Landslide Susceptibility Zonation Mapping Using Machine Learning Algorithms and Statistical Prediction at Hunza Watershed Basin, Pakistan
2024
A. Khan, G. Khan, M. Minhas, S. A. Hussain Gardezi, J. Ahmed and N. Abbas
The mountainous region of the Hunza River watershed basin, especially along the Karakorum highway, and also known as a third pole for the high accumulation of glaciers, which leads to huge devastating landslides occurring every year. Landslide susceptibility mapping was carried out using two deep machine learning techniques (DeeplabV3+ & universal network U-Net) and two statistical models (Intuitionistic Fuzzy divergence IF-D & Frequency ratio FR). The landslide susceptibility mapping is conducted using landslide inventory data and twelve conditional factors. The landslide susceptibility maps obtained from the two statistical models were compared with those generated by two deep machine learning models based on prediction accuracy measures, such as the Area Under the Curve (AUC) and Seed Cell Area Index (SCAI). The Success Rate Curve (SRC) was obtained using the training dataset, and the AUC values for the four models were as follows: 76.9% for IF-D, 76.9% for FR, 80.4% for DeeplabV3+, and 76.3% for U-Net. In terms of the Prediction Rate Curve (PRC) obtained from the validation dataset, the AUC values were found to be 80.8% for IF-D, 80.8% for FR, 81% for DeeplabV3+, and 77.8% for U-Net. To assess the classification ability of the models, the Seed Cell Area Index (SCAI) test was conducted. The results indicated that the SCAI (D-value) was 7.3 for U-Net, 10 for DeeplabV3+, 7.6 for IF-D, and 9.1 for FR. Overall, the findings revealed that DeeplabV3+ exhibited the highest prediction accuracy and classification ability, making it the most suitable choice for landslide susceptibility mapping in the relevant study area.
显示更多 [+] 显示较少 [-]