Prediction in trait-based ecology: global simulations of specific leaf area using a trait-based dynamic vegetation model

AbstractPredicting plant community functional traits is considered a Holy Grail of trait-based ecology because traits underpin ecosystem processes. Previous statistical, machine learning, and optimality approaches have produced global plant trait predictions. However, the utility of trait-based vegetation models, which include demographic processes and can represent trait diversity, remains unexplored at this scale. We use aDGVM2-LL, a trait- and individual-based dynamic global vegetation model (DGVM). aDGVM2-LL simulates community assembly, which is driven by natural selection, biotic, and abiotic conditions; simulated specific leaf area (SLA) is an emergent outcome of community assembly. We examine: 1) how well aDGVM2-LL can simulate global SLA by examining deviations from trait data, and 2) explore drivers of strong deviations. Compared to GBIF-derived SLA data, aDGVM2-LL displays mean SLA differences of -2.9 (m2/kg)(GBIF range ca. 4 - 35 m2/kg), a root mean square error (RMSE) of 7.25, and normalised mean absolute error (nMAE) of 26.54%. Published statistical, machine learning, and optimality approaches displayed differences with GBIF-derived trait data which range between (mean : -4.83 - 2.67, RMSE: 4.41 - 6.68, nMAE: 13.41% - 25.20%). Thus, aDGVM2-LL mean differences are comparable with published predictions while RMSEs and nMAEs are higher. Large aDGVM2-LL mismatches occur in areas where the model incorrectly simulates the relative abundances of deciduous vs. evergreen leaf phenologies. Correcting mismatches in leaf phenological abundances strongly reduces the range of mean SLA differences (-0.14 - 0.43), RMSEs (5.85 - 5.90), and nMAEs (15.44% - 20.61%). These results show that an eco-evolutionary, process-based approach can reasonably simulate global SLA values, particularly when leaf phenological abundances are accurate. Our results highlight the general importance of the global drivers of leaf phenology for leaf traits. The correct simulation of the relative abundances of deciduous and evergreen leaf phenologies is crucial to predict contemporary and future SLA.