Evaluation of four regression techniques for stem taper modeling of Dahurian larch (Larix gmelinii) in Northeastern China

Estimating stem volume and biomass in forests is fundamental to both economic and ecological assessments of forest ecosystem structure and function. Stem taper models were widely used to calculate stem volume and biomass, but it can be challenging to get an accurate and convenient technique for taper models in practice. This study evaluated ordinary nonlinear least squares (ONLS), fixed-effects model (FIXED), quantile regression (QR), and generalized additive model (GAM) to predict tree diameter, volume, and merchantable height of Dahurian larch (Larix gmelinii) in Northeastern China. As far as we know, a comprehensive analysis of these four techniques is limited for taper data. Therefore, our main objectives were to compare these four techniques at an equitable level without calibration and select a single and widely applied technique for the taper model. The dataset comprising 1372 felled-trees from Dahurian larch natural forest were used to evaluate the techniques with a leave-one-out cross-validation approach. Evaluation statistics and box plots showed that the GAM performed better than other techniques for stem profile description and volume estimation. Results also revealed that all techniques had a bias in estimating merchantable height. However, this limitation does not significantly affect the overall performance and applied use of the GAM for diameter and volume prediction. When intuitive interpretations are not needed, the GAM can serve as an accurate and convenient technique for prediction.