, we can find the best way to estimatr the rainfall distribution. After that, probability distribution, correlation coefficient, and significance of the landslide causative factors were checked by P-P plot, Person related factor and Bayesian discriminant analysis. At last, we use nine causative factors in the susceptibility analysis, one is from the above rainfall indices, and the others including elevation, slope, aspect, Ids, distance to road, distance to river, distance to fault, and NDVI/GI, to produce the the logistic regression model. Moreover, the landslide susceptibility models apply the ROC curves for their validation and comparison for their accuracy.
The results reveal that：According RMES, Simple Kriging with elevation is the best spatial distribution for rainfall intensity and estimated rainfall, and Simple Kriging perform better on the number of short duration rainfall lead to landslide disasters. Secondly, the correlation coefficient between rainfall indices is always over 0.8, also due to the method applied in rainfall estimation including elevation fator, rainfall indices could be closely related to elevation. Last but not least, the landslide susceptibility models apply the ROC curves for validation show that, the accuracy of the logistic regression model by using total event estimated rainfall (ET) is the best proformance one. And the number of short duration rainfall lead to landslide disasters performe (ND) is lower than the others. However, these four models can provide good reference susceptibility predictions.
|
|---|
