Plant Regrowth as a Driver of Recent Enhancement of Terrestrial CO₂ Uptake

Abstract

The increasing strength of land CO₂ uptake in the 2000s has been attributed to a stimulating effect of rising atmospheric CO₂ on photosynthesis (CO₂ fertilization). Using terrestrial biosphere models, we show that enhanced CO₂ uptake is induced not only by CO₂ fertilization but also an increasing uptake by plant regrowth (accounting for 0.33 ± 0.10 Pg C/year increase of CO₂ uptake in the 2000s compared with the 1960s–1990s) with its effect most pronounced in eastern North America, southern-eastern Europe, and southeastern temperate Eurasia. Our analysis indicates that ecosystems in North America and Europe have established the current productive state through regrowth since the 1960s, and those in temperate Eurasia are still in a stage from regrowth following active afforestation in the 1980s–1990s. As the strength of model representation of CO₂ fertilization is still in debate, plant regrowth might have a greater potential to sequester carbon than indicated by this study.

Plain Language Summary

The recent enhancement of CO₂ uptake by the terrestrial biosphere is slowing down an acceleration of the atmospheric CO₂ increase. A stimulating effect of rising atmospheric CO₂ on plant photosynthesis (CO₂ fertilization) provides the most pronounced impact on the enhanced CO₂ uptake. However, the question remains on how much of the enhanced uptake CO₂ fertilization accounts for and a possible contribution from past land use change. Here using multiple terrestrial biosphere models, we demonstrate that despite a large contribution from CO₂ fertilization, the enhanced CO₂ uptake in the 2000s cannot be fully explained without an increasing uptake by land use change, in particular, plant regrowth. The regrowth effect is most pronounced in North America, Europe, and temperate Eurasia, and they account for 94% of the global total CO₂ uptake enhancement by plant regrowth. The strengthening trends in both CO₂ fertilization and plant regrowth suggest that the deceleration of the atmospheric CO₂ increase continues in the future.

1 Introduction

CO₂ accumulates in the atmosphere as a result of greater anthropogenic emissions due to fossil fuel consumption and cement production compared to the net uptake by the land and ocean (Le Quéré et al., 2016). Although atmospheric CO₂ has been consistently increasing from the industrial era, the airborne fraction has declined from the early 2000s because of an enhancement in CO₂ uptake by the land and ocean (Keenan et al., 2016; Sarmiento et al., 2010), which has doubled during the past 50 years and is predicted to remain strong hereafter (Ballantyne et al., 2012). Mechanisms behind the enhanced CO₂ uptake involve physiological and biogeochemical processes on both the land and ocean (Ballantyne et al., 2017; DeVries et al., 2017; Keeling et al., 2017; Keenan et al., 2016; Sarmiento et al., 2010), but the land is of primary importance because it has a larger control on the interannual growth rate of atmospheric CO₂ (Cox et al., 2013; Wang et al., 2013), and is thus believed to be more responsible for the recent slowing down of surface warming (Fyfe et al., 2016; Shevliakova et al., 2013).

Growing evidence suggests that the enhancement of CO₂ uptake by the land is primarily due to the effect of CO₂ fertilization (Fisher et al., 2013; Keenan et al., 2016; Sun et al., 2014), which has led to a greening of a large fraction of the terrestrial biosphere (Zhu et al., 2016) and compensated for large CO₂ emissions resulting from tropical land use (Schimel et al., 2015). An experiment with an Earth system model suggests that the observed rise of ~115 ppm in atmospheric CO₂ since the preindustrial era might have been higher by ~85 ppm without the effect of CO₂ fertilization (Shevliakova et al., 2013), implying a large contribution of CO₂ fertilization to net CO₂ flux (balance between CO₂ uptake and release by the terrestrial biosphere). However, it is still arguable whether the CO₂ fertilization is a dominant cause for the recent enhancement of CO₂ uptake because, in addition to the level of atmospheric CO₂, the terrestrial biosphere has undergone historical changes through land use and management (Erb et al., 2013, 2018). CO₂ emissions resulting from land use change (LUC) activities account for ~9% of the total global anthropogenic CO₂ emissions (Le Quéré et al., 2016); therefore, changes in LUC could affect the course of the net sink-source pattern of CO₂ over time. The recent declining trend in global LUC activities (Houghton & Nassikas, 2017) implies likely reductions in CO₂ release from land use and land cover change (LUC emissions, hereafter) and increases in uptake by plants recovering from past LUC (regrowth flux, hereafter). Pacala et al. (2001) demonstrated that forest regrowth in the eastern U.S. accounted for much of the land uptake in the region during the 1980s, thus identifying regrowth as a potentially globally significant flux. However, quantification of such changes over the recent period has not been fully addressed before and contribution of LUC fluxes to the recent terrestrial CO₂ uptake is not clearly understood. Neglecting contributions from LUC fluxes would lead to incomplete understanding of processes involved in the climate-carbon cycle feedback and future pathways to climate change mitigation.

For a better understanding of mechanisms behind the recent enhancement of land CO₂ uptake, we investigate global and regional patterns of relative contributions to net CO₂ uptake through an attribution study using an ensemble of biosphere models from TRENDY, in conjunction with independent net CO₂ flux estimates that are estimated to be optimally consistent with atmospheric CO₂ measurements (atmospheric CO₂ inversion) and CO₂ growth rate (a residual land uptake from Global Carbon Project: GCP). Through the evaluation of the relative contributions (i.e., CO₂ fertilization effect, climate effect, LUC emissions, and regrowth flux) to the past and current CO₂ uptake, we address the role of historical LUC in the recent uptake enhancement.

2 Methods

2.1 Sign Convention for Net CO₂ Flux

In this analysis, we chose the sign convention for net CO₂ flux that is commonly used in top-down analyses: the negative sign (−) for a net sink to the land and the positive sign (+) for a net source to the atmosphere. This sign convention is used for all components of this study and thus applied to terms for CO₂ exchange such as net biome production (NBP) and net ecosystem production (NEP). It should be noted that this convention is opposite to the one commonly used in bottom-up analyses (Chapin et al., 2006).

2.2 Terrestrial Biosphere Models

2.2.1 TRENDY Models

Simulations of the biosphere models used in this study are from the TRENDY v2 (Sitch et al., 2015; Zhao et al., 2016). The TRENDY models were run with a consistent forcing data set: (1) atmospheric CO₂ mixing ratio for 1860–2012 based on ice core measurements and station observations, (2) climate data set for 1901–2012 based on a merging between Climate Research Unit TS3.2 0.5° × 0.5° monthly climate data (Harris et al., 2014) and National Centers for Environmental Prediction and National Center for Atmospheric Research Reanalysis 2.5° × 2.5° 6-hourly climate data (Kistler et al., 2001), and (3) 0.5° × 0.5° gridded annual LUC data set for 1860–2012. The TRENDY models were run following a common protocol: simulation that considers variability in atmospheric CO₂ only (S1), simulation that considers variability in CO₂ and climate (S2), and simulation that considers variability in CO₂, climate, and historical LUC (S3). For each simulation, the models were first spun-up to an equilibrium state of carbon balance forced with the 1860 CO₂ mixing ratio (287.14 ppm), recycling climate mean, and variability from the early decades of the 20th century (i.e., 1901–1920), and using constant 1860 crop and pasture distribution. S1, S2, and S3 simulations were then conducted for a transient period 1861–2012 after initialization from these spin-up runs.

2.2.2 Attributions to Net CO₂ Flux

Attributions to net CO₂ flux were extracted by separating flux signals in the simulations S1, S2, and S3 (Table 1). NBP of the S3 (forced with varying CO₂, climate, and LUC) represents a best estimate of the actual net CO₂ flux of the terrestrial biosphere. NBP from the S1 and S2 simulations represent partial contributions to net CO₂ flux, representing the CO₂ (fertilization) effect and CO₂ + climate effects on net CO₂ flux, respectively. The climate effect was extracted by subtracting NBP of the S1 from that of the S2; their difference leaves out the effect of CO₂ fertilization, and only the effect of climate remains (Table 1).

Table 1. Descriptions of Flux Terminologies and Calculation Methods

Terminology	Calculation methoda^a NBP: Net biome production (photosynthesis-autotrophic and heterotrophic respirations-natural disturbances-LUC emissions); NEP: net ecosystem productivity (photosynthesis-autotrophic and heterotrophic respirations). ^,b^b S3: Simulation forced with varying CO2, climate, and LUC; S2: simulation forced with varying CO2 and climate; and S1: simulation forced with varying CO2.	Description	Symbol for ΔF
Net CO2 flux (1 + 2)	S3 NBP	Net exchange of CO2 uptake and release between land and atmosphere, accounting the spatio-temporal variability in historical CO2, climate, and LUC.	ΔFnet
1. CO2 + climate effect (1a + 1b)	S2 NBP	Partial net exchange of CO2 accounting for spatio-temporal variability in historical CO2, and climate.	ΔFCO2 + climate
1a. CO2 effect	S1 NBP	Partial net exchange of CO2 accounting for spatio-temporal variability in historical CO2 only.	ΔFCO2
1b. Climate effect	S2 NBP-S1 NBP	Partial net exchange of CO2 accounting for spatio-temporal variability in historical climate only.	ΔFclimate
2. Net LUC flux (2a + 2b)	S3 NBP-S2 NBP	Partial net exchange of CO2 accounting for spatio-temporal variability in historical LUC only. This flux constitutes of CO2 uptake and release by LUC and plant regrowth.	ΔFLUC
2a. LUC emissions	(S3 NBP-S2 NBP) – (S3 NEP-S2 NEP)	CO2 emissions from wood storages removed by LUC. It is the dominant component of gross LUC source.	ΔFLUCe
2b. Regrowth flux	S3 NEP-S2 NEP	Exchange of CO2 uptake and release during the process of plant regrowth after LUC. This flux is the dominant component of gross LUC sink, but it also includes emissions from decomposition of woody residues (i.e., litters) remaining on sites.	ΔFreg

• ^a NBP: Net biome production (photosynthesis-autotrophic and heterotrophic respirations-natural disturbances-LUC emissions); NEP: net ecosystem productivity (photosynthesis-autotrophic and heterotrophic respirations).
• ^b S3: Simulation forced with varying CO₂, climate, and LUC; S2: simulation forced with varying CO₂ and climate; and S1: simulation forced with varying CO₂.

Net LUC flux (a partial contribution to net CO₂ flux associated with LUC) was extracted by subtracting NBP of the S2 from that of the S3; their difference leaves out the effects of CO₂ fertilization and climate, and only the effect of LUC remains (to be precise, residuals of the CO₂ and climate effects remain due to changing land cover types). Further, we decomposed net LUC flux into regrowth flux and LUC emissions. Regrowth flux represents the post LUC effect on ecosystem CO₂ exchange (i.e., NEP); thus, it was extracted by subtracting NEP of the S2 from that of the S3 (NEP differs from NBP by excluding disturbance fluxes from fire and LUC). The rest of net LUC flux components (i.e., emissions from removed wood products) were defined as LUC emissions, which was estimated by subtracting regrowth flux from net LUC flux (Table 1).

2.2.3 Land Use Change Forcing

The LUC forcing for the TRENDY models provides gridded information of land cover changes between cropland, pastureland, and primary and secondary lands, based on the UN Food and Agricultural Organization (FAO) national statistics. The initial land cover changes (annual transitions of cropland and pastureland at the spatial resolution of 5′) were calculated using allocation algorithms and time-dependent weighting maps based on global historical population density, soil suitability, distance to rivers, lakes, slopes, and biome distributions (HistorY Database of the Global Environment: HYDE v3.1; Klein Goldewijk et al., 2011). The LUH v1, an extended version of HYDE, then combined the HYDE cropland and pastureland status with the wood harvest information from the FAO national statistics with an empirically estimated biomass density map produced at the spatial resolution of 0.5° (Hurtt et al., 2011). The LUH v1 provides the full annual transition matrix of primary and secondary lands in addition to those of cropland and pastureland.

The implementation of the LUC forcing was left to the discretion of each TRENDY modeling group because of differences in fundamental assumptions and levels of complexity in LUC modeling, for instance, distinction of primary and secondary lands, implementation of wood and crop harvests, consideration of residue carbon after deforestation, and turnover rates of a product pool (Table S1; more details shown in Le Quéré et al., 2015). Despite these differences in LUC schemes, land cover changes predefined by the LUC forcing data ensure relatively consistent forest area changes among the TRENDY models (minor differences occur, e.g., due to dynamic vegetation).

2.3 Independent Estimates of Net CO₂ Flux

2.3.1 Atmospheric CO₂ Inversions

Atmospheric CO₂ inversions estimate net land-atmosphere CO₂ flux from the continuous and discrete atmospheric CO₂ measurements from global networks, for example, National Oceanic and Atmospheric Administration Earth System Research Laboratory (https://www.esrl.noaa.gov/gmd/ccgg/trends/full.html), World Data Centre for Greenhouse Gases (http://ds.data.jma.go.jp/gmd/wdcgg/wdcgg.html), and Comprehensive Observation Network for TRace gases by AIrLiner (CONTRAIL: http://www.cger.nies.go.jp/contrail/), and the prior fluxes (information on land and ocean fluxes, fire emissions, and anthropogenic CO₂ emissions). In this study, an ensemble of six atmospheric CO₂ inversions was used for validation of the biosphere models providing quasi-independent data for net CO₂ flux. Outputs of four inversions are from Thompson et al. (2016): ACTM v5.7b (Saeki & Patra, 2017), CCAM (Rayner et al., 2008), JMA-CDTM (Maki et al., 2010), and MACC v14r2 (Chevallier et al., 2010). Two others are from Peylin et al. (2013): JENA s81 v3.8 (Rödenbeck et al., 2003) and NICAM-TM (Niwa et al., 2012). A choice of CO₂ measurements and prior fluxes for each inversion system was left to the discretion of modeling groups, as well as spatial resolution and time period of inverted fluxes. Details of a transport model, prior fluxes, and CO₂ measurement data for these inversions are described in Thompson et al. (2016) and Peylin et al. (2013), and corresponding literature for each inversion. Using data from the six atmospheric CO₂ inversions, net CO₂ flux for the period 1980–2009 was estimated by an ensemble average for overlapping time periods (ACTM covers the period for 1990–2011, JENA and MACC for 1980–2014, CCAM for 1993–2012, JMA for 1985–2012, and NICAM-TM for 1988–2007).

2.3.2 Residual Method

The residual method from GCP (Le Quéré et al., 2015, 2016) provides the global annual budget of land CO₂ uptake calculated as the difference of the other terms of the global carbon budget such as the CO₂ growth rate (National Oceanic and Atmospheric Administration Earth System Research Laboratory), fossil fuel emissions from Carbon Dioxide Information Analysis Center (http://cdiac.ess-dive.lbl.gov/) and United Nations Framework Convention on Climate Change (http://unfccc.int/ghg_data/items/3800.php), net ocean flux from ocean biogeochemistry models, and net LUC flux from the book-keeping model (Giglio et al., 2013; Houghton et al., 2012), that is, land flux = CO₂ growth rate − fossil fuel emissions − ocean flux − net LUC flux. The land uptake calculated in the above-mentioned method does not account for the effect of LUC (that is provided by the LUC book-keeping model); thus, it represents an attribution from the CO₂ and climate effects on net CO₂ flux (broadly comparable to NBP of the TRENDY S2 simulations). Net CO₂ flux of GCP was estimated as a sum of the residual land uptake and net LUC flux from the book-keeping model, that is, land flux + net LUC flux (comparable to the atmospheric CO₂ inversions and NBP from TRENDY S3 simulations). These land uptake estimates are referred to as GCP, hereafter.

2.4 Screening of Biosphere Models

In this study, we evaluate the relative contributions in terms of the difference between mean annual CO₂ fluxes for the 2000s and 1960s–1990s (termed ΔF), for the key components to net CO₂ flux (Table 1): climatological components (CO₂ fertilization effect, climate effect, and their net effect termed CO₂ + climate effect), and LUC components (LUC emissions, regrowth flux, and their net flux termed net LUC flux). As ΔF is the key variable of the analysis, accurate simulations of CO₂ budgets for the 2000s and 1960s–1990s are required. Therefore, we examined the degree of agreement between the independent estimates of net CO₂ flux (GCP and atmospheric CO₂ inversions) and the eight biosphere models of TRENDY: the Community Land Model v4.5: CLM (Lawrence et al., 2011), Integrated Science Assessment Model: ISAM (Jain et al., 2013), Joint UK Land Environment Simulator v3.2: JULES (Clark et al., 2011), Lund-Potsdam-Jena DGVM wsl: LPJ (Sitch et al., 2003), LPJ-GUESS (Smith et al., 2001), LPX (Stocker et al., 2014), ORCHIDEE-CN: O-CN (Zaehle & Friend, 2010), and Vegetation Integrative SImulator for Trace gases: VISIT (Ito, 2010).

For the period 1960–2012, all the TRENDY models were relatively consistent in patterns of interannual variability (IAV) and trends of global net CO₂ flux with respect to the GCP and atmospheric CO₂ inversions, but for some the consistency was particularly notable (Figure S1). To quantify the level of consistency, we examined a residual sum of squares (RSS) between the TRENDY models and GCP for the periods 1960–2012 and 2000–2012 (Figure S2). Four models (CLM, JULES, O-CN, and VISIT) yielded a substantially lower RSS than the others for both time periods, and an ensemble of the four models resulted in highly consistent IAV in net CO₂ flux with respect to the GCP (r = 0.70, p < 0.01) and atmospheric CO₂ inversions (r = 0.75, p < 0.01; Figure S3).

We cross-checked mean annual CO₂ budgets (from S3 NBP and S2 NBP) for the 2000s and 1960s–1990s between the biosphere models with lower RSS and others (Figure S4). Decadal CO₂ budgets by an ensemble of the four models with lower RSS were consistent with the GCP and atmospheric CO₂ inversions, whereas an ensemble of the other models yielded a weaker sink compared with the independent estimates. Based on these evaluations, we selected the four models, CLM, JULES, O-CN, and VISIT, for the following analysis.

3 Results

3.1 Increasing CO₂ Uptake and Contribution of Regrowth Flux

The biosphere models of this analysis (the four models evaluated against the GCP and atmospheric CO₂ inversions) support the recent increase in CO₂ uptake by the terrestrial biosphere (Figure 1a). Decadal variability in global net CO₂ flux by the ensemble of the biosphere models (S3 NBP) indicates a tendency toward a net source during the 1910s–1950s and a transition toward a net sink during the 1960s–2000s (Figures 1a and S5). The transition from a net source to a net sink in the 1960s is in line with that simulated by an Earth system model (Shevliakova et al., 2013). The increasing CO₂ uptake since the 1960s results in the 2000s displaying a larger decadal CO₂ uptake than at any time during the preceding century, −1.52 ± 0.31 Pg C/year (average ± 1σ as model-by-model variability).

Figure 1

Increasing pattern of global CO₂ uptake and contributions of component fluxes. (a) Decadal variability of global net CO₂ flux from the ensemble mean of the four TRENDY models: (S3 NBP: orange) for the 1910s–2000s, Global Carbon Project (GCP: grey) for the 1960s–2000s, and the ensemble mean of the atmospheric CO₂ inversions (cyan) for the 1980s–2000s. Negative values in net CO₂ flux represent a net sink, and positive values a net source. The error bars indicate 1σ variations among models. The top-right panel shows correlation coefficients (r) between interannual variability of the three net CO₂ flux estimates for the overlapping periods (1980–2009 for the TRENDY and atmospheric CO₂ inversions, 1960–2009 for the TRENDY and GCP, and 1980–2009 for the atmospheric CO₂ inversions and GCP), and statistical significance is indicated by ** (p < 0.01). The middle panel shows mean annual CO₂ budgets of attributing factors to net CO₂ flux (CO₂ effect, climate effect, and net LUC flux) for the periods 1910–1959 and 1960–2009. (b) Changes of global CO₂ uptake in the 2000s with respect to that during the 1960s–1999s (indicated by ΔF: difference between mean annual CO₂ budget for 2000–2009 and that for 1960–1999). ΔF for net CO₂ flux and component fluxes (refer to Table 1 for descriptions of component fluxes) by the TRENDY models (orange bars and colored lines) are showed along with estimates by the GCP (grey bars). Negative values in ΔF represent that CO₂ flux in the 2000s is more toward a net sink than that in the 1960s–1990s, and positive values indicate the opposite. (c) ΔF for net CO₂ flux (open bars) and regrowth flux (green bars) from the individual TRENDY models (CLM, JULES, O-CN, and VISIT) and their ensemble mean. (d) Interannual variability of net LUC flux (red line), LUC emissions (purple line), and regrowth flux (green line) by the TRENDY models in the form of anomaly with a base period 1960–2009. For each flux, shading indicates 1σ variations among models. The dashed lines are linear regressions on the data for 1960–1999 and 1990–2009, and statistical significance is determined by Mann-Kendall test and indicated by ** (p < 0.01).

We found that both climatological and LUC components (ΔF_CO2 + clim and ΔF_LUC, respectively: Table 1) contributed to the recent enhancement of global CO₂ uptake (indicated by ΔF_net), which amounted to −1.27 ± 0.34 Pg C/year (Figure 1b). Components of net CO₂ flux by the GCP agree with the pattern of relative contributions by the biosphere models (Figure 1b; see Figure S6 for individual biosphere model results). Examining the individual relative contributions further, we found that despite its large contribution, the CO₂ fertilization effect (ΔF_CO2) does not fully explain the recent enhancement in CO₂ uptake. A relative contribution from ΔF_CO2 to ΔF_net, −1.11 ± 0.25 Pg C/year, is reduced to −0.92 ± 0.29 Pg C/year when combined with climate effect (ΔF_clim), which induced a shift toward a net source in the 2000s (Figures 1b and S6). Importantly, the remainder of ΔF_net is accounted for by the net LUC flux (ΔF_LUC), −0.37 ± 0.21 Pg C/year, of which regrowth flux (ΔF_reg) is the primary constituent at −0.33 ± 0.10 Pg C/year. The pattern of the relative contribution from ΔF_reg is considered robust because the ratio of ΔF_reg to ΔF_net is consistent between the individual biosphere models with a range of 23–30% (Figure 1c) and is accompanied by a consistent trend toward a net sink throughout the past 50 years (−0.01 Pg C/year², p < 0.01 by Mann-Kendall test; Figure 1d). As a result, regrowth flux appears to have mitigated the increasing trend of LUC emissions during the 1960s–1990s and further facilitated the decreasing trend in LUC emissions during the 1990s–2000s (Figure 1d).

3.2 Spatial Pattern and Hot Spots of the Uptake Enhancement by Plant Regrowth

A closer look at regional patterns of the relative contributions reveals a clear distinction in locations responsible for the uptake enhancement between the climatological and LUC components. As illustrated in the spatial distribution of ΔF_net, the uptake enhancement has occurred over large proportions of vegetated area across the globe (Figure 2a), but with substantial regional variations (the regional classification is shown in Figure S7). The contribution from ΔF_{CO2 + climate} was widespread from boreal Eurasia to tropical regions such as coastal regions of South America, central Africa, and tropical Asia (Figure 2b). In contrast, the contribution from ΔF_LUC was concentrated in three particular regions: an eastern part of North America, southern and eastern parts of Europe (including European Russia), and a southeastern part of Temperate Eurasia (hereafter, hot spots in ΔF_LUC: Figure 2c). It is important to note that these hot spots in ΔF_LUC largely coincide with locations where a large contribution from ΔF_reg is found (especially in North America and Europe; Figure S8), and these patterns are consistent between the biosphere models (Figures S9 and S10). The three regions characterized by large ΔF_reg accounted for 94% of the global total (Table S2), with the largest contribution from North America (−0.17 ± 0.03 Pg C/year). In North America, we found that a large fraction of ΔF_CO2 (−0.21 ± 0.04 Pg C/year) was canceled by ΔF_climate (0.13 ± 0.08 Pg C/year), which clearly demonstrates that the enhanced uptake indicated by ΔF_net (−0.24 ± 0.06 Pg C/year) cannot be explained without the contribution from regrowth flux during the 2000s (Table S2).

Figure 2

Spatial patterns of ΔF for net CO₂ flux and component fluxes. Spatial variability in ΔF for (a) net CO₂ flux (ΔF_net), (b) CO₂ + climate effect (ΔF_{CO2 + climate}), and (c) net LUC flux (ΔF_LUC) by the ensemble mean of the TRENDY models. Along with spatial maps, regional budgets of ΔF based on the RECCAP land classification (Figure S7) are shown for net CO₂ flux, CO₂ and climate effects (ΔF_CO2 and ΔF_climate, respectively), regrowth flux (ΔF_reg), and LUC emissions (ΔF_LUCe). Negative values in ΔF represent that CO₂ flux in the 2000s is more toward a net sink than that in the 1960s–1990s, and positive values indicate the opposite.

Focusing on the hot spots in ΔF_LUC (colored grid cells in Figure 3), we found that IAVs in net LUC flux and NEP in the North American and European hot spots have a similar tendency toward a net sink for the past 50 years when the effect of land use and land over changes is taken into account for NEP, that is, S3 NEP (Figures 3a and 3b). Zonally averaged fluxes indicate that the shift from a net source to net sink in net LUC flux between the 1960s and 2000s in Europe and North America corresponds closely to the emergence of a strong regrowth sink in those locations over this time (Figures 3d and 3e). Contrary to North America and Europe, the hot spot in temperate Eurasia indicates a relatively less uptake from regrowth flux during the 2000s (Figures 3c and 3f), suggesting that a decrease in LUC emissions is the factor also responsible for the change in net LUC flux (Figures S8c and S8d).

Figure 3

Temporal transition of regrowth flux in the three hot spot regions of ΔF_LUC. Temporal variability (five-year averaged) of net LUC flux (red bar) and NEP with and without considering variability in LUC (green and grey lines, respectively) for the three hot spot regions of ΔF_LUC: (a) North America, (b) Europe, and (c) temperate Eurasia. The error bars and shading indicate 1σ variations among models. A spatial map in background is from Figure 2c, and the three regional hot spots characterized by large negative ΔF_LUC are highlighted with different colors. Decadal changes in longitudinal averaged net LUC flux (red gradient lines) and regrowth flux (green gradient lines) over the three hot spot regions: (d) North America, (e) Europe, and (f) temperate Eurasia. All results are from the TRENDY models.

4 Discussion and Conclusions

Our approach for attribution of the net CO₂ flux revealed that both regrowth after LUC and growth enhancement due to CO₂ fertilization are responsible for the recent enhancement of CO₂ uptake, but the quantification of these effects still presents potential large uncertainties. A recent synthesis of biosphere models argues that LUC emissions may previously have been underestimated, due to the neglect, until very recently, of processes such as shifting cultivation, wood harvest, and cropland management (Arneth et al., 2017). Arneth et al. (2017) suggests that such an underestimation implies a larger land CO₂ uptake than previously thought. Because of a large contribution to the net CO₂ balance (Figure 1), the CO₂ fertilization may be a strong candidate for this additional CO₂ uptake. However, physiological evidences from long-term inventory (Clark et al., 2010) and carbon isotope measurements (van der Sleen et al., 2015) criticize a strong CO₂ fertilization effect in tropics, posing a question on its dominate role in the recent uptake enhancement.

Local studies support reliability of the hot spots of plant regrowth found in this study. Regional analyses of extensive forest inventory measurements have reported that a large fraction of the current forest carbon stock accumulations in the eastern part of North America (specifically, the eastern United States) and European countries originates from the large-scale reforestation and afforestation during the postwar period in the 1960s (Ciais et al., 2008; Pacala et al., 2001; Woodall et al., 2015). The LUC forcing used for the biosphere models reflects these regional characteristics, indicating a decadal land conversion with a substantial increase in secondary forests and the corresponding decrease in cropland between 1960 and 2000 (Figures S11 and S12). This corroboration of historical LUC increases a confidence in the modeled increase in CO₂ uptake due to plant regrowth during recent years, and the likely continuation of forest conservation in U.S. and European countries (Forest Europe, 2015; USDA Forest Service, 2016) suggests a further increase in CO₂ uptake by plant regrowth in the future.

In addition to plant regrowth, the decrease in LUC emissions also contributed to the change in net LUC flux in temperate Eurasia. However, this causality should be interpreted with caution. Large-scale afforestation programs have been initiated in eastern China since the 1980s, which led to an increase in forest area at 1.6% per year over the 1990s–2000s (Peng et al., 2014; Piao et al., 2012). Nevertheless, the LUC forcing for the biosphere models does not indicate any notable increase in secondary forests during the past 50 years in this region, instead a large fraction of primary forests is replaced by croplands and pastures (Figures S11 and S12). This mismatch between the real event and LUC forcing in temperate Eurasia might have caused an underestimation of CO₂ uptake in the absence of regrowth of secondary forests, and it calls for an immediate improvement of the LUC forcing for this region.

Although the biogeochemical effects of plant regrowth from historical land use and management have likely moderated rates of present-day climate change, the biophysical effect of land cover changes may act in the opposite direction, especially on the local-regional scale (Alkama & Cescatti, 2016). For example, in Europe, continuous afforestation from past has led to an increase in land CO₂ uptake, but species change from broadleaf to needleleaf forests resulted in a regional increase of the summertime temperature because of a decrease in evapotranspiration (Naudts et al., 2016). Thus, the net effect of plant regrowth on climate is complex and scale-dependent, and further work is required integrating over both biogeochemical and biophysical effects of plant regrowth at both regional and global scales. This will require complementing the existing data sets that identify wood harvest and transitions between forests, croplands, and pastures, with estimates of forest age, and tree species changes due to management.

Acknowledgments

M. K., K. I., and P. K. P. acknowledge Environment Research and Technology Development Funds of the Ministry of the Environment of Japan (2-1401) and of the Environmental Restoration and Conservation Agency (2-1701). C. K. acknowledges support from the US DOE BER through the RUBISCO SFA and NGEE-Tropics projects. T. A. M. P. acknowledges funding from European Commission's Seventh Framework Programme, under grant agreement 603542 (LUC4C). L. C. acknowledges support from the National Aeronautics and Space Administration Earth and Space Science Fellowship, under grant NNX16AP86H. The TRENDY data are available via Stephen Sitch, Exeter University ([e-mail]). MACC and JENA inversion data are available from the web sites (MACC: http://apps.ecmwf.int/datasets/data/macc-ghg-inversions/, JENA: http://www.bgc-jena.mpg.de/CarboScope/s/main.html). ACTM, JMA, and CCAM inversion data used in this study are from Asia-Pacific Network for Global Change Research (APN: grant ARCP2011-11NMY-Patra/Canadell) and available by contacting Prabir K. Patra ([e-mail]). NICAM inversion data are available by contacting Yosuke Niwa ([e-mail] or [e-mail]). Flux data of Global Carbon Project are available from the website (http://cdiac.ess-dive.lbl.gov/GCP/).