Tracking 21st century anthropogenic and natural carbon fluxes through model-data integration

Abstract

<p>Reliable estimates of terrestrial CO<sub>2</sub> sources and sinks are crucial for the successful implementation of the Global Stocktake under the Paris Agreement and for estimating future climate mitigation potentials. However, current estimates of terrestrial CO<sub>2</sub> fluxes by process-based and semi-empirical models exhibit large uncertainties. The spread in model-based estimates is partly caused by different assumptions regarding the amount of carbon stored in vegetation and soils per unit area (=carbon density).</p> <p>Here we aim at reducing these uncertainties by assimilating an observation-based time series of 21<sup>st</sup> century carbon densities of global forests and woodlands into the semi-empirical bookkeeping model BLUE. Our novel approach enables us to distinguish the direct effects of human land use on global CO<sub>2</sub> fluxes from environmental processes. The assimilation of observational data on vegetation carbon allows us to include all impacts on CO<sub>2</sub> fluxes, including processes that are commonly not considered in model-based approaches (e.g., forest degradation). We subsequently compare our results to estimates from 13 Dynamic Global Vegetation Models (DGVMs) from the Trendy Model Intercomparison Project. Further, we identify sources of uncertainty in the bookkeeping model by comparing the vegetation carbon stocks from the observed dataset to the estimates from our data assimilation approach.</p> <p>The results from our approach show that the consideration of indirect anthropogenic influences (e.g., increasing atmospheric CO<sub>2</sub>, wildfires, climate) and their synergies on vegetation carbon leads to much higher global emissions from land use and (land-use induced) land cover changes (=E<sub>LUC</sub>) compared to only considering direct anthropogenic influences on vegetation carbon. This effect is currently excluded from common budgeting of anthropogenic influences on the carbon cycle.</p> <p>The spread in E<sub>LUC</sub> between estimates based on our assimilation approach and other bookkeeping models and DGVMs is reduced by up to 88% compared to earlier estimates of the bookkeeping model, i.e., multi-model uncertainties are substantially reduced. Despite the improvement in E<sub>LUC</sub> estimates, our analysis reveals that the land use forcing and its implementation in the bookkeeping model contribute strongly to uncertainties in the estimated carbon fluxes.</p> <p>Considering only carbon fluxes from woody vegetation due to environmental processes (=natural CO<sub>2</sub> sink), our data assimilation approach reveals that the natural CO<sub>2</sub> sink in forests and woodlands is subject to stronger annual fluctuations and responds more sensitively to extreme events like droughts than estimated by the DGVMs.</p> <p>These findings highlight the need to advance model-data integration to improve estimates of the terrestrial carbon cycle under the Global Stocktake and to foster our understanding of the impacts of environmental change on terrestrial vegetation.</p>