Results

In the Table: Model fit results by variable types and variable selection criteria, the best 8 variable models illustrate moderate fits with RMSE values below 0.15 in only 1 model and R²_adj values above 0.55 in 2 models. Between the no MIFI variables pre- and poststorm models, there was a dramatic increase in R²_adj from 0.176 in the prestorm model to 0.439 in the pre- and poststorm situation along with a matching magnitude reduction in RMSE from 0.219 to 0.181, respectively. In order to compare these no MIFI models to MIFI models, which were created in multiples of three matching the three MIFI forest type designations, a set of pooled RMSE and R²_adj values were created. This creation occurred by combining the three models’ error sum of squares, in the case of RMSE values, and by combining the three error and total corrected mean sum of squares considering all three models’ degrees of freedom, in the case of the R²_adj values. Comparing these fit values for the eight variable models again demonstrated the drastic improvement in model fit. With the use of poststorm imagery, we observed R²_adj values increasing from 0.381 to 0.506 and RMSE decreasing from 0.190 to 0.170. Additional gains were made in using the MIFI data in models by increasing prestorm pooled versus no MIFI R²_adj values from 0.176 to 0.381 and reducing RMSE from 0.219 to 0.190. Similarly, by incorporating MIFI thematic data, poststorm models were affected with an R²_adj increase from 0.439 to 0.506 and an RMSE reduction from 0.181 to 0.170.

The models derived using Hocking’s (1976) method (Table: Model fit results by variable types and variable selection criteria) illustrate the same general trend as the eight variable models. Again, improvements with the addition of MIFI and poststorm data were noted in overall (with no MIFI data) and pooled (with MIFI data) R²_adj and RMSE values, except in cases where the application of Hocking’s method yielded models with very large variable counts (2 instances used 37 variables). In an attempt to rectify this problem, the other models in Table: Model fit results by variable types and variable selection criteria were created for models that still used a large number of variables. These further reduced models corresponded to the Hocking’s identified MIFI, prestorm, and poststorm variable models in the number of employed variables for mixed and hardwood, MIFI, and prestorm models. Examination of these modified models indicates one difference from the eight variable comparisons. The pooled fit values for the prestorm and MIFI models (R²_adj=0.599 and RMSE=0.153), in tandem with the pre-storm, poststorm, and MIFI models (R²_adj0.708 and RMSE=0.130), when compared to the pre- and poststorm model (R²_adj=0.492 and RMSE=0.172) and prestorm model (R²_adj=0.277 and RMSE=0.205) indicate an increased advantage from the 8 variable models with regard to using MIFI data as opposed to poststorm image data.