[Assessing spatial pattern of carbon density of understory vegetation in Xiaoxing'anling area based on SGAM model]

We estimated carbon densities of the sapling, shrub, and herb layers in the Yichun region of the Xiao-xing'anling area in Heilongjiang Province, based on field measurements from 1194 natural and plantation forest plots in 2021. Combined with terrain, climate, and forest factors, we used multiple stepwise regression (SMR), generalized additive model (GAM), and spatial generalized additive model (SGAM) to fit carbon density for understory vegetation across different layers. We then analyzed the regional differentiations of understory vegetation by spatial interpolation analysis. Results showed that the average carbon density of understory vegetation was 0.418 t·hm^-2, comprising 0.223, 0.172, and 0.023 t·hm^-2 for the sapling, shrub, and herbaceous layers, respectively. The sapling and shrub layers were the primary contributors to understory carbon density, accounting for 94.3% of the total, while herbaceous layer contributed 5.7%. Stand density, canopy cover, closure, annual precipitation, and altitude were the primary factors influencing understory carbon density. Model comparison revealed that the SGAM model outperformed both SMR and GAM across all three strata, with determination coefficients of 0.79, 0.71, and 0.62, respectively. Spatial interpolation validation revealed an overall spatial distribution pattern of understory carbon density being higher in the north and lower in the south, with greater abundance in the east and less in the west. The root mean square error values for empirical Bayesian kriging interpolation were 5.47, 3.57, and 1.90 t·hm^-2 for the sapling, shrub, and herb layers, respectively, demonstrating superior prediction accuracy compared to inverse distance weighting, radial basis function, and ordinary kriging methods. By integrating the SGAM model, which accounted for both nonlinear relationships and spatial effects, with empirical Bayesian kriging interpolation, we elucidated the drivers and spatial distribution patterns of carbon density in understory. These fin-dings would provide a scientific basis for forest carbon sink assessments and understory management practices.