Date of Award:

1988

Document Type:

Dissertation

Degree Name:

Doctor of Philosophy (PhD)

Department:

Natural Resources

Advisor/Chair:

NA

Abstract

Multiple regression and discriminant analysis procedures are commonly used to develop forest site quality models. 'When they contain many independent variables relative to sample size, these models may be subject to predicton bias. Fit statistics such as R2 in regression and classification tables in discriminant analysis show the apparent model accuracy but this may be a biased estimate of the model's actual accuracy. Sample splitting methods such as cross-validation and the bootstrap can be used to get an unbiased actual accuracy estimate. A discriminant procedure called classification tree analysis uses cross-validation to build the classifier with the greatest estimated actual accuracy. Because cross-validation is used in model development, the model is less likely to be over-fit with insignificant variables when compared with stepwise linear discriminant analysis. Classification tree analysis and linear discriminant analysis were used to develop models that discriminate prime vs. nonprime ponderosa pine (Pinus ponderosa) sites. Prime sites are defined as having site index 25 greater than 7.6 meters; nonprime sites have site index 25 less than 7.6 meters. Forest habitat type, percent sand content, and soil pH were incorporated in both models. The cross-valiation estimate of classification tree actual accuracy was 88 percent. A random bootstrap estimate of the linear discriminant function actual accuracy was 80 percent. viii A multiple regression model developed with random plots revealed little useful information and was biased when applied to prime site plots. The conventional regression approach using random plots may be misleading if one is interested in identifying relatively rare prime sites. Forest habitat types within the ponderosa pine series in southern Utah were examined as site quality indicators. The site index range within any one habitat type was broad. However, the best ponderosa pine sites consistently occurred in only Pinus ponderosa/Quercus gambelii, and Pinus ponderosa/Symphoricarpos oreophilus habitat types; or in habitat types within the Pseudotsuga menziesii or Abies concolor series. Therefore forest habitat type when used with other site variables may be useful in predicting prime sites. The effect of aspect at the upper elevational limit of ponderosa pine was examined by comparing mean site index and mean initial 10 year diameter increment on southerly and northerly slopes from two cinder cones. Southerly aspects on both cinder cones had greater mean diameter increment. Southerly aspects on the highest elevation cinder cone had the greatest mean site index. There was no significant difference in mean site index on the lower elevation cinder cone. Optimal aspect for height and diameter growth may differ due to l) the effect of density on diameter increment; and/or 2) available soil water limiting height growth during the spring and ambient temperature/solar radiation limiting diameter growth in late summer. Optimal aspect for forest production is not constant but varies with tree species, elevation, latitude, and other factors affecting site microclimate.

Checksum

3191105853ce4a5a01aa4a6a19a06d8c

Share

COinS