Evaluation of parametric and non-parametric stem taper modeling approaches: A case study for Betula platyphylla in Northeast China

Abstract

The accurate estimation of tree volume and biomass are crucial for forest inventories, forest sales, and carbon stock evaluations. Taper models were widely used and regarded as a standard method for estimating tree volume, but in practice it can be challenging to get an accurate and convenient method for taper models when calibration data is not available. Therefore, our main objectives were to compare fitting methods and present a reliable method to be used in forest inventories. In this work, the parametric modeling approaches ordinary nonlinear least squares, fixed-effects models at single-level and multilevel effects, and non-parametric modeling approach quantile generalized additive models were evaluated by predicting diameter, volume (total, partial, and merchantable), and merchantable height of 868 Betula platyphylla samples taken in three regions in Northeast China. The fixed-effects models were evaluated with fixed part parameters estimated from mixed-effects models at tree-level, plot-tree-level, and region-plot-tree-level. Evaluation statistics and box plots revealed that considering plot and region as random effects didn’t get the prediction accuracy improved for fixed-effects taper models. Therefore, both single-level and multilevel mixed-effects models are not recommended for prediction if the calibration data is not available. Although all models showed a bias in predicting merchantable height, this limitation does not significantly affect the performance for diameter and volume prediction. Results showed that non-parametric approach quantile generalized additive models had the highest prediction accuracy for stem diameter and volume prediction, closely followed by the ordinary nonlinear least squares and then the fixed-effects models at single-level. Therefore, the quantile generalized additive models are highly recommended in practice. The recommended model in this study can estimate the diameter and volume for a large area. Moreover, it has broader applicability in the situation when measuring upper-stem diameter may not be realized in practice.