Abstract:

Aims: Knowing how to measure tree height conveniently and accurately has always been a concern for the fields of forestry and community ecology. Since wood density is closely related with tree growth, building a tree height curve model based on wood density could provide a new method for measuring tree height. This method would provide data support for vegetation investigation of forest dynamics plots and exploration of spatial differences in the radial and vertical distribution of community species resources.

Methods: Here, we explored tree height using a curve model based on branch wood density classification using tree height data, diameter at breast height (DBH), and wood density of 4,032 individuals belonging to 119 species in a 1.44 ha plot in a south subtropical evergreen broad-leaved forest in Dinghushan (DHS). First, we randomly sampled individuals, and divided them into model development (70% of the total sample size) and model validation (30% of the total sample size). We then classified wood density of all individuals into one of several categories using a cluster analysis. Second, we built a tree height-DBH model for different classifications based on modeling samples using five common theoretical growth equations (Richards, Korf, Logistic, Gompertz and Weibull equations). We estimated the fitting accuracy using the root mean squared error (RMSE) and Akaike information criterion (AIC). A smaller RMSE index and AIC index indicated the best fitting effect. Third, we determined the most optimal models based on the one model with the smallest mean average absolute error (MAE) and RMSE index. Finally, we established tree height curve models using species classification and compared the differences between models based on wood density and species classifications using the MAE index and RMSE index.

Result: Results suggest that when the classification order of cluster analysis was 4, the SSI (simple structure index) value was the largest, so the individual wood density of the plot was unevenly divided into four categories: [0.06, 0.31), [0.31, 0.45), [0.45, 0.57), and [0.57, 0.82]. There was little difference when fitting the five equations and all the parameter values were extremely significant. Models based on wood density classification corresponding to the MAE index and RMSE index were consistent with the results of the modeling samples. The Gompertz equation and Weibull equation were selected as the optimal tree height models and the Weibull equation had the highest frequency equation for the DHS plot. Moreover, when comparing models based on wood density classification with species classification, the MAE and RMSE indices of the two models in 17 species were less different. In addition, since the estimation accuracy of models based on wood density classification and species classification was low, the tree height of Caryota maxima, Schima superbaand Castanea henryi was hardly to estimated.

Conclusions: The tree height curve model based on wood density classification has a well-fitting effect and high estimation accuracy. It is also more convenient and generally used than the species classification model, which can realize the establishment of tree height curve models for many species easily. What’s more, models based on wood density classification directly reflect plant response to the environment from a mechanistic perspective, and represents the ecological trade-off among individuals with different wood densities in the vertical growth of trees. In summary, this model based on wood density classification provides a new method for tree height prediction and can better serve production practices such as forest surveys and help with understanding scientific issues.

Key words: plant functional traits, wood density classification, theoretical growth equations, nonlinear regression analysis, height-diameter model