prilagođeno pretraživanje po punom tekstu

ŠUMARSKI LIST 1-2/2021 str. 39     <-- 39 -->        PDF

commonly used three-parameter function in forest growth studies (Pretzsch 2001). Given its flexibility and suitability for describing various height growth patterns, many researches (Mamo and Sterba 2006, Batho and García 2014, Pyo 2017 etc.) have tested and employed only this growth function to develop site index curves. The results obtained in our study confirmed that the CHR model performed best, both concerning the significance and prediction accuracy of the model parameters and the biological justification of the models.
Regarding the accuracy of the model parameters of the mean height growth curve (guide curve) for each site class, it is first evident that although the values of RSE for the CHR model are slightly higher than for the others, all the parameters of the CHR model are more significant compared to the Korsun and the Korf models. Nonetheless, the Korf model produces unrealistic estimates of the parameters. Namely, the estimated value for the asymptote parameter of the Korf model (137 m – ŽA and 236 m – RU) for the first site class was extremely large. Conversely, a more realistic prediction of the asymptote parameters (≈44 m) was detected for the model of CHR, taking into account that the maximum beech tree heights recorded in the studied regions were about 40 m. To be more precise, by applying the Korf (for both regions) and the Korsun (for RU) models, the calculated tree height values at the age of 150 were overestimated, indicating values over 40 m at site class I, which do not correspond to the values in reality. On the other hand, the CHR model, in both research sites, shows a height below 40 m at the age of 150 at the same site class (Figure 2).
Another criterion for the selection of the appropriate model to be used in our study was the residual statistics of the tested height growth curves (guide curves) for each site class in relation to their homogeneity. According to the obtained arithmetic mean of residuals (close to zero), the standard deviation values (within a value limit between -2 m and 2 m corresponding to the width of site class of 4 m) and the variation range of standard deviation, skewness and kurtosis coefficients (the smallest for the CHR model) for all five site classes, it can be concluded that the distribution of residuals for the CHR model showed the smallest deviations from normality, while the remaining two models expressed a degree of heteroscedasticity in some site classes.
The established site index curves point to some more and less important differences in beech height growth between regions (Figure 3). These disparities are primarily reflected in the attained tree height values at a certain age and in the