| Landslide is one of the most threatening geological hazards,and China is severely affected by landslides in the world.The area is located in a subtropical region with dense mountainous terrain and abundant rainfall,which provides good conditions for breeding landslide hazards.At present,global warming is exacerbating the instability of the climate system,and the frequency and intensity of extreme hazards such as regional heavy rainfall is increasing,leading to a trend of increased landslide risk in the future.An in-depth study on landslide susceptibility under different future climate and extreme rainfall patterns is conducted to explore the spatial distribution and changing trends of future landslides in the subtropical region.It can provide a theoretical basis for landslide management,landslide prevention and landslide disaster loss reduction.Therefore,this paper selects Hunan and Yunnan provinces in the subtropical regions of China as the study area,and collects landslide influencing factors to evaluate the landslide susceptibility.On this basis,the CMIP6(Coupled Model Intercomparison Project(Phase 6))rainfall dataset was used to obtain the future rainfall distribution in the study area under different models and to simulate the future landslide susceptibility trends in the study areas by combining the results of landslide susceptibility assessment.The main work and innovations of this research are as follows:(1)In this thesis,we selected raster cells as the mapping unit according to the characteristics of the study area,and selected 11 types of landslide influencing factors that are representative of the study area to build a landslide susceptibility evaluation index system.At the same time,to comprehensively consider the influencing factors of landslides and highlight the degree of contribution of rainfall elements to landslides,this thesis proposes a landslide susceptibility evaluation method combining machine learning and I-D threshold model.Three machine learning models,logistic regression,random forest and support vector machine,were used to evaluate the initial landslide susceptibility,and the ratio of the simulated rainfall to the critical rainfall threshold was used to determine the weight,fusing machine learning with the I-D threshold model to evaluate the results.The receiver operating characteristic curve showed that the RF model combined with the I-D thresholds had the highest accuracy in evaluating the results.(2)This thesis uses the CMIP6 historical rainfall data from 1990 to2020 and the World Clim historical rainfall data in the same period to downscale the daily rainfall data under three future development scenarios(SSP1-1.9,SSP2-2.5,SSP5-8.5)to improve its spatial resolution.The spatial resolution was increased to km to obtain the future extreme rainfall distribution,and then evaluate the future landslide susceptibility to explore the future trend of landslide susceptibility.This thesis classifies the landslide susceptibility of the study area from high to low into five classes:very high,high,medium,low and very low,and counts the area of each class of susceptibility.The results show that:(1)Changes in the distribution of landslide prone areas are evident: the simulation data results for 2021-2060 show that the trend of landslide prone changes within Hunan and Yunnan provinces is more obvious,and the area of landslide prone areas in both provinces generally shows an increasing trend under different carbon emission models.Under the carbon emission model SSP5-8.5,the landslide prone area in Hunan Province peaks at 171,730 square kilometers in 2040,an increase of about 13,200 square kilometers or 8.35% compared to the base period.The peak area of landslide prone areas in Yunnan Province occurs in the SSP1-1.9 model,and compared to the base period,the area of landslide prone areas in the province in 2045 is 340,870 square kilometers,an increase in area of approximately 1.84%.(2)The area of landslide prone areas in each administrative region of the two provinces also shows an increase in all three models.The area of landslide prone areas in Hunan Province also shows an overall increasing trend over the next half century,with a total average area increase of 4504 square kilometers,which is an average increase of 2.84% compared to the current total area of landslide prone areas in Hunan Province.The two carbon emission scenarios(SSP1-1.9 and SSP5-8.5)have the greatest impact on the landslide susceptibility in Hunan Province,with an average increase of 3.93% and 2.71% in landslide susceptibility area,respectively.The three regions with the largest average increase in landslide prone area in Yunnan Province under the three carbon emission scenarios are Wenshan,Dehong and Xishuangbanna.And the total average increase in landslide prone areas in Yunnan Province in the next half century is about 5001 square kilometers,with an average increase of 1.51%.The SSP1-1.9 and SSP5-8.5 models have the largest increase in landslide prone area in Yunnan Province in the next half century.(3)The areas of the two provinces with large increments of landslide prone areas are relatively concentrated in one local area.In Hunan Province,the areas with large increments of landslide prone areas are concentrated to the east of the Yueyang-Yongzhou line,while in Yunnan Province they are concentrated to the south of the Dehong-Honghe line. |
