A Manova-based and Object-oriented Statistical Method for Extraction of Impervious Surface Area

Impervious surface area (ISA), one of the consequences of suburban sprawl, has emerged as a key indicator to explain and predict ecosystem health in relationship to watershed management. Quantifying precisely the spatial locations and distributions of ISA is essential for environmental monitoring and management. Classification of high spatial resolution remote sensing data is an important step towards obtaining ISA information. In this study, we developed a Multivariate Analysis of Variance (MANOVA)-based classification algorithm for the purpose of extracting ISA information from high spatial resolution remote sensing data. This classification algorithm took account the variability in both the training objects and the objects to be classified, as well as the correlations among different spectral bands in high spatial resolution remote sensing data. We tested the algorithm using three types of high spatial resolution imageries including true color Orthophoto, QuickBird-2 and IKONOS satellite imagery data. Based on this algorithm, we extracted ISA from the high spatial resolution orthophoto data for the state of Rhode Island. The result indicates that 10% of the state land are covered by the ISA. Ten towns have ISA percentage over 20%. Twelve towns have ISA percentage between 10% and 20%. Only sixteen towns in the state have ISA percentage less than 10%. The distribution patterns indicate that the ISA are mainly concentrated along the coastal lines in the southern and the eastern sections of the state. The extracted information of ISA provides the most updated and precise information for coastal and watershed management, as well as for environmental monitoring and modeling.