Extracting The Urban Impervious Surface And Its Application

The massive increasing of city impervious surface brought by urbanization has becoming a threat to the city ecological environment. Extracting instant impervious surface information has an essential impact on environmental management, urban planning and region sustainable development. This study is based upon Remote Sensing Cognitive Theory, discussing pixel level city Impervious Surface Area(ISA) and sub-pixel level Impervious Surface Percentage extracting methodology, at the mean while analysis the impervious surface information of Jingjiang. Below is a major concepts and conclusion:(1) Extracting the urban impervious surface area at the pixel level.based on support vector machine Studied which can do well in multi-dimensional,made ground temperature by atmospheric correction, normalized difference impervious surface index by band math, Image feature information by 3*3 windows analysis, to optimize the sample set, and in order to estimate the optimization ability, used composite separation index, precision, kappa. Made the composite separation index from high to low order based on optimization ability of spacial parameters,it was image feature information, normalized difference impervious surface index,ground temperature. But proved the composite separation index of the sample set not mean that would be proved precision. The precision was 89.44% of ISA which extracted from spectrum which never add spacial parameters, the kappa was 0.867, the precision was 90.97% of ISA which extracted from spectrum and ground temperature, the kappa was 0.867,it often mixed the way and bridge and so on, the precision was of ISA which extracted from spectrum and normalized difference impervious surface index 93.10%,the kappa was 0.913,the precision was of them was 96.10%and0.939. So the best done is normalized difference impervious surface index.(2) Extracting the urban impervious surface percentage at the sub-pixel level. Built two control group which have unconstrained, parial-constrained,fully constrained,used root mean square to measure error(RMSE). The normalized be expanded after normalized, but reduce the difference between end-member, made ISP between 0.003-0.004, so whether if normalized was not important. Linear spectral separation model appear to increasing RMSE with more restrictions, but after normalizing the linear regression of low albedo and high albedo ISP, this model emerge the best ISP result with least noisy spot and reasonable distribution.(3) Extracting 2001 year and 2009 year ISA of Jinjiang based on spectrum and image feature information,and calculate landscape pattern index,based on fully constrained extracted ISP. From 2001 to 2009, Jinjiang City, impervious surface extended 26.49km2, gradually formed in the old city as the center, the main trend to expand to the City West and south-west direction along the road of impervious surface.