Research On Impervious Surface Extraction Method For Complex Urban Areas Based On Multi-source Remote Sensing Data

Impervious surface is a type of artificial surface through which surface water cannot penetrate into the soil.It mainly includes buildings,roads,and squares.The rapid expansion of urban impervious surface causes urban heat island,flood disaster and water pollution and other ecological and environmental problems.Therefore,high precision remote sensing monitoring of impervious surface is of great significance for urban planning and urban ecological and environmental management.Remote sensing has the advantages of high efficiency,low cost and large area synchronous observation,and has become an important means of automatic extraction of urban impervious surface.In recent ten years,with the rapid development of earth observation technology,various satellite remote sensing data(such as optics,radar,night light,etc.)have been widely used in urban impervious water extraction.However,due to urban planning policies,culture and geographical environment,urban impervious water surface is complex and diversified,and a single remote sensing data source is difficult to solve the problem of mixing of different types of bare soil and impervious water surface and the interference of building shadow.In this study,the advantages of optical,radar,night light and other modal remote sensing data will be comprehensively utilized to propose the impervious surface extraction method considering building height and construct the impervious surface index based on logistic regression analysis method.Sentinel-1/2and NPP-VIIRS were used as data sources,Google Earth Engine cloud platform was used as experimental computing platform,and Beijing,Shanghai,Guangzhou,Wuhan,Xi’an and Kunming were selected as research areas to verify the extraction accuracy and effectiveness of the proposed method.The main research achievements of this paper include the following aspects:1.Extraction method of urban building height based on Sentinel-1/2 data.Firstly,Sentinel-1A/B and Sentinel-2A/B time series data were used to extract multi-dimensional features such as spectrum,polarization,time and space.Then,a Support Vector Regression(SVR)model with multidimensional features and building height information was constructed.Finally,the model is used to extract the building height and verify the extraction accuracy of the model.The experimental results show that:(1)the prediction results of this model have good stability,with the root mean square error(RMSE)ranging from 6.21m to 9.05m and R~2 ranging from 0.50 to 0.59.(2)The height of buildings in the six cities decreases gradually from the city center to the city edge,and the height of buildings along the higher-grade streets is higher.2.Urban impervious water extraction method considering building height characteristics.By using the remote sensing data of Sentine-l A/B,Sentine-2A/B and NPP-VIIRS,the polarization,spectrum and building height features were extracted,and the multi-dimensional features were formed by stacking.The impervious water surface was extracted based on the multi-dimensional features and Random Forest(RF)classifier.The experimental results show that:(1)The identification accuracy of the impermeable surface can be effectively improved through the fusion of spectral,polarization and building height features,and the overall accuracy of the six urban areas is above 95%.Compared with the method based on the two-dimensional feature extraction of spectrum,polarization and texture,the overall accuracy is improved by about 3%on average,and the error rate and leakage rate are reduced by about 2.5%.(2)The relative importance of multiple features was analyzed through the feature contribution estimated by random forest classifier,and the conclusion was drawn that the building height feature had a great contribution to the improvement of classification accuracy;(3)Building height characteristics can effectively improve the mixing between buildings and bare soil,and inhibit the interference of building shadows on impervious water extraction.3.Construct the urban impervious surface remote sensing index based on logistic regression analysis method.Firstly,SEa TH algorithm is used to analyze the optical image features and SAR image features of urban surface objects.Then,a VISWac waterpermeable surface index model based on vegetation index(NDVI),urban index(PII),bare soil index(BAI),water index(m NDWI),Anisotropy and coherence coefficient(Coh)was established by logistic regression.The results show that:(1)Compared with other indexes(NDBI,ENDISI,PII,PISI,BCI,BANUI,MISI,LISI),the overall accuracy of the method in this paper is more than 92%,and the average overall accuracy is about 2.75%higher than that of the index model with the best effect(BANUI,LISI).The misseparation and leakage rate is reduced by about 3.5%,which effectively reduces the mixing of bright bare soil and impervious surface of metal material.(2)Compared with the existing land use products(GLC＿FCS30,GISD30,Esri＿Land＿Cover and World Cover),the index model can completely extract the road contour and reduce the mixing with dark bare soil and vegetation.The method presented in this paper can achieve high precision extraction of impervious surface in complex urban areas,and provide data basis and technical support for urban planning and ecological environment management.