Research On Risk Assessment And Prediction Of Urban Waterlogging In Nanjing Based On Multi-source Dat

In the process of my country’s sustainable development,waterlogging occurs frequently in urban areas,which has seriously affected urban transportation,people’s lives and the ecological environment.Therefore,the prevention and control of urban waterlogging has become the key object of governance in my country in the 21 st century,and it is also the core content of risk business development.However,most of the current studies focus on waterlogging risk assessment,that is,only focus on the intensity of waterlogging risk itself,ignoring the corresponding prediction work.Taking Nanjing as the research area,this paper first obtains the meteorological and geographical data that affect urban waterlogging from 2016 to 2020,including nine influencing factors such as elevation,slope,aspect,impermeable surface,NDVI,and land use type.Load the Sentinel-1 satellite data on the Engine,use the images before and after the disaster to extract the submerged range of Nanjing City from 2016 to 2020,and use the frequency ratio method to analyze the correlation between the submerged range and the impact factors in each year;secondly Use the Analytic Hierarchy Process(AHP)to calculate the internal grading weights of each factor,multiply and add the obtained weights and frequency ratios,and finally obtain the scores of each factor,and obtain the waterlogging risk map through risk calculation,and then compare with the submerged range and waterlogging points Verification and correlation analysis;finally,the selected waterlogged points(184)and nonwaterlogged points(129)were connected with the corresponding attributes of relevant meteorological and geographical data,and then based on the support vector machine,XGBoost model and random forest three The machine learning method constructs the prediction model for the waterlogging points in Nanjing,calculates the accuracy,precision,recall rate,F1 value of the three prediction models and draws the ROC curve to calculate the AUC value,and calculates the confusion matrix of the three models to predict the results Compared.The main conclusions are as follows:(1)In the Landsat-8 remote sensing image with a spatial resolution of 30m×30m,Nanjing City is composed of 6,818,400 grids,and the submerged grids from 2016 to 2020 are 153,238,134,642,77,596,120,222,and 158,079,respectively.Among the various influencing factors,the frequency ratio of building land is the highest,and the frequency ratios from 2016 to 2020 are 5.22,4.99,4.97,3.89,and 3.31,which are much higher than the frequency ratios of cultivated land,forest land,and grassland.Compared with other influencing factors,the calculated results are the most average,so it can be concluded that the influence of slope aspect on waterlogging risk is small.(2)Through the weight calculation of the AHP for each influencing factor,the results show that the land use has the most score,indicating that the land use type has the greatest impact on the waterlogging risk;the slope aspect has the least score,indicating that the slope aspect has the greatest impact on the waterlogging risk Minimal.Then the scoring results of the impact factors are scored by the percentage system.The percentage scores of land use,elevation,slope,aspect,vegetation index,rainstorm,topographic moisture index,slope length and impermeable surface are 20.05%,9.22%,and 6.62 respectively.%,4.15%,17.49%,8.72%,13.17%,12.01%and 8.56%,and finally divide the waterlogging risk results into 5 levels through the risk calculation formula.(3)There is a correlation between the change trend of the risk percentage and the waterlogging percentage.The results show that the higher the risk area,the more waterlogged points will be distributed,but the low-risk area cannot guarantee that the waterlogged point in the area is 0,so the distribution of waterlogged points determines the number of waterlogged points.Depending on the level of the risk level,when the number of waterlogging reaches a certain level,the risk level is higher;at the same time,the probability of waterlogging in the high-risk area is higher through the verification of the frequency ratio between the risk area and the submerged area,which proves the rationality of the risk calculation and feasibility.Finally,the correlation shows that the waterlogging problem is not local,but has a certain spatial continuity and diffusion.(4)Based on the waterlogged points,the XGBoost algorithm,random forest and support vector machine is used to predict them.The accuracy rates of the three are 82.54%,81.39% and75.88%,respectively.better than support vector machines.Finally,the risk division is gridded,and it is concluded that the three algorithms have differences in the medium risk level.Among them,the XGBoost algorithm accounts for 59% and 41% of the predicted waterlogged points and non-waterlogged points;the random forest is 56%.and 44%;SVM is 47% and 53%.