Predicting Douglas-fir knot size in the stand: a random forest model based on CT and field measurements

Bruna L Longo,Udo H Sauter,Gero Becker,Franka Brüchert

doi:10.1007/s00226-021-01332-3

Bruna L Longo, Udo H Sauter + Show 2 more

Open Access

https://doi.org/10.1007/s00226-021-01332-3

Copy DOI

Abstract

Branches are not only of vital importance to tree physiology and growth but are also one of the most influential features in wood quality. To improve the availability of data throughout the forest-to-industry production, information on internal quality (e.g. knots) of both felled and standing trees in the forest would be desirable. This study presents models for predicting the internal knot diameter of Douglas-fir logs based on characteristics measured in the field. The data were composed of 87 trees (aged from 32 to 78 years), collected from six trial sites in southwest Germany, and cut into 4–5 m logs on-site. The internal knot diameter was obtained by applying a knot detection algorithm to the CT images of the logs. Applying the Random Forest (RF) technique, two models were developed: (1) MBD: to predict the branch diameter (BD) at different radial positions within the stem, and (2) MBDmax: to predict the maximum internal branch diameter (BDmax). Both models presented a good performance, predicting BD with an RMSE of 4.26 mm (R2 = 0.84) and BDmax with an RMSE of 5.65 mm (R2 = 0.78). In this context, the innovative combination of CT technology and RF modelling technique showed promising potential to be used in future investigations, as it provided a good performance while being flexible in terms of input data structure and also allowing the inclusion of otherwise underexplored databases. This study showed a possibility to predict the internal diameter of branches from field measurements, introducing an advance towards connecting forest and sawmill.

Highlights

Branches are vital for tree development and as such, constitute an inherent feature in solid timber products, influencing their utilization from a wood technology perspective
According to the RFECV results, Random Forest Regression (RFR) models with more than nine (MBD) and eight (MBDmax) selected variables do not improve the model performance by more than 1%
The variables selected within the RFECV analysis for MBD were CBH, CR, depth into crown (DINC), DWH, HD, RDINC, RPOS, WH, WRH; and for MBDmax were CBH, diameter at breast height (DBH), DINC, Fig. 3 Spearman correlation matrix of the variables assessed in the training data for MBD (a) and MBDmax (b)

Summary

Introduction

Branches are vital for tree development and as such, constitute an inherent feature in solid timber products, influencing their utilization from a wood technology perspective. Douglas-fir (Pseudotsuga menziesii (Mirb.) Franco), as most coniferous species, typically develops its branches according to its annual longitudinal growth, forming “whorls” (groups of branches growing at the same longitudinal position in the stem). Douglas-fir may develop branches between the whorls, which originate from adventive buds in an irregular longitudinal pattern, and these “internodal branches” are typically smaller in dimension (length and diameter) than “whorl branches”. In the wood products context, knots (the portion of the branch enclosed within a log) represent a disruption of the longitudinally oriented wood matrix of the stem. When processing logs to sawn timber or veneer, knots normally have a negative effect on volume (Fahey et al 1991) and value recovery (Barbour and Parry 2001; Gartner 2005)

Objectives

Methods

Results

Conclusion