Depicting impervious surfaces as a continuous variable: estimates and uncertainties using lnear spectral mixture analysis and random forest

Abstract

Impervious surfaces correspond to the areas where soil water infiltration is impeded due to the paving of roads, sidewalks, parking lots, buildings, roofs, etc. Impervious surface area (ISA) constitutes a major human alteration of the land surface, playing important impact on the deterioration of environmental quality. Because of its relevance, this project aims at mapping ISA in the metropolitan region of Sao Paulo by using remotely sensed dataset and image processing techniques. To this end, the methodology explores the synergism of the detailed land information obtained from the high-resolution aerial photography and the broad coverage of the multispectral images (in terms of area and frequency) generated by the OLI instrument, onboard the Landsat-8 satellite. Owing to the great diversity of material presented in the urban plot, which may hamper an accurate classification, this study treats ISA as a continuous variable of soil impermeability. We will test two techniques for estimating ISA: the linear spectral mixture model and the random forest algorithm. The first model is well known in remote sensing applications, but it is still little used for mapping impervious areas. Random forest, in turn, is the state-of-the-art of the predictive modelling, and study addressed to mapping water percolation at the intra-pixel scale is lacking. (AU)