Multitemporal analysis of urban reflectance

Spatial and temporal changes in urban reflectance have a strong influence on energy flux through the urban environment. Optical sensors on operational satellites provide self-consistent time series of urban reflectance variations, but quantitative analyses are complicated by spectral heterogeneity at sensor instantaneous field of view (IFOV) scales and by temporal changes in illumination and atmospheric conditions. These complications can be minimized by combining a multitemporal radiometric rectification with a physically based reflectance analysis. Spectral Mixture Analysis (SMA) provides a physically based approach to quantifying spatial and temporal changes in spectrally heterogeneous urban reflectance. Multitemporal analysis of Landsat Thematic Mapper (TM) imagery of the New York metropolitan area suggests that urban reflectance can be described with a three-component linear mixture model spanned by high albedo, low albedo, and vegetation endmembers. The topology of the spectral mixing space indicates that mixing fractions are well constrained for the vegetation endmember and that nonlinear mixing occurs primarily between the high and low albedo endmembers. Selection of pseudoinvariant (PIV) image endmembers allows radiometric rectification of multitemporal imagery to a common set of endmembers, thereby minimizing variations in radiance that are unrelated to changes in surface reflectance. Inversion of the three-component linear mixture model for the New York metro area produces robust, consistent fraction estimates for different combinations of rectifications and inversion constraints. Temporal variation of the presumed invariant endmember sites provides a measure of uncertainty for the endmember fraction estimates. The resulting vegetation fraction estimates agree with high-resolution reference measurements to within 10% for a 1996 midsummer validation and PIV endmember fraction estimates vary by less than 7% over the course of the 1996 growing season. In contrast, intraurban spatial variations in vegetation fraction span several tens of percent, suggesting that the measured changes significantly exceed the uncertainty of the estimates. These results suggest that Landsat TM imagery may be used to monitor seasonal to interannual variations in urban reflectance and vegetation abundance. (C) 2002 Published by Elsevier Science Inc.