KMS XINJIANG INSTITUTE OF ECOLOGY AND GEOGRAPHY，CAS
| 中东非洲卢旺达滑坡易感性模拟 | |
| Alternative Title | MODELING LANDSLIDE SUSCEPTIBILITY IN RWANDA,CENTRAL-EASTERN AFRICA |
| Jean Baptiste Nsengiyumva | |
| Subtype | 博士 |
| Thesis Advisor | 罗格平 |
| 2019-06-30 | |
| Degree Grantor | 中国科学院大学 |
| Place of Conferral | 北京 |
| Degree Discipline | 理学博士 |
| Keyword | 灾害 山体滑坡灾害 卢旺达 敏感性制图 空间多标准评估模型（SMCE） 统计指数法（SI） 稳定性指数制图（SINMAP） 证据权重（WOE） 逻辑回归法(LR） Disaster Landslide hazard Rwanda Susceptibility mapping Spatial-muticriteria evaluation model (SMCE) Statistical index method (SI) Stability index mapping (SINMAP) Weights of evidence (WOE) Logistic regression method (LR) |
| Abstract | 位于非洲东部的内陆型国家卢旺达对气候变化敏感，经常发生洪水、山体滑坡、风暴、雷暴、干旱、火山活动和地震等自然灾害事件。这些自然灾害会引发严重的人员伤亡、 巨大的财产损失， 对各种基础设施损坏和环境造成严重的破坏。 其中， 频繁、剧烈的山体滑坡所造成的危害最大， 是卢旺达境内最为严重的自然灾害。为了实现降低滑坡风险和可持续的管理，很有必要进行敏感性制图， 以检测滑坡易发区域并为灾害风险监测提供信息。众所周知，滑坡敏感性评估和制图是滑坡风险分析中的关键步骤。因此，在滑坡风险管理中，应用适当的方法来生成准确的滑坡敏感图至关重要。本研究旨在利用半定量模型（SMCE-空间多标准评估方法）、 统计模型（证据权重方法 WoE、统计指数法 SI、频率比 FR 以及逻辑回归 LR）和基于过程的方法（稳定性指数映射-SINMAP模型） 对卢旺达进行详细的滑坡敏感性建模， 对不同建模结果进行分析比较并评估其性能。本研究将卢旺达作为研究区，在该区域内进行了实验和实地调查， 准备了 15 个滑坡因果因子图并应用于生成滑坡敏感性图，其中包括地形湿度指数（TWI）、坡度、海拔、 距主要道路的距离、坡度、土地覆被/土地利用、岩性、地形因子（LS）、归一化植被指数（NDVI）、降水量、曲率、河流溪流距离、土壤质地和土壤深度。SINMAP 模型结合 30 米空间分辨率数字高程模型（DEM），使用土壤物理参数（内摩擦角， 渗透性和粘聚力）。基于土壤的这些参数，计算出安全系数（FS）。因此，SMCE 方法采用了七层调节因子，包括土地利用/土地覆被、岩性、土壤深度、坡度、降水、 地震活动和土壤质地。在研究区域内，通过实地调查和已有的记录共确定了 980个历史滑坡遗址并绘制相应的地图。进而将其用于构建模型（训练点）并对模型的性能（测试数据集）进行验证。 应用 SMCE 模型，计算了库存地图调节因子之间的空间相关性。结果表明，约有 42.3％的研究区域呈现为中度和极高的敏感性。滑坡敏感型区域在研究区覆盖总人口的 49.3％。此外，易受影响程度高至极高的区域分别为西部，北部和南部（分别为 40.4％， 22.8％和 21.5％）。然而，卢旺达东部地区受滑坡灾害的影响较小， 滑坡敏感性低（87.8％）， 不存在滑坡高敏感性区域（0％）。研究结果表明，所采用的模型对滑坡敏感性的计算结果合理且准确。 此外， 49.5％的历史滑坡点在模拟结果中属于极高滑坡敏感性区， 进一步证实了该模型在研究区具有良好的适用性和预测能力。通过对 SMCE 和 SINMAP 模型的比较表明， SMCE 方法在预测敏感性方面比SINMAP 方法具有更好的性能。 本研究利用受试者操作特性曲线（ROC / AUC）和其他统计评价因子（包括准确性、 精度和均方根误差（RMSE））验证和比较模型的预测能力。 SMCE 和 SINMAP 的 AUC 值分别为 87.92％和 78.09％。 而且， SMCE 模型具有更高的准确度和精度，分别为 0.77 和 0.734。对于 RMSE 值， SMCE 较 SINMAP 更好，分别为 0.332 和 0.398。总体上， SINMAP 和 SMCE 模型的模拟结果都较为可靠。最后，对四个概率统计模型（FR、 SI、 LR 和 WoE）进行了比较，生成了卢旺达的滑坡敏感性图，并使用受试者操作特性曲线（ROC / AUC） 对其进行了验证。从该研究的结果可以看出， WoE， FR， LR 和 SI 模型对滑坡敏感性的预测率分别为 92.7％、86.9％、 81.2％和 79.5％。 其中， WoE 具有最高的 AUC 值（92.7％），而 SI 的 AUC 值最低（79.5％）。此外，卢旺达 20.42％的（5,048.07 平方公里）区域在模拟结果中为高度易受滑坡影响区，西部地区与其他地区相比更为敏感。 因此，通过对生成的四幅地图的比较可以看出，所有应用模型对卢旺达滑坡敏感性研究都是可靠的。本研究的结果可能对研究区域和具有相似地形和地貌条件的其他区域的滑坡风险缓解有很大的价值。在今后的研究中应该考虑更多的与敏感性制图相关的其他重要条件和因素，尤其是人为因素和降雨强度。该研究提供的结果可为今后的滑坡风险降低和管理提供相关的参考依据。 |
| Other Abstract | Being a land locked nation, Rwanda is situated in the east-African region. Thecountry is highly vulnerable du climate variability resulting into recurrent and disastrousevents. Rwanda hazard profile is mainly dominated by floods, landslides, windstorms,thunderstorms, localized droughts, volcanic activities and earthquakes. These natural hazardsinduce a serious web of impacts including death, injuries, property and infrastructuredamages as well as environment destructions. Predominantly, the country faces intense andfrequent slope movements which result into considerable impacts. For sustainable landsliderisk reduction and management, susceptibility mapping must be conducted to detect proneareas and inform disaster risk monitoring. Landslide susceptibility analysis and mapping isknown to be the critical phase of landslide risk reduction and management. Thus, theapplication of appropriate methods to produce accurate susceptibility maps (LSM) is veryrecognizable in landslide risk management. Therefore, various approaches exist to handle thistask. Therefore, this study mainly aimed to effectively conduct a detailed landslidesusceptibility modeling for Rwanda using exploration and comparative analysis ofsemi-quantitative models (SMCE-Spatial Multi-Criteria Evaluation method), statisticalmodels (weights of evidence methods-WoE, statistical index-SI, frequency ratio-FR andlogistic regression-LR) and the process-based methods (the stability index mapping-SINMAPmodel) and assess their performances. Rwanda was used as a case study to carry outexperiments and field investigations. Overall, 15 maps of landslide causal factors wereprepared and applied to generate landslide susceptibility maps including: topographicwetness index (TWI), slope degree, elevation, distance from main roads, slope aspect, landcover land use, lithology, topographic factor (LS), the normalized difference vegetation index(NDVI), precipitation, curvature, distance from rivers/streams, seismicity/earthquake, soiltexture and soil depth.The SINMAP model used the soil physical parameters (internal friction, conductivityand cohesion angles) in combination with the 30m spatial resolution digital elevation model(DEM). Based on these soil physical aspects, the factor of safety (FS) was calculated. Thus theSMCE approach employed seven layers of conditioning factors including land-use/ land cover,lithology, soil depth, slope, precipitation, seismicity/earthquake, and soil texture. In total, 980historical landslide sites have been identified and mapped in the study area through fieldsurveys and past records to derive the inventory map. This landslide inventory was applied to build the model (training points) and validate the model‘s performance (testing datasets). Forthe SMCE model, the spatial correlation between the inventory map and the conditioningfactors was calculated.The results of the first analysis showed about 42.3% of the area under study iscategorized from moderate susceptibility to very high susceptibility. It can be seen thatsusceptibility to landslide is geographically dispersed nationwide and these susceptibilityclasses account for 49.3% of the total population. Moreover, regions falling into very high tohigh susceptibility are Western, Northern and Southern provinces (40.4%, 22.8% and 21.5%,respectively). However, the Eastern Rwanda was found the low susceptible to landslidehazards (87.8%) and has no area falling into very high susceptibility (0%). As evidenced bythe findings of this research, the used approach generated reasonable and accurate results interms of landslide susceptibility. Besides, 49.5% of historical landslide locations fell into thevery high susceptibility class, which testifies the model‘s predictive capability andperformance for the study area.The comparison of SMCE and SINMAP models showed that the SMCE methodproduced better performance in predicting susceptibility than SINMAP method. Thus, thereceiver operating characteristic (ROC/AUC) and other statistical estimators, includingaccuracy, precision, and the root mean square error (RMSE) were used for validation andcomparison of the predictive capabilities of the models. Therefore, the AUC values were87.92 and 78.09% respectively for SMCE and SINMAP. In addition, the SMCE modelgenerated better accuracy and precision values of 0.77 and 0.734 respectively. For the RMSEvalues, the SMCE produced better prediction than SINMAP (0.332 and 0.398 respectively).The overall comparison of results confirmed that both SINMAP and SMCE models generatedreasonable results.Finally, the comparison of four statistical and probabilistic models (FR, SI, LR, andWoE) was done to generate the landslide susceptibility maps for Rwanda and the validation ofthe derived maps of landslide susceptibility was done using the receiver operatingcharacteristic curves (ROC/AUC). From the findings of this research, it was revealed that therates of susceptibility prediction were 92.7%, 86.9%, 81.2% and 79.5% for the weights ofevidence, frequency ratio, logistic regression and statistical index methods, in a respectiveorder. Obviously, the WoE model has attained the peak AUC value (92.7%) whereas thestatistical index yielded the least value of AUC (79.5%). Furthermore, 20.42% of the area under investigation (5,048.07km2) fell into high landslide susceptibility class with the westernRwanda being the very vulnerable relating to other parts of Rwanda. Consequently, based onthe comparison of the four generated maps, the present research showed that all usedapproaches are reliable susceptibility methods for Rwanda. The study outcomes may bevaluable and applicable for geological hazard risk prevention and mitigation in the study areaand in other parts of Africa presenting equal topographic and geomorphic settings.Supplementary researches are highly recommended to incorporate other significant landslideconditioning factors pertinent for susceptibility modeling notably human-induced predictionfactors as well as rainfall intensity factor. The current scientific research has provided resultsthat can be instrumental and baseline for further initiatives landslide risk reduction andmanagement initiatives. |
| Subject Area | 地图学与地理信息系统 |
| Language | 英语 |
| Document Type | 学位论文 |
| Identifier | http://ir.xjlas.org/handle/365004/15276 |
| Collection | 中国科学院新疆生态与地理研究所 研究系统 |
| Affiliation | 中国科学院新疆生态与地理研究所 |
| First Author Affilication | 中国科学院新疆生态与地理研究所 |
| Recommended Citation GB/T 7714 | Jean Baptiste Nsengiyumva. 中东非洲卢旺达滑坡易感性模拟[D]. 北京. 中国科学院大学,2019. |
| Files in This Item: | There are no files associated with this item. | |||||
Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.
Edit Comment