Two-stage ingrowth models for four major tree species in Alberta

Abstract

A two-stage modelling approach was used for developing ingrowth models for four major tree species in Alberta: aspen, lodgepole pine, black spruce and white spruce. The probability of ingrowth presence was modelled first, followed by the modelling of annual amount of ingrowth conditional on ingrowth being present. To handle variable measurement intervals and plot sizes typical of repeatedly measured forestry data, two generalized logistic models were evaluated at the first stage: a traditional model with measurement interval length and plot size factor incorporated as an exponent of a standard logistic function and an alternative model with measurement interval length and plot size factor incorporated as a denominator. Two criteria were evaluated for defining optimal threshold values for plot classification: criterion 1 to maximize the summation of sensitivity and specificity and criterion 2 to maximize the overall rate of correct plot classification. It was found that criterion 1 was superior to criterion 2 for providing better plot classifications. The two ingrowth probability models behaved similarly based on model fitting and model validation results. Annual amount of ingrowth was modelled at the second stage by exponential functions.