Primerjava različnih regresijskih modelov za napovedovanje debelinskega priraščanja jelke

Abstract

We present seven alternative statistical models for modelling tree diameter increment with data from permanent sampling plots. In addition to the polynomial regression model, we present a regression model with added random noise, a mixed linear model, regression with natural splines, and three models with limited dependent variables: truncated regression, tobit regression and grouped data regression. The models may be used when dealing with truncated or censored variables, biased estimation of the increment due to censoring and rounding down, or when having multilevel data. The parametrization of the models was done using 21,013 fir trees on 4,405 plots in the period 1990–2014 in uneven-aged Dinaric fir-beech forests. All models showed a similar effect of tree diameter, stand basal area, basal area of larger trees, diameter structure diversity, altitude and slope. There were only minor differences in the regression coefficients and fit measures. The highest increment predictions were given by the tobit model. The mixed model fit the data best and, compared to the other models, predicted a slower decrease in the growth of large-diameter trees after growth culmination.