| This dissertation describes the evaluation and revision of the Inland Empire variant of the Prognosis large-tree basal area increment model, and the assessment of an extension of linear modelling.; The current public release of the model was tested using independent data comparable to those used in original development. The results showed that the model was biased, under-predicting 10-year basal area increment by 12% and volume increment by 2.7%. Results differed by species and geographic location. This may be meaningful to forest managers, depending on management objectives, but in a scientific sense is less important, and is evidence that the model is relatively robust. Revision to the way climate is represented is warranted.; The model formulation was revised to test the hypotheses that: adding climate variables and soil parent material imputed from digital maps would increase model precision, and that revisions to the way competition variables were calculated would be useful. No revision produced practical improvements in model precision, but revisions that improved model properties were suggested for all three classes of predictors. The outcome was sensitive to the distribution of the sample of fitting data across levels of classification variables, which should be investigated further.; Finally, two alternative statistical techniques for implementing the model were compared. In theory, the structural based prediction (SBP) method reduces bias that occurs under ordinary least squares (OLS) when predictors are subject to variable sampling errors. Two hypotheses were tested: first, that SBP produces results according to theory when tested on independent data, and second, that comparisons of model revisions are not sensitive to choice of method. The results supported both hypotheses. The ability to judge revisions using OLS is useful because OLS is simpler to implement than SBP. The theoretical and practical benefit of SBP is useful because model accuracy can be improved. |
PDF Full Text Request