Results and Discussion

Fig. 1 - Predicted average yearly decline

The final 3-term predictive model included aspect and slope (Equation 1). This model accounted for a little over a third of the variability in decline rates for the 21-plot calibration set with a p=0.02, R ²=0.35, R²adjusted=0.27 and RMSE=0.19 (Figure 1). A PRESS statistic for jack-knifed residuals of 1.04 indicates that, on average, if each plot was left out of the calibration and retained individually for validation, the average error would equal approximately 0.23 for the 0 to 10 scale.

Decline_rate=-0.154+ (aspect*0.002)– (slope*0.118)

Equation 1: A landscape-variables only model selected aspect (calculated as degrees from south) and slope (degrees) to predict the rate of decline expected following HWA infestation. This model accounted for 35 percent of the variability in the calibration set.

Fig. 2 - Predicted average yearly
decline following infestation

When we added foliar nitrogen concentrations to the mixed stepwise linear regression, model accuracy improved significantly. The resulting model based on slope, aspect, physiography, and foliar nitrogen concentration produced a p=0.0009, R ²=0.79, R ²adjusted=0.69, and RMSE=0.13 (Figure 2). Jack-knifed residuals resulted in a PRESS statistic of 0.55, or an average error of approximately 0.16 on the 0 to 10 decline rating scale.

Decline_rate=0.643–(aspect*0.0003)–(slope*0.158)+(physiography*0.049)–(foliarN*0.425)+([aspect* slope]*0.001)+([aspect *foliarN]*0.022)

Fig. 3 - Susceptibility Risk:
Slope/Aspect/Foliar N

Equation 2. The full GIS model again selected aspect (calculated as degrees from southwest) and slope (degrees) to predict the rate of decline expected following HWA infestation, with the addition of physiographic position and foliar nitrogen concentration. This model accounted for 79.percent of the variability in the calibration set.

Fig. 4 - Susceptibility Risk:
Slope/Aspect/Foliar N

Using this final model based on both landscape variables and foliar N concentrations, we combined coverages of key variables to create a map of relative hemlock vulnerability to decline following HWA infestation for the Catskills region of New York (Figure 3). The availability of a hemlock distribution coverage from previous work (Pontius and others 2005) allows us to isolate only those areas dominated by hemlock (greater than 40 percent hemlock basal area) for the final risk coverage (Figure 4). The resulting coverage of hemlock and its relative vulnerability to infestation should aid land managers in targeting management activities in the region.

Fig. 5 - Physiography and Susceptibilty

Fig. 6 - Slope and HWA Induced
Decline Since Infestation

Fig. 7 - General Aspect and
HWA Induced Decline

Although these quantitative models were statistically significant, we wanted to ensure that there was a theoretical basis for why these variables might exert influence on hemlock decline rates. The inclusion of landscape characteristics in our risk models has a strong theoretical basis in the literature. Similar to previous HWA research discussed in the introduction, we found that stands with a demonstrated resistance to long-term HWA infestation typically occupy lower physiographic positions, such as stream beds, flats, and toe-slopes (p=0.0076, Figure 5).

In addition to physiography, resistant stands were consistently found on less steep terrain than susceptible stands across our calibration data set (p=0.008, Figure 6). A weak yet significant correlation between aspect (in degrees from southwestern exposure) and the rate of decline (r=0.24, p=0.04) was also seen, with more rapid decline on southern facing exposures (p=0.012, Figure 7). Significant interactions between aspect/slope (p=0.06) and aspect/nitrogen (p=0.002) indicate that aspect may be more significant when other stressors (such as steeper slopes or higher nitrogen concentrations) are involved.

Fig. 8 - Factors determining
hemlock decline rates

The existing literature suggests that inherently low N concentration may limit HWA success, which, in turn, may impart some measure of resistance for host trees. Under low nitrogen conditions, concentrations may not be sufficient to maintain viable HWA populations. The data presented here support this "palatability-based" relationship between nitrogen and decline rates (Figure 8). The strongest correlate with hemlock decline rates across the region was the percent infestation (r=-0.67, p=0.008), with higher infestation levels associated with more rapid decline rates. In turn, the strongest correlate with HWA infestation levels was foliar nitrogen concentration (r=0.396, p=0.005), with higher infestation levels associated with higher nitrogen concentrations. This may explain the significant relationship between hemlock decline rates and foliar nitrogen concentrations (r=-0.46, p=0.004). Higher nitrogen levels support a larger, more successful adelgid population, which is able to deplete hemlock of photosynthate more rapidly, leading to more rapid decline.

Fig. 9 - Predicted average yearly decline
following infested landscape
variables and nitrogen- all plots.

Whereas these statistics and jack-knifed residuals suggest that this final landscape and foliar nitrogen model is robust enough to apply to new input data, independent validation provides a better assessment of model accuracy. As a preliminary test, this model was applied to all regional plots, regardless of infestation history, resulting in an R²=0.51 and RMSE=0.142 (Figure 9). This reduction in model accuracy when newly infested plots are added is most likely due to a nonlinear decline response over the duration of infestation. Newly infested trees may decline only slightly in the first or second years because photosynthate reserves are available for injury response, defensive reaction, and continued productivity. Once these reserves have been reduced, decline becomes much more rapid. To validate this model, we will continue to track hemlock decline in the remaining plots over the next several years.