Research-Eon The Extraction Ofing Urban Impervious Surfaces Using From Spectral Information In Synergy With Spatial Heterogeneity Of Landsat Remotely Sensed Imagery And Its Thermal Environment Effect

The distribution of urban impervious surfaces(UIS)acts as an important indicator for assessing human activity intensity and ecological environment change,playing an underlying data support role in the strategic development of urban sustainable management.Remote sensing data is widely used for the extraction of UIS because of its easy data acquisition,continuous monitoring,high resolution,and wide coverage characteristics.Using the Landsat 8 OLI images,acquired on 8/11/2013 and 6/6/2018,based on the Pixel Purity Index(PPI),and an improved spatial pixel purity index(SPPI),four types of endmembers including vegetation,bare soil,high albedo UIS and low albedo UIS were identified.Then,based on the extracted two suites of endmembers derived from PPI and SPPI respectively,the linear mixed spectral mixed model,the bilinear mixed spectral model,the BP neural network and the support vector machine were implemented to extract UIS in Nanjing,followed by a validation of the extracted UIS abundance using visually interpreted results of 50 random sample windows with a size of 90 m×90 m from high spatial resolution Google Earth images in 2018.Additionally,the Landsat 8 TIRS images coupled with the thermal radiation transfer equation were used to retrieve surface temperature of Nanjing City,and combinedfollowed by a linear regression validation with the hourly temperature statistical dataobservations of the 5 discrete Mmeteorological Administrationstations for linear regression.The results were summarized as follows:(1)Using the spatial heterogeneity of the panchromatic band in partnership with multi-spectral information was able to more effectively and accurately extract endmembers,and to reduce the computational workload simultaneously.(2)Based on two suites of endmembers extracted from PPI and SPPI,the Linear mixed spectral model,bilinear mixed spectral model,BP neural network and support vector machine algorithms were applied to extract UIS.In 2018,the SPPI-based support vector machine method performed best,with the highest accuracy at 91.39%,followed by the SPPI-based BP neural network(90.45%),and the PPI-based linear mixed spectral model achieved the lowest accuracy at 80.62%.And it was concluded that SPPI based models were always better than PPI based corresponding models in the UIS extraction.In addition,no matter which method used for endmembers extraction,the support vector machine method always had the highest extraction accuracy of UIS,followed by BPNN,bilinear spectral mixed model and linear spectral mixed model.The SPPI-based support vector machine method was finalized to map UIS of Nanjing in 2013 and 2018,and it was found that UIS of Nanjing increased from 28.03%in 2013 to 29.40%in 2018.And Lishui district pioneered the UIS growth rate at 2.62%,the UIS growth rate of Pukou district was at 2.25%,leaving Gulou district,Xuanwu district,Qinhuai district in the downtown region with less than 1%growth rate.(3)Based on the Landsat 8 TIRS remote sensing images,acquired on 2013 and 2018,the surface temperature of Nanjing City was retrieved by using the thermal radiation transfer equation.After linear regression with the hourly temperature data,the coefficient of determination accuracyR square iswas at 0.8594 and 0.8297,respectively.(4)Correlation analysis between the extracted UIS of Nanjing and the surface temperature results of Nanjing city obtained from the inversion in 2013 and 2018 showed that the correlation coefficient of the them were 0.7090 and 0.7268,respectively.And the urban heat island effect needs to be mitigated by increasing urban vegetation,changing the underlying surface of the city etc.