Strategies for improving the accuracy and specificity of large-area, satellite-based land cover inventories
Researchers and cooperators associated with the University of Wisconsin-Madison Environmental Remote Sensing Center (ERSC) are part of a larger team effort aimed at planning and developing the technical format of two interrelated, large-area, satellite-based land cover mapping efforts. These are: 1) the Wisconsin Initiative for Statewide Cooperation on Land Cover Analysis and Data (WISCLAND), concerned with statewide land cover mapping in Wisconsin, and 2) the Upper Midwest Biodiversity Assessment in Michigan, Minnesota, and Wisconsin being conducted as part of the national Gap Analysis Program (GAP). Such statewide, to multistate, land cover inventories afford numerous technical challenges, particularly when high levels of classification detail and accuracy are required. This paper provides an overview of the basic philosophies underlying the design of the remote sensing activities attendant to the WISCLAND program. It is envisioned that the Upper Midwest Gap Analysis effort will use the same general approaches, or appropriate modifications of same, to accomplish the land cover classification in the other two states involved in the tri-state effort. Among other technical approaches being proposed are the use of an extendable classification system, multi-date GIS-assisted classification, stratification of the entire study area into spectrally consistent geographic areas that are subsequently classified independently, the use of hybrid guided clustering to affect major portions of the classification, and the use of geographically stratified systematic non-aligned sampling for collecting training and accuracy assessment data. It should be noted that these procedures are presently undergoing preliminary testing and are subject to considerable change as the above programs become fully operational.