Research On Monitoring And Prediction Of Flood Disaster In Shouguang City Based On Seminel-1/2 Data

Flooding is one of the most frequent and destructive natural hazards with causing property and life loss.Therefore,monitoring and evaluating flood disasters have important social significance.Geographic Information System(GIS)and Remote Sensing(RS)have become the main means of flood disaster extraction and prediction.Synthetic aperture radar(SAR)systems is the preferred data for flood monitoring due to its characteristics of all-time and allweather,and optical images have high accuracy in ground object classification with rich band information.However,the backscattering characteristics of flood monitoring based on sentinel-1 SAR is unclear,and there are limitations in flood disaster monitoring methods.With Sentinel-1 as the main data and Sentinel-2 as the auxiliary data,this thesis focuses and studies three key issues: backscattering mechanism of flood monitoring using C-band,flood monitoring methods,flood prediction models and risk assessment.The main works and results are summarized as follows:(1)The backscattering characteristics and variation rules of flood monitoring are revealed using Sentinel-1 SAR image.The monthly variation rules of ground objects' backscattering coefficient and the effects of polarization and orbit on the backscattering coefficient of ground objects were obtained.The flood depths were estimated by GIS and hydrological analysis,the change trend of the backscattering coefficients and characteristics of ground objects with different flood depths was obtained.The flood extraction rules of five ground objects based on sentinel-1 image was obtained.The effects of monthly variation and orbit difference of backscattering coefficient on flood extraction were evaluated quantitatively,which provides a basis for selecting pre-disaster reference images and plays the foundation for flood disaster area extraction and degree evaluation based on Sentinel-1 images as main data.(2)A flood extraction and degree evaluation method with supervised classification and image fusion as the core are proposed.A Sentinel-1 SAR and optical image fusion model is proposed using Sentinel-1 SAR images and the fused products as classification features.The image fusion model not only improves the accuracy of image classification,but also enhances the backscattering features of ground objects from different angles.Compared with SAR image classification,the fusion image classification accuracy is improved by 16.15%.Through the flood monitoring based on image fusion and supervised classification(FMIFSC)method,the change information of the backscattering coefficients of ground objects is obtained by detecting the changes of the image classification,and the range and degree of inundation are determined according to the change law of the scattering characteristics of the ground objects.It avoids the defect of the threshold method without covering all land types,and makes up for the deficiency of the traditional flood monitoring by RS without achieving inundation degree estimation.Compared with other commonly used methods,the recognition accuracy of completely submerged area is improved by 26.98% and 5.20%,respectively,and the accuracy of partially submerged vegetation is improved by 5.02% and 24.09%.(3)A flood prediction model based on data-driven and Bagging ensemble is constructed,and the detailed factor design scheme is given.The optimization of rainfall factors was carried out from the three dimensions of previous rainfall,rainfall intensity,and rainfall process,and the prediction accuracy of the optimized rainfall factor scheme was increased by 3.00%.Aiming at the problem of limited rainfall samples,a Bagging ensemble algorithm based on a decision tree is constructed to improve the generalization ability of the model and suppress the influence of noise,whose performance is stable and accuracy is the highest.The flood prediction model was used to simulate two historical and four designed rainfalls,and the prediction accuracy of the model is up to 79.89%.Using Information Gain(IG)theory to analyze the causes of floods,it was found that the total amount of rainfall is the most critical factor leading to floods,and the areas with low vegetation index and close to rivers are prone to flooding.Through the flood disaster simulation and cause analysis as well as rainfall and flood risk assessment,the results show that the area proportion of medium-risk is the highest under the four rainfall scenarios.The spatial distribution of superior-risk and high-risk areas is similar,which is mainly distributed in farmland along the river and a small part of roads.