Modeling suggests fossil fuel emissions have been driving increased land carbon uptake since the turn of the 20th Century
Abstract
Introduction
Methods
Approach
We use historical reconstructions of gross primary productivity (GPP) from offline and fully coupled state-of-the-art Earth system models (ESMs). Offline reconstructions are drawn from 11 terrestrial biosphere simulators in the Multi-scale synthesis and Terrestrial Model Intercomparison Project (MsTMIP Version 1; https://doi.org/10.3334/ORNLDAAC/1225) (ref. 11,39). MsTMIP uses a constrained simulation protocol (driving data, vegetation cover, boundary conditions, and steady-state spin-up protocol are all standardized; only model structure varies). All runs use observation-based driving d ata to replicate 1901 to 2010 historical conditions34 with a semi-factorial design where time-varying factors of climate (SG1), land cover and land use change (LULCC) (SG2), [CO2] (SG3), and nitrogen deposition (BG1) are sequentially enabled after steady-state (RG1)—the reference run11,39 based on a repeated cycle of a randomized 30-yr block of trend-free weather—is achieved (MsTMIP simulation name in parenthesis). MsTMIP models used for this study are CLM, CLM4VIC, DLEM, GTEC, ISAM, LPJ-wsl, ORCHIDEE-LSCE, SiBCASA, TEM6, VEGAS2.1 and VISIT. Only CLM, CLM4VIC, DLEM and TEM6 simulate BG1, i.e., have carbon-nitrogen coupling. Fully coupled reconstructions for 1901 to 2010 for 13 ESMs (Extended Data Table 1) are taken from the CMIP5 archive (http://cmip-pcmdi.llnl.gov/cmip5/data_portal.html). CMIP5, the fifth phase of the Coupled Model Intercomparison Project, serves as a central repository for ESM simulations that inform the Fifth Assessment Report (AR5) of the Intergovernmental Panel on Climate Change (IPCC) (https://ipcc.ch/report/ar5/). CMIP5 is designed to provide a multi-model framework to investigate model differences in carbon cycle and cloud-based feedbacks as well as climate predictability and the range in ESM responses from similar forcing inputs12. For this study ESMs are chosen based on simulation experiment availability. Each ESM has, at a minimum, a preindustrial control (piControl) and historical (historical) experiments (CMIP5 simulation experiment name in parenthesis; see Extended Data Table 1). In addition, experiments that differ from piControl or historical by a single factor (or factor set, e.g., all anthropogenic forcings) are also included: (1) with greenhouse gas forcing only (historicalGHG), (2) with natural forcing only (historicalNat), (3) with LULCC only and/or (4) with anthropogenic forcings only (variants of historicalMisc). Finally, two fully coupled carbon/climate experiments are included if available: (1) the carbon cycle sees preindustrial levels of CO2 but the radiation code sees historical conditions (esmFdbk2) and (2) the radiation code sees preindustrial levels of CO2 but the carbon cycle sees historical CO2 (esmFixClim2). In all cases only one realization per ESM is used; land carbon cycling is driven by model structure and therefore insensitive to initial conditions. For both offline and coupled reconstructions we analyze annual values as global aggregates as well as by grid cell. For mapped values CMIP5 reconstructions are resampled to a half-degree spatial resolution to match the MsTMIP land mask and use the barren mask from a contemporary upscaled eddy covariance product2. Our results are based on integration over the full ensemble of ESMs—the consensus ensemble mean40—using “one-model-one-vote” (ref. 41) and assume that an unweighted multi-model mean is the best estimate42,43,44. We note that not all reconstructions share the same set of simulation experiments. Uncertainty is calculated as ensemble spread based on bootstrapping with 1000 bootstrap replicates and is expressed as 90% confidence intervals throughout.Attribution
Attribution is based on differencing9. For CMIP5 and MsTMIP, we assume the difference between two simulations is solely attributable to the relevant factor, e.g., subtracting SG2 from SG3 recovers the effect of time-varying [CO2] as both MsTMIP simulations are identical apart from SG3 having dynamic [CO2] enabled. Similarly, in CMIP5, subtracting piControl from historicalNat GPP recovers the effect of natural forcings only as both simulations are identical apart from historicalNat including changes in natural forcings–solar irradiance and volcanic aerosols. For climatic factors (near-surface air temperature, precipitation, and downwelling shortwave radiation) the effect of each is recovered using a machine learning-based emulator of MsTMIP run SG1 (time-varying climate only). Here the random forest algorithm45 is trained with SG1 GPP as the target and near-surface air temperature, precipitation, and downwelling shortwave radiation as explanatory variables. The explanatory variables are those used in forcing all offline MsTMIP runs and are identical across all MsTMIP models. Training is done by grid cell at monthly time step for each MsTMIP model individually. For the SG1 emulation (Extended Data Fig. 4) median variance explained, in a least-squares sense based on those observations not used in training (out-of-bag data points), is at least 93% across all models and all grid cells. As the climate space used to force RG1 is a randomized subset of that used to force SG1 (ref. 33,34) the random forest algorithm does not have to extrapolate beyond the limits of the training data. As such the emulation of RG1 shows equivalent skill (Extended Data Fig. 4) with median variance explained of 96%. For both emulations, skill across space is highly similar with the lowest skill in the Indonesian tropics and Australia and the highest skill across the extratropical Northern Hemisphere. The effect of each climatic factor is then calculated by differencing after sequentially enabling downwelling shortwave radiation, precipitation, and then near-surface air temperature in the emulator. As an illustration, the effect of downwelling shortwave radiation is recovered based on the difference between two emulations: the RG1 case (all climate explanatory variables use RG1 climate) subtracted from the emulation where SG1 downwelling shortwave radiation is paired with RG1 precipitation and RG1 near-surface air temperature. While the emulator offers a computationally inexpensive approach to attribute changes in GPP to individual climate drivers it is not process-aware but rather maps changes in climate to instantaneous changes in GPP. As such, the emulator provides only a first-order assessment (interaction terms are excluded) of how each climate driver impacts GPP without having to formally execute additional MsTMIP simulations. It is important to note that climatic effects cannot be precisely attributed to anthropogenic or natural forcings. In MsTMIP prescribed forcing data is based on historical information, i.e., contains a mix of anthropogenic changes as well as natural variability in Earth’s climate system. In total we attribute changes in GPP from 1901 to 2010 to seven single factors and two factor sets: climate, LULCC, [CO2], nitrogen deposition, near-surface air temperature, precipitation, and downwelling shortwave radiation as well as all natural forcings (changes in solar irradiance and volcanic aerosols) vs. all anthropogenic forcings (with emphasis on well-mixed greenhouse gases and LULCC; see Extended Data Table 1). We note that attributed changes at 1901, the start of the analysis period, are not a priori zero as anthropogenic impacts on carbon cycling predate 1901.Data availability
References
- 1.
Le Quéré, C. et al. Global Carbon Budget 2017, Earth Syst. Sci. Data 10, 405–448, https://doi.org/10.5194/essd-10-405-2018 (2018).
- 2.
Jung, M. et al. Compensatory water effects link yearly global land CO2 sink changes to temperature. Nature 541(7638), 516–520 (2017).
- 3.
Hegerl, G.C. et al. Good Practice Guidance Paper on Detection and Attribution Related to Anthropogenic Climate Change. In: Meeting Report of the Intergovernmental Panel on Climate Change Expert Meeting on Detection and Attribution of Anthropogenic Climate Change [Stocker, T.F., C.B. Field, D. Qin, V. Barros, G.-K. Plattner, M. Tignor, P.M. Midgley, and K.L. Ebi (eds.)]. IPCC Working Group I Technical Support Unit, University of Bern, Bern, Switzerland, (2010).
- 4.
Bellassen, V. et al. Reconstruction and attribution of the carbon sink of European forests between 1950 and 2000. Global Change Biology 17(11), 3274–3292 (2011).
- 5.
Fernández-Martínez, M. et al. Global trends in carbon sinks and their relationships with CO 2 and temperature. Nature Climate Change 9, 73–79 (2018).
- 6.
Ribes, A., Zwiers, F. W., Azaïs, J. M. & Naveau, P. A new statistical approach to climate change detection and attribution. Climate Dynamics 48(1–2), 367–386 (2017).
- 7.
Tian, H. et al. China’s terrestrial carbon balance: Contributions from multiple global change factors. Global Biogeochem. Cycles 25, GB1007, https://doi.org/10.1029/2010GB003838 (2011).
- 8.
Williams, C. A. et al. Interannual variability of photosynthesis across Africa and its attribution. J. Geophys. Res. 113, G04015, https://doi.org/10.1029/2008JG000718 (2008).
- 9.
Huntzinger, D. N. et al. Uncertainty in the response of terrestrial carbon sink to environmental drivers undermines carbon-climate feedback predictions. Scientific Reports 7(1), 4765 (2017).
- 10.
Friedlingstein, P. et al. Uncertainties in CMIP5 climate projections due to carbon cycle feedbacks. Journal of Climate 27(2), 511–526 (2014).
- 11.
Huntzinger, D. N. et al. The North American Carbon Program Multi-scale synthesis and Terrestrial Model Intercomparison Project – Part 1: Overview and experimental design, Geosci. Model Dev. 6, 2121–2133, https://doi.org/10.5194/gmd-6-2121-2013 (2013).
- 12.
Taylor, K. E., Stouffer, R. J. & Meehl, G. A. An overview of CMIP5 and the experiment design. Bulletin of the American Meteorological Society 93(4), 485–498 (2012).
- 13.
Campbell, J. E. et al. Large historical growth in global terrestrial gross primary production. Nature 544(7648), 84–87 (2017).
- 14.
Ahlström, A. et al. The dominant role of semi-arid ecosystems in the trend and variability of the land CO2 sink. Science 348(6237), 895–899 (2015).
- 15.
Crutzen, P. J. Geology of mankind. Nature 415(6867), 23–23 (2002).
- 16.
Zhou, S., et al. (2017). Dominant role of plant physiology in trend and variability of gross primary productivity in North America. Scientific Reports, 7 (2017).
- 17.
Zhu, P. et al. Elevated atmospheric CO2 negatively impacts photosynthesis through radiative forcing and physiology-mediated climate feedback. Geophysical Research Letters 44(4), 1956–1963 (2017).
- 18.
Sun, Y. et al. Impact of mesophyll diffusion on estimated global land CO2 fertilization. Proceedings of the National Academy of Sciences 111(44), 15774–15779 (2014).
- 19.
Smith, W. K. et al. Large divergence of satellite and Earth system model estimates of global terrestrial CO2 fertilization. Nature Climate Change 6(3), 306 (2016).
- 20.
van der Sleen, P. et al. No growth stimulation of tropical trees by 150 years of CO2 fertilization but water-use efficiency increased. Nature Geoscience 8(1), 24 (2015).
- 21.
Hararuk, O., Campbell, E. M., Antos, J. A., & Parish, R. Tree rings provide no evidence of a CO2 fertilization effect in old-growth subalpine forests of western Canada. Global change biology, https://doi.org/10.1111/gcb.14561 (2018).
- 22.
Ehlers, I. et al. Detecting long-term metabolic shifts using isotopomers: CO2-driven suppression of photorespiration in C3 plants over the 20th century. Proceedings of the National Academy of Sciences 112(51), 15585–15590 (2015).
- 23.
Girardin, M. P. et al. No growth stimulation of Canada’s boreal forest under half-century of combined warming and CO2 fertilization. Proceedings of the National Academy of Sciences 113(52), E8406–E8414 (2016).
- 24.
Gómez-Guerrero, A. et al. Growth decline and divergent tree ring isotopic composition (δ13C and δ18O) contradict predictions of CO2 stimulation in high altitudinal forests. Global Change Biology 19(6), 1748–1758 (2013).
- 25.
Zhu, K., Zhang, J., Niu, S., Chu, C. & Luo, Y. Limits to growth of forest biomass carbon sink under climate change. Nature Communications 9(1), 2709 (2018).
- 26.
Bobbink, R., Ashmore, M., Braun, S., Flückiger, W., & Van den Wyngaert, I. J. Empirical nitrogen critical loads for natural and semi-natural ecosystems: 2002 update. Pages 43–107 in B. Achermann and R. Bobbick, editors. Empirical critical loads for nitrogen. Environmental Documentation Number 164. Swiss Agency for the Environment, Forests, and Landscape, Berne, Switzerland (2003).
- 27.
Pardo, L. H. et al. Effects of nitrogen deposition and empirical nitrogen critical loads for ecoregions of the United States. Ecological Applications 21(8), 3049–3082 (2011).
- 28.
Cárate‐Tandalla, D., Camenzind, T., Leuschner, C. & Homeier, J. Contrasting species responses to continued nitrogen and phosphorus addition in tropical montane forest tree seedlings. Biotropica 50(2), 234–245 (2018).
- 29.
Reay, D. S., Dentener, F., Smith, P., Grace, J. & Feely, R. A. Global nitrogen deposition and carbon sinks. Nature Geoscience 1(7), 430 (2008).
- 30.
Turner, B. L., Brenes-Arguedas, T. & Condit, R. Pervasive phosphorus limitation of tree species but not communities in tropical forests. Nature 555(7696), 367 (2018).
- 31.
Arora, V. K. et al. Carbon–concentration and carbon–climate feedbacks in CMIP5 Earth system models. Journal of Climate 26(15), 5289–5314 (2013).
- 32.
Guanter, L. et al. Global and time-resolved monitoring of crop photosynthesis with chlorophyll fluorescence. Proceedings of the National Academy of Sciences 111(14), E1327–E1333 (2014).
- 33.
Zhang, Y., Song, C., Band, L. E. & Sun, G. No proportional increase of terrestrial gross carbon sequestration from the greening Earth. Journal of Geophysical Research: Biogeosciences 124(8), 2540–2553 (2019).
- 34.
Lawrence, D. M. et al. The Community Land Model version 5: Description of new features, benchmarking, and impact of forcing uncertainty. Journal of Advances in Modeling Earth Systems (2019).
- 35.
Fisher, J. B., Huntzinger, D. N., Schwalm, C. R. & Sitch, S. Modeling the terrestrial biosphere. Annual Review of Environment and Resources 39, 91–123 (2014).
- 36.
Fisher, R. A., & Koven, C. D. Perspectives on the future of Land Surface Models and the challenges of representing complex terrestrial systems. Journal of Advances in Modeling Earth Systems (2020).
- 37.
Pongratz, J. et al. Models meet data: Challenges and opportunities in implementing land management in Earth system models. Global Change Biology 24(4), 1470–1487 (2018).
- 38.
Zhang, Q., Wang, Y. P., Matear, R. J., Pitman, A. J. & Dai, Y. J. Nitrogen and phosphorous limitations significantly reduce future allowable CO2 emissions. Geophysical Research Letters 41(2), 632–637 (2014).
- 39.
Wei, Y. et al. The North American Carbon Program Multi-scale Synthesis and Terrestrial Model Intercomparison Project–Part 2: Environmental driver data. Geoscientific Model. Development 7(6), 2875–2893 (2014).
- 40.
Phillips, T. J., Bonfils, C. J. & Zhang, C. Model consensus projections of US regional hydroclimates under greenhouse warming. Environmental Research Letters 14(1), 014005 (2019).
- 41.
Knutti, R. The end of model democracy? Climatic Change 102, 395–404 (2010).
- 42.
Bishop, C. H. & Abramowitz, G. Climate model dependence and the replicate Earth paradigm. Climate dynamics 41(3-4), 885–900 (2013).
- 43.
Haughton, N., Abramowitz, G., Pitman, A. & Phipps, S. J. On the generation of climate model ensembles. Climate dynamics 43(7-8), 2297–2308 (2014).
- 44.
Schwalm, C. R. et al. Toward “optimal” integration of terrestrial biosphere models. Geophys. Res. Lett. 42, 4418–4428, https://doi.org/10.1002/2015GL064002 (2015).
- 45.
Liaw, A. & Wiener, M. Classification and regression by randomForest. R news, 373 2(3), 18–22 (2002).