Driving factors of knot size in Larix olgensis plantations: a multi-model ensemble by using GAM, machine learning, and SEM

Abstract

Knots are common wood defects that severely reduce the mechanical performance and market value of timber. To identify the key drivers of knot diameter in Larix olgensis and determine its nonlinear response mechanisms, this study, within the framework of tree resource allocation trade-offs and stand competition, integrates a generalized additive model (GAM), SHapley Additive exPlanations (SHAP)–based machine learning models, and structural equation modelling (SEM) to build a multi-dimensional analytical framework spanning nonlinear effect identification, threshold detection, and causal pathway analysis. The results show that diameter at breast height (DBH), tree height (H), sound knot length (SKL), knot height (KH), and knot tilt angle (TA) are the main positive drivers of KD, whereas increasing stand density (SD) exerts a significant suppression on knot expansion through intensified resource competition. By organically combining predictive and mechanistic models, this study reveals several biologically meaningful thresholds and further clarifies how stand structure influences knot development. These findings provide a quantitative basis for optimizing stand density and crown structure to control knot size, thereby improving wood quality and promoting the sustainable management of L. olgensis plantations, and also offer a transferable methodological framework for quantitative studies of knot characteristics in other plantation tree species.

Keywords

Larix olgensis / Knot diameter / GAM / Machine learning / SEM

References

[1] Ali S, Khan SM, Ahmad Z, Siddiq Z, Ullah A, Yoo S, Han H, Raposo A. Carbon sequestration potential of different forest types in Pakistan and its role in regulating services for public health. Front Public Health, 2023, 10: 1064586

[2] Benjamin JG, Kershaw JAJrWeiskittel AR, Chui YH, Zhang SY. External knot size and frequency in black spruce trees from an initial spacing trial in Thunder Bay, Ontario. For Chron, 2009, 85(4): 618-624

[3] Bladen K, Cutler DR. Assessing agreement between permutation and dropout variable importance methods for regression and random forest models. Era, 2024, 32(7): 4495-4514

[4] Bufford JL, Brandt AJ, Ausseil AG, Black A, Case BS, Forbes AS, Kirby C, Sprague R, Wreford A, Peltzer DA. Conservation futures 2050: developing future scenarios to explore potential socio-economic developments and their impact on biodiversity. PLoS ONE, 2024, 19(12): e0311361

[5] Cattaneo N, Puliti S, Fischer C, Astrup R. Estimating wood quality attributes from dense airborne LiDAR point clouds. For Ecosyst, 2024, 11: 100184

[6] Chen TL, Pang ZQ, He SM, Li Y, Shrestha S, Little JM, Yang HC, Chung TC, Sun JY, Whitley HC, Lee IC, Woehl TJ, Li T, Hu LB, Chen PY. Machine intelligence-accelerated discovery of all-natural plastic substitutes. Nat Nanotechnol, 2024, 19(6): 782-791

[7] Deckmyn G, Mali B, Kraigher H, Torelli N, Op de Beeck M, Ceulemans R. Using the process-based stand model ANAFORE including Bayesian optimisation to predict wood quality and quantity and their uncertainty in Slovenian beech. Silva Fenn, 2009, 43(3): 523-534

[8] Ding Y, Zang RG, Letcher SG, Liu WD, Lu XH. Aboveground and belowground competition affect seedling performance and allometry in a tropical monsoon forest. New for, 2016, 47(4): 529-540

[9] Drew DM, Downes GM, Seifert T, Eckes-Shepard A, Achim A. A review of progress and applications in wood quality modelling. Curr Forestry Rep, 2022, 8(4): 317-332

[10] Duchateau E, Auty D, Mothe F, Longuetaud F, Ung CH, Achim A. Models of knot and stem development in black spruce trees indicate a shift in allocation priority to branches when growth is limited. PeerJ, 2015, 3: e873

[11] Fajardo A, Llancabure JC, Moreno PC. Assessing forest degradation using multivariate and machine-learning methods in the Patagonian temperate rain forest. Ecol Appl, 2022, 32(2): e2495

[12] Fredriksson M, Cool J, Duchesne I, Belley D. Knot detection in computed tomography images of partially dried jack pine (Pinus banksiana) and white spruce (Picea glauca) logs from a Nelder type plantation. Can J for Res, 2017, 47(7): 910-915

[13] Gao HL, Liu QF, Song Y, Jiang MZ, Yin Y. Modeling primary branch diameter and length for planted Pinus koraiensis by incorporating neighbor competition in Northeast China. Forests, 2022, 13(6): 912

[14] Guo HT, Jia WW, Li DD, Sun YM, Wang F, Zhang XY. Modelling branch growth of Korean pine plantations based on stand conditions and climatic factors. For Ecol Manag, 2023, 546: 121318

[15] Hao YJ, Sun ZY, Tan ZH. Biomass allocation of China’s forests as indicated by a literature-based allometry database. Forests, 2024, 15(6): 942

[16] He X, Lei XD, Xu QG, Lan J, Wu BY, Guo H, Gao WQ, Liu D. Combining multiple feature selection methods and structural equation modelling for exploring factors affecting stand biomass of natural coniferous-broad leaved mixed forests. Sci Total Environ, 2024, 954: 176669

[17] Hein S. Knot attributes and occlusion of naturally pruned branches of Fagus sylvatica. For Ecol Manag, 2008, 256(12): 2046-2057

[18] Hein S, Spiecker H. Comparative analysis of occluded branch characteristics for Fraxinus excelsior and Acer pseudoplatanus with natural and artificial pruning. Can J for Res, 2007, 37(8): 1414-1426

[19] Høibø O, Turnblom EC. Models of knot characteristics in young coastal US Douglas-fir: are the effects of tree and site data visibly rendered in the annual ring width pattern at breast height?. For Prod J, 2017, 67(1–2): 29-38

[20] Holloway J, Mengersen K. Statistical machine learning methods and remote sensing for sustainable development goals: a review. Remote Sens, 2018, 10(9): 1365

[21] Hong Chun T, Hashim UR, Ahmad S, Salahuddin L, Choon NH, Kanchymalay K. A review of the automated timber defect identification approach. Int J Electr Comput Eng, 2023, 13(2): 2156

[22] Hwang SW, Lee T, Kim H, Chung H, Choi JG, Yeo H. Classification of wood knots using artificial neural networks with texture and local feature-based image descriptors. Holzforschung, 2022, 76(1): 1-13

[23] Ji AR, Cool J, Duchesne I. Using X-ray CT scanned reconstructed logs to predict knot characteristics and tree value. Forests, 2021, 12(6): 720

[24] Ji L, Yu JB, Zhang XZ, Liu Y, Yang LX. Differential responses of soil bacterial and fungal community to short-term crop tree management in a Larix gmelinii plantation. Forests, 2021, 12(10): 1411

[25] Ji M, Zhang W, Diao XL, Wang GF, Miao H. Intelligent automation manufacturing for Betula solid timber based on machine vision detection and optimization grading system applied to building materials. Forests, 2023, 14(7): 1510

[26] Ji M, Zhang W, Wang GF, Diao XL, Miao H, Gao R. Machine vision for knot detection and location in Chinese fir lumber. For Prod J, 2024, 74(2): 185-202

[27] Jia F, Wu W. A comparison of multiple imputation strategies to deal with missing nonnormal data in structural equation modeling. Behav Res, 2023, 55(6): 3100-3119

[28] Kamperidou V, Aidinidis E, Barboutis I. Impact of structural defects on the surface quality of hardwood species sliced veneers. Appl Sci, 2020, 10(18): 6265

[29] Kershaw JAJrBenjamin JG, Weiskittel AR. Approaches for modeling vertical distribution of maximum knot size in black spruce: a comparison of fixed- and mixed-effects nonlinear models. For Sci, 2009, 55(3): 230-237

[30] Kint V, Hein S, Campioli M, Muys B. Modelling self-pruning and branch attributes for young Quercus robur L. and Fagus sylvatica L. trees. For Ecol Manage, 2010, 260(11): 2023-2034

[31] Kovács I, Tóth B, Schally G, Csányi S, Bleier N. The assessment of wildlife damage estimation methods in maize with simulation in GIS environment. Crop Prot, 2020, 127: 104971

[32] Li ZL, Jia WW, Li FR, Zhao Y, Guo HT, Wang F. A study on the variation of knot width in Larix olgensis based on a mixed-effects model. Comput Electron Agric, 2025, 234: 110215

[33] Liu L, Zeng FP, Song TQ, Wang KL, Du H. Stand structure and abiotic factors modulate karst forest biomass in southwest China. Forests, 2020, 11(4): 443

[34] Liu YK, Wang DZ, Zhang ZD, Liu Q, Zhang DY, Xu ZQ. Modeling free branch growth with the competition index for a Larix principis-rupprechtii plantation. Forests, 2023, 14(7): 1495

[35] Liu GZ, Zhao K, Zhang SQ, Liang YM, Yue YJ, Liu GH, Qin FC. Biomass allocation and allometric relationship of Salix gordejevii branches in sandy habitats heterogeneity in northern China. Sustainability, 2024, 16(13): 5483

[36] Liu HX, Li MY, Han RY, Zhang XL, Ge JN, Hao GY. Increased temperatures contribute to early aging of plantation-grown Mongolian pine in introduced areas at lower latitudes. J Forestry Res, 2024, 35: 122

[37] Liu JR, Yang ZH, Li L, Chu XX, Wei SG, Lian JY. Predicting aboveground carbon storage in different types of forests in south subtropical regions using machine learning models. Ecol Evol, 2025, 15(5): e71499

[38] Liu TR, Peng DL, Ye SM, Yang M, Tan ZJ, Zhang YX, Guo JP. Effects of stand density regulation on soil carbon pools in different-aged Larix principis-rupprechtii plantations and soil respiration model enhancement. J for Res, 2025, 36(1): 141

[39] Longo BL, Brüchert F, Becker G, Sauter UH. Predicting Douglas-fir knot size in the stand: a random forest model based on CT and field measurements. Wood Sci Technol, 2022, 56(2): 531-552

[40] Mäkelä A, Grace J, Deckmyn G, Kantola A, Kint V. Simulating wood quality in forest management models. For Syst, 2010, 19: 48-68

[41] Manso R, McLean JP, Ash A, Achim A. Estimation of individual knot volumes by mixed-effects modelling. Can J for Res, 2020, 50(2): 81-88

[42] Meng CN, Xu W, Qiao Y, Qin LJ, Su P, Liao XL. Predicting vulnerability through hybrid modeling combining GAM and XGBoost—a case of affected population vulnerability to tropical cyclone in Hainan Province. Int J Disaster Risk Reduct, 2024, 111: 104732

[43] Miao Z, Zhao XH, Jiang YM, Dong LH, Li FR. Survival modeling of branch–knot developmental dynamics in Korean Pine plantations: linking climate, region, and artificial pruning. For Ecol Manage, 2025, 597: 123160

[44] Mirski R, Wieruszewski M, Trociński A, Kawalerczyk J, Łabęda K. Elastic moduli of butt-end logs and the variable knots distribution in Scots pine from Western Poland. BioResources, 2020, 16(1): 1842-1853

[45] Moberg L. Predicting knot properties of Picea abies and Pinus sylvestris from generic tree descriptors. Scand J for Res, 2006, 21(S7): 49-62

[46] Neuville R, Bates JS, Jonard F. Estimating forest structure from UAV-mounted LiDAR point cloud using machine learning. Remote Sens, 2021, 13(3): 352

[47] Newton M, Lachenbruch B, Robbins JM, Cole EC. Branch diameter and longevity linked to plantation spacing and rectangularity in young Douglas-fir. For Ecol Manag, 2012, 266: 75-82

[48] Pehkonen M, Vastaranta M, Holopainen M, Hyyppä J, Kukko A, Pyörälä J. Identification and segmentation of branch whorls and sawlogs in standing timber using terrestrial laser scanning and deep learning. Forestry, 2025, 98(5): 712-725

[49] Peng W, Pukkala T, Jin XJ, Li FR. Optimal management of larch (Larix olgensis A. Henry) plantations in Northeast China when timber production and carbon stock are considered. Ann for Sci, 2018, 75(2): 63

[50] Picard N. Asymmetric competition can shape the size distribution of trees in a natural tropical forest. For Sci, 2019, 65(5): 562-569

[51] Pinto I, Pereira H, Usenius A. Analysis of log shape and internal knots in twenty Maritime pine (Pinus pinaster Ait.) stems based on visual scanning and computer aided reconstruction. Ann for Sci, 2003, 60(2): 137-144

[52] Pretzsch H, Rais A. Wood quality in complex forests versus even-aged monocultures: review and perspectives. Wood Sci Technol, 2016, 50(4): 845-880

[53] Ranjbaran G, Recupero DR, Roy CK, Schneider KA. C-SHAP: a hybrid method for fast and efficient interpretability. Appl Sci, 2025, 15(2): 672

[54] Reyes-Santías F, Cordova-Arevalo O, Rivo-Lopez E. Using flexible regression models for calculating hospital’s production functions. BMC Health Serv Res, 2020, 20: 641

[55] Sattler DF, Comeau PG, Achim A. Branch models for white spruce (Picea glauca (Moench) Voss) in naturally regenerated stands. For Ecol Manag, 2014, 325: 74-89

[56] Schmucker J, Uhl E, Steckel M, Pretzsch H. Crown allometry and growing space requirements of four rare domestic tree species compared to oak and beech: implications for adaptive forest management. Eur J Forest Res, 2022, 141(4): 587-604

[57] Shendryk Y, Gorrod E. Leveraging airborne LiDAR data and gradient boosting for mapping the density of different sized trees. IEEE J Sel Top Appl Earth Obs Remote Sens, 2021, 14: 1572-1579

[58] Šilinskas B, Varnagirytė-Kabašinskienė I, Aleinikovas M, Beniušienė L, Aleinikovienė J, Škėma M. Scots pine and Norway spruce wood properties at sites with different stand densities. Forests, 2020, 11(5): 587

[59] Steele PH, Hartley I, Cooper J, Conners T, King R. The differential thermal response of knots and clear wood following rapid heating. Res Nondestruct Eval, 2010, 21(1): 30-47

[60] Sterck FJ, Schieving F, Lemmens A, Pons TL. Performance of trees in forest canopies: explorations with a bottom-up functional–structural plant growth model. New Phytol, 2005, 166(3): 827-843

[61] Streukens S, Leroi-Werelds S. Bootstrapping and PLS-SEM: a step-by-step guide to get more out of your bootstrap results. Eur Manag J, 2016, 34(6): 618-632

[62] Tian A, Wang YH, Webb AA, Yu PT, Wang X, Liu ZB. Modelling the response of larch growth to age, density, and elevation and the implications for multifunctional management in Northwest China. J for Res, 2023, 34(5): 1423-1436

[63] Tipu RK, Batra V, Suman PVR, Pandya KS. Predictive modelling of surface chloride concentration in marine concrete structures: a comparative analysis of machine learning approaches. Asian J Civ Eng, 2024, 25(2): 1443-1465

[64] Tong YW, Chen DS, Gao HL, Xie YH, Liu JT. Developing branch diameter and length models for the planted Larix kaempferi from different latitude regions in China. J Forestry Res, 2025, 36(1): 77

[65] Usmani M, Ali Memon Z, Zulfiqar A, Qureshi R. Preptimize: automation of time series data preprocessing and forecasting. Algorithms, 2024, 17(8): 332

[66] Villasante A, Íñiguez-González G, Puigdomenech L. Comparison of various multivariate models to estimate structural properties by means of non-destructive techniques (NDTs) in Pinus sylvestris L. timber. Holzforschung, 2019, 73(4): 331-338

[67] Wang CS, Tang C, Hein S, Guo JJ, Zhao ZG, Zeng J. Branch development of five-year-old Betula alnoides plantations in response to planting density. Forests, 2018, 9(1): 42

[68] Wang ZH, Yu N, Li RS, Zou WT, Yang JC, Yin GT, Peng TL. Branch development of eight-year-old Mytilaria laosensis plantations in response to planting density. Austrian J for Sci, 2020, 137(4): 247-266

[69] Wernicke J, Körner M, Möller R, Seltmann CT, Jetschke G, Martens S. The potential of generalized additive modelling for the prediction of radial growth of Norway spruce from Central Germany. Dendrochronologia, 2020, 63: 125743

[70] Widyati E, Nuroniah HS, Tata HL, Mindawati N, Lisnawati Y, Darwo AL, Lelana NE, Mawazin OD, Prameswari D, Rachmat HH, Sutiyono DW, Wardani M, Kalima T, Yuliantivan Noordwijk M. Soil degradation due to conversion from natural to plantation forests in Indonesia. Forests, 2022, 13(11): 1913

[71] Wood SN. Stable and efficient multiple smoothing parameter estimation for generalized additive models. J Am Stat Assoc, 2004, 99(467): 673-686

[72] Wood SN, Pya N, Säfken B. Smoothing parameter and model selection for general smooth models. J Am Stat Assoc, 2016, 111(516): 1548-1563

[73] Wu JJ, Wang ZH, Zhang D, Gong C, Zhai C, Wang YY. Effects of structural and diversity attributes on biomass in different types of urban forests in Changchun, Northeast China, and suggestions for urban forest planning. Forests, 2022, 13(11): 1805

[74] Wu Y, Xu HP, Wu JM, Ji YC. The low-cost technical system of rural sustainable wood construction and its suitability based on the small-diameter round timber of fast-growing Northeast Larch. Case Stud Constr Mater, 2025, 22: e04276

[75] Wu YW, Rademacher T, Liu HY, Manzanedo R, Liang BY, Yu ST, Guo ZF, Du MC, Zhang XL. Balancing individual growth and stand carbon dynamics: optimizing age-dependent density management for Larix principis-rupprechtii plantations in semi-humid to semiarid regions. J for Res, 2026, 37(1): 74

[76] Xiao YL, Zhang HJ, Zong J, Bai XY, Yang KL, Li XM, Huang X, Hu YF, Bao YT, Razzaq A, Cui D, Wang ZJ, Zhang G. Mercury accumulation in urban forest leaves: dynamics, driving factors, and potential economic value. J Hazard Mater, 2025, 490: 137763

[77] Xu CG, Li HX, Yang JP, Peng YZ, Cai HY, Zhou J, Gu WY, Chen LX. Interpretable prediction of 3-year all-cause mortality in patients with chronic heart failure based on machine learning. BMC Med Inform Decis Mak, 2023, 23: 267

[78] Yang XZ, Zhang WH, He QY. Effects of intraspecific competition on growth, architecture and biomass allocation of Quercus liaotungensis. J Plant Interact, 2019, 14(1): 284-294

[79] Ye N, Mason E, Xu C, Morgenroth J. Estimating individual tree DBH and biomass of durable Eucalyptus using UAV LiDAR. Ecol Inform, 2025, 89: 103169

[80] Zhang HY, Feng ZK, Wang S, Ji WX. Disentangling the factors that contribute to the growth of Betula spp. and Cunninghamia lanceolata in China based on machine learning algorithms. Sustainability, 2022, 14(14): 8346

[81] Zhang J, Wang C, Wang JL, Huang X, Zhou ZL, Zhou ZT, Cheng F. Study on forest growing stock volume in Kunming City considering the relationship between stand density and allometry. Forests, 2025, 16(6): 891

[82] Zhao XL, Xie PL, Xin YT, Fan JF, Wan P, Ma HJ. Optimal management for promoting growth of poplar plantations: insights from canopy structure and light environment. J Forestry Res, 2025, 36(1): 109

[83] Zięba A, Różański W, Szwagrzyk J. Structure of dominance among tree species in relic Swiss stone pine (Pinus cembra L.) forests in Tatra Mountains. Pol J Ecol, 2020, 68(2): 159-171

RIGHTS & PERMISSIONS

Northeast Forestry University