The Cause And Evolution Of Terrestrial Carbon Cycle Modeling Uncertainty In Earth System Models

Anthropogenic carbon emissions have led to the continuous accumulation of greenhouse gases in the atmosphere,resulting in global climate change.Earth System Models(ESMs)serve as crucial tools for studying global climate change and are widely used for risk assessment and addressing climate-related issues.However,model uncertainty stemming from the terrestrial carbon cycle has long been a limiting factor,which undermines the value of the model outputs but has also stimulated model development and improvement.Over the past decades,unremitting efforts have been made from experimental,observational and modeling sides to reduce uncertainty in the terrestrial carbon cycle within ESMs.With more and more processes being incorporated into models,the simulation of the terrestrial carbon cycle in ESMs has become more intricate.However,it remains unclear how the evolution and improvement of models will influence the uncertainty of the terrestrial carbon cycle.In this study,we expanded the traceability analysis framework for the terrestrial carbon cycle and combined it with numerous observational datasets.We then utilized the fifth and sixth phases of the Coupled Model Intercomparison Project(CMIP5 and CMIP6)and a Carbon-NitrogenPhosphorus coupled model,CABLE,to assess the sources and evolution of uncertainty in the terrestrial carbon cycle over the model development process.The main findings are as follows:(1)The modeled global terrestrial carbon storage has converged among ESMs from CMIP5(1936.9 ± 739.3 Pg C)to CMIP6(1774.4 ± 439.0 Pg C)but is persistently lower than the observation-based estimates(2285 ± 669 Pg C).Although over half of the CMIP6 models have explicitly included the nutrient limitation on terrestrial carbon processes,and some models have added vertical representation of soil biogeochemistry,CMIP6 models still underestimated the carbon storage in polar ecosystems.Outside of the circumpolar region,the modeled carbon stocks in two CMIPs fell within the ranges of observation-based estimates.By further decomposing terrestrial carbon storage into net primary production(NPP)and ecosystem carbon residence time(),we found that the decreased inter-model spread in land carbon storage primarily resulted from more accurate simulations on NPP among ESMs from CMIP5 to CMIP6.The persistent underestimation of land carbon storage was caused by the biased.In CMIP5(28.4± 7.7 years)and CMIP6(31.1 ± 8.3 years),the modeled was far shorter than the observation-based estimates(35.9 ～ 52.9 years).Thus,became the essential source for model ensemble bias and inter-model spread in terrestrial carbon storage.(2)We further investigated whether and where the new-generation models in CMIP6 projected robust and uncertain land carbon change,and how the robustness and uncertainty evolved from CMIP5 to CMIP6.Almost no model projections on land carbon change in CMIP5 passed the robustness criteria under low,moderate and high carbon emission scenarios.In CMIP6,we detected robust land carbon increases under three climate change scenarios,mostly in boreal and temperate forests.The CMIP6 ESMs showed a robust projection of land carbon increase on 6.6% of the land surface under the low carbon emission scenario.The fraction decreased to 4.9% and 2.7% under moderate and high carbon emission scenarios,respectively.Under three climate change scenarios,projected land carbon changes between 1986-2005 and 2081-2100 on nearly60% of the land surface were uncertain.We assessed the leading uncertainty source over the uncertain regions.The NPP-driven ecosystem carbon change was the leading uncertainty source over nearly 50% of the uncertain regions.Over parts of Brazil,western and eastern Africa tropics,the-driven term was the leading source of model uncertainty in ecosystem carbon change.The disequilibrium-driven term dominated model uncertainty over boreal and tundra regions.(3)One of the key reasons affecting the simulation and projection of the terrestrial carbon cycle in CMIP5 to CMIP6 was that: more models in CMIP6 models have coupled carbon-nitrogen interactions.We then developed an approach to study the disequilibrium magnitude of the terrestrial carbon cycle,and applied with a CarbonNitrogen-Phosphorus coupled model CABLE.We evaluated the impact of coupling nutrient limitations on the simulations of terrestrial carbon cycle.Considering nutrient limitations significantly reduced the disequilibrium magnitude of the terrestrial carbon cycle.Over the modeled period of 1901-2013,absolute change in disequilibrium magnitude was 497.6 Pg C without nutrient limitations,while decreased to 155.6 and124.3 Pg C under nitrogen and nitrogen-phosphorus limitations,respectively.To understand the underlying reasons,we further disaggregated the changes of disequilibrium magnitude into changes in steady-state carbon storage and transit carbon storage,with the former being decomposed into NPP-driven change,-driven change,and change induced by NPP-interactions.We found that nutrient constrained the increase of disequilibrium magnitude primarily by dampening the NPP-driven changes in the steady-state carbon storage.In summary,this study enhanced the traceability analysis framework of the terrestrial carbon cycle.For the historical simulations,the framework can trace the uncertainty sources of inter-model spread and diagnose the key process dominating model-data bias.For the future projections,it can rigorously differentiate regions where the model ensemble shows robust or uncertain projections of land carbon cycle.At the same time,we developed an approach by combining a process-based numerical model and an analytical framework to evaluate the role of nutrient limitation on the disequilibrium magnitude.The proposed framework can help investigate how changes in NPP,and land carbon storage could affect disequilibrium magnitude under different model assumptions of carbon-nutrient coupled models.This study gives a new perspective for the future study of modeling uncertainty in terrestrial carbon cycle.From the aspect of reducing model uncertainty,this study has detected where model performance needs further improvement.This provides important support for improving models and enhancing the accuracy,precision,and reliability of simulations,ultimately contributing to research on global climate change.