Weighting Function Alternatives for a Subpixel Allocation Model

This study investigates the “pixel-swapping” optimization algorithm proposed by Atkinson for predicting subpixel landcover distribution. Two limitations of this method are assessed: the arbitrary spatial range value and the arbitrary exponential model for characterizing spatial autocorrelation. Various alternative weighting functions are evaluated. For this assessment, two different simulation models are employed to develop spatially autocorrelated binary class raster maps. These rasters are then resampled to generate sets of representative medium-resolution class maps. Prior to conducting the subpixel allocation, the relationship between cell resolution and spatial autocorrelation, as measured by Moran’s I, is evaluated. It is discovered that the form of this relationship depends upon the simulation model. For all tested weighting functions (Nearest Neighbor, Gaussian, Exponential, and IDW), the pixel swapping method increased classification accuracy compared with the initial random allocation of subpixels. Nearest Neighbor allocation performs as well as the more complex models of spatial structure.