WEIGHTED LINEAR REGRESSION USING (DH)-H-2 AND D-2 AS THE INDEPENDENT VARIABLES

Several error structures for weighted regression equations used for pr edicting volume were examined for 2 large data sets of felled and stan ding loblolly pine trees ( Pinus taeda L.). The generally accepted mod el with variance of error proportional to the value of the covariate s quared ((DH)-H-2 = diameter squared times height or D-2 = diameter squ ared) remains the best. Although (DH)-H-2 is a better covariate than D -2, we found no significant difference between them when testing model accuracy for felled trees, but there were significant differences for standing trees. When we predicted the total volume of a population us ing equations based on felled tree data, assuming known frequencies fo r diameter classes and using (DH)-H-2 as the covariate, we obtained es sentially the same estimate as that predicted using D-2 (0.1% differen ce). Using the conventional approach of (DH)-H-2 for all trees (standi ng and felled) yielded an estimate of volume of 5.6% less than using t he equation with (DH)-H-2 for felled trees only. Trees are more accura tely measured for volume when felled, and total heights are often not measured accurately on standing trees. Therefore, we recommend that vo lume equations be based on felled tree data only and that when they ar e intended to be applied to standing trees, D-2 be used as the covaria te in prediction.