| Under the background of global climate change,the damage caused by rainfall induced landslides has increased significantly in recent years.It is of great significance to establish the meteorological early warning system of landslides for regional disaster prevention and reduction.However,in the mountainous area with high incidence of landslide disasters(especially in the alpine canyon area),due to the limited data,the existing early warning model has many problems,such as poor spatial indication and lack of comparative analysis data,resulting in low early warning accuracy and high false alarm rate of the established landslide meteorological early warning model.Therefore,the establishment of an effective landslide meteorological early warning model in the mountainous area with scarce data is the key problem to be solved in landslide disaster prevention and reduction.Taking the middle and lower reaches of Bailong River Basin,one of the four high-risk areas of geological disasters in China,as an example,the remote sensing satellite data of tropical rainfall measurement mission(TRMM)with a resolution of 0.25°(about 27 km)is downscaled to a daily resolution of 1 km by geographical weighted regression method,and fused with the rainfall records of ground stations to obtain high-precision rainfall data;At the same time,the temporal and spatial catalogue of rainfall induced landslide events in the area is carried out.On the basis of revealing the temporal and spatial distribution characteristics of landslide,the rainfall threshold curve based on Bayesian theorem and frequency method is established;Further integrate the landslide sensitivity map based on depth learning,and construct the regional landslide meteorological early warning model based on rainfall threshold;Finally,the recent mass landslide events in the study area are used to verify the accuracy of the model.The main research results are as follows:(1)In order to overcome the scarcity of rainfall data and the lack of rainfall landslide data in Alpine canyon area,1 km daily rainfall data is obtained based on geographic weighted regression method and conditional fusion method.According to the accuracy evaluation index,compared with downscale rainfall and rain gauge interpolation data,the conditional fusion data with resolution of 1 km daily rainfall data has better overall performance and higher accuracy.In addition,combined with data investigation and remote sensing interpretation,80 rainfall induced landslides with detailed time and place records and 120 mass landslides for verification in the study area from 2003 to 2019 were collected.Scientific rainfall data preprocessing and landslide data collection lay the data foundation for the subsequent rainfall threshold research.(2)Based on the high correlation between rainfall events and rainfall induced landslides,the critical rainfall inducing landslide rainfall events is studied.The rainfall duration,rainfall intensity,cumulative rainfall and previous rainfall induced by landslide are extracted.Based on one-dimensional Bayesian model,the a priori probability of rainfall induced landslide is studied.The results show that the interpretation ability of rainfall duration and cumulative rainfall variables is higher.Based on the frequency method,the threshold curve T5 of 5%in the rainfall threshold grade curve is E=(5.01±0.06)*D(0.64±0.09);The threshold curve T20 of 20%is E=(7.08±0.67)*D(0.64±0.09);The threshold curve T50 of 50%is E=(15.14±1.15)*D(0.64±0.09).Combined with frequency method and Bayesian theorem,event rainfall and duration threshold curves with probability of 5%,20%and 50%are generated in double logarithmic coordinates respectively.The results show that when the probability of landslide occurrence is 50%,there are differences between them,but the overall trend is the same.(3)Considering the influence of spatial difference of geological environment background on rainfall threshold curve,landslide sensitivity zoning evaluation is carried out.Based on the analysis of the spatial distribution law of landslide,this paper reveals the influence and contribution of landform factors,geological factors,hydrological factors and human activities on the development of landslide,and further constructs the landslide sensitivity evaluation model based on one-dimensional convolution neural network model.High susceptibility areas are mainly in areas with strong tectonic activities,soft rock strata and frequent human activities,and are mainly distributed in Wudu,along the main stream and main tributaries of Bailong River.The overall accuracy evaluation results show that the area under the curve(AUC)of the model is 0.95.(4)By combining the sensitivity zoning of rainfall threshold with temporal information and spatial information,the landslide meteorological early warning model in the study area is established,and the accuracy of the model is tested by typical rainfall landslide events.The results show that the early warning ability of the model combined with rainfall threshold and landslide sensitivity map is better than that of the model using rainfall threshold alone,and the false alarm rate is low.The research results deepen the understanding of the disaster mechanism of rainfall induced landslide,promote the development of landslide meteorological early warning method system,and provide scientific and technological support for early warning and risk management of geological disasters in mountainous areas with lack of data. |
PDF Full Text Request