A multiprocess model of adaptable complexity for impervious surface detection

Accurately estimating impervious surfaces is of considerable importance towards understanding human impacts on the environment. In this article, a sequential multiprocess model is presented to estimate the distribution of impervious surfaces using Landsat Enhanced Thematic Mapper Plus (ETM+) satellite imagery. Unlike typical single-thread classification processes, the multiprocess model consists of a series of classification steps. By separating the classification task into several sub-tasks of variable difficulty, multiple classifiers are integrated while adapting to different levels of complexity. The results show statistically significant improvements for the proposed model. An interesting finding is the uneven distribution of simple classifiers in either highly rural or suburban/urban areas, suggesting that the classification difficulty is concentrated in rural areas. The proposed methodology also supports spatially explicit accuracy metrics, facilitating incorporation of the results obtained in interdisciplinary modelling efforts and providing a guide for further algorithmic refinement.