Nutrient resorption is a key process through which plants optimize resource use in nutrient-limited environments. Global nutrient addition experiments and meta-analyses have revealed high variability in nutrient resorption efficiency (NuRE) and its influencing factors. However, most existing ecosystem models utilize static constants for NuRE, which fail to capture this variability and limit the model’s ability to accurately describe nutrient cycling. Here, we introduce two innovations: (i) a multifactor allometric model that extends conventional single-factor formulations alongside a hybrid framework that couples the allometric core with random forest (RF) residual correction to capture nonlinearities and interactions; and (ii) utilization of the resorbed nutrient amount (ReNu), rather than the ratio-based NuRE, as a more robust modeling target to reduce uncertainty and improve predictability. Using independent datasets from China and a global meta-analysis, nutrient resorption exhibited substantial variability. Allometric modeling predicted ReNu with R 2 > 0.7, outperforming NuRE ( R 2 < 0.3), and hybrid modeling further reduced prediction error. NuRE is less robust and relies more on RF residual correction. By combining the interpretability of parametric allometry with the flexibility of data-driven learning, our framework provides a more accurate and dynamic representation of nutrient resorption for modeling forest nutrient cycling under global environmental change.
Hua et al. (Sun,) studied this question.