{
  "documents": [
    {
      "chunkSource": "",
      "description": "No description found.",
      "docAuthor": "Springer",
      "docSource": "pdf file uploaded by the user.",
      "id": "405e54bd-7513-4f86-a49d-28d82ff2734c",
      "location": "custom-documents/2022-01-01-Test-Document.pdf-405e54bd-7513-4f86-a49d-28d82ff2734c.json",
      "pageContent": "ARTICLE \nVegetation structural change since \n1981 signi fi cantly enhanced the terrestrial \ncarbon sink \nJing M. Chen \n1,2 \n, Weimin Ju \n2,3 \n, Philippe Ciais \n4 \n, Nicolas Viovy \n4 \n, Ronggao Liu \n5 \n, Yang Liu \n5 \n& \nXuehe Lu \n2 \nSatellite observations show that leaf area index (LAI) has increased globally since 1981, but \nthe impact of this vegetation structural change on the global terrestrial carbon cycle has not \nbeen systematically evaluated. Through process-based diagnostic ecosystem modeling, we \nfi nd that the increase in LAI alone was responsible for 12.4% of the accumulated terrestrial \ncarbon sink (95 ± 5 Pg C) from 1981 to 2016, whereas other drivers of CO \n2 \nfertilization, \nnitrogen deposition, and climate change (temperature, radiation, and precipitation) con- \ntributed to 47.0%, 1.1%, and − 28.6% of the sink, respectively. The legacy effects of past \nchanges in these drivers prior to 1981 are responsible for the remaining 65.5% of the \naccumulated sink from 1981 to 2016. These results re fi ne the attribution of the land sink to \nthe  various  drivers  and  would  help  constrain  prognostic  models  that  often  have  large \nuncertainties  in  simulating  changes  in  vegetation  and  their  impacts  on  the  global \ncarbon cycle. \nhttps://doi.org/10.1038/s41467-019-12257-8 \nOPEN \n1 \nDepartment of Geography and Program in Planning, University of Toronto, Toronto, ON M5S 3G3, Canada. \n2 \nJiangsu Provincial Key Laboratory of \nGeographic Information Science and Technology, International Institute for Earth System Science, Nanjing University, Nanjing 210023, China. \n3 \nJiangsu \nCenter for Collaborative Innovation in Geographic Information Resource Development and Application, Nanjing 210023, China. \n4 \nLaboratoire des Sciences du \nClimat et de l ’ Environnement, LSCE/IPSL, CEA-CNRS-UVSQ, Universite Paris-Saclay, F-91191 Gif-sur-Yvette, France. \n5 \nState Key Laboratory of Resources and \nEnvironmental Information System, Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing 100101, China. \nCorrespondence and requests for materials should be addressed to W.J. (email: juweimin@nju.edu.cn ) \nNATURE COMMUNICATIONS |         (2019) 10:4259  | https://doi.org/10.1038/s41467-019-12257-8 | www.nature.com/naturecommunications 1 \n1234567890():,;T \nerrestrial ecosystems are an important part of the climate \nsystem and their effect on the global carbon cycle is one of \nthe  largest  uncertainties  in  the  projection  of  future \nclimate \n1 , 2 \n. The estimation of the terrestrial carbon cycle is com- \nplicated by large spatial and temporal variability of the vegetation \ncover, as well as complex biological, climate, and soil controls on \nplant growth and organic matter decomposition \n3 \n. Because of this \ncomplexity, the estimates of the global terrestrial carbon sink \nusing prognostic models differ considerably with the residual of \nthe global carbon budget in terms of both the mean and inter- \nannual variations \n4 \n. The attribution of the global land carbon sink \nto various drivers also differs greatly among models \n5 \n. For the \npurpose of projecting future climate, we need to better under- \nstand the mechanisms controlling the terrestrial carbon cycle so \nthat we can reliably estimate the terrestrial carbon cycle under \nfuture climatic and atmospheric conditions \n6 , 7 \n. \nWith human ’ s perturbation to the climate system through \ngreenhouse gas emissions to the atmosphere and land-use change, \nthe terrestrial carbon cycle has been greatly altered since pre- \nindustrial time \n8 , 9 \n. Studies have shown that increased atmospheric \nCO \n2 \nconcentration since 1850 has enhanced plant growth and \nhence induced carbon sinks \n10 \n, although the magnitudes of this \nenhancement  vary  among  model  estimates,  satellite-based \nassessments, and Free Air Enrichment Experiments performed \nat a limited number of sites \n3 \n. Climate change and atmospheric \nnitrogen deposition also played important roles in modulating the \nterrestrial carbon sink \n11 , 12 \n. In addition to direct effects of these \ndrivers on the carbon cycle, they have also induced changes in \nvegetation structure, i.e., leaf area index (LAI), de fi ned as one half \nthe total leaf area per unit ground surface area \n13 – 15 \n, which in turn \nalso changes the carbon cycle. This feedback of vegetation LAI to \nthe carbon cycle has not yet been systematically studied, although \nthe increase in LAI over the last several decades has been found to \nbe signi fi cant and dubbed global greening. \nCurrently,  prognostic  models,  which  simulate  vegetation \nstructure, growth, and carbon cycle under given climatic and \nedaphic conditions, are used, e.g., by the Global Carbon Project \n(GCP) as the main tools to estimate the terrestrial carbon sink \n4 \n. \nThe simulated results vary among models due to different \nassumptions and parameter settings, causing uncertainties \n3 \n. One \nof the largest variations among these models is the simulation of \nvegetation structural change with time. Without accurate assess- \nment of this change, it is highly uncertain to attribute the land \nsink to the various drivers, even if the total land sink is adjusted to \nan appropriate range and constrained by measured atmospheric \nCO \n2 \nconcentration. \nReliable satellite measurements of LAI are available to assess \nvegetation changes at the global scale since 1981. This source of \ninformation is underutilized in constraining the estimation of the \nterrestrial carbon sink and closing the global carbon budget, \nalthough many studies showed the usefulness of LAI products in \noptimizing several ecosystem parameters \n16 \n. To use this infor- \nmation beyond assessing vegetation greening/browning trend \n14 \n, \ndiagnostic models that assimilate remotely sensed vegetation \nstructural information to simulate physical, biological, and eco- \nlogical processes in vegetation are effective tools to estimate the \nimpact of LAI changes on the carbon cycle. In this study, we use a \nmodel of this type, which is named Boreal Ecosystem Productivity \nSimulator (BEPS) \n17 \n. This model was initially developed for boreal \necosystems and has been adapted for all ecosystems over the \nglobe \n18 \n. BEPS mechanistically includes the impacts of various \ndrivers on gross primary productivity (GPP) (climate, CO \n2 \ncon- \ncentration, and nitrogen deposition) and assimilates vegetation \nstructure (LAI) satellite data. It differs from light-use ef fi ciency \n(LUE) models, which estimate GPP based on radiation absorbed \nby the canopy and prescribed LUE functions that may or may not \ninclude CO \n2 \nand nutrient effects \n3 \n. BEPS also simulates the \ndynamics of carbon pools beyond GPP and uses a spin-up pro- \ncedure to prescribe soil carbon pools for estimating autotrophic \nrespiration (AR) and heterotrophic respiration (HR) (see Meth- \nods). It is therefore a diagnostic process model for estimating the \nfull carbon cycle using remote-sensing data and suitable for \nascribing land carbon sinks to the various drivers. \nBased on three LAI time series derived from satellite data, we \nfi nd that vegetation structural change re fl ected by the trend of \nLAI contributed 12.4% to the accumulated total terrestrial carbon \nsink (95 ± 5 Pg C) from 1981 to 2016. This is small, but sig- \nni fi cant, compared with the contributions of CO \n2 \nfertilization \n( + 47.0%) and climate change ( − 28.6%) in the same period. This \nfi nding suggests the importance in tracking this vegetation \nstructural parameter using satellite data in global carbon cycle \nresearch. Quantifying this separate effect of vegetation structural \nchange on the land sink helps attribute the sink to the various \ndrivers including CO \n2 \nfertilization, climate change, and nitrogen \ndeposition, and may also help rectify some differences among \nprognostic models. \nResults \nLAI data analysis . A global LAI time series from 1981 to 2016 at 8 \nkm resolution and 16-day (1981 – 2000) and 8-day (2001 – 2016) \nintervals was produced using Advanced Very High Resolution \nRadiometer (AVHRR) and Moderate Resolution Imaging Spec- \ntroradiometer (MODIS) satellite data with the GLOBMAP algo- \nrithm \n19 \n. The global distribution of the temporal trend of LAI over \nthis period is shown in Fig. 1 . About 74.2% of the land surface \nshows an increasing trend, among which 52.7% is signi fi cant at \np = 0.05 level (two-tailed). Globally, the increase in the maximum \nLAI in the peak growing season is about twice as large as that in the \nannual mean, suggesting that growing season lengthening is not the \nmain factor explaining the greening trend. This general increase in \nLAI resulted from the combined effects of various drivers including \nCO \n2 \n, climate, and nitrogen deposition over the same period, and \ntherefore provides a new base for separating the effects of these \ndrivers on vegetation structure and growth. The focus of this study \nis on the increase of LAI on plant growth and the land carbon sink. \nConsidering the uncertainty in the temporal trend of this LAI time \nseries, we used two other LAI products: GLASS LAI \n20 \nand GIMMS \nLAI3g \n21 \n(Supplementary Fig. 1 and Table 1). \nAnalysis of the residual land sink . Driven by one of LAI time \nseries at a time as well as climate, CO \n2 \n, soil, and nitrogen \ndeposition data, BEPS is used to simulate GPP, AR, and HR at \ndaily time intervals for each pixel. The sum of simulated net \necosystem productivity (NEP), taken as GPP-AR-HR, for all land \nareas is compared with the global residual land sink (RLS) \nreported by the GCP \n4 \n, which is computed as the sum of emissions \nfrom fossil fuel consumption, cement production, and land-use \nchange minus the sum of CO \n2 \naccumulated each year in the \natmosphere and ocean. The modeled annual NEP as average from \nthe BEPS model forced by each of the three LAI products closely \nfollows the trend and interannual variability of the residual land \ncarbon sink (Fig. 2 ), although it does not capture well extremely \nlow and high values in some years, such as 1987, 1991, 2000, \n2002, and 2009. Over the 1981 – 2016 period, the modeled accu- \nmulated NEP is 95 ± 5 Pg C, whereas the accumulated RLS is \n94 ± 5 Pg C. \nIn comparison with 15 prognostic models used by GCP \n4 \n, BEPS \nis among the best in terms of Pearson ’ s coef fi cient ( R \n2 \n), root \nmean  square  error  (RMSE)  between  simulated  and  the \nobservation-based annual RLS, and the accumulated sink from \n1981 to 2016 in comparison with RLS (Supplementary Table 2). \nARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-12257-8 \n2 NATURE COMMUNICATIONS |         (2019) 10:4259  | https://doi.org/10.1038/s41467-019-12257-8 | www.nature.com/naturecommunicationsBEPS has an R \n2 \nvalue of 0.56 (Supplementary Fig. 2), which is \nslightly lower than those of Community Atmosphere Biosphere \nLand Exchange (CABLE) and Lund – Potsdam – Jena (LPJ), but \nthe RMSE of BEPS is lower than that of all prognostic models. \nThe modeled accumulated sink by BEPS is also very close to the \naccumulated RLS, whereas estimates from the 15 prognostic \nmodels differ by a wide range. The results of these comparisons \nshow that BEPS as a diagnostic model driven by remote-sensing \ndata can do a similar or better job than fully prognostic models in \nsimulating the past land sink, because vegetation structural \nchanges observed by satellites provide a critical constraint to the \nsink estimation. We do not include other diagnostic models in \nthese comparisons \n22 , 23 \n, because BEPS so far may be the only \nprocess-based diagnostic model that calculates the full carbon \ncycle  at  the  global  scale  for  multiple  decades.  With  the \ncomputation of the land sink close to the RLS, BEPS can then \nbe used to attribute the sink to the various drivers including \nvegetation structural change. \nThe time-varying maps of satellite LAI constrain the effect of \nvegetation structural change on the land sink since 1981. \nHowever, the land sink in recent decades results from the \naccumulated changes in climate, CO \n2 \nconcentration, nitrogen \ndeposition, and land use since the preindustrial period that \noccurred before 1981. From 1981 to 2016, global land ecosystems \nabsorbed 95 ± 5 Pg C from the atmosphere (i.e., accumulated \nNEP), which is 32.8 Pg C larger than the baseline (62.1 Pg C) \nde fi ned by NEP simulated assuming without changes of these \ndrivers after 1981 (Supplementary Table 3). Several conditions \nwere set for simulating the baseline. First, CO \n2 \nconcentration and \nnitrogen deposition were kept at the 1981 levels; second, LAI was \ntaken as the mean value in 1982 – 1986; and third, meteorology in \na  year  between  1981  and  2016  is  randomly  taken  from \nmeteorology in a year within the 1971 – 1979 period (Supplemen- \ntary Table 3), so that no climate trend exists over the 1981 – 2016 \nperiod. Under the baseline conditions, the land sink over the \nperiod from 1981 to 2016 is thus caused only by the legacy of \nchanges in the drivers prior to 1981, given the residence time of \ncarbon in ecosystems. The mean legacy effect over 1981 – 2016 was \nlarger in forests, especially in evergreen broadleaf forests over \ntropical regions (Supplementary Fig. 3). We therefore refer to the \nsink increase from the mean legacy effect as the sink enhance- \nment due to changes of the drivers after 1981. \nAttribution of the land sink . The increase in atmospheric CO \n2 \nconcentration is modeled to be the dominant diver for the land \nsink enhancement during the 1981 – 2016 period. The CO \n2 \ncon- \ncentration increase after 1981 alone enhanced the global land sink \nby 44.6 Pg C accounting for 47.0% of the total accumulated sink \nenhancement after 1981 (Fig. 3 ). The simulated global total net \nprimary productivity (NPP) increased by 11.6% with the increase \nof CO \n2 \nconcentration from 340.13 p.p.m. in 1981 to 404.20 p.p.m. \nin 2016, whereas climate, LAI, and N deposition remain at \nbaseline values. The sensitivity of simulated total NPP in the \nnorthern hemisphere to atmospheric CO \n2 \nconcentration ( β -fac- \ntor \n3 \n) was 18.6%/100 p.p.m. during 1981 – 2016. This β -factor is \nAccumulated simulated land sink \n* \nMean accumulated land sink from TRENDY models \n*** \nSimulated annual land sink  \nAccumulated simulated land sink \nAccumulated simulated land sink \n** \nAnnual residual land sink \nAccumulated residual land sink \n0 \n140 \n120 \n100 \n80 \n60 \n40 \n20 \nAccumulated  \ng \nlobal land sink (P \ng \n C) \n4 \n3 \n2 \n1 \n0 \n–1 \n5 \n6 \nAnnual global land sink (Pg C yr \n–1 \n) \n1981  1984  1987  1990  1993  1996  1999  2002  2005 2008   2011  2014 \nYear \nFig. 2 Comparison of the simulated annual land sink (NEP) by BEPS and the \nresidual land sink (RLS) estimated by the Global Carbon Project. The pink \nred error bars are the SDs of the annual land sink simulated using three \ndifferent LAI datasets. The solid red line indicates the accumulated carbon \nsink simulated using three LAI datasets. For the accumulated simulated \nland sink* (light solid blue line), pixels with >20% areal changes in short \nvegetation or tree canopy are excluded in the accumulation. For the \naccumulated simulated land sink** (dashed red line), pixels with >30% \nchanges are excluded. The solid dark line indicates the accumulated \nresidual land sink estimated by the Global Carbon Project. The solid blue \nline is the mean of accumulated land sinks simulated by 15 TRENDY \nmodels, and the shaded gray area represents its uncertainty (mean ± SD). \nThe shaded light yellow area represents the range of accumulated land \nsinks simulated using three different LAI datasets \np  < 0.01 p  < 0.05 p  > 0.05 p  < 0.05 p  < 0.01 \nChanging signifiance \nDreasing Increasing \nArea fraction (%) \n50 \n40 \n30 \n20 \n10 \n0 \nLAI trend (10 \n–1 \n m \n2 \n m \n–2 \n yr \n–1 \n) \n–0.1 0 0.1 \nFig. 1 Global map indicating the trend of LAI from 1981 to 2016. The gray color indicates non-vegetated areas and the white color denotes that the trend is \nstatistically insigni fi cant ( p > 0.05). Positive values indicate increasing trends of growing season mean LAI and vice versa \nNATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-12257-8 ARTICLE \nNATURE COMMUNICATIONS |         (2019) 10:4259  | https://doi.org/10.1038/s41467-019-12257-8 | www.nature.com/naturecommunications 3about 6.1% smaller than the average value of prognostic models \nparticipating in the Coupled Model Intercomparison Project 5 \n(CMIP5) \n3 \n. Our β -factor excludes the effects of CO \n2 \nfertilization \non LAI, whereas CMIP5 models lump vegetation LAI responses \nto CO \n2 \nincrease into the β -factor. If the in fl uence of LAI change is \nincluded in our β -factor calculation, it would be 31.8%/100 p.p.m. \nfor the northern hemisphere, whereas the global average value is \n27.0%, because LAI increased much less in the southern hemi- \nsphere than in the northern hemisphere. Both BEPS and CMIP5 \nmodels include the in fl uences of CO \n2 \nincrease on stomatal con- \nductance affecting photosynthesis, transpiration, and soil moist- \nure. Our β -factor is considerably larger than the values inferred \nfrom remote-sensing NPP models \n3 \n, because these models are \nmostly based on empirical LUEs that do not explicitly consider \nenhanced LUE at higher CO \n2 \nlevels \n24 \n(see Supplementary Fig. 4 \nfor further comparisons). CO \n2 \nfertilization enhanced the carbon \nsink in all regions (Fig. 4 ), especially in regions with high NPP \n(15°S – 30°N and 45°N – 60°N) (Supplementary Fig. 5). \nThe impacts of other drivers on the accumulated sink shown in \nFig. 3 are also calculated by changing one driver at a time, while \nholding other drivers at the baseline level. The accumulated global \nsink enhancement due to vegetation structure (LAI) change over \nthe 1981 – 2016 period is 11.7 Pg C, which is 12.4% of the total sink \nor 35.7% of the enhanced sink in the same period. Over this \nperiod, global average LAI increased from 1.6 to 1.7, enhancing \nGPP by 1.2% and NEP by 0.3% relative to GPP. The spatial \ndistribution of NEP enhancement (Supplementary Fig. 6) is \nsimilar to that of the LAI trend shown in Fig. 1 , where positive \ntrends of LAI induced sinks, whereas negative trends caused \nsources, suggesting that LAI might have acted as a surrogate for \nthe impacts of changes in other factors such as soil moisture and \ntemperature. As most areas show positive trends, the overall effect \nof the vegetation structural change is a large sink enhancement. \nChanges in atmospheric nitrogen deposition made global land \necosystems sequester 1.1 Pg C more carbon during the 1981 – 2016 \nperiod than the baseline (Fig. 3 ). Nitrogen deposition contributed \nto 1.1% of the total accumulated sink since 1981. Based on global \nnitrogen  deposition  data,  global  total  nitrogen  deposition \nincreased from 42.3 Tg N per year in 1981 to 58.9 Tg N per year \nin 2016 (Supplementary Fig. 7), with a total additional cumulative \ninput of 0.30 Pg N into land ecosystems above the 1981 baseline. \nOur simulated carbon sink enhancement per unit of deposited \nnitrogen is 3.7 g C/g N, which is slightly lower than the range of \n4.3 – 4.8 g C/g N by previous global simulations \n25 , 26 \n. The NEP \nenhancement during the 1981 – 2016 period by nitrogen deposition \nmainly occurred in Asia and in part of Europe, where nitrogen \ndeposition continuously increased (Supplementary Fig. 8). \nGlobally, climate change weakened the land sink during the \n1981 – 2016 period (Fig. 3 ), when it ’ s effect on LAI, such as longer \ngrowing  season,  is  excluded.  Climate  change  induced  an \naccumulated GPP reduction of 37.6 Pg C, whereas the accumu- \nlated decrease of ecosystem respiration was 10.5 Pg C during the \n1981 – 2016 period. Consequently, the climate change caused a net \nreduction  of  27.1 Pg C  ( − 28.6%)  in  the  accumulated  sink \nenhancement since 1981. The decrease of the land sink due to \nclimate change occurred almost in all regions (Fig. 4 , Supple- \nmentary Fig. 9). The sum of the effects of changes in LAI, CO \n2 \n, \nnitrogen deposition, and climate mainly enhanced carbon sink in \nEurasia, southeastern China, eastern North America, central \nAfrica, and southeast Asia (Fig. 4 , Supplementary Fig. 10). The \ndominant driver affecting the land sink varies spatially (Supple- \nmentary Fig. 11). Climate had the most dominant-negative \nimpact on the accumulated carbon sink in 14.2% of the total \nvegetated area of the globe. LAI is the dominant-positive driver \nfor 43.6% of the area, whereas it is negative for 4.6% of the area. \nCO \n2 \nand nitrogen deposition are both positive, dominant factors \nover 36.4% and 0.2% of the area, respectively. \nDiscussion \nIn principle, NEP from BEPS should not equal the RLS, because \nNEP excludes the net emission from anthropogenic and natural \ndisturbances (land cover and land-use change, harvest, planta- \ntion, fi res, and insects) (Supplementary Discussion). Those dis- \nturbances  can  induce  both  immediate  LAI  reduction  and \nsubsequent gradual LAI increase due to regrowth. In BEPS, a LAI \nreduction induces additional transfer of the same portions of \nbiomass pools to soil organic matter, which is subsequently \nrespired  as  a  source  of  carbon  to  the  atmosphere.  This \ndisturbance-enhanced carbon loss by respiration is mostly com- \npensated by regrowth, which is driven by observed LAI series in \nClimate change \nN deposition change \nLAI change \nCO \n2 \n  \nchange \nCO \n2  \n+ LAI + N deposition + climate change \nAccumulated NEP enhancement (Pg C) \n1981  1984  1987  1990  1993  1996  1999  2002  2005  2008  2011  2014 \nYear \n50 \n40 \n30 \n20 \n10 \n0 \n–10 \n–20 \n–30 \n–40 \nFig. 3 The accumulated terrestrial carbon sink from 1981 to 2016, after \nsubtracting the baseline sink of 62.1 Pg C due to long-term changes (legacy \neffects). Through numerical experiments, the variations of the accumulated \nsink with time due to the various drivers are calculated, showing that CO \n2 \nfertilization (without LAI changes), LAI, nitrogen deposition, and climate are \nresponsible for the sinks of 44.6, 11.7, 1.1, and − 27.1 Pg C over this period, \nrespectively. These sinks contribute 47.0%, 12.4%, 1.1%, and − 28.6% to \nthe total accumulated residual land sink (95 ± 5 Pg C) over this period, \nrespectively. For example, the accumulated CO \n2 \nenhancement is calculated \nby holding other drivers at the baseline, while changing CO \n2 \naccording to \nthe global mean CO \n2 \ndata \n2.0 \n1.6 \n1.2 \n0.8 \n0.4 \n0.0 \n–0.4 \n–0.8 \nRegion \nAsia North Europe Africa South    Oceania   Global \ntotal \nAmerica America \nEnhancement on NEP (Pg C yr \n–1 \n) \nCO \n2 \nLAI \nN deposition \nClimate \nFig. 4 Attribution of the global land sink to various factors by region \naveraged over the period from 1981 to 2016. The values shown are regional \nand global totals of mean land sinks simulated with an individual diver \n(CO \n2 \n, LAI, nitrogen deposition, and climate) changing historically and other \nfactors being remained temporally unchanged minus the lase sinks due to \nlong-term changes (legacy effects) \nARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-12257-8 \n4 NATURE COMMUNICATIONS |         (2019) 10:4259  | https://doi.org/10.1038/s41467-019-12257-8 | www.nature.com/naturecommunicationsthe same period, whereas NEP equals RLS for undisturbed areas. \nIf the carbon gain through regrowth equaled the carbon loss due \nto enhanced respiration, the simulated NEP would equal RLS for \ndisturbed areas. This would be true only if the disturbance and \nrecovery were historically maintained at constant rates. However, \nthere are considerable differences between NEP modeled by BEPS \nand RLS for areas with large changes in short vegetation (SV), \ntree cover (TC), and bare land. To assess these differences, we also \nshow in Fig. 2 the simulated accumulated NEP curves after \nexcluding pixels (8 km resolution) with changes in SV or TC \ncoverage over 20% and 30%, respectively, in the accumulation, \nbased on SV and TC high-resolution (30 m) map produced from \nLandsat data \n27 \n. Excluding these areas with large SV or TC \nchanges results in small decreases in accumulated NEP by 8.0% \nand 2.5% for these two cases, respectively, suggesting that \ncumulative NEP approximately equals cumulative RLS at the \nglobal scale since 1981. These effects of disturbance on simulated \nNEP are small, because pixels with 20% and 30% disturbance in \nSV or TC are only 6.7% and 2.0% of the land area, respectively \n(Supplementary Fig. 12), and the LAI observations used to drive \nBEPS can capture a large portion of the impacts of disturbance on \nphotosynthesis  and  ecosystem  respiration.  These  exclusions \nalways cause smaller NEP values, because the contributions from \nundisturbed portions within the excluded pixels are discounted. \nChanges from SV to TC or from bare land to SV or TC within the \nexcluded pixels are also discounted, although they may incur \nsmall sinks (Supplementary Fig. 13), which are associated with \nLAI changes (Supplementary Fig. 14). For simplicity, we do not \nexclude disturbed pixels in the results shown in Figs. 2 and 3 . \nIn conclusion, the results of this study show the value of \nassimilating observed LAI over the last three decades to quantify \nthe land carbon sink and separate the effect of increasing LAI \nalone vs. the effects of changes in environmental drivers alone. It \nshould be kept in mind, however, that the observed increase of \nLAI also results from environmental drivers as well as of land use \nand land management. \nMethods \nGPP modeling methods . The BEPS model \n17 , 18 \nused in this study is a process- \nbased diagnostic model driven by remotely sensed vegetation parameters, including \nLAI, clumping index, and land cover type, as well as meteorological and soil data. It \nsimulates photosynthesis, energy balance, and hydrological and soil biogeochemical \nprocesses at daily time steps \n17 , 28 \n. For GPP simulation, BEPS uses the leaf-level \nbiochemical model \n29 \nwith a two-leaf upscaling scheme from leaf to canopy: \n17 \nGPP ¼ GPP \nsun \nLAI \nsun \nþ GPP \nshaded \nLAI \nshaded \nð 1 Þ \nwhere GPP \nsun \nand GPP \nshaded \nare the GPP per unit area of sunlit and shaded leaves, \nrespectively. LAI \nsun \nand LAI \nshaded \nare the LAI of sunlit and shaded leaves, \nrespectively, and are estimated as: \nLAI \nsun \n¼ 2  ́ cos θ  ́ 1 \u0002 exp \u0002 0 : 5  ́ Ω  ́ \nLAI \ncos θ \n\u0012\u0013 \u0014\u0015 \nð 2 Þ \nLAI \nshade \n¼ LAI \u0002 LAI \nsun \nð 3 Þ \nwhere Ω is the clumping index derived from MODIS data at 500 m resolution \n30 \nand θ is the daily mean solar zenith angle. \nGPP values of sunlit and shaded leaves are calculated using the Farquhar ’ s \nmodel \n29 \nwith consideration of the large difference in incident solar irradiance and \nthe small difference in the carboxylation rate between these two-leaf groups \n18 \n. \nStomatal conductances of sunlit and shaded leaves are determined separately \naccording to photosynthesis rates of these leaves, atmospheric CO \n2 \nconcentration, \nand soil moisture \n28 , 31 \n, and are used to estimate water consumption by \nevapotranspiration. Although initially developed to simulate GPP in boreal \necosystems in Canada, BEPS has been adopted and has been widely used to \nestimate terrestrial carbon and water fl uxes in China \n32 , 33 \n, North America \n34 , 35 \n, \nEurope \n36 \n, East Asia \n37 \n, and the globe \n18 \n. \nFull carbon cycle modeling methods . BEPS includes modules to calculate HR and \nNEP \n28 \n. Based on a modi fi ed Century model \n38 , 39 \n, it strati fi es the biomass carbon \nstock into four pools (leaf, stem, coarse root, and fi ne root pools) and the soil \ncarbon stock into nine pools (surface structural litter, surface metabolic litter, soil \nstructural litter, soil metabolic litter, coarse woody litter, surface microbe, soil \nmicrobe, slow, and passive carbon pools). These carbon pools are initialized in the \nfollowing way. The model fi rst calculates NPP in 1901 using N deposition and CO \n2 \nconcentration in 1901, a random year of climate data selected in the period from \n1901 to 1910, seasonally variable LAI averaged over the 1982 – 1986 period, and \ndefault C:N ratios for all carbon pools. The nine soil and four biomass carbon pools \nare then estimated under the assumption that the carbon cycle of terrestrial eco- \nsystems was in dynamic equilibrium in 1901. With this assumption, all carbon \npools are determined by solving a set of equations describing the dynamics of \ncarbon pools \n40 \n. For the period from 1901 to 1980, the model is run using historical \ndata of N deposition, CO \n2 \nconcentration, and climate, and the average LAI during \n1982 to 1986. Due to lack of data, we assume that LAI in 1982 – 1986 represents that \nin 1901 – 1981. If LAI increased in this period, NPP in 1901 – 1910 would be over- \nestimated, leading to larger soil carbon pools in 1901 and smaller NEP in \n1981 – 2016. To address this issue, we conducted a set of simulations by extending \nthe LAI time series to 1901 according to atmospheric CO \n2 \nconcentration with the \nrate of LAI change with CO \n2 \ndetermined using 1981 – 2016 data. We fi nd that the \nenhancement of the land sink due to LAI change in 1981 – 2016 decreases from \n12.4% to 10.5% relative to the accumulated sink in the same period when the \npossible LAI increase from 1901 to 1981 is considered. This decrease is due to \nhigher accumulated NEP by 6.8 Pg C during 1981 – 2016, resulting from lower \ninitial soil carbon pools at 1901 when LAI is smaller than our simulations without \nconsidering LAI change over 1901 – 1981. This set of simulations suggests that the \nimpact of possible LAI changes prior to 1981 on the role of LAI after 1981 is within \na few percent and does not affect our conclusion on the signi fi cance of LAI increase \nafter 1981 in enhancing the land sink. Although extrapolating LAI according to \nCO \n2 \nconcentration is overly simplistic, it may be considered as setting the upper \nbound of the possible error due to LAI changes prior to 1981, because climate \nchange could have been negative on plant growth and LAI. \nThe decomposition of soil carbon and mineralization of soil nitrogen are \nregulated by soil temperature, moisture, texture, and chemical property of soil \ncarbon pools. Nitrogen available for vegetation growth consists of the total of \nmineralized and deposited nitrogen. The uptake of nitrogen by vegetation is \nsimulated according to temperature, total amounts of soil carbon and nitrogen, and \nvegetation demand. The absorbed nitrogen is allocated daily to different vegetation \ncarbon pools based on the C:N ratios and NPP allocation coef fi cients. The nitrogen \ncontent of leaves is used to adjust the parameter Vcmax at 25 °C, which \nconsequently affects the photosynthesis rates of sunlit and shade leaves in the \nFarqhuar model \n29 \n. AR consists of maintenance and growth respiration, and \nmaintenance respiration depends on foliage, stem and root biomass, and \ntemperature, whereas growth respiration is taken as a fraction (25%) of GPP. NEP \nof each modeling grid equals GPP-AR-HR. \nLAI data . LAI is an input into the BEPS model for the simulation of the carbon \nfl ux. Three LAI time series, GLOBMAP-V2, GLASS, and LAI3g are used in this \nstudy and are shown in Supplementary Fig. 1 in comparison with other LAI time \nseries. The GLOBMAP _V2 product is the basis for our simulations, whereas \nGLASS and LAI3g are used to assess uncertainties in carbon sink estimation due to \nthe choice of LAI products. GLOBMAP_v2 over the period from 1981 to 2016 was \ngenerated through fusing LAI inverted from MODIS re fl ectance data with AVHRR \nGIMMS NDVI data. LAI from 2001 to 2016 was fi rst derived from the MOD09A1 \nC6 land surface re fl ectance and the associated illumination and view angles based \non the GLOBCARBON LAI algorithm \n19 , 41 \n, which was developed on the basis of \nthe 4-Scale geometric optical model \n42 \n. This algorithm explicitly considers the \neffects of the bidirectional re fl ectance distribution function on re fl ectance over the \ncanopy as measured by the sensors \n41 \n. For the fusion of MODIS and AVHRR \nremote-sensing data, the relationships between GIMMS NDVI and MODIS LAI \nwere established pixel by pixel over a period (2001 – 2006) \n19 \nthat they overlap. Then \nthe AVHRR LAI from 1981 to 2000 was generated using these relationships, to \nensure the temporal consistency between these two sensors. The spatial resolution \nof the LAI series is 0.072727° × 0.072727° and temporal resolution varies from \n16 days (1981 to 2000) to 8 days (2001 to 2016). In the simulation, these 16 days \nand 8 days LAI values were interpolated into daily values. GLASS and LAI3g have \nthe similar temporal coverage. All three long-term LAI series used in this study \nhave similar magnitudes, because they have all considered the three-dimensional \ncanopy structure, as characterized by the clumping index, in their retrieval algo- \nrithms \n43 \n. Both GLOBMAP_V2 and GLASS used the same global clumping index \nmap \n30 \n, whereas LAI3g considered clumping in a different way \n21 \n. For accurate \nsimulation of sunlit and shaded leaf area and GPP, both LAI and clumping index \nare needed. \nAll these three products used the processed AVHRR data (GIMMS). The issues \nwith possible artifacts and errors in the AVHRR data series (GIMMS), such as \nsensor degradation, sensor intercalibration, orbital drift causing changes in sun- \ntarget-view geometry, distortions by clouds, and abnormal aerosol absorption by \ntwo major volcanic eruptions, have been fully considered and recti fi ed to a large \nextent, ensuring the useful signals in the trend being extracted \n44 \n. The GIMMS time \nseries has quality fl ags with values from 0 to 6. GLOBMAP used only the top \nquality 0 or 1. Depending on the strength of quality control, different LAI products \ncould show different interannual variations, although they are consistent in their \nincreasing trends (Supplementary Table 1). \nNATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-12257-8 ARTICLE \nNATURE COMMUNICATIONS |         (2019) 10:4259  | https://doi.org/10.1038/s41467-019-12257-8 | www.nature.com/naturecommunications 5LAI data have been assimilated into diagnostic ecosystem models to optimize \nplant carbon allocation, stock, and residence time, as well as carbon-use \nef fi ciency \n16 , 45 \n. In our study, we similarly optimized some of these parameters \nthrough a spin-up procedure and also used LAI data to calculate the long-term \nglobal carbon cycle. \nMeteorological data . Meteorological data required to force the BEPS model \ninclude daily maximum and minimum temperatures, downward solar radiation, \nrelative humidity, and precipitation. These data are interpolated from the 0.5° × \n0.5° CRUNCEP V8.0 dataset, which is a combination of CRU monthly climatology \nand 6 hourly NCEP reanalysis meteorological data \n46 \n. Daily maximum and mini- \nmum temperatures, downward solar radiation, and precipitation are directly \nretrieved from the CRUNCEP V8.0 dataset. Relative humidity is calculated from \ntemperatures, speci fi c humidity, and pressure from the CRUNCEP V8.0 dataset. \nThe 0.5° × 0.5° meteorological data are interpolated into 0.072727° × 0.072727° \nresolution using a bilinear interpolation method. These data are the same as those \nused by the TRENDY models. \nSoil data . Fractions of clay, silt, and sand are retrieved from the harmonized global \nsoil database ( http://www.fao.org/nr/lman/abst/lman_080701_en.htm ) and are \nused to determine soil physical parameters, including wilting point, fi eld capacity, \nporosity, hydrological conductance, exponent of the moisture release equation, \nand so on. \nNitrogen deposition data . The yearly global nitrogen deposition data at 0.5° × 0.5° \nresolution over the period from 1960 to 2009 are estimated from tropospheric NO \n2 \ncolumn density retrieved from Global Ozone Monitoring Experiment and Scan- \nning Imaging Absorption Spectrometer for Atmospheric Cartography sensors, \nmeteorological data, and NOx emission inventory data \n47 \n. For the years from 2010 \nto 2016, nitrogen data are extrapolated using the estimated nitrogen data over the \nperiod from 2000 to 2009. For the period from 1901 to 1959, nitrogen data are \nextrapolated based on the change rates of nitrogen deposition over the period from \n1960 to 1969. The 0.5° × 0.5° nitrogen deposition data are interpolated into the \n0.072727° × 0.072727° resolution using a bilinear interpolation method. \nThe N deposition dataset used in this study is compared with that used by \nTRENDY models (Supplementary Fig. 2). These two datasets are similar before \n1990, as both are based on measurements, but the increasing trends after 1990 are \ndifferent, because the data used by the models are based on linear extrapolation \nfrom 1990 to 2050, at which the nitrogen deposition is estimated based on \nprojected anthropogenic sources and other assumptions \n48 \n, whereas satellite \nmeasurements from 2000 to 2009 are used in our dataset \n47 \nand could follow the \nrealistic trend more closely than the linearly extrapolated trend used by the models. \nOver the 1981 – 2016 period, the total N deposition is 301 Tg N in our study, \nwhereas it is 403 Tg N in TRENDY. The difference could be due to the overall \ndecrease in N deposition in North America and other regions in this period. \nUncertainty assessment . The uncertainty of model results shown in Figs. 2 and 3 \nis estimated based on differences among simulated results using the three LAI \nproducts. This uncertainty is of a similar magnitude to those estimated from the \nuncertainties of model parameters that in fl uence the simulated temporal trends, \nbecause bias errors are mostly constrained by the atmospheric CO \n2 \nconcentration \nand the main interest of this study is to attribute the land sink to the various drivers \nthrough their in fl uences on the temporal trend. The uncertainty for the CO \n2 \nfer- \ntilization effect, e.g., is mostly caused by the uncertainties in the slope and intercept \nof the linear relationship between stomatal conductance and photosynthesis rate. \nThe trend of NEP against climate is mostly controlled by the sensitivities of AR and \nHR to temperature. \nData availability \nThe global clumping data are available at http://globalmapping.org/CI/ . The global \nGLOBMAPV3 LAI dataset for the period from 1981 to 2016 can be downloaded at \nhttp://globalmapping.org/globalLAI/ . \nCode availability \nThe code used in this study is available from the corresponding author on request. \nReceived: 29 April 2019 Accepted: 30 August 2019 \nReferences \n1.   Fiedlingstein, P. et al. Uncertainties in CMIP5 climate projections due to \ncarbon cycle feedbacks. J. Clim. 27 , 511 – 526 (2014). \n2.   Langley, J. A. & Hungate B. A. Plant community feedbacks and long-term \necosystem responses to multi-factored global change. AoB Plants 6 , https://doi. \norg/10.1093/aobpla/plu035 (2014). \n3.   Smith, W. K. et al. Large divergence of satellite and Earth system model \nestimates of global terrestrial CO \n2 \nfertilization. Nat. Clim. Change 6 , 306 – 310 \n(2016). \n4.   Le Quere, C. Global carbon budget 2017. Earth Syst. Sci. Data 10 , 405 – 448 \n(2018). \n5.   Huntzinger D. N. et al. Uncertainty in the response of terrestrial carbon sink \nto environmental drivers undermines carbon-climate feedback predictions. \nSci. Rep. 7 , 4765, https://doi.org/10.1038/s41598-017-03818-2 (2017). \n6.   Piao, S. et al. Evaluation of terrestrial carbon cycle models for their response to \nclimate variability and to CO \n2 \ntrends. Glob. Change Biol. 14 , 2015 – 2039 \n(2013). \n7.   Weider, W. R., Cleveland, C. C., Smith, W. K. & Todd-Brown, K. Future \nproductivity and carbon storage limited by terrestrial nutrient availability. Nat. \nGeosci. 8 , 441 – 444 (2015). \n8.   Clark, P. U. et al. Consequences of twenty- fi rst century policy for multi- \nmillennial climate and sea-level change. Nat. Clim. Change 6 , 360 – 369 (2016). \n9.   Schimel, D. et al. Observing terrestrial ecosystems and the carbon cycle from \nspace. Glob. Change Biol. 21 , 1762 – 1776 (2015). \n10.  Bayer, A. D., Pugh, T. A. M., Krause, A. & Arneth, A. Historical and future \nquanti fi cation of terrestrial carbon sequestration from a greenhouse-gas-value \nperspective. Glob. Environ. Change 32 , 153 – 164 (2015). \n11.  Arora, V. K. et al. Carbon concentration and carbon climate feedbacks in \nCMIP5 Earth System Models. J. Clim. 26 , 5289 – 5314 (2013). \n12.  Thomas, R. Q., Canham, C. D., Weathers, K. C. & Goodale, C. L. Increased \ntree carbon storage in response to nitrogen deposition in the US. Nat. Geosci. \n3 ,13 – 17 (2009). \n13.  Chen, J. M. & Black, T. A. De fi ning leaf area index for non- fl at leaves. Plant \nCell Environ. 15 , 421 – 429 (1992). \n14.  Zhu, Z. et al. Greening of the Earth and its drivers. Nat. Clim. Change 6 , \n791 – 795 (2016). \n15.  Jiang, C. et al. Inconsistencies of Interannual Variability and trends in long- \nterm satellite leaf area index products. Glob. Change Biol. 23 , 4133 – 4146 \n(2017). \n16.  Exbrayat, J. F. et al. Understanding the land carbon cycle with space data: \ncurrent status and prospects. Surv. Geophys. 40 , 735 – 755 (2019). \n17.  Chen, J. M., Liu, J., Cihlar, J. & Guolden, M. L. Daily canopy photosynthesis \nmodel through temporal and spatial scaling for remote sensing applications. \nEcol. Model. 124 ,99 – 119 (1999). \n18.  Chen, J. M. et al. Effects of foliage clumping on global terrestrial gross primary \nproductivity. Glob. Biogeochem. Cycles 26 , GB1019 (2012). \n19.  Liu, Y., Liu, R. & Chen, J. M. Retrospective retrieval of long-term consistent \nglobal leaf area index (1981-2010) maps from combined AVHRR and MODIS \ndata. J. Geophys. Res. Biogeosci. 117 , G04003 (2012). \n20.  Xiao, Z. et al. Long-time-series global land surface satellite leaf area index \nproduct derived from MODIS and AVHRR surface re fl ectance. IEEE Trans. \nGeosci. Remote Sens. 54 , 5301 – 5318 (2016). \n21.  Zhu, Z. et al. Global data sets of vegetation leaf area index (LAI) 3 g and \nFraction of Photosynthetically Active Radiation (FPAR) 3g derived from \nGlobal Inventory Modeling and Mapping Studies (GIMMS) Normalized \nDifference Vegetation Index (NDVI3 g) for the period 1981-2011. Remote \nSens. 5 , 927 – 948 (2013). \n22.  Ryu, Y. et al. Integration of MODIS land and atmosphere products with a \ncoupled-process model to estimate gross primary productivity and \nevapotranspiration from 1 km to global scales. Glob. Biogeochem. Cycles 25 , \nGB4017 (2011). \n23.  Potter, C. S. et al. Terrestrial ecosystem production: a process model based \non global satellite and surface data. Glob. Biogeochem. Cycles 7 , 811 – 841 \n(1993). \n24.  De Kauwe, M. G., Keenan, T. F., Medlyn, B. E., Prentice, I. C. & Terrer, C. \nSatellite based estimates underestimate the effect of CO \n2 \nfertilization on net \nprimary productivity. Nat. Clim. Change 6 , 892 – 893 (2016). \n25.  Zaehle, S. Terrestrial nitrogen-carbon cycles interactions at the global scale. \nPhilos. T. Roy. Soc. B 368 , https://doi.org/10.1098/rstb.2013.0125 (2013). \n26.  Churkina, G. et al. Synergy of rising nitrogen deposition and atmospheric CO \n2 \non land carbon uptake moderately offset global warming. Glob. Biogeochem. \nCy . 23 , GB4027 (2009). \n27.  Song, X. P. et al. Global land change from 1982 to 2016. Nature 560 , 639 – 643 \n(2018). \n28.  Ju, W. M. et al. Modelling multi-year coupled carbon and water fl uxes in a \nboreal aspen forest. Agr. For. Meteorol. 140 , 136 – 151 (2006). \n29.  Farquhar, G. D., von Caemmerer, S. & Berry, J. A. A biochemical model of \nphotosynthetic CO \n2 \nassimilation in leaves of C3 species. Planta 149 ,78 – 90 \n(1980). \n30.  He, L., Chen, J. M., Pisek, J., Schaaf, C. B. & Strahler, A. H. Global clumping \nindex map derived from the MODIS BRDF product. Remote Sens. Environ. \n119 , 118 – 130 (2012). \n31.  Ball, J. T. An analysis of stomatal conductance, Ph.D Thesis , Standford \nUniversity (1988). \nARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-12257-8 \n6 NATURE COMMUNICATIONS |         (2019) 10:4259  | https://doi.org/10.1038/s41467-019-12257-8 | www.nature.com/naturecommunications32.  Feng, X. et al. Net primary productivity of China ’ s terrestrial ecosystems from a \nprocess model driven by remote sensing. J. Environ. Manag. 85 , 563 – 573 (2007). \n33.  Liu, Y. et al. Evapotranspiration and water yield over China ’ s landmass from \n2000 to 2010. Hydrol. Earth Syst. Sci. 17 , 4957 – 4980 (2013). \n34.  Schwalm, C. R. et al. A model-data intercomparison of CO \n2 \nexchange across \nNorth America: results from the North American Carbon Program site \nsynthesis. J. Geophys. Res. Biogeosci. 115 , https://doi.org/10.1029/ \n2009jg001229 (2010). \n35.  Sprintsin, M., Chen, J. M., Desai, A. & Gough, C. M. Evaluation of leaf-to- \ncanopy upscaling methodologies against carbon fl ux data in North America. J. \nGeophys. Res. Biogeosci. 117 , https://doi.org/10.1029/2010jg001407 (2012). \n36.  Wang, Q., Tenhunen, J., Falge, E., Bernhofer, C. & Granier, A. Simulation and \nscaling of temporal variation in gross primary production for coniferous and \ndeciduous temperate forests. Glob. Change Biol. 10 ,37 – 51 (2004). \n37.  Matsushita, B. & Tamura, M. Integrating remotely sensed data with an \necosystem model to estimate net primary productivity in East Asia. Remote \nSens. Environ. 81 ,58 – 66 (2002). \n38.  Ju, W. M., Chen, J. M., Harvey, D. & Wang, S. Future carbon balance of \nChina ’ s forests under climate change and increasing CO \n2 \n. J. Environ. Manag. \n85 , 538 – 562 (2007). \n39.  Parton, W. J. et al. Observations and modeling of biomass and soil organic- \nmatter dynamics for the grassland biome worldwide. Glob. Biogeochem. Cy. 7 , \n785 – 809 (1993). \n40.  Chen, J. M. et al. Spatial distribution of carbon sources and sinks in Canada ’ s \nforests. Tellus B 55 , 622 – 642 (2003). \n41.  Deng, F., Chen, J. M., Plummer, S., Chen, M. Z. & Pisek, J. Algorithm for \nglobal leaf area index retrieval using satellite imagery. IEEE Trons. Geosci. \nRemote Sens. 44 , 2219 – 2229 (2006). \n42.  Chen, J. M. & Leblanc, S. G. A four-scale bidirectional re fl ectance model based \non canopy architecture. IEEE Trans. Geosci. Remote Sens. 35 , 1316 – 1337 (1997). \n43.  Chen, J. M. in Comprehensive Remote Sensing , Vol. 3 (ed. Liang, S.) Pages \n53 – 77, ISBN 9780128032213, https://doi.org/10.1016/B978-0-12-409548- \n9.10540-82018 (Elsevier, Oxford, 2017). \n44.  Tucker, C. J. et al. An extended AVHRR 8-km NDVI dataset compatible with \nMODIS and SPOT vegetation NDVI data. Int. J. Remote Sens. 26 , 4485 – 4498 \n(2005). \n45.  Bloom, A. A., Exbrayat, J. F., van der Velde, I. R., Feng, L. & Williams, M. The \ndecadal state of the terrestrial carbon cycle: global retrievals of terrestrial \ncarbon allocation, pools and residence times. Proc. Natl Acad. Sci. USA 113 , \n1285 – 1290 (2015). \n46.  Viovy, N. CRUNCEP data set, available at: ftp://nacp.ornl.gov/synthesis/2009/ \nfrescati/temp/land_use_change/original/readme.htm , last access (2016). \n47.  Lu, X. H. et al. Estimated global nitrogen deposition using NO \n2 \ncolumn \ndensity. Int. J. Remote Sens. 34 , 8893 – 8906 (2014). \n48.  Dentener, F. J. Global maps of atmospheric nitrogen deposition, 1860, 1993, \nand 2050. Data set. Available on-line [ http://daac.ornl.gov/ ] from Oak Ridge \nNational Laboratory Distributed Active Archive Center, Oak Ridge, \nTennessee, USA. https://doi.org/10.3334/ORNLDAAC/830 (2006). \nAcknowledgements \nThis study is funded by the National Key R&D Program of China (2016YFA0600202) \nand in part by the National Natural Science Foundation of China (41671343) and by the \nFundamental Research Funds for the Central Universities. P.C. acknowledges support \nfrom the European Research Council Synergy project SyG-2013-610028 IMBALANCE-P \nand the ANR CLAND Convergence Institute \nAuthor contributions \nC.J.M. and J.W.M. designed this research and wrote the fi rst draft of the manuscript. \nJ.W.M. carried out all computations. C.P. helped with the assessment of disturbance \neffects and provided constructive comments on the manuscript and helped. V.N. assisted \nin the use of meteorological data and in writing part of the manuscript. L.R.G. and L.Y. \ngenerated the GLOMAP LAI dataset. L.X.H. provided the major part of the nitrogen \ndeposition dataset. \nAdditional information \nSupplementary Information accompanies this paper at https://doi.org/10.1038/s41467- \n019-12257-8 . \nCompeting interests: The authors declare no competing interests. \nReprints and permission information is available online at http://npg.nature.com/ \nreprintsandpermissions/ \nPeer review information Nature Communications thanks Chris Jones and the other, \nanonymous, reviewer(s) for their contribution to the peer review of this work. Peer \nreviewer reports are available. \nPublisher ’ s note Springer Nature remains neutral with regard to jurisdictional claims in \npublished maps and institutional af fi liations. \nOpen Access This article is licensed under a Creative Commons \nAttribution 4.0 International License, which permits use, sharing, \nadaptation, distribution and reproduction in any medium or format, as long as you give \nappropriate credit to the original author(s) and the source, provide a link to the Creative \nCommons license, and indicate if changes were made. The images or other third party \nmaterial in this article are included in the article ’ s Creative Commons license, unless \nindicated otherwise in a credit line to the material. If material is not included in the \narticle ’ s Creative Commons license and your intended use is not permitted by statutory \nregulation or exceeds the permitted use, you will need to obtain permission directly from \nthe copyright holder. To view a copy of this license, visit http://creativecommons.org/ \nlicenses/by/4.0/ . \n© The Author(s) 2019 \nNATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-12257-8 ARTICLE \nNATURE COMMUNICATIONS |         (2019) 10:4259  | https://doi.org/10.1038/s41467-019-12257-8 | www.nature.com/naturecommunications 7",
      "published": "3/14/2024, 10:57:10 AM",
      "title": "2022-01-01-Test-Document.pdf",
      "token_count_estimate": 13442,
      "url": "file:///app/collector/hotdir/2022-01-01-Test-Document.pdf",
      "wordCount": 8259
    }
  ],
  "error": null,
  "success": true
}