Results

Gradients in Species Composition

Daubenmire (1952) noted the axiomatic relationship between climate and vegetation. Our CCA modeling results are consistent with this observation and suggest that climatic variables, as well as topographic and edaphic variables indirectly related to temperature and moisture, strongly influence the patterns of species composition. The gradients described by the three CCA models were comprised of the dominant patterns in temperature and precipitation. In Washington, elevation, precipitation frequency, and brightness in MODIS 8-day composites separated warmer drought-tolerant conifers from both high-elevation wet and dry forests. In California, species were separated by September growing degree days, September average air temperature, April cooling degree days, and water vapor pressure variability in July. In Oregon, the dominant environmental variables were August mean precipitation, June cooling degree days, soil permeability, and June standard deviation of water vapor pressure.

Species Mapping Performance of GNN

Species performance from confusion matrices are listed for all species occurring with a frequency of at least 5 percent in the plot data set. California had the highest average Kappa statistics at 0.53 for the 30-m study and 0.48 for the 250-m study (Table: California Species presence/absence results). Washington was second with 0.46 and 0.43 (Table: Washington Species presence/absence results), followed by Oregon with 0.32 and 0.25 (Table: Oregon Species presence/absence results). Patterns for producers and users accuracy were similar across sites, as were the actual values. In each case, the 30-m study had higher producers accuracy than the 250-m study by about 22 percent, whereas for users accuracy, the 250-m study was actually about 3 percent higher with an average across sites of 55 percent.

Structure Mapping Performance of GNN

To date we have only developed structure models for the Washington and Oregon study sites. Second nearest neighbor correlations for structure variables were generally low in both sites. In Washington , total basal area had an r-square of 0.06 compared to 0.17 for the 30-m analysis, 0.04 for snags-per-hectare compared to 0.16, and 0.01 for quadratic mean diameter compared to an almost equally random 0.05 for the 30-m analysis. In Oregon, where we had quite good results for structure with 30-m data, we mapped basal area with an r-square of 0.09 compared to 0.59 for the 30-m analysis, 0.03 for snags-per-hectare compared to 0.09, and 0.08 for quadratic mean diameter compared to 0.69.