Forest/non-forest stratification in Georgia with Landsat Thematic Mapper data

Geographically accurate Forest Inventory and Analysis (FIA) data may be useful for training, classification, and accuracy assessment of Landsat Thematic Mapper (TM) data. Minimum expectation for maps derived from Landsat data is accurate discrimination of several land cover classes. Landsat TM costs have decreased dramatically, but acquiring cloud-free scenes at optimum seasons for vegetation discrimination is still problematic. FIA plot locations determined from hand-held GPS units can vary ± 5-20 m. Landsat pixels can also vary ± 25 m. These spatial inaccuracies restrict the use of pixels on feature edges and decrease the usefulness of plots that have split conditions. Current research at the USDA Forest Service's, Southern Research Station involves aggregating forest types in the lab based on field plot measurements of dominant, co-dominant, and intermediate trees. We believe this methodology is most appropriate for tying FLA field plot data to the satellite imagery. We are testing methodological approaches for image processing that can satisfy the dual goals of repeatability and timeliness.