Remote Sensing and Quantitative Geogrids in PHASES [ Pixel Hyperclusters As Segmented Environmental Signals

PHASES constitute an alternative paradigm for working with multiband image data and layered grids of quantitative multivariate spatial data. PHASES differ substantially from classical counterparts in being geared toward extracting dominant characteristics of spatial mosaics in a condensed manner which facilitates and simplifies storage, analysis, presentation, and distribution. Since most landscapes and images have such a mosaic pattern, the approach is quite broadly applicable. PHASE is a multiple acronym concerning pixel hyperclusters. The PHASES approach is applicable to multiband image data and/or layered (multivariate) grids (rasters) of quantitative spatial data representing phenomena that have a mosaic pattern of patches, with characteristic patch types. The fundamental strategy is to exploit the patchiness for collapsing (compressing) the multiple layers into one layer that captures the patch types through integer identifiers which index patch characteristics in tabular relational files keyed to the identifiers. Such a compression does not require that the number or character of patch types be known, nor does it yield conclusive determinations in this regard. Although some degree of information loss is inherent to PHASE compression, one PHASE layer captures much of the information when there is strong spatial pattern among 250 or fewer patch types. A PHASE (Pixel Hyperclusters As Segmentation Elements) is a heuristic sorting of the cell (pixel) population into groups on the basis of similarity of values in the several bands or layers. In the terminology of multivariate statistics, such a sorting process lies within the purview of clustering, with the groups being clusters. Clustering has a long history of use in the image analysis literature under the umbrella of "unsupervised classification." The current usage is different, however, since the focus is on approximation rather than classification. The "hyper" in "hypercluster" refers to extraction of of considerably more clusters (groups) than has typically been done for purposes of unsupervised classification. Variability between cluster groups is retained as information for the PHASE approximation. Variability within the groups of a PHASE is described statistically. One PHASE layer thus provides an approximating surrogate to the entire dataset for many application purposes, including tentative thematic classification. PHASE embryonics lie in the hypercluster work of Kelly and White with the Khoros Group at the Los Alamos National Laboratory and their Spectrum software for interactive classification of image data, but their intent is fairly limited relative to the conceptual scope of the PHASE approach. In the broader image analysis literature, PHASE compression …