Sub-pixelmapping (SPM; a.k.a. super-resolutionmapping) is a sub-field of remote sensing focused on mapping the spatial distribution of land covers from fractional land cover maps, such as those produced by spectral unmixing. A necessary consideration for developing universally applicable SPM methods is that they are able to function across multiple target scales (i.e. zoom factors). Yet most studies test only a single scale, likely due to the challenges of testing a SPM technique that is transferable across multiple scales. This study develops a simple adjustment to the widely used confusion-errormatrix that can account for the uncertainties introduced when reference data must be aggregated to permit testing multiple downscaling factors. Results show how off-diagonal components (e.g. errors of commission and omission) can be adjusted to account for the overestimation of disagreement that occurs during scale translations. Furthermore, results show how supplemental accuracy can be accounted for between the SPM result and areas with low fractional proportions, which is not possible using currently available methods. This simple yet effective way to compare accuracies across scale factors provides a means to develop and test SPM methods applicable across multiple scales.
ASJC Scopus subject areas
- Earth and Planetary Sciences (miscellaneous)
- Electrical and Electronic Engineering