Automatic Extraction Of Underlying Surface Based On RS Technology And Flood Simulation

As an important technology for predicting the inner raft and optimizing the rainwater pipe network in the construction of sponge city,the flood model is of great significance.The extraction of underlying surface vector data is the basis of flood model construction.However,the mixed pixels of medium-low spatial resolution RS data,and the spectral limitations of high spatial resolution RS data,which lead to a phenomenon of “the same material has different spectrums,different materials have the same spectrum” in the process of automatically extracting the underlying surface data.This study is based on RS technology and GIS technology.The spectral,texture and geometric feature layers of the RS data are expanded and optimized,and object-oriented method is used to automatically extract the underlying surface of the study area.Based on Digital Water and Arcgis platforms,the flood models are constructed based on pixels and objects respectively.The simulation results are compared and verified with the measured datas in the D district of Rongqiao City,Nan'an District,Chongqing.The main research contents and results are as follows:(1)Combining the principle of sensor,the characteristics and application range of different types of RS data were compared and studied.In order to obtain acceptable signal-to-noise ratio,the spatial resolution and spectral resolution of RS data are mutually constrained.The digitization process of pixels,the general storage format corresponding to pixels and the generation process of objects were analyzed.(2)After obtaining high spatial resolution RS data of the research area from Google Earth,the RS data was preprocessed.The pre-processed RS data was subjected to K-means iterative clustering,and automatic extraction of the underlying surface based on pixels was completed.The results showed that the proportion of the surface of the impervious surface was 33.079%,and the proportion of forest land and grassland in the permeable area were 19.037% and 47.884%,respectively.Among them,grassland and undeveloped land produced mixed phenomenon,and some areas had the phenomenon of “the same material has different spectrums,different materials have the same spectrum”.(3)To compensate for the spectral limitations of RS data and achieve object-oriented automatic extraction of underlying surface.Firstly,the band data layer Band U was expanded in the IDL language environment.Subsequently,Based on the eCognition Developer platform,the object layer generated by the fractal network evolution method was utilized and the optimal combination analysis of the features under different dimensions was performed,and the seven features of Band U,Border Index,Band 3,GLCM-A,Compactness,Band 2,and Band 1 were optimized.Finally,based on the preferred feature data layer,the K-nearest neighbor calculation was used to extract the underlying surface to achieve refined extraction of 982 terrains in the study area.The statistical calculation results showed that grassland accounted for 41.719%,the impervious surface accounts for 32.697%,the undeveloped land accounted for 4.581%,and the forest land accounted for 21.003%.In order to apply the two extraction methods to large-scale areas,the underlying surface of the Nan'an area was extracted.Based on the extraction results of the pixels,the grassland accounted for 29.039%,the impervious surface accounted for 21.534%,and the undeveloped land accounted for 2.670%,forest land accounted for 46.757%.Based on the object-oriented method,the results showed that the grassland accounted for 27.492%,the impervious surface accounted for 22.666%,the undeveloped land accounted for 4.020%,and the forest land accounted for 45.822%.(4)Based on Digital Water and Arcgis software,a rain flood model based on underlying surface vector data was constructed.Firstly,the raster data was converted.The extracted geocoding was called,and the underlying surface property library was automatically created using the Python language.Subsequently,meteorological and hydrological data were imported and the construction of the rain flood model was completed,and the rainfall sequence from September 14 th,2018 to 14:15~16:15 was used as the simulation scenario.Based on pixels,the peak flow error was 12.34%,the peak time was 15:30,and the Nash efficiency coefficient was 0.842.Based on object-oriented method,The peak flow error was 9.56%,the peak time was 15:30,and the Nash efficiency coefficient was 0.917.Compared with the flood model constructed by the former,based on object-oriented method,the peak flow error was reduced by 2.78%,and the Nash efficiency coefficient was increased by 8.91%.