Results

The analysis of Monte Carlo results indicated that all estimates stabilized to within less than 1 percent by 25,000 simulations. Therefore, 25,000 simulations were used when applying Procedure C to estimate the contributions of the various sources of uncertainty.

The forest/nonforest maps constructed using logistic regression model predictions produced realistic spatial distributions, although no independent accuracy assessment was conducted (Figure 1a). However, considerable detail was revealed in the uncertainty map; e.g., the field structure, road networks, and highway on/off ramps (Figure 1b). Considerably less detail was revealed in the ash tree count-per-hectare map, but this result was expected because of the fewer FIA plots available and the more continuous nature of the layer (Figure 2a). As expected, the greatest uncertainty in the latter map occurred in the same locations as the greatest estimated values (Figure 2b). This phenomenon is common to biological analyses.

The estimates obtained using Procedure C revealed in dramatic fashion that the source of uncertainty making the greatest contribution to uncertainties in the estimates of both F_total and A_total was spatial correlation in the realizations of the forest/nonforest maps (Table: Monte Carlo simulation estimates from Procedure C). The magnitude of this effect is highlighted by noting that when uncertainty from this source was included, 95-percent confidence intervals for both F_total and A_total included or were close to including 0. The contribution of the uncertainty in the underlying ash tree count-per-hectare layer to was much less than was the contribution due to the uncertainty in the forest/nonforest layer.