correction s_chemes: no correction, and a correction of In(Y/Y), where Y is mean basal area 



growth and Y is mean predicted basal area growth with no correction for log bias. I chose 

 the latter because the correction was about halfway between the original correction and zero 

 correction . 



Six validation runs were made on each plot, each run representing a different log cor- 

 rection and one of the two methods for applying the equations (that is, average basal area 

 growth or basal area growth in thirds) . From these runs I concluded that the best log bias 

 correction was no correction, and the best method for applying the equations was to divide the 

 diameter classes into thirds and project each third. 



The conclusion that a zero (when added to intercept terms of the log model) correction is 

 best of the three methods should not be construed to mean that a zero correction is the best 

 of all possible ones. A more likely interpretation is that the correction should be smaller 

 than the two values tested. Obviously, the correction method posed in appendix D does not 

 produce appropriate corrections for log bias. Until a more theoretical basis for correcting 

 log bias in the nonnormal case is developed, the most reasonable approach may be to not correct 

 the equations directly for log bias, but rather, to build in a calibrating routine that will 

 correct for log bias, site differences, and so forth. 



Table 18 aids comparison of the effects of adding predictive components upon accuracy (as 

 measured by average differences between predicted and actual number of trees in a diameter 

 class) and precision (as measured by the variance of the differences) . The first two columns 

 of values provide the mean difference and the variance of differences for each plot and growth 

 period (since initialization) for the set of runs using the final predicted upgrowth model and 

 actual mortality, conversion, and ingrowth. If a model predicted upgrowth perfectly, the 

 resulting values would be zero. While the values do differ from zero, their magnitudes indi- 

 cate that upgrowth predictions are reasonably good. 



As expected, results also indicate that, as the number of growth periods from initial- 

 ization increases, the mean differences and the variance of the differences also increase. 



PREDICTED UPGROWTH AND MORTALITY; ACTUAL CONVERSION AND INGROWTH 



I tested two blackjack pine mortality equations in this set of runs: one equation was 

 developed using the uneven-aged data, and the other using both the even- and uneven-aged data. 

 I developed the single yellow pine equation using uneven-aged data. 



Analysis of the runs showed the even- and uneven-aged blackjack pine equation evidently 

 inferior to the uneven-aged blackjack pine equation for predicting uneven-aged mortality. 

 The former equation also did not predict even-aged mortality well. This finding was not 

 surprising because of the high level of snowbreak loss on the even-aged plots. 



A different approach for predicting even-aged mortality was tried. I reclassified the 

 snowbreak mortality as a "cutting" loss (that is, not endemic), and predicted the remaining 

 mortality using the uneven-aged mortality equations. The result was greatly improved runs 

 over those using even- and uneven-aged mortality equations. 



This finding indicates that if catastrophic losses (such as snowbreak, fire, insects, or 

 disease) are treated in the same fashion as cutting losses, endemic losses in even- and uneven- 

 aged stands could be predicted using the same mortality equations, if the equations are 

 developed appropriately. 



The third and fourth columns of table 18 provide the mean difference and the variance of 

 differences for the runs using the final upgrowth and mortality models and actual conversion 

 and ingrowth. A comparison with the values derived from the previously discussed set of runs 

 indicates that adding the mortality functions has not greatly affected accuracy and precision. 



34 



