Pennsylvania Pennsylvania’s Gap Analysis Project is moving along unconventional paths toward the common goals of GAP, and uncovering some novel things en route. We are tying down our land cover map, which is one aspect of GAP that has been approached in fairly conventional fashion. Some of the smaller and more difficult pieces are still being addressed, and the metadata work is also continuing. Unlike many other states, we have clung tenaciously to the idea of basing as much of our other mapping work as possible on thematic mapper data. Coupling Gap Analysis with a more theoretically oriented project on "multiscale statistical approaches to critical areas in watersheds and landscapes" (jointly funded by NSF and EPA) has enabled us to rethink ways of handling such image data. We have used the Spectrum idea (but not Spectrum itself) as a springboard for developing new scenarios of compressed image analysis and portrayal. We hypercluster differently and treat the result as a hybrid dataform between a multispectral image and a grid coverage. Our raster consists of a single "layer" of (byte) cluster IDs linked to relational tables of multispectral properties. This gives us a nonproprietary image-based dataform that accommodates substantial analysis in the table domain while being compatible with ArcView and mappable on an HP DesignJet with ArcPress. With sponsorship by Digital Equipment Corporation and cooperation of other spatial data centers at Penn State, we are putting ten compressed images covering Pennsylvania along with vector coverages of roads, streams, major watersheds, floodplains, physiographic provinces and counties on a single CD-ROM for integration with ArcView or simple imaging via an onboard viewer for PCs. This provides reasonably comprehensive landscape-level viewing of Pennsylvania from a single source. Our renderings of the compressed images have been very well received. The above gives rise to a new version of land cover mapping. We label hyperclusters in eight vegetative cover stages by a kind of supervised classification of centroids which does not require that individual pixels be processed. "Training clusters" are chosen from an image display, and suitability of thematic assignments is likewise assessed interactively. The development aspect of land use is captured by on-screen digitizing and analytically combined with the vegetative cover raster to give a four-class vector overlay showing high intensity developed, low intensity developed, rural nonforest, and rural forest. A further breakdown of vegetation with regard to forest types is on tap for summer 1997. Our mode of accuracy checking is through aerial videography. The aforegoing products have considerable spatial detail, from whence comes our final challenge of how to conduct a reasonable map generalization to the level that vectorization becomes feasible. Here again, our other project kicks in with its "multiscale" component. Our habitat models for birds, mammals, and herps are well along. We do, however, continue to struggle with how (and how much of) the surrounding landscape should be linked with watercourses and waterbodies with respect to generalized models for fish. We are really seeking a landscape perspective on fish rather than a detailed channel analysis of stream reaches. A substantial database of fish collection records is available, but this does not resolve the landscape linkage problem. Is there a happy medium between painting entire watersheds or just coloring streamlines? Insightful suggestions would be appreciated. |