Predicting bark thickness with one- and two-stage regression models for three hardwood species in the southeastern US

Abstract

Tree bark, as the outermost protective layer of tree stems, is an important indicator to evaluate the fire resistance properties of trees and to assess the tree mortality induced by fire. Despite its importance, many existing bark thickness models were not primarily developed for predicting bark thickness directly, i.e. with bark thickness as a response variable, and most past studies were focused on modeling bark thickness in conifers. Thus, the objective of this study was to compare the efficacy of various bark thickness models/methods for three common hardwood species in the southeastern US. A total number of 47,281 measurements from 2,070 trees were used in analysis.Results showed that bark thickness at breast height (1.37 m or 4.5 ft above ground) varies by tree size and species, which can be predicted by a species-specific linear regression model with DBH as a single predictor. To predict bark thickness profile, a combination of stem taper function and bark thickness model, a two-stage method, is suggested, which generally performs better than a single bark thickness function (one-stage method) in terms of bias and precision. For a given model form, the two-stage method produced more reliable prediction of bark thickness at upper and lower portions of tree stem than the one-stage method. With the three species examined, the segmented stem taper functions provided more accurate predictions than the variable-exponent function. The results of this study can provide guidance for ecologists and forest managers when selecting appropriate approaches to predict bark thickness.