Accuracy and biological plausibility of a machine learning-based transition matrix growth model for long-term mixed forest projections under climate change

Abstract

Modeling mixed forest growth is challenging due to multiple tree species, uneven-aged structures, and complex environmental interactions. Traditional transition matrix models, widely used for predicting ingrowth, upgrowth, and mortality in mixed forests, rely on statistical approaches but have limited capacity to handle high-dimensional data. Machine learning has the potential, but there is limited application and a lack of comparison, especially in terms of model accuracy and biological plausibility. This study developed a climate-sensitive transition matrix growth model with the integration of machine learning (ML-Matrix) for mixed conifer-broad-leaved forests in northeastern China. Forest data were obtained from permanent plots in the 5th–9th National Forest Inventories. We evaluated five machine learning algorithms—random forest (RF), boosted regression trees (BRT), artificial neural network (ANN), support vector machine (SVM), and k‑nearest neighbor (KNN)—to model upgrowth, two-stage mortality, and recruitment. Results showed BRT excelled in modeling upgrowth of Pinus koraiensis and the first-stage mortality, while ANN excelled for upgrowth of other tree species. For the second-stage mortality, ANN performed best for two broad-leaved groups, while SVM outperformed for three coniferous groups. Optimal recruitment algorithms (SVM, BRT, or ANN) varied by species. Stand variables were the most important predictors with the relative important values of 85.1%–99.8% for upgrowth, 49.0%–88.4% for mortality, and 46.4%–84.3% for recruitment. Compared to the traditional statistical transition matrix (TS-Matrix), ML-Matrix achieved higher test-set accuracy for four tree species groups, excluding Pinus koraiensis. However, long-term simulations revealed TS-Matrix produced more biologically reasonable prediction of stand density, basal area and volume. This study highlighted the potential of ML-Matrix and the need for caution with its long-term projections for mixed forests under climate change.

Graphical abstract

Keywords

Forest dynamics / Machine learning / Upgrowth / Ingrowth / Mortality / Climate change

References

[1] Aertsen W, Kint V, van Orshoven J, Özkan K, Muys B. Comparison and ranking of different modelling techniques for prediction of site index in Mediterranean mountain forests. Ecol Model. 2010, 221(8): 1119-1130.

[2] Aguirre F, Sebastian A, Le Gallo M, Song W, Wang T, Yang JJ, Lu W, Chang M-F, Ielmini D, Yang Y, Mehonic A, Kenyon A, Villena MA, Roldan JB, Wu Y, Hsu H-H, Raghavan N, Sune J, Miranda E, Eltawil A, Setti G, Smagulova K, Salama KN, Krestinskaya O, Yan X, Ang K-W, Jain S, Li S, Alharbi O, Pazos S, Lanza M. Hardware implementation of memristor-based artificial neural networks. Nat Commun. 2024, 15(1. 1974

[3] Allen CD, Breshears DD, McDowell NG. On underestimation of global vulnerability to tree mortality and forest die-off from hotter drought in the Anthropocene. Ecosphere. 2015, 68. 129

[4] Altman NS. An introduction to kernel and nearest-neighbor nonparametric regression. Am Stat. 1992, 463): 175-185.

[5] Anderson-Teixeira KJ, Herrmann V, Rollinson CR, Gonzalez B, Gonzalez-Akre EB, Pederson N, Alexander MR, Allen CD, Alfaro-Sanchez R, Awada T, Baltzer JL, Baker PJ, Birch JD, Bunyavejchewin S, Cherubini P, Davies SJ, Dow C, Helcoski R, Kaspar J, Lutz JA, Margolis EQ, Maxwell JT, McMahon SM, Piponiot C, Russo SE, Samonil P, Sniderhan AE, Tepley AJ, Vasickova I, Vlam M, Zuidema PA. Joint effects of climate, tree size, and year on annual tree growth derived from tree-ring records of ten globally distributed forests. Glob Change Biol. 2022, 28(1): 245-266.

[6] Bojorquez A, Martinez-Yrizar A, Alvarez-Yepiz JC. A landscape assessment of frost damage in the northmost Neotropical dry forest. Agric for Meteorol. 2021, 308. 108562

[7] Breiman L. Random forests. Mach Learn. 2001, 45(1): 5-32.

[8] Cao QV. Linking individual-tree and whole-stand models for forest growth and yield prediction. For Ecosyst. 2014, 1(18): 177-184

[9] Chen Z, Dayananda B, Fu B, Li Z, Jia Z, Hu Y, Cao J, Liu Y, Xie L, Chen Y, Wu S. Research on the potential of forestry's carbon-neutral contribution in China from 2021 to 2060. Sustainability. 2022, 14(9): 5444.

[10] Clark JS, Beckage B, Camill P, Cleveland B, HilleRisLambers J, Lichter J, McLachlan J, Mohan J, Wyckoff P. Interpreting recruitment limitation in forests. Am J Bot. 1999, 8611-16.

[11] Colombi T, Walder F, Buechi L, Sommer M, Liu K, Six J, van der Heijden MGA, Charles R, Keller T. On-farm study reveals positive relationship between gas transport capacity and organic carbon content in arable soil. SoiL. 2019, 5(1): 91-105.

[12] Cover TM, Hart PE. Nearest neighbor pattern classification. IEEE Trans Inf Theory. 1967, 13(1): 21-27.

[13] dos Reis AA, Carvalho MC, de Mello JM, Gomide LR, Ferraz Filho AC, Acerbi Junior FW. Spatial prediction of basal area and volume in Eucalyptus stands using Landsat TM data: an assessment of prediction methods. N Z J for Sci. 2018, 48. 1

[14] Du X, Chen XY, Zeng WS, Meng JH. A climate-sensitive transition matrix growth model for uneven-aged mixed-species oak forests in North China. Forestry. 2021, 94(2): 258-277.

[15] Du X, Wang X, Meng JH. A climate-sensitive transition matrix growth model for Masson pine (Pinus massoniana Lamb.) natural forests in Hunan Province, South-Central China. Forests. 2023, 14(8. 1539

[16] Du X, Lei X, He X, Lan J, Guo H, Xu Q. Ecosystem service multifunctionality of mixed conifer-broad-leaved forests under climate change and forest management based on matrix growth modelling. For Ecosyst. 2024, 11. 100231

[17] Dumais D, Prévost M. Physiology and growth of advance Picea rubens and Abies balsamea regeneration following different canopy openings. Tree Physiol. 2014, 342): 194-204.

[18] Elith J, Leathwick JR, Hastie T. A working guide to boosted regression trees. J Anim Ecol. 2008, 77(4): 802-813.

[19] Friedman JH. Greedy function approximation: a gradient boosting machine. Ann Stat. 2001, 29(5): 1189-1232.

[20] Goldstein A, Kapelner A, Bleich J, Pitkin E. Peeking inside the black box: visualizing statistical learning with plots of individual conditional expectation. J Comput Graph Stat. 2015, 24(1): 44-65.

[21] Gong XF, Yuan DY, Zhu LJ, Li ZS, Wang XC. Long-term changes in radial growth of seven tree species in the mixed broadleaf-Korean pine forest in Northeast China: are deciduous trees favored by climate change?. J for Res. 2024, 35(1): 70.

[22] Greenwell BM. pdp: an R package for constructing partial dependence plots. R J. 2017, 91421-436.

[23] Greenwell BM, Boehmke BC. Variable importance plots: an introduction to the vip package. R J. 2020, 12(1): 343-366.

[24] Hamidi SK, Zenner EK, Bayat M, Fallah A. Analysis of plot-level volume increment models developed from machine learning methods applied to an uneven-aged mixed forest. Ann for Sci. 2021, 78(1. 4

[25] Hamner B, Frasco M (2018) Metrics: evaluation metrics for machine learning. R package version 0.1.4

[26] Hao QY, Meng FR, Zhou YP, Wang JX. Determining the optimal selective harvest strategy for mixed-species stands with a transition matrix growth model. New for. 2005, 293): 207-219.

[27] Hao QY, Meng FR, Zhou YP, Wang JX. A transition matrix growth model for uneven-aged mixed-species forests in the Changbai Mountains, northeastern China. New for. 2005, 29(3): 221-231.

[28] He Y, Liu H, Yang Q, Cao Y, Yin H, Zhou Z, Yu Q, Wang X. Neighborhood effects on tree mortality depend on life stage of neighbors. Front Plant Sci. 2022, 13. 838046

[29] He X, Lei XD, Liu D, Lei YC. Developing machine learning models with multiple environmental data to predict stand biomass in natural coniferous-broad leaved mixed forests in Jilin Province of China. Comput Electron Agric. 2023, 212. 108162

[30] He X, Lei XD, Liu D, Lei YC, Gao WQ, Lan J. Understanding the contribution of structural diversity to stand biomass for carbon management of mixed forests using machine learning algorithms. Eur J for Res. 2025, 144(1): 13-28.

[31] Hengl T, de Jesus JM, Heuvelink GBM, Gonzalez MR, Kilibarda M, Blagotic A, Shangguan W, Wright MN, Geng XY, Bauer-Marschallinger B, Guevara MA, Vargas R, MacMillan RA, Batjes NH, Leenaars JGB, Ribeiro E, Wheeler I, Mantel S, Kempen B. SoilGrids250m: global gridded soil information based on machine learning. PLoS ONE. 2017, 12(2. e0169748

[32] Käber Y, Meyer P, Stillhard J, De Lombaerde E, Zell J, Stadelmann G, Bugmann H, Bigler C. Tree recruitment is determined by stand structure and shade tolerance with uncertain role of climate and water relations. Ecol Evol. 2021, 1117): 12182-12203.

[33] Karatzoglou A, Smola A, Hornik K, Zeileis A. kernlab-an S4 package for kernel methods in R. J Stat Softw. 2004, 11(9): 1-20.

[34] Kuhn M. Building predictive models in R using the caret package. J Stat Softw. 2008, 28(5): 1-26.

[35] Lacroix EM, Aeppli M, Boye K, Brodie E, Fendorf S, Keiluweit M, Naughton HR, Noel V, Sihi D. Consider the anoxic microsite: acknowledging and appreciating spatiotemporal redox heterogeneity in soils and sediments. ACS Earth Space Chem. 2023, 7(9): 1592-1609.

[36] Lei X. Applications of machine learning algorithms in forest growth and yield prediction. J Beijing For Univ. 2019, 41(12): 23-36in Chinese

[37] Li Y, Feng Z, Chen S, Zhao Z, Wang F. Application of the artificial neural network and support vector machines in forest fire prediction in the Guangxi autonomous region, China. Discrete Dyn Nat Soc. 2020, 2020. 5612650

[38] Liang J. Dynamics and management of Alaska boreal forest: an all-aged multi-species matrix growth model. For Ecol Manag. 2010, 260(4): 491-501.

[39] Liang J, Picard N. Matrix model of forest dynamics: an overview and outlook. For Sci. 2013, 59(3): 359-378.

[40] Liang J, Zhou M. A geospatial model of forest dynamics with controlled trend surface. Ecol Model. 2010, 221(19): 2339-2352.

[41] Liaw A, Wiener M. Classification and regression by randomForest. R News. 2002, 2(3): 18-22

[42] Liu Z, Peng C, Work T, Candau J-N, DesRochers A, Kneeshaw D. Application of machine-learning methods in forest ecology: recent progress and future challenges. Environ Rev. 2018, 26(4): 339-350.

[43] Long S, Zeng S, Shi Z, Yang S. Estimating the self-thinning boundary line for oak mixed forests in central China by using stochastic frontier analysis and a proposed variable density model. Ecol Evol. 2022, 12(9. e9064

[44] Luo Y, Chen HYH. Competition, species interaction and ageing control tree mortality in boreal forests. J Ecol. 2011, 996): 1470-1480.

[45] Ma W, Liang JJ, Cumming JR, Lee E, Welsh AB, Watson JV, Zhou M. Fundamental shifts of central hardwood forests under climate change. Ecol Model. 2016, 332: 28-41.

[46] Ma W, Woodall CW, Domke GM, D'Amato AW, Walters BF. Stand age versus tree diameter as a driver of forest carbon inventory simulations in the northeastern U.S. Can J for Res. 2018, 48(10): 1135-1147.

[47] Ma W, Zhou X, Liang J, Zhou M. Coastal Alaska forests under climate change: what to expect?. For Ecol Manag. 2019, 448: 432-444.

[48] Ma W, Lin G, Liang J. Estimating dynamics of central hardwood forests using random forests. Ecol Model. 2020, 419. 108947

[49] Matiza C, Mutanga O, Peerbhay K, Odindi J, Lottering R. A systematic review of remote sensing and machine learning approaches for accurate carbon storage estimation in natural forests. South for. 2023, 85(3–4): 123-141.

[50] Matsushita M, Takata K, Hitsuma G, Yagihashi T, Noguchi M, Shibata M, Masaki T. A novel growth model evaluating age-size effect on long-term trends in tree growth. Funct Ecol. 2015, 29(10): 1250-1259.

[51] Meyer D, Dimitriadou E, Hornik K, Weingessel A, Leisch F (2024) e1071: misc functions of the department of statistics, probability theory group (Formerly: E1071), TU Wien. R package version 1.7-16. In. Vienna TU Wien

[52] Namaalwa J, Eid T, Sankhayan P. A multi-species density-dependent matrix growth model for the dry woodlands of Uganda. For Ecol Manag. 2005, 213(1–3): 312-327.

[53] Olden JD, Lawler JJ, Poff NL. Machine learning methods without tears: a primer for ecologists. Q Rev Biol. 2008, 83(2): 171-193.

[54] Ou QX, Lei XD, Shen CC. Individual tree diameter growth models of larch-spruce-fir mixed forests based on machine learning algorithms. Forests. 2019, 10(2. 187

[55] Padilla-Martinez JR, Paul C, Husmann K, Corral-Rivas JJ, von Gadow K. Grouping tree species to estimate basal area increment in temperate multispecies forests in Durango, Mexico. For Ecosyst. 2024, 11. 100158

[56] Peng CH, Liu JX, Dang QL, Apps MJ, Jiang H. TRIPLEX: a generic hybrid model for predicting forest growth and carbon and nitrogen dynamics. Ecol Model. 2002, 1531–2): 109-130.

[57] Peterson RL, Liang J, Barrett TM. Modeling population dynamics and woody biomass in Alaska coastal forest. For Sci. 2014, 60(2): 391-401.

[58] Petrenko TY, Korznikov KA, Kislov DE, Belyaeva NG, Krestov PV. Modeling of cold-temperate tree Pinus koraiensis (Pinaceae) distribution in the Asia-Pacific region: climate change impact. For Ecosyst. 2022, 9. 100015

[59] Pretzsch H, Block J, Dieler J, Dong PH, Kohnle U, Nagel J, Spellmann H, Zingg A. Comparison between the productivity of pure and mixed stands of Norway spruce and European beech along an ecological gradient. Ann for Sci. 2010, 67(7. 712

[60] Qin Y, Wu B, Lei X, Feng L. Prediction of tree crown width in natural mixed forests using deep learning algorithm. For Ecosyst. 2023, 10. 100109

[61] R Core Team. R: A Language and Environment for Statistical Computing. 2024, Vienna, Austria

[62] Ralstona R, Buongiornoa J, Schultea B, Friedb J. Non-linear matrix modeling of forest growth with permanent plot data: the case of uneven-aged Douglas-fir stands. Int Trans Oper Res. 2003, 105): 461-482.

[63] Rawls WJ, Pachepsky YA, Ritchie JC, Sobecki TM, Bloodworth H. Effect of soil organic carbon on soil water retention. Geoderma. 2003, 116(1–2): 61-76.

[64] Ridgeway G. gbm: generalized boosted regression models. R Pack Vers. 2024, 2(2): 2

[65] Robin X, Turck N, Hainard A, Tiberti N, Lisacek F, Sanchez J-C, Müller M. pROC: an open-source package for R and S+ to analyze and compare ROC curves. BMC Bioinform. 2011, 12. 77

[66] Rosa R, Soares P, Tomé M. Evaluating the economic potential of uneven-aged maritime pine forests. Ecol Econ. 2018, 143: 210-217.

[67] Rozendaal DMA, Phillips OL, Lewis SL, Affum-Baffoe K, Alvarez-Davila E, Andrade A, Aragao LEOC, Araujo-Murakami A, Baker TR, Banki O, Brienen RJW, Camargo JLC, Comiskey JA, Djuikouo Kamdem MN, Fauset S, Feldpausch TR, Killeen TJ, Laurance WF, Laurance SGW, Lovejoy T, Malhi Y, Marimon BS, Marimon Junior B-H, Marshall AR, Neill DA, Nunez Vargas P, Pitman NCA, Poorter L, Reitsma J, Silveira M, Sonke B, Sunderland T, Taedoumg H, Ter Steege H, Terborgh JW, Umetsu RK, van Der Heijden GMF, Vilanova E, Vos V, White LJT, Willcock S, Zemagho L, Vanderwel MC. Competition influences tree growth, but not mortality, across environmental gradients in Amazonia and tropical Africa. Ecology. 2020, 1017. e03052

[68] Schapire RE (2003) The boosting approach to machine learning: an overview. In, Workshop on nonlinear estimation and classification. Berkeley, CA, pp. 149–171

[69] Schölkopf B, Smola AJ, Williamson RC, Bartlett PL. New support vector algorithms. Neural Comput. 2000, 1251207-1245.

[70] Shabani S, Pourghasemi HR, Blaschke T. Forest stand susceptibility mapping during harvesting using logistic regression and boosted regression tree machine learning models. Glob Ecol Conserv. 2020, 22. e00974

[71] Song L, Yang B, Liu LL, Mo YX, Liu WJ, Meng XJ, Lu HZ, Li Y, Zakari S, Tan ZH, Fan ZX, Zhang YJ. Spatial-temporal differentiations in water use of coexisting trees from a subtropical evergreen broadleaved forest in Southwest China. Agric for Meteorol. 2022, 316. 108862

[72] Spathelf P, Durlo MA. Transition matrix for modeling the dynamics of a subtropical seminatural forest in southern Brazil. For Ecol Manag. 2001, 151(1–3): 139-149.

[73] Stovall AEL, Shugart H, Yang X. Tree height explains mortality risk during an intense drought. Nat Commun. 2019, 10: 4385.

[74] Tang Z, Tang J, Liu W, Chen G, Feng C, Zhang A. Predictive value of gradient boosting decision trees for postoperative atelectasis complications in patients with pulmonary destruction. Am J Transl Res. 2024, 16(7): 2864-2876.

[75] Teng M, Li S, Yang J, Wang S, Fan C, Ding Y, Dong J, Lin H, Wang S. Long-term PM_2.5 concentration prediction based on improved empirical mode decomposition and deep neural network combined with noise reduction auto-encoder- a case study in Beijing. J Clean Prod. 2023, 428. 139449

[76] Tian HL, Zhu JH, He X, Chen XY, Jian ZJ, Li CY, Ou QX, Li Q, Huang GS, Liu CF, Xiao WF. Using machine learning algorithms to estimate stand volume growth of Larix and Quercus forests based on national-scale Forest Inventory data in China. For Ecosyst. 2022, 9. 100037

[77] Toïgo M, Castagneyrol B, Jactel H, Morin X, Meredieu C. Effects of tree mixture on forest productivity: tree species addition versus substitution. Eur J for Res. 2022, 1411165-175.

[78] Urgoiti J, Messier C, Keeton WS, Belluau M, Paquette A. Functional diversity and identity influence the self-thinning process in young forest communities. J Ecol. 2023, 111(9): 2010-2022.

[79] Usher MB. A matrix model for forest management. Biometrics. 1969, 25(2): 309.

[80] Venables WN, Ripley BD. Modern applied statistics with S. 20024New York, Springer.

[81] Wang T, Hamann A, Spittlehouse DL, Murdock TQ. ClimateWNA-High-Resolution spatial climate data for Western North America. J Appl Meteorol Climatol. 2012, 51116-29.

[82] Wang T, Wang G, Innes JL, Seely B, Chen B. ClimateAP: an application for dynamic local downscaling of historical and future climate data in Asia Pacific. Front Agric Sci Eng. 2017, 4(4): 448-458.

[83] Wang X, Pederson N, Chen Z, Lawton K, Zhu C, Han S. Recent rising temperatures drive younger and southern Korean pine growth decline. Sci Total Environ. 2019, 649: 1105-1116.

[84] Wang W, Wang J, Meng J. A climate-sensitive mixed-effects tree recruitment model for oaks ( Quercus spp.) in Hunan Province, south-central China. For Ecol Manag. 2023, 528. 120631

[85] Welle T, Aschenbrenner L, Kuonath K, Kirmaier S, Franke J. Mapping Dominant Tree Species of German Forests. Remote Sens. 2022, 14(14. 3330

[86] Wickham H. ggplot2: Elegant graphics for data analysis. 2016, New York, Springer-Verlag.

[87] Wu YC, Feng JW. Development and application of artificial neural network. Wirel Pers Commun. 2018, 1022): 1645-1656.

[88] Wu C, Chen Y, Hong X, Liu Z, Peng C. Evaluating soil nutrients of Dacrydium pectinatum in China using machine learning techniques. For Ecosyst. 2020, 7(1. 30

[89] Xu L, Yu H, Chen Z, Du W, Chen N, Huang M. Hybrid deep learning and S2S model for improved sub-seasonal surface and root-zone soil moisture forecasting. Remote Sensing. 2023, 1513. 3410

[90] Yu H, Cooper AR, Infante DM. Improving species distribution model predictive accuracy using species abundance: application with boosted regression trees. Ecol Model. 2020, 432. 109202

[91] Yu Y, Zhou ZY, Sharma RP, Zhang LJ, Du MY, Zhang HR. Comparing crown ratio models for spruce-fir broadleaved mixed forests using beta regression and random forest algorithm. Comput Electron Agric. 2024, 225. 109302

[92] Zhang L, Wang L, Zhang X, Liu S, Sun P, Wang T. The basic principle of random forest and its applications in ecology: a case study of Pinus yunnanensis. Acta Ecol Sin. 2014, 343): 650-659

[93] Zhou ZY, Zhang HR, Sharma RP, Zhang XH, Feng LY, Du MY, Zhang LJ, Feng HY, Hu XF, Yu Y. Modelling height to crown base using non-parametric methods for mixed forests in China. Ecol Inform. 2025, 85. 102957

[94] Zuur AF, Ieno EN, Smith GM. Analysing Ecological Data. 2007, New York, USA, Springer.

RIGHTS & PERMISSIONS

Northeast Forestry University