Sub-pixel Target Mapping from Soft-classified, Remotely Sensed Imagery

A simple, efficient algorithm is presented for sub-pixel target mapping from remotely-sensed images. Following an initial random allocation of “soft” pixel proportions to “hard” subpixel binary classes, the algorithm works in a series of iterations, each of which contains three stages. For each pixel, for all sub-pixel locations, a distance-weighted function of neighboring sub-pixels is computed. Then, for each pixel, the sub-pixel representing the target class with the minimum value of the function, and the sub-pixel representing the background with the maximum value of the function are found. Third, these two sub-pixels are swapped if the swap results in an increase in spatial correlation between sub-pixels. The new algorithm predicted accurately when applied to simple simulated and real images. It represents an accessible tool that can be coded and applied readily by remote sensing investigators. Introduction Land cover is a fundamental variable that underpins much scientific research. For example, data on land cover are required to provide boundary conditions for climate models (e.g., global circulation models (van den Hurk et al., 2003)) and hydrological and hydraulic models (e.g., the provision of spatially distributed friction coefficients) (Mason et al., 2003; Wilson and Atkinson, 2003). Despite their importance, informative, and accurate data on land cover are both difficult and expensive to provide. Therefore, much of the land cover data currently being used in scientific research are of inadequate quality. For example, much land cover data may be (a) incomplete spatially, (b) out-of-date, or (c) inaccurate. Remote sensing is capable of providing synoptic and complete coverage of potentially very large areas. Furthermore, multi-temporal images can be used to monitor changes in land cover over time. For these reasons, remote sensing has been of great value for land cover mapping and monitoring. Despite the obvious utility of remote sensing for land cover mapping, many problems remain. For example, it is often difficult to ensure appropriate spatial coverage at an appropriate spatial resolution. Furthermore, it can be difficult to provide temporal coverage with sufficient frequency for monitoring purposes because of the limited revisit time of the satellite, or obscuration of the scene due to persistent cloud cover (e.g., in tropical regions). Even where sufficient spatial and temporal coverage is possible, the accuracy of classification algorithms is frequently limited to around 80 to 90 percent (Congalton, 1991; Foody, 2002). Increasing the accuracy of land cover classification has been the subject of intensive research for many years Sub-pixel Target Mapping from Soft-classified, Remotely Sensed Imagery