USGS - science for a changing world

Northern Prairie Wildlife Research Center

  Home About NPWRC Our Science Staff Employment Contacts Common Questions About the Site

Fuel Model Selection for BEHAVE
in Midwestern Oak Savannas

Results and Discussion


Our analysis indicated that BEHAVE rate-of-spread (ROS) predictions using Fuel Model 2 (Timber and Grass) reliably predicted prescribed fire behavior in oak savannas. ROS predictions using Fuel Models 1 (Short Grass), 3 (Tall Grass), and 9 (Hardwood Litter) did not reliably predict oak savanna fire behavior. However, a casual examination of the data and regression equations would have suggested that any of the four fuel models would have provided adequate information if simply R2 values were considered (Figure 2). All R2 values exceeded 0.80, and one of them was almost 0.90.

Figure 2: Four line graphs illustrating the 'Line of perfect fit' and  'Regression line' for each of the studied fuel models
Figure 2.  Mean observed and predicted rates-of-spread by study site.

BEHAVE predictions using FM 2 appeared to reliably predict individual and mean ROS data when the data was visually inspected (Table 4), prediction error was calculated (Table 5), and percent prediction error was calculated (Figure 3). Regression analysis for the mean ROS data indicated that FM 2 was reliable (Figure 2). The intercept of the regression equation was not different from 0 (α = 0.05, P = 0.60) and slope of the regression line was not different than 1 (α = 0.05, P = 0.0775) for the mean data. The regression for the individual data indicated that the intercept was not different within 50% prediction error. In the current study, we found 79% of the observed rates of spread were within 50% prediction error. Linear regression analysis of the observed and predicted fire behavior calculated an R2 = 0.84, this is comparable to other tests of this fire behavior model (Table 7).

Figure 3: Four pie charts showing percentages for observed rate-of-speed (ROS)
Figure 3.  Percentage of rate-of-spread (ROS) observations within three classes of prediction error. Prediction error is calculated as the absolute difference between predicted and observed ROS values divided by the larger of the two values (Andrews 1980).

BEHAVE predictions using FM 1 were not reliable when the data was visually inspected (Table 4). This observation was supported by calculating prediction error (Table 5), percent prediction error (Figure 3), and linear regression analysis (Figure 2). Fuel Model 1 usually overpredicted individual (9.7 ft/min.) and mean (8.6 ft/min.) ROS data (Table 5 and Figure 2). Seventy-three percent of all observed ROS were greater than +/-50% of the prediction error using FM 1. Linear regression analysis indicated that the regression intercept was not different than 0 [α = 0.05, P = 0.39 (individual) and P = 0.38 (mean)] but the slope of the regression was different than 1 [α = 0.05, P > 0.0001 (individual and mean)]. The grassy fuels characterized by FM 1 are finer and more porous than fuels typically found in established oak savannas — 86% (2.72 tons/ac) of the 1 hr fuel load was comprised of leaf litter with no 10 or 100 hr fuels (Table 6). This may account for most of the variation between predicted and observed fire behavior.

BEHAVE predictions using FM 3 were not reliable when the data was visually inspected (Table 4). This observation was supported by calculating prediction error (Table 5), percent prediction error (Figure 3), and linear regression analysis (Figure 2). Fuel Model 3 consistently overpredicted individual (20.4 ft/min.) and mean (17.6 ft/min.) ROS (Table 5 and Figure 2). Eighty-four percent of all observed ROS were greater than +/-50% of the prediction error using FM 3. Regression analysis confirmed that FM3 was not reliable; the intercept was not different from 0 [α = 0.05, P = 0.69 (individual) and P = 0.96 (mean)], but the slope of the regression equation was different from 1 [α = 0.05, P > 0.0001 (individual and mean)]. FM 3 is exclusively composed of 1 hr fuels with grass fuel loads slightly over 3 tons/ac. Savanna fuel matrices contain some grassy fuels (0.43 tons/ac), but leaf litter was the dominant fuel and burning leaf litter appeared to influence fire behavior more than the grassy fuels.

BEHAVE predictions using FM 9 were not reliable when the data was visually inspected (Table 4). This observation was supported by calculating prediction error (Table 5), percent prediction error (Figure 3), and linear regression analysis [slope different from 1 (α = 0.05, P > 0.0001 (individual and mean))] (Figure 2). Fuel Model 9 underpredicted individual (9.7 ft/min.) and mean (8.7 ft/min.) ROS (Table 5 and Figure 2). In addition, 81.6% of the observed ROS were greater than +/-50% of the prediction error using FM 9. Intuitively, FM 9 (Hardwood Litter) may have seemed the most appropriate choice given the mix of fuels in the model when compared to the actual fuel loading (Table 6). While the fuels described by FM 9 appear to closely match the fuel conditions on the selected savannas, FM 9 consistently underpredicted ROS. The total amounts of 1 and 10 hr fuels found on the savannas and used in FM 9 were surprisingly close (see Table 1 and 6). However, the savannas typically contained a fine grassy fuel component, no live fuels, and had roughly eight times the amount of 100 hr fuels (Table 6). Leaf litter dominates both savanna and hardwood litter fuels, but grass litter fuel loads were enough in the savanna to increase prescribed fire rate-of-spread.


Previous Section -- Methods
Return to Contents
Next Section -- Conclusions

Accessibility FOIA Privacy Policies and Notices

Take Pride in America logo USA.gov logo U.S. Department of the Interior | U.S. Geological Survey
URL: http://www.npwrc.usgs.gov/resource/habitat/behave/results.htm
Page Contact Information: Webmaster
Page Last Modified: Saturday, 02-Feb-2013 04:38:43 EST
Sioux Falls, SD [sdww55]