Subpixel Urban Land Cover Estimation: Comparing Cubist, Random Forests, and Support Vector Regression

Three machine learning subpixel estimation methods (Cubist, Random Forests, and support vector regression) were applied to estimate urban cover. Urban forest canopy cover and impervious surface cover were estimated from Landsat-7 ETM imagery using a higher resolution cover map resampled to 30 m as training and reference data. Three different band combinations (reflectance, tasseled cap, and both reflectance and tasseled cap plus thermal) were compared for their effectiveness with each of the methods. Thirty different training site number and size combinations were also tested. Support vector regression on the tasseled cap bands was found to be the best estimator for urban forest canopy cover, while Cubist performed best using the reflectance plus tasseled cap band combination when predicting impervious surface cover. More training data partitioned in many small training sites generally produces better estimation results. Introduction Describing the urban environment using remote sensing techniques has been an active area of research for many years. One popular method for characterizing urban landcover is the V-I-S (vegetation-impervious surface-soil) model presented by Ridd (1995). The V-I-S model describes the complex urban landscape as a tripartite mixture of the fundamental components of an urban ecosystem: vegetation, impervious surfaces, and exposed soil (ignoring water surfaces). Ridd calls for “a standardized way to define these urban building blocks and to detect and map them in repetitive and consistent terms.” This study compares three machine learning regression methods that can be used to map individual urban land constituents. Mapping landscape components in urban areas using traditional hard classification techniques is impeded by the large proportion of mixed pixels. In moderate resolution imagery, such as Landsat ETM , mixed pixels predominate because of the heterogeneous combination of landscape features that are smaller than the ground instantaneous field of view (GIFOV) of the sensor. Estimating subpixel proportions of land-cover components overcomes the difficulty of assigning a pixel to one thematic class and yields a better Subpixel Urban Land Cover Estimation: Comparing Cubist, Random Forests, and Support Vector Regression