<![CDATA[Abstract. Meunpong P, Takuathung CN, Rianthakool L, Yingdee J, Kaewgrajang T, Sriarkarin S, Jumwong N, Sangvisitpirom P, Chonlapap N, Chumsangsri T. 2026. Generalized additive models vs. traditional models for teak biomass estimation in Northern Thailand. Biodiversitas 27 (1): d270119. https://doi.org/10.13057/biodiv/d270119. Accurate estimation of aboveground biomass is essential for carbon accounting and sustainable management of teak (Tectona grandis) plantations, yet traditional allometric equations may not capture nonlinear variation in stem form and crown biomass allocation. We developed and evaluated generalized additive models (GAMs) incorporating internal stem geometry (mid-height and mid-diameter) using destructive sampling of 30 trees (n = 30) from managed teak plantations in northern Thailand. Model performance was assessed using leave-one-out cross-validation (LOOCV) and compared with a conventional log-linear DBH-only allometric model. The GAM substantially outperformed the traditional allometric model, reducing prediction error by approximately 44% (RMSE = 28.47 kg vs. 51.28 kg) and increasing predictive accuracy (R² = 0.975 vs. 0.918). GAM smoothers revealed interpretable nonlinear effects for DBH and mid-height, while mid-diameter was penalized toward linearity. Residual diagnostics indicated adequate model fit, and sensitivity analysis showed that GAM performance was robust to changes in smoother complexity. Branch biomass exhibited lower deviance explained but moderate LOOCV accuracy. Leaf biomass showed low predictability due to inherently high biological variability. Monte Carlo simulations propagated tree-level uncertainty to plot-level biomass. The GAM produced substantially narrower 95% confidence intervals (0.91-1.85 t ha-¹) than the allometric model (1.76-4.14 t ha-¹), indicating improved stability for operational carbon assessments. These results demonstrate that flexible modeling approaches incorporating internal stem geometry can significantly enhance both accuracy and precision of teak biomass estimation. Generalized additive models provide a robust alternative to traditional allometries, particularly where nonlinear stem structure influences biomass allocation, and offer clear advantages for plantation-scale carbon accounting and sustainable forest management. ]]>
