Evaluation of vegetation above-ground biomass carbon from CMIP6 ESMs using global satellite-based estimates

Abstract

Accurate simulation of vegetation above-ground biomass carbon (AGC) by Earth System Models (ESMs) is crucial for reliable predictions of the terrestrial carbon sink in different scenarios. However, significant uncertainties and inter-model spread persist in current ESMs. This study provides a comprehensive evaluation of forest AGC simulations from 30 CMIP6 ESMs under the historical scenario (2010~2014) against the ESA CCI Biomass v7 product as a benchmark. Model performance was assessed at global and biome scales, spatial patterns of model-data discrepancies were quantified, and their underlying influencing factors were investigated through clustering analysis and explainable machine learning. Our results revealed an over-estimation of global forest AGC by the multi-model ensemble mean (about 14% higher than reference data), with substantial inter-model variability. Model performance was highly biome-dependent, with the magnitude of model-data discrepancies primarily associated with vegetation attributes of canopy height and forest age, as well as, local hydrothermal conditions. Specifically, simulated AGC tended to be lower than CCI Biomass estimates in tall forests and higher in shorter forests, with larger positive model-data differences also occurring under warmer and wetter conditions. Process-orientated analysis further suggested that these model-data discrepancies were linked to differences in the representation of key terrestrial processes in land models. Many over-estimating models were characterized by unrealistic AGC–net primary productivity (NPP) relationships and deficiencies in the balance of carbon stock and fluxes. Conversely, AGC under-estimation in other model clusters was primarily associated with lower simulated NPP and reduced carbon use efficiency (CUE). These diagnostic patterns suggest potential limitations in the representation of carbon input, allocation, and loss processes in current ESMs, which can guide future controlled model experiments to assess their individual contributions to model-data discrepancies. This evaluation helps prioritize model development efforts to reduce uncertainties in terrestrial carbon cycle projections.