Predicting Douglas-fir wood density by artificial neural networks (ANN) based on progeny testing information

Abstract A heuristic wood density prediction model has been developed by means of artificial neural networks (ANNs). Four populations of 32-year-old coastal Douglas-fir (Pseudotsuga menziesii (Mirb.) Franco var. menziesi) trees representing 20 full-sib families growing on comparable sites were in focus of this study. Tree height, diameter, volume, wood density, and acoustic velocity data from 632 trees were considered for the calculations. Two different ANN platforms were developed employing different classes and architectures, namely, the multilayer feed-forward (MLFF) and modular (MOD) models. After establishing the optimal configuration of the model, a MLFF network and a MOD neural network (with the obtained optimal structure) were developed and tested without cross-validation by employing a typical training and testing set methodology. To this purpose, the data set was divided in 480 trees for training and 152 trees for validation. A significant relationship between actual and predicted wood density was obtained with R2 values of 0.50 and 0.52 for the two networks, respectively, demonstrating their predictive potential for wood density estimation. A classic multiple regression analysis produced substantially lower predictive power with an R2 of 0.23. The application of ANNs as a viable predictive tool in determining wood density using growth and acoustic velocity data without additional intrusive sampling and laboratory work was demonstrated. An additional work including other species is required for these approaches.

[1]  F. Schweingruber,et al.  RESISTOGRAPH and X-Ray Density Charts of Wood. Comparative Evaluation of Drill Resistance Profiles and X-ray Density Charts of Different Wood Species , 1996 .

[2]  D. J. Cown,et al.  Comparison of the pilodyn and torsiometer methods for the rapid assessment of wood density in living trees , 1978 .

[3]  Claudio Moraga,et al.  A Fuzzy Risk Model and Its Matrix Algorithm , 2002, Int. J. Uncertain. Fuzziness Knowl. Based Syst..

[4]  R. Evans,et al.  Predicting the longitudinal modulus of elasticity of Sitka spruce from cellulose orientation and abundance , 2010 .

[5]  K. Kang,et al.  The effects of initial spacing on wood density, fibre and pulp properties in jack pine (Pinus banksiana Lamb.) , 2004 .

[6]  K. Kang,et al.  Heritability and phenotypic and genetic correlations of coastal Douglas-fir (Pseudotsuga menziesii) wood quality traits , 2008 .

[7]  P. Saranpää,et al.  Effects of timing and intensity of thinning on wood structure and chemistry in Norway spruce , 2009 .

[8]  Dan W. Patterson,et al.  Artificial Neural Networks: Theory and Applications , 1998 .

[9]  Geoffrey E. Hinton,et al.  Learning representations by back-propagating errors , 1986, Nature.

[10]  S. Srivastava,et al.  Principles of Plant Breeding , 1960 .

[11]  Yoshua Bengio,et al.  Pattern Recognition and Neural Networks , 1995 .

[12]  H. Grüneberg,et al.  Introduction to quantitative genetics , 1960 .

[13]  Robert Callan,et al.  The essence of neural networks , 1998 .

[14]  H. Mäkinen,et al.  Wood density in Norway spruce : changes with thinning intensity and tree age , 2005 .

[15]  S. Mansfield,et al.  In situ wood quality assessment in Douglas-fir , 2011, Tree Genetics & Genomes.

[16]  Brian D. Ripley,et al.  Pattern Recognition and Neural Networks , 1996 .

[17]  S. Pang,et al.  Changes in chemistry, color, dimensional stability and fungal resistance of Pinus radiata D. Don wood with oil heat-treatment , 2011 .

[18]  R. Jordan,et al.  Classification of wood species by neural network analysis of ultrasonic signals , 1998 .

[19]  Robert Evans,et al.  Variation of microfibril angle, density and fibre orientation in twenty-nine Eucalyptus nitens trees , 2000 .

[20]  F. Schumacher.  Logarithmic expression of timber-tree volume , 1933 .

[21]  Lazaros S. Iliadis,et al.  Wood-water sorption isotherm prediction with artificial neural networks: A preliminary study , 2005 .

[22]  Lazaros S. Iliadis,et al.  Predicting Wood Thermal Conductivity Using Artificial Neural Networks , 2007 .

[23]  Lazaros S. Iliadis,et al.  Neural network prediction of bending strength and stiffness in western hemlock (Tsuga heterophylla Raf.) , 2007 .

[24]  H. Mäkinen,et al.  Effects of thinning and fertilisation on tracheid dimensions and lignin content of Norway spruce , 2007 .

[25]  H. Brix.  Fertilization and thinning effect on a Douglas-fir ecosystem at Shawnigan Lake: a synthesis of project results. , 1992 .

[26]  Gene Namkoong,et al.  Tree Breeding: Principles and Strategies , 1988, Monographs on Theoretical and Applied Genetics.

[27]  B. Zobel,et al.  Applied Forest Tree Improvement , 1984 .

[28]  Wei Xu,et al.  Application of a drill resistance technique for density profile measurement in wood composite panels , 1995 .

[29]  G. Namkoong.  Introduction to quantitative genetics in forestry , 1981 .

[30]  Fikret Isik,et al.  Rapid assessment of wood density of live trees using the Resistograph for selection in tree improvement programs , 2003 .

[31]  S. Linder,et al.  Effect of Growth Rate on Fibre Characteristics in Norway Spruce (Picea abies (L.) Karst.) , 2002 .

[32]  Yung C. Shin,et al.  Intelligent Systems: Modeling, Optimization, and Control , 2008 .

[33]  Timothy M. Young,et al.  Prediction of internal bond strength in a medium density fiberboard process using multivariate statistical methods and variable selection , 2008, Wood Science and Technology.

[34]  S. Linder,et al.  Wood-density variation of Norway spruce in relation to nutrient optimization and fibre dimensions , 2002 .

[35]  T. Pakkanen,et al.  Effect of growth site type, felling season, storage time and kiln drying on contents and distributions of phenolic extractives and low molar mass carbohydrates in secondary xylem of silver birch Betula pendula , 2004 .

[36]  C. Beadle,et al.  Genetic and environmental variation in heartwood colour of Australian blackwood (Acacia melanoxylon R.Br.) , 2011 .

[37]  A. Yanchuk.  General and specific combining ability from disconnected partial diallels of coastal Douglas-fir , 1996 .

[38]  Peng Zhang,et al.  Research on Wood Density Detection by X-Ray Based on Neural Network , 2009, 2009 Fifth International Conference on Natural Computation.

[39]  L. García Esteban,et al.  Prediction of standard particleboard mechanical properties utilizing an artificial neural network and subsequent comparison with a multivariate regression model , 2008 .

[40]  Ali A. Minai,et al.  Back-propagation heuristics: a study of the extended delta-bar-delta algorithm , 1990, 1990 IJCNN International Joint Conference on Neural Networks.

[41]  Stavros Avramidis,et al.  Wood dielectric loss factor prediction with artificial neural networks , 2006, Wood Science and Technology.

[42]  Lazaros S. Iliadis,et al.  Predicting the strength of Populus spp. clones using artificial neural networks and ε-regression support vector machines (ε-rSVM) , 2011 .

[43]  Michael I. Jordan,et al.  Hierarchies of Adaptive Experts , 1991, NIPS.

[44]  A. Polle,et al.  FTIR spectroscopy in combination with principal component analysis or cluster analysis as a tool to distinguish beech (Fagus sylvatica L.) trees grown at different sites , 2008 .

[45]  L. Bouffier,et al.  What are the consequences of growth selection on wood density in the French maritime pine breeding programme? , 2008, Tree Genetics & Genomes.

[46]  John C. F. Walker,et al.  Non-destructive assessment of green density and moisture condition in plantation-grown radiata pine (Pinus radiata D. Don.) by increment core measurements , 2010 .

[47]  Geoffrey E. Hinton,et al.  Adaptive Mixtures of Local Experts , 1991, Neural Computation.

[48]  Robert J. Ross,et al.  Acoustic testing to enhance western forest values and meet customer wood quality needs , 2005 .

[49]  Simon Haykin,et al.  Neural Networks: A Comprehensive Foundation , 1998 .