| Abrupt geological hazard (i.e. rainfall-triggered landslides) is one of the prominent natural disasters in southeast China. It is due to the reasons of widespread hilly mountains, vulnerable geological environment conditions, extensive engineering activities and heavy rainfalls. The developing social economic and high-density population adds the vulnerability to the hazards. In the recent years, the increasing number of rainfall-triggered landslides was observed in Zhejiang Province. According to records, the number of geological hazard during the 12th Five-Year Plan period (i.e. 2011-2015) is 127% more than that in 11th Five-Year Plan period (i.e.2006-2010). They have caused huge casualties and severe economic losses. The Chun’an County in west Zhejiang province is one of the places where landslides most developed. The potential landslides in rain season pose an increasing threat on people’s life and property security. Therefore, it is essential to study the development regularity of rainfall-triggered landslides and assess landslide hazard in Chun’an to strategically mitigate the landslide risks, which is considered to have great impacts on scientific and realistic values.Based on the theory of regional landslide hazard assessment and aided by various technologies including geographic information system (GIS), remote sensing, spatial database, statistical analysis and computer programming, this dissertation aims to develop quantitative regional landslide hazard assessment methodologies and technical framework suitable for southeast China. It includes the research contents of development regularity analysis, remote sensing interpretation, field investigation, spatiotemporal database establishment, triggering rainfall threshold analysis, comparative study on susceptibility mapping methods, time frequency and magnitude analysis etc. The main research outcomes and conclusions are achieved as follows:(1) This study designed and developed a landslide spatiotemporal database based on remote sensing and temporal geographic information system (TGIS). Considering the landslides in this study area are characterized by small magnitude, high frequency and being affected by rainfall and human activities, a methodology of implicit landslide remote sensing interpretation was presented. It aimed to interpret the slopes where the landslide is prone to occurrence (named as implicit landslide), which consisted the steps of establishing GIS-based working environment, preliminary susceptibility analysis, construction of interoperation rules, determination of interoperation points, zones and important survey regions and field survey and result verification. The accuracy (in terms of AUC) of implicit landslide remote sensing interpretation was 92.9%, which suggested that the proposed method could effectively guide the field investigation of landslide and improve the work efficiency.The landslide spatiotemporal data model, also named as multilevel base state with amendments model (DMBSAE), was proposed based on TGIS. An event-based spatiotemporal database of landslide was designed and constructed. The associated key technical problems like four database structure (temporary library, current library, historical library and archive library), spatiotemporal data storage, historical state recovering and trace of hazard object’s change were discussed. The outcomes of geological disasters detailed investigation in 2013 in Chun’an County was used as base state in database establishment. There are in total 644 landslides after two years (i.e. 2014 and 2015) operation and update of the established database.(2) This study investigated the relationship between landslides and rainfall events. In the plum-rain dominated area, the inter-monthly distribution of landslide occurrence appeared to have a "single-peak" in June, with 65%of landslides occurred. The landslide occurrence was strongly correlated with heavy rainfall events, which having a relatively longer durations (e.g. less than 11 days). The landslides related with five-day rainfall accounted for 38%. The intensity (Ⅰ)-duration (D) rainfall threshold curves of triggered landslides in high, middle and low susceptibility lithology were derived, respectively. Based on these, the rainfall thresholds for regional landslide forecast were proposed.(3) This study compared Information Value Model (IVM), Logistic Regression (ANN), Artificial Neural Networks (LR) and Support Vector Machine (SVM) in landslide susceptibility mapping. The uncertainty originated from selected methods and combination of 9 condition factors including elevation, slope, aspect, curvature, lithology, distance to fault, distance to road, land use and vegetation was discussed. The results showed that the success rate and prediction rate increased with the increasing of condition factor numbers, but the optimum was not the one with maximum number of conditions factors. It indicated that the affection of different combination of conditioning factors was significant. Generally, based on AUC, the performance of the four methods was ranked as ANN, IVM, SVM and LR from high to low. The ANN (with conditioning factors of slope, lithology, road, elevation, land use and vegetation) was considered to be the most reasonable model for modelling landslide susceptibility in this study area. The achieved AUC for verification and prediction was 88.34% and 88.13%, respectively.(4) The landslide hazard assessment was implemented to combine spatial probability, temporal frequency and landslide magnitude. The results of magnitude analysis showed that the occurrence probability of landslide with a magnitude varying from -6.5 to -5.4 was 0.47. The achieved landslide activity intensity index (R) was RⅠ (0.03), RⅡ (0.47), RⅢ (4.30) and RⅣ (9.21) for classifying various level of hazard zones, respectively. The percentage of the testing samples fallen into high susceptible area was 56.25% and that in super-low susceptible area accounted for 64.3%. The results of the landslide hazard assessment showed that the high and middle level risk regions were distributed mainly over the slope areas with road and valley aside of them, where the developed landslides have the characteristics of high density and large magnitude. Nevertheless, the landslide risk in flat or high-vegetation-covered arears was mitigated. Finally, based on the findings in this dissertation, the measurements and advices in response and adaptation to landslide hazard mitigation were proposed. |
PDF Full Text Request