Anthropogenic CO2 emissions have steadily risen over the past 60 years. A significant portion (approximately 25%) of these emissions has been absorbed by the land biosphere, creating a natural land sink that mitigates climate change. Predicting future global carbon cycle dynamics (and thus climate response) hinges on understanding the processes and timescales driving this contemporary land sink. Multiple interconnected mechanisms regulate carbon flow into, through, and out of the land, with timescales ranging from days to centuries. The interplay of these processes (photosynthesis, allocation, plant growth, litterfall, plant mortality, and soil turnover) determines changes in land carbon storage. Recent research indicates that the global net land sink is concentrated in northern latitudes, likely due to increasing atmospheric CO2, reactive nitrogen deposition, and warming. Tropical regions are closer to a net-zero carbon balance because large LULCC carbon losses offset the natural sink. The Amazon, for instance, appears carbon neutral. While tropical regions may exhibit relatively low net sink values, they harbor the largest gross carbon fluxes between land and atmosphere. Therefore, changes in tropical ecosystem functioning have significant global repercussions. Process-based dynamic global vegetation models (DGVMs) are valuable tools for investigating the roles of individual drivers (atmospheric CO2, climate, nutrient deposition, and LULCC), attributing changes to specific processes, and quantifying regional sinks over relevant timescales, going beyond the limitations of current empirical and remotely sensed data.

Existing literature highlights the significant role of the land biosphere as a carbon sink in mitigating climate change. Studies have shown a strong correlation between rising atmospheric CO2 and enhanced terrestrial carbon uptake. The impact of nitrogen deposition on carbon sequestration has also been extensively documented. Moreover, the impact of LULCC is well established as a significant source of carbon emissions, particularly in tropical regions. However, the partitioning of the carbon sink between vegetation and soil remains poorly understood, and models often disagree on the direction and magnitude of changes in carbon stocks. Previous research has also emphasized the importance of plant productivity, allocation, and turnover, as well as soil processes, in regulating carbon sequestration. However, existing models struggle to consistently represent the complexity of these processes.

This study utilizes a triple (3D-matrix) approach to analyze the sources of uncertainty in DGVM estimations of the global land carbon sink. It considers external drivers (atmospheric CO2, nitrogen deposition, climate, and LULCC), main regions (tropics and extratropics), and processes (production vs. turnover) to determine their contributions to changes in the global net land carbon sink. The analysis uses a suite of 18 DGVMs from the GCB2021 (TRENDYv10). These models were run with four simulations to separate the impacts of different drivers. A process attribution framework was used to decompose changes in carbon stocks into productivity and turnover components. This enabled direct comparison between productivity and turnover contributions. The global net land sink was estimated by subtracting the atmospheric CO2 accumulation rate and ocean carbon uptake from fossil fuel emissions. This was compared to the model outputs to evaluate model performance. For spatial patterns, the models' agreement in terms of the direction of change in ecosystem carbon was assessed. Further, to better understand the underlying processes driving changes in the carbon sink, the attribution framework linked changes in vegetation and soil carbon to changes in inputs (NPP for vegetation, litterfall for soil) and outputs (turnover rates for vegetation and soil). The influence of CO2, climate change, and LULCC on these processes were identified for each model. Uncertainty in the estimates were quantified using the spread of model results. Data analysis included the calculation of global and regional annual mean values for several variables, such as NBP, NPP, heterotrophic respiration, vegetation carbon, and soil carbon. A steady-state approximation of carbon pool dynamics, corrected for non-steady-state conditions, was used to partition changes into input and turnover components, improving upon existing approaches. This analysis involved using equations to model changes in vegetation and soil carbon based on changes in inputs and turnover times, with a correction for discrepancies between the steady-state approximation and actual model simulations. The model uncertainty was characterized using the spread in model estimates for key variables.

The study found a robust agreement among DGVMs that rising atmospheric CO2 and nitrogen deposition drive the land carbon sink, while climate change and LULCC contribute to carbon emissions. The DGVMs confirmed that the net land sink is concentrated in northern latitudes, with tropical regions showing near-neutral carbon balance. However, models showed significant disagreement on the sink's partitioning between vegetation and soil. Thirteen of eighteen models indicated increased global vegetation and soil carbon stocks, but the magnitude of this increase varied greatly (0 to 80 PgC over 60 years). Five models showed a net decrease in either vegetation or soil stocks. The study's process attribution framework revealed that changes in baseline vegetation turnover rates and turnover responses to rising CO2 and nitrogen deposition were the primary sources of uncertainty in modeled vegetation carbon changes. Uncertainty in the increase of NPP and biomass due to rising CO2 was also considerable. LULCC resulted in decreased NPP and woody carbon stocks, accounting for substantial vegetation carbon losses, although the magnitude of the loss was uncertain across models. This was impacted by different representations of land management practices, such as forest management and shifting cultivation. In soil carbon, the models showed that increased production and biomass loss enhanced soil inputs, but again, the magnitude was highly uncertain. Increases in atmospheric CO2 increased soil turnover (a phenomenon called 'false priming'). LULCC primarily caused soil carbon loss via reduced litterfall due to forest-to-agricultural land conversion, particularly in the tropics. The effect of climate change on soil carbon was highly uncertain, with increases in respiration potentially offsetting increases from productivity. The overall soil carbon sink was mainly due to enhanced litter inputs from CO2-driven vegetation growth, but the model's ability to represent underlying soil processes significantly affected this estimate. Regrowth in northern regions played a significant but uncertain role in the carbon sink. The study also highlighted that DGVMs do not fully capture climate-induced mortality and the impacts of long-term forest degradation.

This research provides strong evidence that DGVMs accurately capture the global land carbon sink's long-term evolution, but important uncertainties remain in fully understanding the underlying processes. The most significant uncertainties relate to the partitioning of the sink between vegetation and soil, highlighting the shortcomings of current DGVMs in representing internal carbon cycling. The large spread in estimates across different models underscores the need for improved process representation, particularly in plant allocation, tissue lifespan, mortality, and soil carbon and nutrient cycling. Further refinements in modeling land management practices, especially forest management and shifting cultivation, are also needed. Discrepancies between model results and observational data, particularly regarding tropical forest dynamics, warrant attention. The findings emphasize the need for more comprehensive models that capture the complex interactions between various drivers and processes that affect carbon dynamics in both vegetation and soil.

This study demonstrates that while the overall global net land carbon sink is reasonably well-represented by state-of-the-art DGVMs, significant uncertainties remain in understanding the underlying processes, particularly concerning the allocation of the sink between vegetation and soil. The discrepancies between models emphasize the need for improvements in several key areas, including the treatment of plant allocation, turnover, mortality, and soil carbon cycling. Further research is needed to refine the representation of land management practices in DGVMs. Improved data and improved representation of key processes, such as climate-induced mortality and land degradation, are vital for reducing uncertainties in projections of future land carbon sinks.

The study acknowledges the limitations of relying on a large ensemble of DGVMs, each with its own structural differences and parameterizations. The lack of long-term, high-quality observational data on global vegetation and soil carbon stocks limits the ability to fully constrain model estimates. The simplified representation of soil carbon processes in many DGVMs introduces uncertainties, particularly with respect to the quantification of turnover-driven changes. Additionally, some aspects of land management practices are not consistently represented across all models, introducing inconsistencies into LULCC estimates.