Conclusions

The Gradient Nearest Neighbor method (GNN) applies direct gradient analysis and nearest neighbor imputation to ascribe detailed ground attributes of vegetation to each patch in a regional landscape. Similar to other Landsat-TM-based methods, predicted vegetation maps are appropriate for regional-scale analyses but are insufficiently accurate for most site-level applications. 

GNN has several advantages over existing methods for mapping forest vegetation. Data and methods are consistent over a multi-ownership region. Resulting maps thus are “repeatable,” accuracy assessments apply to the entire region, and valid subregional comparisons can be made. Map accuracy can be quantified in a variety of ways, and can be tailored to specific objectives. Information content of resulting maps (e.g. species identities, understory characteristics) is more detailed than in image classifications. Because vegetation attributes are represented as individual continuous variables, maps and classifications can be constructed for specific analytical purposes. 

GNN can be applied to any region where field plot and spatial data are available. Currently, we are using vegetation maps predicted with GNN in the Oregon coastal province to initialize conditions for simulating landscape change under alternative land-use policies and to characterize regional patterns of biodiversity.