Results and Discussion

Several model variables were found to have statistically significant differences between the two populations (blowdown and nonblowdown). However, not all statistically significant differences were in the direction expected. Positive differences indicate statistically significant population-to-population differences that are in the direction that is expected on the basis of preliminary model research. Negative differences indicate statistically significant population-to-population differences that are in a direction that is opposite the direction expected on the basis of preliminary model research.

The consistency analysis utilized a one-sample t-test. Variables that displayed the same relationship in all 10 iterations cannot be tested for consistency in this manner. But, variables that displayed the same relationship in all 10 iterations are considered inherently consistent. The thinning, elevation, composite stand, and all 10 cumulative risk variables displayed a positive difference between the population means through all 10 iterations and are considered statistically consistent. Results are displayed in (Table:Results for Individual Component Variables) and (Table:Results for Composite Risk Variables).

For both tables, positive difference indicates that the mean risk value for the population of stands with recorded wind damage was higher than the mean risk value for the population of stands without recorded wind damage. Positive differences agree with the assumptions used during model construction. Negative difference indicates that the mean risk value for the population of stands with recorded wind damage was lower than the mean risk value for the population of stands without recorded wind damage. Negative differences do not agree with the assumptions used during model construction.

The density variable (Table: Results for Individual Component Variables) did not produce significant differences between the population means in any of the iterations. The assumption based on the wind vulnerability literature (Gardiner and others 1997. Lohmander and Helles 1987) was that the less dense stands would be more susceptible. This was thought to be the case in an area with a long management history of natural regeneration and frequent stand entry.

The edge variable (Table: Results for Individual Component Variables) had a negative difference between population means 30 percent of the time, not frequently enough to be considered statistically consistent. Most of the stands in the landscape being evaluated are in the two tallest height classes. This trend results in a landscape with very little edge in general. The edge that is present may be in areas at lower risk to wind or edge may not be a critical factor in this landscape.

The height variable (Table: Results for Individual Component Variables) had a positive difference between population means 90 percent of the time. This statistically consistent difference likely reflects the increased vulnerability to wind damage with increased tree size (Lohmander and Helles 1987, Peltola and others 1997, Smith and others 1987). The species variable (Table: Results for Individual Component Variables) did not have significant differences between the population means in any of the iterations. This was surprising considering managers all cited softwoods and, most notably, balsam fir, as being the most sensitive to wind disturbance. It may indicate homogeneity within the landscape or an insensitivity of the index to differences in composition.

The thinning variable (Table: Results for Individual Component Variables) had statistically significant positive differences between the population means 100 percent of the time. This agrees with conventional wisdom of the land managers—that windthrow is much more common in previously thinned stands. An evaluation of the landscape shows that 99.72 percent of the recorded blowdown occurred in thinned stands.

The composite stand grid, STAND, (Table: Results for Composite Risk Variables) also had significant positive differences between the population means 100percent of the time. This may be primarily driven by the combination of the height and thinning components that comprise this composite variable. The thinning variable’s binary property makes it a relatively powerful component of the composite grid.

Differences between population means for the topographic exposure variables (Table: Results for Individual Component Variables) were never statistically consistent. This trend was noticed by Mitchell and others (2001) when an analysis of their model revealed a level of contribution from topographic variables to the model lower than expected. Simple terrain variables may not adequately describe airflow phenomena induced by complex terrain (Mitchell and others 2001).

The elevation variable (Table: Results for Individual Component Variables) had a positive difference between population means in the model 100 percent of the time. This agrees with the assumption that susceptibility increases in higher areas of the landscape, where exposure and wind speed are greater (Bair 1992).

The soil variable (Table: Results for Individual Component Variables) had a negative difference between population means. This trend was statistically consistent, occurring 70 percent of the time. This is counter to the original assumption of the model that forests growing in areas with more restricted rooting depths would be more vulnerable to wind disturbance (Day 1950, Mergen 1954). An analysis of the correlation between soil depth and elevation in this landscape yields a mean Pearson correlation of 0.344 for the 10 iterations. This is substantially larger than the test statistic (0.088) for an alpha of 0.05 and a sample size greater than 100. This result indicates a statistically significant positive correlation between the two variables (Zar 1984); deeper soils are correlated with higher elevations in the landscape being evaluated.

Differences between population means for the composite site variables (Table: Results for Composite Risk Variables) were either positive or non-significant. Differences between population means were statistically consistent for three exposure variants, both site grids with nondirectionally weighted exposure input variables (SITE_1500 and SITE_1000) and SITE_se, the site variant modeling topographic exposure to the southeast.

The positive difference between the population means of all site variables (with the exception of the north, northeast, and northwest exposure variants) suggests that topographic exposure is important, even though significant differences between the population means were not consistently detected for exposure as an individual variable.

Although the data within the GIS database is fairly coarse, the general spatial model we developed associates moderate to high vulnerability ratings with reported wind damage in the landscape. All of the final risk assessment variables, CMLTV_direction (Table: Results for Composite Risk Variables), have a positive difference between the population means. This validates the model’s ability to differentiate between vulnerability of damaged stands and vulnerability of undamaged stands. The difference between the population means in the cumulative risk variables is highly significant with p-values of 0.000 recorded in all 10 iterations for all 10 exposure variants.