| Landslide is a natural phenomenon in which the soil or rock on a slope,affected by river erosion,rainwater immersion,earthquake and other factors,slides down the slope as a whole or scattered along a certain weak surface or weak zone under the effect of gravity.Especially after strong earthquakes,which trigger landslides of large number and scale,Ludian County is located in Zhaotong City,Yunnan Province,and on August 3,2014,a Ms6.5 earthquake occurred in Ludian County,causing a total of617 deaths,and as of 2022,eight years have passed since the Ludian earthquake,but for a long time,the evolution of landslides triggered after this earthquake is still unknown,and continuous monitoring of the long time series dynamic evolution of landslides after earthquakes It will not only help to improve the preventive measures before the earthquake,but also be indicative for long-term post-earthquake recovery,land planning,and seismic hazard prevention and control.To this end,the thesis takes the Ludian earthquake as the research object,uses a variety of remote sensing images as the data source,performs coarse identification of post-earthquake landslides based on the deep learning module of the ENVI software platform,followed by the use of topographic slope features,geometric features,and texture features for the rejection and correction of misidentified landslides,obtains landslide fine identification results,and constructs a multi-temporal post-earthquake landslide data inventory.On this basis,firstly,the spatial and temporal evolution of landslides is analyzed from three perspectives: spatial and temporal distribution characteristics,spatial and temporal distribution patterns and spatial and temporal area changes,and secondly,the degree of sensitivity of landslides to various factors is analyzed by selecting 12 factors among seismic factors,topographic factors,geological factors,human activity factors and river factors.The main findings of the paper are as follows:(1)An inventory of multi-temporal post-earthquake landslide data was constructed.The results of the recognition were verified using precision,recall,F1 score,wrong score and missed score for four selected sub-regions.The results show that the accuracy rate of each region is above 85%,the highest recall rate is 84.39%,the F1 score of landslide is above 80%,the highest is 85.06%,the phenomenon of wrong score is not obvious,although there are some missed scores of landslide,but the overall identification accuracy of landslide is good.Finally,according to the whole identification process,513,374,466,386,222,147,188,107,and 59 landslides in October 2014,2015,2016,2017,2018,2019,2020,2021,and 2022 were extracted,and a list of multi-temporal landslide data was established.(2)Analysis of the spatial and temporal evolutionary characteristics of landslides.The spatial and temporal evolutionary characteristics show that the subsequent landslides show an overall decreasing trend in number area compared to the same earthquake landslide in August 2014.The landslides in all years showed obvious clustering patterns and clustering areas both globally and spatially locally,with clustering points located in the northeast area of Predigou Village,the northeast and southeast corners of Babao Village,the south and southwest corners of Mule Mouth Community,and the northeast and southeast directions of Longjing Village.And over time,the number of landslides in the study area showed a cold spot,and the total area showed a gradually decreasing hot spot,specifically the total area around the large red rock landslide area has been the hot spot.Using the August 2014 co-earthquake landslide as the baseline for comparison,by landslide activity area and activity rate,the landslide area and activity rate in general showed a trend of first decreasing and then increasing and then continuously decreasing in the 8-year post-earthquake time,and as of 2022,the active landslide has been reduced to 6.08%,and the co-earthquake landslide is gradually recovering.(3)The sensitivity of landslide impact factors was analyzed.Through impact factor selection,impact factor grading,and also based on Certainty factor(CF)analysis method,the area of isoseismic landslide,expanded landslide,new landslide and restoration area in different factor grading intervals are counted,and their CF values are also calculated.For coseismic landslides,the CF values of distance from the epicenter and peak ground acceleration(PGA)increase and then decrease with increasing distance,indicating that the sensitivity tends to increase and then decrease with increasing value.The CF values increase with increasing elevation,slope,and topographic relief,and the factor sensitivity increases.When the seismic intensity isⅨ,the slope direction is south,the distance from the fault is 2000-2500 m,the lithology is dolomite,and the distance from the settlement is 600-1400 m,the factor sensitivity is strongest.When the seismic intensity is Ⅸ,the slope direction is south,the distance from the fault is 2000-2500 m,the lithology is dolomite,and the distance from the settlement is 600-1400 m,the sensitivity to the factor is the stronges.In addition,the closer the distance from the road and river,the larger the CF value is,the stronger the sensitivity to the landslide is.The sensitivity of expanding landslide to elevation,distance from road and distance from river tends to decrease gradually with the increase of factor value.While within a certain range,the sensitivity of expanding and new landslide increases with the increase of slope and topographic relief,and the trend of landslide expansion and addition is obvious.For other influencing factors,the sensitivity is less regular,with jumpiness and diversity.In addition,analysis of the distribution of the restored area among the factor classifications revealed that the distribution trends corresponded to and were similar to those of the co-seismic landslides among the various factor classifications,indicating that the co-seismic landslides among the factors showed a gradual restoration. |
