Atmos. Chem. Phys., 19, 12361–12375, 2019 https://doi.org/10.5194/acp-19-12361-2019 © Author(s) 2019. This work is distributed under the Creative Commons Attribution 4.0 License. Contrasting effects of CO2 fertilization, land-use change and warming on seasonal amplitude of Northern Hemisphere CO2 exchange Ana Bastos1 , Philippe Ciais2 , Frédéric Chevallier2 , Christian Rödenbeck3 , Ashley P. Ballantyne2,4 , Fabienne Maignan2 , Yi Yin5 , Marcos Fernández-Martínez6 , Pierre Friedlingstein7 , Josep Peñuelas8,9 , Shilong L. Piao10 , Stephen Sitch11 , William K. Smith12 , Xuhui Wang2 , Zaichun Zhu13 , Vanessa Haverd14 , Etsushi Kato15 , Atul K. Jain16 , Sebastian Lienert17 , Danica Lombardozzi18 , Julia E. M. S. Nabel19 , Philippe Peylin2 , Benjamin Poulter20 , and Dan Zhu2 1 Department of Geography, Ludwig-Maximilians-Universität, Luisenstr. 37, 80333, Munich, Germany 2 Laboratoire des Sciences du Climat et de l’Environnement, LSCE/IPSL, CEA-CNRS-UVSQ, Université Paris-Saclay, 91191 Gif-sur-Yvette, France 3 Max Planck Institute for Biogeochemistry, 07745, Jena, Germany 4 Department of Ecosystem and Conservation Sciences, University of Montana, Missoula, Montana 59812, USA 5 Division of Geological and Planetary Sciences, California Institute of Technology, Pasadena, CA 91125, USA 6 Centre of Excellence PLECO (Plants and Ecosystems), Department of Biology, University of Antwerp, 2610 Wilrijk, Belgium 7 College of Engineering, Mathematics and Physical Sciences, University of Exeter, Exeter EX4 4QF, UK 8 CSIC, Global Ecology Unit CREAF–CEAB–UAB, Bellaterra, 08193, Catalonia, Spain 9 CREAF, Cerdanyola del Vallès, 08193, Catalonia, Spain 10 Sino-French Institute for Earth System Science, College of Urban and Environmental Sciences, Peking University, Beijing, China 11 College of Life and Environmental Sciences, University of Exeter, Exeter EX4 4RJ, UK 12 School of Natural Resources and the Environment, University of Arizona, Tucson, AZ 85721, USA 13 School of Urban Planning and Design, Shenzhen Graduate School, Peking University, Shenzhen, 518055, China 14 CSIRO Oceans and Atmosphere, Canberra, 2601, Australia 15 Institute of Applied Energy (IAE), Minato, Tokyo 105-0003, Japan 16 Department of Atmospheric Sciences, University of Illinois, Urbana, IL 61801, USA 17 Climate and Environmental Physics, Physics Institute and Oeschger Centre for Climate Change Research, University of Bern, Bern 3012, Switzerland 18 Climate and Global Dynamics Laboratory, National Center for Atmospheric Research, Boulder, CO 80302, USA 19 Max Planck Institute for Meteorology, 20146 Hamburg, Germany 20 Biospheric Sciences Laboratory, NASA Goddard Space Flight Center, Greenbelt, MD 20816, USA Correspondence: Ana Bastos (ana.bastos@lmu.de) Received: 19 March 2019 – Discussion started: 29 March 2019 Revised: 10 August 2019 – Accepted: 2 September 2019 – Published: 7 October 2019 Published by Copernicus Publications on behalf of the European Geosciences Union. 12362 A. Bastos et al.: Seasonal-cycle amplitude change Abstract. Continuous atmospheric CO2 monitoring data in- surements (Dargaville et al., 2002; Forkel et al., 2016; Piao dicate an increase in the amplitude of seasonal CO2 -cycle et al., 2017) or over latitudinal transects (Graven et al., 2013; exchange (SCANBP ) in northern high latitudes. The major Thomas et al., 2016). These studies have shown that LSMs drivers of enhanced SCANBP remain unclear and intensely systematically underestimated SCA trends, but it is not clear debated, with land-use change, CO2 fertilization and warm- whether these biases are due to LSM uncertainties or due to ing being identified as likely contributors. We integrated trends or errors in the ATM (Dargaville et al., 2002). Piao CO2 -flux data from two atmospheric inversions (consistent et al. (2017) addressed these problems by designing system- with atmospheric records) and from 11 state-of-the-art land- atic model experiments to compare observed CO2 concen- surface models (LSMs) to evaluate the relative importance of trations at multiple sites with ATM simulations forced by an individual contributors to trends and drivers of the SCANBP ensemble of NBP from different LSMs and an ocean bio- of CO2 fluxes for 1980–2015. The LSMs generally reproduce geochemistry model. Point Barrow was the only site where the latitudinal increase in SCANBP trends within the inver- nearly all models accurately described the trend in SCA, sions range. Inversions and LSMs attribute SCANBP increase while at other sites, LSMs generally captured the sign of to boreal Asia and Europe due to enhanced vegetation pro- the trend in SCA but either underestimated or overestimated ductivity (in LSMs) and point to contrasting effects of CO2 its magnitude. Piao et al. (2017) further reported that CO2 fertilization (positive) and warming (negative) on SCANBP . fertilization and climate change drove the increase in SCA Our results do not support land-use change as a key contrib- for sites > 50◦ N, but at mid-latitude sites, land use, oceanic utor to the increase in SCANBP . The sensitivity of simulated fluxes, fossil-fuel emissions and trends in atmospheric trans- microbial respiration to temperature in LSMs explained bi- port may have contributed to the SCA trends. ases in SCANBP trends, which suggests that SCANBP could Attributing changes in the seasonal amplitude of atmo- help to constrain model turnover times. spheric CO2 to specific processes requires analysing net sur- face fluxes as a function of changes in gross fluxes (photo- synthesis, respiration and disturbance), which LSMs can pro- vide. However, quantifying a bias in CO2 concentration at a 1 Introduction given site from a bias in fluxes simulated by a land-surface model (LSM) is difficult, since the biases can be affected The increase in the amplitude of seasonal atmospheric CO2 by many other factors such as transport model characteris- concentrations at northern high latitudes is one of the most tics, forcing data used, etc. Atmospheric inversions provide intriguing patterns of change in the global carbon (C) cycle. a consistent framework for assimilating in situ CO2 concen- The seasonal-cycle amplitude (SCA) of atmospheric CO2 in tration observations to estimate net CO2 surface fluxes while the lower troposphere at the high-latitude monitoring site of accounting for errors in the prior fluxes and for some errors Point Barrow, Alaska, has increased by about 50 % since the in the ATM (Peylin et al., 2013). At large spatial scales, the 1960s (Keeling et al., 1996; Dargaville et al., 2002). Increas- trends in SCA can be related to trends in the seasonal ampli- ing SCA has also been registered at other high-latitude sites, tude of CO2 fluxes (i.e. SCA of NBP – SCANBP ). Such an mostly above 50◦ N (Piao et al., 2017), and appears to be approach has been used to analyse trends in net CO2 uptake driven primarily by changes in seasonal growth dynamics of in boreal regions (Welp et al., 2016). terrestrial ecosystems (i.e. net biome productivity – NBP), The spatiotemporal distribution of terrestrial and oceanic but uncertainty remains about the relative contributions from surface fluxes estimated by inversions provides thus direct different continents and mechanisms. insight about the regional patterns of SCANBP that is fully Some studies proposed that the trend in SCA is primar- consistent with the amplitude of CO2 concentrations in all ily driven by increased natural vegetation growth and forest stations of the observational network used and constitutes expansion at high latitudes due to CO2 fertilization and cli- a direct benchmark for SCANBP simulated by LSMs. Here, mate change (Graven et al., 2013; Forkel et al., 2016; Piao we use top-down (inversions; Chevallier et al., 2010; Röden- et al., 2017). Others (Gray et al., 2014; Zeng et al., 2014) beck, 2005) and bottom-up (TRENDYv6 LSMs; Le Quéré suggested that agricultural expansion and intensification re- et al., 2018) estimates of terrestrial CO2 fluxes at northern sulted in increased productivity and thus enhanced the sea- extra-tropical latitudes between 1980 and 2015 to (i) assess sonal exchange in cultivated areas at mid-latitudes. However, the ability of those LSMs to simulate inversion-based trends evidence suggests that crop productivity stagnated after the in SCANBP , (ii) attribute the trends in SCANBP to specific re- 1980s in many regions in the Northern Hemisphere (Grassini gions in the Northern Hemisphere and (iii) attribute the rela- et al., 2013), which is not reflected in SCA trends in recent tive importance of drivers using the ensemble model frame- decades (Yin et al., 2018). work. Studies using land-surface models (LSMs) to attribute Trends in SCANBP from the inversions are based on mul- trends to the suggested processes usually convert simulated tiple in situ measurements and therefore provide a reference fluxes to CO2 concentrations using atmospheric transport (and respective uncertainty) for evaluating the regional at- models (ATMs) and compare the results to in situ mea- tribution by LSMs. Regarding process attribution, LSMs al- Atmos. Chem. Phys., 19, 12361–12375, 2019 www.atmos-chem-phys.net/19/12361/2019/ A. Bastos et al.: Seasonal-cycle amplitude change 12363 low separating the contribution of different drivers through have assimilated observations from 10, 23 and 38 sites since factorial simulations. However, the attribution by LSMs can- 1976, 1985 and 1993, respectively. Surface fluxes (ocean not be easily validated, which is especially problematic given and land) are provided at the latitude–longitude resolution that LSMs underestimate the trends in SCA at the latitudinal of 4◦ × 5◦ of the TM3 atmospheric transport model is used scale (Thomas et al., 2016). Thus, we compare (i) the pro- (Rödenbeck, 2005). In this version, the atmospheric model is cess attribution by LSMs (as e.g. in Thomas et al., 2016; Piao forced by the National Centers for Environmental Prediction et al., 2017), (ii) the statistical attribution based on inversion (NCEP) meteorological fields. fluxes, (iii) the statistical attribution based on LSMs fluxes, CarboScope further provides a sensitivity analysis of the directly comparable to the inversion results, and (iv) the sta- s85 version fluxes to different parameters of the inversion. tistical attribution based on the differences between factorial The sensitivity tests performed are as follows: “oc” – fix- simulations (cross evaluation of i and iii). ing the ocean prior, “eraI” – forcing the inversion with fields Our approach allows us thus to constrain SCANBP trends from ERA-Interim reanalysis instead of NCEP, “loose” and at hemispheric and regional scales from both top-down (in- “tight” – scaling the a priori sigma for the non-seasonal land versions) and bottom-up (LSMs) methods and to evaluate and ocean flux components by 4 (dampening) and 0.25 (am- the process attribution by LSMs using top-down estimates plification), respectively, “fast” – reducing the length of a pri- of SCANBP . ori temporal correlations, and “short” – reducing the length of a priori spatial correlations. The resulting latitudinally in- tegrated SCANBP and respective trends are shown in Fig. S2 2 Data in the Supplement. Since CAMS includes a larger, but time-varying, num- 2.1 Atmospheric inversions ber of multi-year air-sampling sites as they are available, it constrains better spatial patterns, while CarboScope keeps a The inversion of a transport model to infer surface fluxes fixed set of sites covering a given period, using fewer sites from concentration measurements is an ill-posed problem but avoiding artefacts in the time series related to the appear- due to the dispersive nature of transport in the atmosphere ance or disappearance of measurement sites. and to the finite number of available measurements. This ill- posedness can be compensated for by using some prior in- 2.2 Land-surface models formation about the fluxes to be inferred. This prior infor- mation also drives the separation between natural and fossil Land-surface models (LSMs) provide a bottom-up approach fuel emissions in the estimation. In order to illustrate the di- for evaluating terrestrial CO2 fluxes (i.e. NBP) and allow versity of the inversion results, we take the example of two deeper insight into the mechanisms driving changes in C inversion systems that provide results for the study period stocks and fluxes. The TRENDY intercomparison project between 1980 and 2015. We analysed monthly surface CO2 compiles simulations from state-of-the-art LSMs to evalu- fluxes estimated by the inversion systems from the Coper- ate terrestrial energy, water and CO2 exchanges starting from nicus Atmosphere Monitoring Service (CAMS; Chevallier the pre-industrial period (Sitch et al., 2015; Le Quéré et al., et al., 2005, 2010) and from Jena CarboScope (Rödenbeck 2018). We use LSMs from the TRENDYv6 simulations for et al., 2003; Rödenbeck, 2005). The two inversions used here 1860–2015. To identify the contributions of CO2 fertiliza- solve for fluxes on their ATM grid, thus minimizing aggrega- tion, climate, and land-use and land-cover change (LULCC) tion errors for large regions (Kaminski and Heimann, 2001). and management to the observed changes in SCANBP , we use The CAMS version r16v1 (http://atmosphere.copernicus. outputs from three factorial simulations. eu/, last access: 16 November 2017) (Chevallier, 2017) pro- The models in simulation S3 were forced by (i) atmo- vides estimates of ocean and terrestrial fluxes at 1.9◦ lat- spheric CO2 concentrations from ice core data and observa- itude by 3.75◦ longitude resolution. The CAMS inversion tions, (ii) historical climate reanalysis from the CRU-NCEP system assimilates observations from a variable number of v8 (Viovy, 2016; Harris et al., 2014), and (ii) human-induced atmospheric CO2 monitoring sites (119 in total, providing land-cover changes and management from a recent update at least 5 years of measurements) and uses the transport of the land-use harmonization strategy (Hurtt et al., 2011) model from the LMDz general circulation model (LMDz5A) prepared for the next set of historical CMIP6 simulations, nudged to ECMWF-analysed winds. More details can be LUH2v2h (described below). Most models, though, still do found in Chevallier et al. (2010). not represent many of the management processes included The CarboScope v4.1 (available at http://www.bgc-jena. in LUH2v2h. As summarized in Table A1 in Le Quéré et al. mpg.de/CarboScope/?ID=s, last access: 26 March 2018) pro- (2018), four models do not simulate wood harvest and three vides several versions that assimilate a temporally consistent do not simulate cropland harvest. Two models simulate crop set of observations. We used these versions for the study pe- fertilization, tillage and grazing. riod (1980–2015) to test the influence of the number of as- The models in simulation S2 were forced by (i) and (ii) similated sites on the results. The s76, s85 and s93 versions described above with a fixed land-cover map from 1860. www.atmos-chem-phys.net/19/12361/2019/ Atmos. Chem. Phys., 19, 12361–12375, 2019 12364 A. Bastos et al.: Seasonal-cycle amplitude change Simulation S2 estimates “natural” fluxes, and the difference gust 2017) for 1992–2015. Data are provided for different between S2 and S3 outputs corresponds to anthropogenic vegetation types but were aggregated here for four main land- CO2 fluxes from LULCC. The models in simulation S1 were cover classes: forest, shrubland, grassland and cropland. The forced by changing atmospheric CO2 and no climate change average distribution of these classes (forest and shrubland (recycling 1901–1920 values to simulate interannual vari- aggregated for readability) is shown in Fig. 2a. LUH2v2h ability) or LULCC. S1 thus provided changes in the terres- was used for the statistical analysis of inversion and the LSM trial sink due to CO2 fertilization, and the difference between drivers (because it was the data set used to force the models), S1 and S2 indicates the influence of climate change only. ESA-CCI data were used for the analysis of satellite-based However, management practices (e.g. wood harvest), when vegetation data sets, and results were additionally compared simulated, are already included in S1 and S2 for some mod- with LUH2v2h. els. A baseline simulation with none of these effects (S0) was also performed to check for residual variability and trends. 2.4 Satellite-based vegetation data sets We selected only models providing spatially explicit outputs for the four simulations (S0, S1, S2 and S3) at monthly inter- We further evaluated trends in the activity and growth of vals (to evaluate seasonality; Table S1 in the Supplement). vegetation for the different land-cover classes using three We used NBP outputs selected for the period common satellite-based data sets: leaf-area index (LAI), net primary to the inversion data, i.e. 1980–2015. NBP corresponds to production (NPP) and above-ground biomass (AGB) stocks. the simulated net atmosphere–land flux (positive sign for a The LAI data set was calculated from satellite imagery from CO2 sink), i.e. gross primary productivity (GPP) minus total Global Inventory Modeling and Mapping Studies (GIMMS ecosystem respiration (TER), fire emissions, and fluxes from LAI3g) described by Zhu et al. (2015) for 1982–2015. LAI LULCC and management (e.g. deforestation, agricultural data were provided in two time steps per month on a regular and wood harvest, and shifting cultivation). All model out- latitude–longitude grid of 1/12◦ (subsequently aggregated to puts were resampled to a common regular latitude–longitude 0.5◦ ). Smith et al. (2016) used the MODIS NPP algorithm grid of 1◦ ×1◦ . and data for LAI and the fraction of photosynthetically ac- tive radiation from GIMMS to produce a 30-year global NPP 2.3 Land cover and management data set, provided at monthly timescales for 1982–2011 at a latitude–longitude resolution of 1◦ ×1◦ . The data are avail- 2.3.1 LUH2v2h able at the NTSG data portal (https://wkolby.org/data-code/, last access: 4 August 2017). AGB stocks can be derived from The LUH2v2h (Hurtt et al., 2011; available at http://luh.umd. estimates of vegetation optical depth derived from passive- edu/, last access: 7 July 2019) provides historical states and microwave satellite measurements. Liu et al. (2015) pro- transitions of land use and management in a regular latitude– duced a 20-year data set of AGB stocks for 1993–2012 based longitude grid of 0.25◦ ×0.25◦ , covering the period 850– on measurements from a series of passive-microwave sen- 2015 at annual time intervals. Land-use states distinguish be- sors. The data set is provided at a latitude–longitude resolu- tween primary and secondary natural vegetation (and forest tion of 0.25◦ ×0.25◦ in annual time intervals and is available and non-forest subtypes), managed pastures and rangelands, at http://www.wenfo.org/wald/global-biomass/ (last access: and multiple crop functional types. The updated data set in- 13 February 2018). We tracked changes in LAI, NPP and cludes several new layers of agricultural management, such AGB stocks for different land-cover types over time by se- as irrigation, nitrogen fertilization and biofuel management, lecting periods of at least 20 years common to ESA-CCI LC and spatially explicit information about wood harvest con- and the vegetation data sets (1992–2012 for LAI, 1992–2011 strained by LANDSAT data. Each LSM, however, may not for NPP and 1993–2012 for AGB stocks). Vegetation vari- simulate all the processes introduced in LUH2v2h, so the S3 ables were then aggregated for the four land-cover types at results from each simulation might not be directly compara- each time interval to account for land-cover changes. ble. 3 Methods 2.3.2 ESA-CCI land cover 3.1 Trends in seasonal-cycle amplitude (SCANBP ) Land-cover information in LUH2v2h is combined with par- tial information on land use (e.g. rangeland in LUH2v2h The seasonal amplitude of CO2 concentration is modulated can be either grassland or shrubland with low grazing dis- by higher ecosystem CO2 uptake during the growing season turbance). We therefore compared this information to an- and increased emissions during the release period (TER) and nual land-cover maps at a latitude–longitude resolution thus controlled by the seasonal amplitude of NBP. We cal- of 0.5◦ ×0.5◦ based on the 300 m satellite-based land- culated SCANBP as the difference between peak uptake and cover data sets from ESA-CCI land cover (LC) (https:// trough for each year at the pixel scale, shown in Fig. 1a. How- www.esa-landcover-cci.org/?q=node/175, last access: 1 Au- ever, since inversion fluxes have large uncertainty at the pixel Atmos. Chem. Phys., 19, 12361–12375, 2019 www.atmos-chem-phys.net/19/12361/2019/ A. Bastos et al.: Seasonal-cycle amplitude change 12365 level, we focused our analysis on SCANBP trends estimated with climate should be consistent for the fluxes estimated from aggregated NBP over latitudinal bands or TransCom3 by the difference between S2 and S1 (effects of climate). regions (Baker et al., 2006). Because we do not impose the The GLM fit to the partial fluxes for the effects of LULCC timing of peak and trough, changes in SCANBP can be af- (S3 − S2), climate (S2 − S1) and CO2 fertilization (S1 − S0) fected by the relative phase changes of GPP versus TER. are shown in Fig. S6. The trend in SCANBP was calculated by a least-squares linear fit of annual values for 1980–2015, and confidence in- tervals were calculated based on Student’s t distribution. We 4 Results tested the robustness of estimated trends of inversions and LSMs for shorter periods by removing the first and last 1– 4.1 Large-scale patterns 10 years and trends of interannual variability by randomly 4.1.1 Top-down estimates removing 5 and 10 years of data 104 times. The significance of these trends was calculated using a Mann–Kendall test. We Both inversions estimate increasingly positive trends in also compared different versions of CarboScope to evaluate SCANBP with increasing latitude, even though CAMS shows the influence of the assimilated network size on the SCANBP heterogeneous patterns in North America, with strong de- trends (Fig. S1). We further calculated the trends for each of creasing trends for mid-latitudes (Figs. 1a and S1). Both in- the sensitivity tests from CarboScope s85. versions agree on significant positive SCANBP trends north of 40◦ N (defined here as band L>40 N ) and non-significant 3.2 Process attribution trends for 25–40◦ N (band L25−40 N ; Fig. 1b). In the L>40 N band, CAMS and CarboScope s76 v4.1 estimate an SCANBP The three TRENDY experiments allow evaluating separately increase of 17.3 ± 4.5 and 13.3 ± 3.3 Tg C yr−2 , respectively. the effects of CO2 fertilization, climate change, and LULCC The uncertainties given for SCANBP trends represent here the in the models. The differences between S1 and S2 and be- uncertainty of the linear fit due to the year-to-year SCANBP tween S2 and S3, however, could not isolate specific pro- variability (Methods). The difference between the CAMS cesses that may have contributed to the trend (e.g. cropland and CarboScope inversions reflects part of the uncertainty in expansion versus afforestation or precipitation versus tem- inversions due to their different choices in the ATM (includ- perature). Furthermore, the LSMs may miss or simulate cer- ing different atmospheric forcing and spatial resolution), the tain processes poorly that could influence SCANBP . There- set of assimilated CO2 data, the prior fluxes, and the a priori fore, the attribution of drivers by the models is uncertain and spatial and temporal correlation scales and is comparable to should be cross-evaluated. Because inversions do not allow the uncertainty of the linear fit due to interannual variability. such partitioning between processes, a possible solution is This finding is corroborated by two further analyses of in- to compare statistical attribution to drivers in inversions and version uncertainties: LSMs. We therefore compared the sensitivity of SCANBP esti- 1. While both inversions assimilate atmospheric CO2 mea- mated by the inversions and the LSMs by fitting a gen- surements from Point Barrow, CAMS increasingly as- eral linear model (GLM) using the iteratively reweighted similates many other sites in the NH as they become least-squares method to eliminate the influence of outliers available, helping to better constrain the CO2 fluxes (Gill, 2000; Green, 1984). We tested the following vari- in mid-latitudes to high latitudes with time. Assimilat- ables (after unity-based normalization) as predictors: fertil- ing a non-stationary network of stations, however, pos- ization, irrigation, wood harvest, growing-season precipita- sibly leads to spurious additional trends in SCANBP . tion, growing-season temperature, atmospheric CO2 concen- To test this, we use different runs provided by Carbo- tration, and change in the extent of cropland and forest. These Scope using more sites (but still fixed in number for variables were taken from the corresponding data sets used each run; Fig. S2) for more recent periods. The re- to force TRENDYv6 models. All possible combinations of n sults from CarboScope version s85 v4.1 (1985–2015) predictors (n = 1, 2, . . ., 7) were tested, and for each value of are generally consistent with CAMS, but version s93 n, the “best” model (according to Akaike’s information cri- v4.1 (1993–2015) estimates much stronger SCANBP terion) was chosen separately for each data set. Above n = 4 trends (Table S1). A higher SCANBP trend in the period no model showed improved fit compared to the models with 1993–2015 is reported by both CAMS and CarboScope, fewer predictors. The coefficients from the GLM fit for each which estimate very similar trends in L>40 N (19.5 and data set are shown in Fig. S5. 19.2 Tg C yr−2 , respectively). We further tested the robustness of the statistical relation- ships by fitting the GLM to the differences between each 2. CarboScope provides a set of sensitivity runs for TRENDYv6 experiment. The significant predictors in the s85 v4.1, varying some of the inversion’s parameters GLM fit to the LSMs in S3 should be detected in the cor- (Fig. S3). Changes in the meteorological fields driv- responding factorial simulations, e.g. predictors associated ing the transport model and the prior ocean fluxes have www.atmos-chem-phys.net/19/12361/2019/ Atmos. Chem. Phys., 19, 12361–12375, 2019 12366 A. Bastos et al.: Seasonal-cycle amplitude change Figure 1. Variability in seasonal-cycle amplitude and trends from the inversions and LSMs. (a) Geographical distribution of SCANBP trends from the inversions (CAMS and CarboScope) and the multi-model ensemble mean (MMEM) from TRENDYv6 simulation S3 (all forcings). Both inversions estimated predominantly positive trends in SCANBP > 40◦ N (Fig. S1), so we defined two latitudinal bands, L>40 N and L25−40 N , for flux aggregation. (b, c) Aggregated SCANBP time series estimated by the inversions (CAMS in black and CarboScope s76 in grey) and S3 MMEM (red) for L>40 N and L25−40 N , respectively. The dashed lines indicate the linear fits used to calculate the slopes of the trends (corresponding colours), and the slopes and confidence intervals (95 %) are provided. the largest effect on the SCANBP trends, giving L>40 N variability and less so by the amount of information available trends of 8.6 ± 4.9 Tg C yr−2 (ERA-Interim instead of through the atmospheric data or by inversion settings. NCEP) and 13.9 ± 5.6 Tg C yr−2 (fixed ocean), respec- tively, both well within the uncertainty range (interan- 4.1.2 Bottom-up estimates nual variability affecting linear fit to SCANBP trend) es- timated by the standard CarboScope s85 v4.1 (11.7 ± The large-scale patterns of SCANBP trends from the LSM 5.0 Tg C yr−2 ). multi-model ensemble mean (MMEM) of simulation S3 (all forcings) are consistent with CarboScope inversion (Fig. 1a). In summary, the ability of inversions to quantify the SCANBP The MMEM estimates are within the range of the inver- trend is mostly limited by the intrinsic year-to-year SCANBP sions for most latitudes (Fig. S1) but always at the lower Atmos. Chem. Phys., 19, 12361–12375, 2019 www.atmos-chem-phys.net/19/12361/2019/ A. Bastos et al.: Seasonal-cycle amplitude change 12367 Figure 2. Regional distribution of the dominant land-cover types and SCANBP trends. (a) Land-cover map averaged over the study period for the three main land-cover classes (forest–shrubland, grassland and cropland) based on ESA-CCI annual land-cover data (1992–2015 average). (b) The continental regions correspond to the regions defined by Baker et al. (2006) and are delimited by bold lines: boreal and temperate North America (BorNA and TempNA), Europe (Eur), and boreal and temperate Eurasia (BorEA and TempEA). (c) Comparison of the SCANBP trends from the inversions to the trends estimated by the LSM experiments: S3, S2 (no LULCC) and S1 (no LULCC and no climate change). The bars for the inversions and LSMs indicate the average trend over each latitudinal band. The error bars for the inversions indicate the 95 % confidence levels for the trend values, and the vertical lines for the LSMs indicate interquartile ranges of the multi-model ensemble. The 95 % confidence interval for the MMEM was also calculated (see Methods). end of SCANBP trends reported by inversions. Consistent simulation and parameterizations from v4 to v6 (Table 4 in with inversions, LSMs report a significant trend in L>40 N Le Quéré et al., 2018). and a very weak (non-significant) trend in SCANBP in In summary, we showed that the TRENDYv6 ensemble L25−40 N (Fig. 1b). The overall MMEM trend in L>40 N is mean SCANBP trend captures the positive trends in the high significantly lower than in inversions (9.5 ± 3.4 Tg C yr−2 , latitudes and the lack of trend in the mid-latitudes given by i.e. 55 %–71 % of inversions’ estimates. The agreement be- inversions, and underestimates the magnitude of the high lat- tween LSMs and inversions also varies depending on the pe- itudes SCANBP trends by 9 %–45 %, depending on the inver- riod and set of inversions considered (LSMs capture 65 %– sion considered and period analysed. 91 % of inversion trends in 1985–2015 and 74 %–75 % in 1993–2015; Table S1). The MMEM estimate for 1985–2015 4.2 Regional attribution (10.6 ± 4.5 Tg C yr−2 ) is, in fact, even higher than the Carbo- Scope inversion with different meteorological fields (8.6 ± The comparison of SCANBP trends in large latitudinal bands 4.9 Tg C yr−2 ). These results indicate that, despite a general may be useful in diagnosing general patterns but is less useful underestimation of SCANBP trend in L>40 N during 1980– in diagnosing drivers of trends (e.g. climate and agriculture), 2015 when compared to top-down estimates, the LSMs simu- since ecosystem composition, land management and climate late the main spatiotemporal patterns in SCANBP trends con- effects are not necessarily separated along a latitudinal gra- sistent with inversions estimates, especially when accounting dient. However, the comparison of inversions and models at for the uncertainty in the latter. the pixel scale is also not advisable because the sparse at- To understand if recent improvements to the set of LSMs mospheric network does not allow constraining the fluxes at and their forcing in TRENDYv6 may have improved their this scale. We thus compared inversions and LSMs for the performance in reproducing the SCANBP trend, we compared SCANBP trends over five sub-continental scale regions: bo- SCANBP trends from the previous intercomparison round real and temperate Eurasian and North American regions and (TRENDYv4). The MMEM from v6 estimates an SCANBP Europe (“TransCom3” regions; Fig. 2). We then use LSMs trend in L>40 N that is 43 % higher than in than v4 (MMEM for attributing SCANBP trends to different drivers using their shown in Table S1 but evaluated for individual models). The factorial simulations (Methods). specific reasons for improvement are hard to identify because Inversions and LSMs consistently attribute the increase of multiple model-dependent changes in the forcing, process in SCANBP mainly to boreal Eurasia, both in area-specific (Fig. 1a) and integrated values (Fig. 2b, 5.3–7.1 Tg C yr−2 for www.atmos-chem-phys.net/19/12361/2019/ Atmos. Chem. Phys., 19, 12361–12375, 2019 12368 A. Bastos et al.: Seasonal-cycle amplitude change inversions and 4.6 Tg C yr−2 for MMEM, respectively), and latitudinal bands. The coefficients of the GLM fit are shown to Europe (1.9–3.7 and 2.3 Tg C yr−2 ). The LSMs ascribed in Fig. S5. the trends in boreal Eurasia approximately equally to cli- The GLMs provide a better fit the trend of SCANBP in mate change and CO2 fertilization (S1 and S2), with LULCC L>40 N (explaining 57 %–74 % of the variance; Table S2) having a slight negative (i.e. decreasing) effect (compare S2 than for L25−40 N (8 %–49 % only). The GLM fit to inver- and S3), consistent with the results by Piao et al. (2017). sions and to the MMEM identified CO2 fertilization as the In Europe, LSMs indicate negative contributions from both most important factor explaining (statistically) the SCANBP climate and LULCC. The negative effect of climate may be trends in both latitudinal bands, consistent with S1 (Figs. 1, linked to increasingly drier conditions in this region (Greve S4, S5 and S6), although the CO2 -fertilization effect was et al., 2014) and to strong heatwaves in Europe in the early weaker for the GLM fit to LSMs than for inversions in re- 21th century (Seneviratne et al., 2012). The negative con- gion L>40 N . The statistical models for inversions and LSMs tribution of LULCC indicated by LSMs in Europe does not agreed on a significant negative contribution of warming in support the idea that agricultural intensification or expan- both latitudinal bands but a stronger contribution in L25−40 N . sion drove an increase in SCANBP and is discussed further GLM models fitted to LSMs and CarboScope also point to on. In temperate Eurasia, inversions disagree on the sign changes in forest area contributing to increased SCANBP , and of SCANBP trends, and LSMs indicate weak positive trends changes in crop area have a negative effect in SCANBP from dominated by the CO2 -fertilization effect. In boreal North LSMs. In L25−40 N , the GLM fit to LSMs further points to America, LSMs estimate SCANBP trends very close to Car- a small negative contribution of wood harvest to SCANBP boScope estimates, mainly attributed to CO2 followed by cli- trends, and, for CAMS, negative effects of irrigation and mate, whereas CAMS points to a trend close to zero because fertilization are also significant. The statistical attribution of cancelling regional trends with opposing sign (Fig. 1a). of SCANBP trends in LSMs is generally consistent with CAMS and CarboScope point to increasing SCANBP in tem- the factorial simulations but is mostly clearer for the CO2 - perate North America (1.4–1.6 Tg C yr−2 ), but the LSMs do fertilization effect than for the other drivers. In the difference not indicate any significant change (simulation S3). CAMS between factorial simulations (Fig. S6), some drivers appear (which uses prior information with smaller a priori uncertain- to have strong interactive effects, e.g. the effect of CO2 is ties than CarboScope together with a denser network) shows significantly negative for S3 − S2 (LULCC). This could be sharper regional differences than CarboScope, which illus- explained by higher emissions from LULCC under higher trates that there are still substantial differences in the inver- CO2 concentrations from the loss of additional sink capacity sion at the scale of continental regions regarding SCANBP (Pongratz et al., 2014). The key role of CO2 fertilization in trends. the observed changes is in line with Piao et al. (2017), but our Aggregated over the two latitudinal bands (Fig. 2c), the results challenge some of the previously proposed hypothe- MMEM indicates a dominant positive effect (increasing ses to account for the increase in seasonal CO2 exchange, as SCANBP ) of CO2 fertilization both in L25−40 N and L>40 N . addressed below. In L25−40 N , the CO2 effect is offset by other factors: S1 dif- fers significantly from S2 and S3, which have lower trends of SCANBP . In L>40 N , the MMEM points to a positive ef- 5 Discussion fect of climate change in SCANBP trends, thus adding to the CO2 effect. The MMEM suggest a negligible contribution of 5.1 Confronting hypotheses LULCC to the SCANBP trend in both latitudinal bands. The relative contributions of LULCC, climate and CO2 , however, 5.1.1 Contribution of LULCC differ between LSMs (Fig. S4). Most models nevertheless agree on non-significant SCANBP trends in L25−40 N as well Agricultural intensification and expansion occurred mainly as on the predominant role of CO2 fertilization and a non- in latitudes below 45◦ N (Gray et al., 2014), and inversions significant contribution of LULCC to the trends in SCANBP and LSMs reported instead a peak in the amplitude of land- in L>40 N . Interestingly, models including carbon–nitrogen surface CO2 exchange for latitudes above 45◦ N (Figs. 1 and interactions had the weakest SCANBP trends (CABLE, ISAM S1). Furthermore, our regional attribution identifies Eurasia and LPX-Bern), with the exception of CLM4.5, but we can- as the region contributing most to increasing SCANBP ; this not draw conclusions from a small set of carbon–nitrogen region is dominated by natural ecosystems (Fig. S3) and has models. experienced very little land use change (Verburg et al., 2015) over the previous decades. Additionally, factorial LSM sim- ulations indicate a negligible contribution of LULCC and 4.3 Driving processes management to SCANBP trends not only at the latitudinal- band scale but also regionally (Figs. 1c and 2). Figure 3 shows the relative contributions of the predictors This, though, could not in itself falsify the hypothesis that (weighted by their trends) found to SCANBP trends in both agricultural intensification is a key driver of SCANBP trends Atmos. Chem. Phys., 19, 12361–12375, 2019 www.atmos-chem-phys.net/19/12361/2019/ A. Bastos et al.: Seasonal-cycle amplitude change 12369 Figure 3. Statistical attribution of drivers of SCANBP estimated by the inversions and LSMs. The main drivers of SCANBP are presented for (a) L>40 N and (b) L25−40 N and are calculated as the product of the coefficients of a general linear model fit on SCANBP using a number of predictors (normalized) and their corresponding trends. Fertilization, irrigation, wood harvest, growing-season precipitation, growing-season temperature and atmospheric CO2 concentration were tested as predictors, and the best fit was chosen for each data set: CAMS (dark grey), CarboScope s76 (light grey) and the MMEM (red). The bars indicate the contribution of each predictor to the trend in SCANBP , error bars indicate the corresponding 95 % confidence intervals and the symbols indicate significant GLM fits (two asterisks, one asterisk and crosses indicate p < 0.01, p < 0.05 and p < 0.1, respectively). because most LSMs still do not include processes that could to SCANBP . Our results are consistent with those by Smith intensify cropland NPP over time, such as better cultivars, et al. (2014) that show that NPP generally decreased follow- fertilization and irrigation. Still, management practices are ing conversion from natural ecosystems to cropland, except neither a significant predictor for GLM fitted to LSMs nor for in areas of highly intensive agriculture, such as the midwest- inversions, with the exception of CAMS. CarboScope further ern USA. Increasing crop productivity (intensification) could identifies a negative effect of cropland expansion to SCANBP partly explain increasing SCANBP . However, satellite-based in L>40 N rather than a positive one, which partly challenges data for LAI (Zhu et al., 2016), NPP (Smith et al., 2016) the contribution of cropland expansion (Gray et al., 2014) and above-ground biomass (AGB) carbon stocks (Liu et al., www.atmos-chem-phys.net/19/12361/2019/ Atmos. Chem. Phys., 19, 12361–12375, 2019 12370 A. Bastos et al.: Seasonal-cycle amplitude change 2015) for different land-cover classes from ESA-CCI LC between SCA and T has transitioned from positive in the (Fig. S7) indicate that the increase in crop productivity ac- early 1980s to negative in recent decades, reconciling the re- counted for only a small fraction of the hemispheric trends sults by Keeling et al. (1996) and Schneising et al. (2014). in ecosystem productivity, consistent with the crop produc- In Fig. S9 we present a conceptual scheme of the impacts tivity stagnation in Europe and Asia identified by Grassini of warming in SCANBP through its component fluxes. Gen- et al. (2013). We also compared the trends in remote-sensing erally, warming in high latitudes has been associated with variables for land-cover classes from LUH2v2h, with similar a longer growing season and increased GPP (Piao et al., results. 2008), which would contribute to increased SCANBP through Previous studies suggesting a large role of the green rev- increased productivity during the “uptake period” and in- olution in SCA trends focused on a longer period, starting creased decomposition (due to more litter) during the “re- in the 1960s. The acceleration of SCANBP reported by in- lease period”. However, a weakening of this relationship has versions and LSMs (Table S1) concurrent with crop produc- been reported (Piao et al., 2014; Peñuelas et al., 2017). Other tivity stagnation indicates that since the 1980s agriculture processes can, though, contribute to the negative relationship intensification is not likely to be the main driver of the in- between SCANBP and T reported here and in other stud- crease in SCA. Even in the intensive agricultural areas in the ies. The empirical negative relationship between trends in midwestern USA, CAMS estimates contrasting negative and SCANBP and warming at the higher latitudes may be due positive trends (Figs. 1a, S8). Eddy-covariance flux measure- to (i) a stronger effect of T on total ecosystem respiration ments (only for 7–13 years) in the areas of intensive agricul- (TER) than on GPP during the uptake period, (ii) a negative ture in the USA show a weak relationship between trends in response of ecosystem productivity to warming during the NBP and trends in SCANBP , showing mostly non-significant uptake period or (iii) indirect negative effects of T on de- trends in SCANBP (Fig. S8). composition during the release period. Evidence nevertheless supports negative effects of warm- 5.1.2 Contribution of warming ing on SCA trends. Temperature increase in recent decades has been associated with widespread reduction in the extent We found that warming during the growing season had a neg- and depth of snow cover (Kunkel et al., 2016) and in the num- ative effect on SCANBP trends in both latitudinal bands, al- ber of days with snow cover (Callaghan et al., 2011). Snow though this effect is uncertain for LSMs in L>40 N . Annual has an insulating effect, so snow-covered soil during winter temperature used in the statistical models was also negatively can be kept at relatively constant temperatures, several de- correlated with SCANBP , but the correlation was only signif- grees above the air temperature (> 10 ◦ C), which promotes icant for CAMS. respiration of soil C (Nobrega and Grogan, 2007). Soils be- The negative relationship with growing-season tem- come subject to more fluctuations in temperature, and be- perature (T ) at the mid-latitudes may be explained by come colder, as the snow cover recedes or becomes thinner. warmer temperature increasing atmospheric demand for wa- Yu et al. (2016) reported that respiration suppression due to ter (Novick et al., 2016) and inducing soil-moisture deficits a reduction in snow cover in winter may account for as much in water-limited regions in summer (Seneviratne et al., 2010) as 25 % of the increase in the annual CO2 sink of north- or increased fire risk (Peñuelas et al., 2017) that reduces the ern forests. A decrease in respiration in response to warm- summer minimum of SCANBP . This negative effect of tem- ing during the release period could thus decrease SCANBP , perature can explain the negative contribution of climate to but the effect of growing-season temperature was stronger in the simulated SCANBP trends in L25−40 N given by the facto- our study. The expansion of vegetation in Arctic tundra, par- rial simulations (Fig. 2c). ticularly shrubland, has been linked to warming trends but The negative statistical relationship found between the also depends on soil-moisture and permafrost conditions (El- trend of SCANBP and T in L>40 N challenges the assumption mendorf et al., 2012). Many regions of dry tundra and low that a warming-related increase in plant productivity in high arctic shrubland (Walker et al., 2005) experience summer latitudes necessarily increases the seasonal CO2 exchange drought or soil-moisture limitations even though northern (Keeling et al., 1996; Graven et al., 2013; Forkel et al., 2016). regions are usually considered to be energy-limited (Greve Although the MMEM shows a small positive contribution of et al., 2014). Indeed, Myers-Smith et al. (2015) found a climate in L>40 N , LSMs diverge on the contribution of cli- strong soil-moisture limitation of the (positive) sensitivity mate in this latitudinal band (Fig. S4). Moreover, the factorial of shrub growth to temperature in summer, possibly asso- simulations in Fig. 2c allow evaluating the impact of changes ciated with the limitation of growth due to drought and/or in all climate variables (e.g. also rainfall and radiation) in with reduced growth and dieback due to standing water dur- addition to temperature. A negative relationship between T ing thawing. CAMS indicates a decrease in SCANBP in east- and SCA has, though, also been reported by Schneising et al. ern regions in boreal North America (Fig. 1a), where Myers- (2014) for interannual changes in the SCA of total column Smith et al. (2015) reported negative sensitivity of shrub- CO2 for 2004–2010. Yin et al. (2018) have further shown growth to temperature. The coarse network and large cor- that, at latitudes between 60◦ N and 80◦ N, the relationship relation lengths used by CarboScope do not allow such re- Atmos. Chem. Phys., 19, 12361–12375, 2019 www.atmos-chem-phys.net/19/12361/2019/ A. Bastos et al.: Seasonal-cycle amplitude change 12371 Figure 4. Emerging relationships between LSM sensitivities to climate and CO2 and their SCANBP trends. The SCANBP trend for L>40 N estimated by each inversion (grey intervals) and corresponding responses of SCANBP to (a) T and (b) CO2 (as calculated in Fig. 3 but considering the scores of the regression only; shown in Fig. S5) are compared to the results from individual models (simulation S3; coloured markers). In (c) the SCANBP trend for L>40 N estimated by individual models is compared with the simulated sensitivity of TER to T . The shaded areas indicate the inversion ranges, and the distribution of the grey lines shows uncertainty in the relationship between each pair of variables. gional contrasts to be resolved. Most process-based models should increase non-linearly (Piao et al., 2014, 2017; Peñue- lack a detailed representation of processes described above – las et al., 2017), suggesting a progressively dominant (nega- e.g. a realistic effect of snow insulation on soil temperatures, tive) influence of warming on SCANBP . The degree of such soil freezing and thawing (Koven et al., 2009; Peng et al., an offset would likely depend on the thresholds of soil tem- 2016; Guimberteau et al., 2017) – potentially overestimating perature and water limitation that are complex and thus dif- the net sink response to temperature changes (Myers-Smith ficult to assess and require process-based modelling. Our re- et al., 2015). Moreover, soil-moisture limitation due to tem- sults imply that future constraints of productivity based only perature increase could also contribute to decreased TER by on the CO2 effect (as in Wenzel et al., 2016) may overesti- limiting microbial activity, which is currently not simulated mate future GPP. in most LSMs. This may in turn explain why LSMs under- We evaluated whether the differences between the ob- estimate the negative effect of temperature in SCANBP in the served SCANBP trends (significant only in L>40 N ) and those high latitudes compared to CAMS (Figs. 3 and S4). simulated by the LSMs could be associated with the mod- elled sensitivities to atmospheric CO2 concentration (CO2 ) 5.2 Evaluating model biases and growing-season temperature (T ) in L>40 N (Fig. 4). In Fig. 4, only models with too small a sensitivity of SCANBP Wenzel et al. (2016) proposed that the observed sensitivity of to T produce a realistic trend of SCANBP . In contrast, the SCANBP to CO2 was an emergent constraint on future terres- models indicating sensitivities to T and CO2 more similar to trial photosynthesis, but their study focused on simulations those estimated by the inversions tend to underestimate the by an earth-system model that excluded the effects of cli- trend in SCANBP . mate change (i.e. the radiative feedback of CO2 to climate Why are the LSM sensitivities of SCANBP to T positively was not considered). Our results are consistent with a strong correlated with their long-term SCANBP trend (Fig. 4) even increase in the peak uptake due to the effect of CO2 fertil- though CO2 is a stronger driver of the simulated SCANBP ization driven by GPP as proposed by Wenzel et al. (2016). trend (Fig. 3)? We found a clear relationship between the The negative effect of temperature in our study (Fig. 3), model bias in the trend of SCANBP and the sensitivity to CO2 although weaker than the positive effect of absolute CO2 fertilization in S3 (in line with Wenzel et al., 2016), but we concentration, suggested that warming partly cancelled out also found a compensatory effect, where models that overes- the increase in SCANBP expected from the effect of fertil- timate the sensitivity of SCANBP to T tend to underestimate ization alone. We propose that other processes partly con- the sensitivity to CO2 and vice versa. LSMs tend to overes- trol SCANBP trends linked to reduced decomposition under timate the sensitivity of SCANBP to T and underestimate the lower snow cover (Yu et al., 2016) or to emerging limitations sensitivity to CO2 , compared to the observation-based con- to growth in response to water limitation (Elmendorf et al., straints from inversions. LSMs often compensate too strong 2012; Myers-Smith et al., 2015). Additionally, while the sen- (or too weak) a simulated water stress or temperature sen- sitivity of productivity to the CO2 -fertilization effect is ex- sitivity by adjusting photosynthesis parameters (that control pected to decrease, the control of respiration by temperature www.atmos-chem-phys.net/19/12361/2019/ Atmos. Chem. Phys., 19, 12361–12375, 2019 12372 A. Bastos et al.: Seasonal-cycle amplitude change CO2 fertilization) during model optimization to match the Data availability. CAMS v16r1 can be freely available observed net terrestrial sink. This compensatory effect has from the COPERNICUS repository: https://apps.ecmwf.int/ previously been reported by Huntzinger et al. (2012) for the datasets/data/cams-ghg-inversions/ (last access: 16 Novem- mean terrestrial sink; we find that it could also affect the ber 2017) (Chevallier, 2017). CarboScope s85 v4.1 results trends in seasonal CO2 exchange. are available at http://www.bgc-jena.mpg.de/CarboScope/ s/s85_v4.1.html (last access: 26 March 2018) (Röden- We argue that the trend of SCANBP can differ between beck, 2017a, b, c) (https://doi.org/10.17871/CarboScope- models due to (a) differences in their NPP response to T and s85_v4.1, https://doi.org/10.17871/CarboScope-s76_v4.1, and CO2 , (b) differences in turnover times of short-lived C pools https://doi.org/10.17871/CarboScope-s93_v4.1). LUH2v2h by which increased NPP is coupled to increased winter de- data are freely available at http://gsweb1vh2.umd.edu/ composition, and (c) phase shifts between GPP and ecosys- LUH2/LUH2_v2h/states.nc (last access: 7 July 2019). tem respiration. These phase shifts may be associated with ESA-CCI LC maps can be downloaded after registration errors in the phase and amplitude of simulated ecosystem at http://maps.elie.ucl.ac.be/CCI/viewer/download.php (last respiration, arising from factors such as (i) representing soil access: 1 August 2017). NPP from Smith et al. (2016) carbon stocks as pools with discrete turnover times and asso- can be downloaded from https://drive.google.com/open?id= ciated effective soil depths (Koven et al., 2009) and (ii) ne- 1m-srVbFoO4IIFGWd__jIoTW7cQEx6890 (last access: 4 Au- glect of seasonal acclimation effects on autotrophic and het- gust 2017). TRENDYv6 DGVM outputs are available on request to Stephen Sitch (s.a.sitch@exeter.ac.uk) and Pierre Friedlingstein erotrophic respiration. The sensitivities of NPP to CO2 and (p.friedlingstein@exeter.ac.uk). T between models are strongly and consistently correlated with the compensatory effect of the model parameterizations (Fig. S10), but we find no clear relationship between the bi- Supplement. The supplement related to this article is available on- ases of the modelled SCANBP trend and the sensitivity of line at: https://doi.org/10.5194/acp-19-12361-2019-supplement. NPP to T (Fig. S11), suggesting a key role of respiration. In- deed, the models with SCANBP trends closer to observations tend to be associated with a lower sensitivity of ecosystem Author contributions. AB and PC designed the study, conducted respiration to growing-season temperature (Fig. S4c). Too the analysis and wrote the paper. APB, FC, CR, FM, MFM, JP, SLP, large a turnover of short-lived pools in a model should pro- WKS, XW, YY and ZZ contributed with expert knowledge during duce too small an increase in the SCANBP amplitude (i.e. in- the development of the study. SS and PF coordinated the TRENDY creased respiration during the uptake period followed by too simulations and maintained the TRENDYv6 data. FC and CR de- little during the release period) for a given sensitivity of NPP veloped the atmospheric inversion data sets and contributed to the to CO2 or climate. A recent study by Jeong et al. (2018) has analysis of inversions. VH, EK, AKJ, SL, DL, JEMSN, PP, BP and reported that ecosystem carbon-cycle models (not used in DZ performed the TRENDYv6 simulations. All authors contributed to the writing of the paper. this study) underestimated changes in carbon residence times in northern Alaska. The evaluation of the effect of model turnover times in SCANBP requires a deeper analysis of trans- Competing interests. The authors declare that they have no conflict fers between litter and soil organic carbon pools and can be of interest. verifiable in future simulations. 6 Conclusions Special issue statement. This article is part of the special issue “The 10th International Carbon Dioxide Conference (ICDC10) and Based on our assessment of atmospheric observations and the 19th WMO/IAEA Meeting on Carbon Dioxide, other Green- house Gases and Related Measurement Techniques (GGMT-2017) the most advanced land-surface model simulations, the most (AMT/ACP/BG/CP/ESD inter-journal SI)”. It is a result of the 10th likely explanation of the seasonal cycle of atmospheric CO2 International Carbon Dioxide Conference, Interlaken, Switzerland, at high latitudes is the CO2 fertilization of photosynthesis 21–25 August 2017. in unmanaged high-latitude ecosystems, especially in the Eurasian boreal forests. Our study further points to key pro- cesses that need to be developed to better simulate NBP re- Acknowledgements. This work was partly supported by the sponses to changing climate, especially to Arctic warming, European Space Agency Climate Change Initiative (ESA- in particular productivity limitations and the decomposition CCI) RECCAP-2 project (ESRIN/ 4000123002/18/I-NB). Mar- terms. Our results indicated that the signal of the SCANBP cos Fernández-Martinez is a postdoctoral fellow of the Research trend contains valuable information for the turnover times of Foundation – Flanders (FWO). Vanessa Haverd acknowledges sup- short-term pools, which await further investigation. port from the Earth Systems and Climate Change Hub, funded by the Australian government’s National Environmental Science Pro- gram. Atmos. Chem. Phys., 19, 12361–12375, 2019 www.atmos-chem-phys.net/19/12361/2019/ A. Bastos et al.: Seasonal-cycle amplitude change 12373 Financial support. This research has been supported by the Euro- Gill, J.: Generalized linear models: a unified approach, vol. 134, pean Space Agency Climate Change Initiative (RECCAP-2 project Sage Publications, Thousand Oaks, 91320 California, USA, (grant no. 4000123002/18/I-NB)). 2000. Grassini, P., Eskridge, K. M., and Cassman, K. G.: Distinguishing between yield advances and yield plateaus in historical crop pro- Review statement. This paper was edited by Eliza Harris and re- duction trends, Nat. Commun., 4, 2918, 2013. viewed by two anonymous referees. Graven, H. D., Keeling, R. F., Piper, S. C., Patra, P. K., Stephens, B. B., Wofsy, S. C., Welp, L. R., Sweeney, C., Tans, P. P., Kelley, J. J., Daube, B. C., Kort, E. A., Santoni, G. W., and Bent, J. D.: Enhanced Seasonal Exchange of CO2 by Northern Ecosystems Since 1960, Science, 341, 1085–1089, 2013. References Gray, J. M., Frolking, S., Kort, E. A., Ray, D. K., Kucharik, C. J., Ramankutty, N., and Friedl, M. A.: Direct human influence on at- Baker, D. F., Law, R. M., Gurney, K. R., Rayner, P., Peylin, P., mospheric CO2 seasonality from increased cropland productiv- Denning, A. S., Bousquet, P., Bruhwiler, L., Chen, Y. H., Ciais, ity, Nature, 515, 398–401, https://doi.org/10.1038/nature13957, P., Fung, I. Y., Heimann, M., John, J., Maki, T., Maksyutov, 2014. S., Masarie, K., Prather, M., Pak, B., Taguchi, S., and Zhu, Green, P. J.: Iteratively reweighted least squares for maximum like- Z.: TransCom 3 inversion intercomparison: Impact of trans- lihood estimation, and some robust and resistant alternatives, J. port model errors on the interannual variability of regional R. Stat. Soc. B, 46, 149–192, 1984. CO2 fluxes, 1988–2003, Global Biogeochem. Cy., 20, GB1002, Greve, P., Orlowsky, B., Mueller, B., Sheffield, J., Reichstein, M., https://doi.org/10.1029/2004GB002439, 2006. and Seneviratne, S. I.: Global assessment of trends in wetting and Callaghan, T. V., Johansson, M., Brown, R. D., Groisman, P. Y., drying over land, Nat. Geosci., 7, 716–721, 2014. Labba, N., Radionov, V., Barry, R. G., Bulygina, O. N., Essery, R. Guimberteau, M., Zhu, D., Maignan, F., Huang, Y., Yue, C., Dantec- L., Frolov, D. M., and Golubev, V. N.: The changing face of Arc- Nédélec, S., Ottlé, C., Jornet-Puig, A., Bastos, A., Laurent, P., tic snow cover: A synthesis of observed and projected changes, Goll, D., Bowring, S., Chang, J., Guenet, B., Tifafi, M., Peng, AMBIO: A Journal of the Human Environment, 40, 17–31, 2011. S., Krinner, G., Ducharne, A., Wang, F., Wang, T., Wang, X., Chevallier, F.: Copernicus Atmosphere Monitoring Service In- Wang, Y., Yin, Z., Lauerwald, R., Joetzjer, E., Qiu, C., Kim, H., version, available at: https://apps.ecmwf.int/datasets/data/ and Ciais, P.: ORCHIDEE-MICT (v8.4.1), a land surface model cams-ghg-inversions/, last access: 16 November 2017. for the high latitudes: model description and validation, Geosci. Chevallier, F., Fisher, M., Peylin, P., Serrar, S., Bousquet, P., Model Dev., 11, 121–163, https://doi.org/10.5194/gmd-11-121- Bréon, F.-M., Chédin, A., and Ciais, P.: Inferring CO2 sources 2018, 2018. and sinks from satellite observations: Method and appli- Harris, I., Jones, P. D., Osborn, T. J., and Lister, D. H.: Updated cation to TOVS data, J. Geophys. Res., 110, D24309, high-resolution grids of monthly climatic observations – the https://doi.org/10.1029/2005jd006390, 2005. CRU TS3. 10 Dataset, Int. J. Climatol., 34, 623–642, 2014. Chevallier, F., Ciais, P., Conway, T. J., Aalto, T., Anderson, Huntzinger, D. N., Post, W. M., Wei, Y., Michalak, A. M., West, B. E., Bousquet, P., Brunke, E. G., Ciattaglia, L., Esaki, Y., T. O., Jacobson, A. R., Baker, I. T., Chen, J. M., Davis, K. J., Fröhlich, M., Gomez, A., Gomez-Pelaez, A. J., Haszpra, Hayes, D. J., and Hoffman, F. M.: North American Carbon Pro- L., Krummel, P. B., Langenfelds, R. L., Leuenberger, M., gram (NACP) regional interim synthesis: Terrestrial biospheric Machida, T., Maignan, F., Matsueda, H., MorguÃ, J. A., model intercomparison, Ecol. Modell., 232, 144–157, 2012. Mukai, H., Nakazawa, T., Peylin, P., Ramonet, M., Rivier, Hurtt, G. C., Chini, L. P., Frolking, S., Betts, R. A., Feddema, J., L., Sawa, Y., Schmidt, M., Steele, L. P., Vay, S. A., Ver- Fischer, G., Fisk, J. P., Hibbard, K., Houghton, R. A., Janetos, meulen, A. T., Wofsy, S., and Worthy, D.: CO2 surface fluxes A., and Jones, C. D.: Harmonization of land-use scenarios for the at grid point scale estimated from a global 21 year reanalysis period 1500–2100: 600 years of global gridded annual land-use of atmospheric measurements, J. Geophys. Res., 115, D21307, transitions, wood harvest, and resulting secondary lands, Clim. https://doi.org/10.1029/2010JD013887, 2010. Change, 109, 117–161, 2011. Dargaville, R. J., Heimann, M., McGuire, A. D., Prentice, I. C., Jeong, S. J., Bloom, A. A., Schimel, D., Sweeney, C., Para- Kicklighter, D. W., Joos, F., Clein, J. S., Esser, G., Foley, J., Ka- zoo, N. C., Medvigy, D., Schaepman-Strub, G., Zheng, C., plan, J., and Meier, R. A.: Evaluation of terrestrial carbon cy- Schwalm, C. R., Huntzinger, D. N., and Michalak, A. M.: cle models with atmospheric CO2 measurements: Results from Accelerating rates of Arctic carbon cycling revealed by long- transient simulations considering increasing CO2 , climate, and term atmospheric CO2 measurements, Sci. Adv., 4, eaao1167, land-use effects, Global Biogeochem. Cy., 16, 39-1, 2002. https://doi.org/10.1126/sciadv.aao1167, 2018. Elmendorf, S. C., Henry, G. H. R., Hollister, R. D., Björk, R. G., Kaminski, T. and Heimann, M.: Inverse Modeling of Atmospheric Boulanger-Lapointe, N., Cooper, E. J., Cornelissen, J. H. C., Day, Carbon Dioxide Fluxes, Science, 294, 259–259, 2001. T. A., Dorrepaal, E., and Elumeeva, T. G.: Plot-scale evidence of Keeling, C. D., Chin, J. F. S., and Whorf, T. P.: Increased activity tundra vegetation change and links to recent summer warming, of northern vegetation inferred from atmospheric CO2 measure- Nat. Clim. Change, 2, 453–457, 2012. ments, Nature, 382, 146–149, https://doi.org/10.1038/382146a0, Forkel, M., Carvalhais, N., Rödenbeck, C., Keeling, R., Heimann, 1996. M., Thonicke, K., Zaehle, S., and Reichstein, M.: Enhanced sea- Koven, C., Friedlingstein, P., Ciais, P., Khvorostyanov, D., Krinner, sonal CO2 exchange caused by amplified plant productivity in G., and Tarnocai, C.: On the formation of high-latitude soil car- northern ecosystems, Science, 351, 696–699, 2016. www.atmos-chem-phys.net/19/12361/2019/ Atmos. Chem. Phys., 19, 12361–12375, 2019 12374 A. Bastos et al.: Seasonal-cycle amplitude change bon stocks: Effects of cryoturbation and insulation by organic Global atmospheric carbon budget: results from an ensemble of matter in a land surface model, Geophys. Res. Lett., 36, L21501, atmospheric CO2 inversions, Biogeosciences, 10, 6699–6720, https://doi.org/10.1029/2009GL040150, 2009. https://doi.org/10.5194/bg-10-6699-2013, 2013. Kunkel, K. E., Robinson, D. A., Champion, S., Yin, X., Estilow, T., Piao, S., Ciais, P., Friedlingstein, P., Peylin, P., Reichstein, M., and Frankson, R. M.: Trends and extremes in Northern Hemi- Luyssaert, S., Margolis, H., Fang, J., Barr, A., Chen, A., Grelle, sphere snow characteristics, Current Climate Change Reports, 2, A., Hollinger, D. Y., Laurila, T., Lindroth, A., Richardson, 65–73, 2016. A. D., and Vesala, T.: Net carbon dioxide losses of northern Le Quéré, C., Andrew, R. M., Friedlingstein, P., Sitch, S., Pongratz, ecosystems in response to autumn warming, Nature, 451, 49–52, J., Manning, A. C., Korsbakken, J. I., Peters, G. P., Canadell, https://doi.org/10.1038/nature06444, 2008. J. G., Jackson, R. B., Boden, T. A., Tans, P. P., Andrews, O. Piao, S., Nan, H., Huntingford, C., Ciais, P., Friedlingstein, P., Sitch, D., Arora, V. K., Bakker, D. C. E., Barbero, L., Becker, M., S., Peng, S., Ahlström, A., Canadell, J. G., Cong, N., Levis, S., Betts, R. A., Bopp, L., Chevallier, F., Chini, L. P., Ciais, P., Levy, P. E., Liu, L., Lomas, M. R., Mao, J., Myneni, R. B., Peylin, Cosca, C. E., Cross, J., Currie, K., Gasser, T., Harris, I., Hauck, P., Poulter, B., Shi, X., Yin, G., Viovy, N., Wang, T., Wang, J., Haverd, V., Houghton, R. A., Hunt, C. W., Hurtt, G., Ily- X., Zaehle, S., Zeng, N., Zeng, Z., and Chen, A.: Evidence for ina, T., Jain, A. K., Kato, E., Kautz, M., Keeling, R. F., Klein a weakening relationship between interannual temperature vari- Goldewijk, K., Körtzinger, A., Landschützer, P., Lefèvre, N., ability and northern vegetation activity, Nat. Commun., 5, 5018, Lenton, A., Lienert, S., Lima, I., Lombardozzi, D., Metzl, N., https://doi.org/10.1038/ncomms6018, 2014. Millero, F., Monteiro, P. M. S., Munro, D. R., Nabel, J. E. M. Piao, S., Liu, Z., Wang, Y., Ciais, P., Yao, Y., Peng, S., Chevallier, S., Nakaoka, S., Nojiri, Y., Padin, X. A., Peregon, A., Pfeil, B., F., Friedlingstein, P., Janssens, I. A., Peñuelas, J., and Sitch, S.: Pierrot, D., Poulter, B., Rehder, G., Reimer, J., Rödenbeck, C., On the causes of trends in the seasonal amplitude of atmospheric Schwinger, J., Séférian, R., Skjelvan, I., Stocker, B. D., Tian, H., CO2 , Global change biology, Glob. Change Biol., 24, 608–616, Tilbrook, B., Tubiello, F. N., van der Laan-Luijkx, I. T., van der 2017. Werf, G. R., van Heuven, S., Viovy, N., Vuichard, N., Walker, Pongratz, J., Reick, C. H., Houghton, R. A., and House, J. I.: Ter- A. P., Watson, A. J., Wiltshire, A. J., Zaehle, S., and Zhu, D.: minology as a key uncertainty in net land use and land cover Global Carbon Budget 2017, Earth Syst. Sci. Data, 10, 405–448, change carbon flux estimates, Earth Syst. Dynam., 5, 177–195, https://doi.org/10.5194/essd-10-405-2018, 2018. https://doi.org/10.5194/esd-5-177-2014, 2014. Liu, Y. Y., Van Dijk, A. I., De Jeu, R. A., Canadell, J. G., McCabe, Rödenbeck, C.: Estimating CO2 sources and sinks from atmo- M. F., Evans, J. P., and Wang, G.: Recent reversal in loss of global spheric mixing ratio measurements using a global inversion of terrestrial biomass, Nat. Clim. Change, 5, 470–474, 2015. atmospheric transport, Technical Report 6, Max Planck Institute Myers-Smith, I. H., Elmendorf, S. C., Beck, P. S., Wilmking, M., for Biogeochemistry, 07701 Jena, Germany, 2005. Hallinger, M., Blok, D., Tape, K. D., Rayback, S. A., Macias- Rödenbeck, C.: Atmospheric CO2 Inversion, 1985–2016, Fauria, M., Forbes, B. C., and Speed, J. D.: Climate sensitivity https://doi.org/10.17871/CarboScope-s85_v4.1, 2017a. of shrub growth across the tundra biome, Nat. Clim. Change, 5, Rödenbeck, C.: Atmospheric CO2 Inversion, 1976–2016, 887–891, 2015. https://doi.org/10.17871/CarboScope-s76_v4.1, 2017b. Nobrega, S. and Grogan, P.: Deeper snow enhances winter respi- Rödenbeck, C.: Atmospheric CO2 Inversion, 1993–2016, ration from both plant-associated and bulk soil carbon pools in https://doi.org/10.17871/CarboScope-s93_v4.1, 2017c. birch hummock tundra, Ecosystems, 10, 419–431, 2007. Rödenbeck, C., Houweling, S., Gloor, M., and Heimann, M.: CO2 Novick, K. A., Ficklin, D. L., Stoy, P. C., Williams, C. A., Bohrer, flux history 1982–2001 inferred from atmospheric data using a G., Oishi, A., Papuga, S. A., Blanken, P. D., Noormets, A., global inversion of atmospheric transport, Atmos. Chem. Phys., Sulman, B. N., Scott, R. L., Wang, L., and Phillips, R. P.: 3, 1919–1964, https://doi.org/10.5194/acp-3-1919-2003, 2003. The increasing importance of atmospheric demand for ecosys- Schneising, O., Reuter, M., Buchwitz, M., Heymann, J., Bovens- tem water and carbon fluxes, Nat. Clim. Change, 6, 1023, mann, H., and Burrows, J. P.: Terrestrial carbon sink observed https://doi.org/10.1038/nclimate3114, 2016. from space: variation of growth rates and seasonal cycle ampli- Peng, S., Ciais, P., Krinner, G., Wang, T., Gouttevin, I., McGuire, tudes in response to interannual surface temperature variability, A. D., Lawrence, D., Burke, E., Chen, X., Decharme, B., Koven, Atmos. Chem. Phys., 14, 133–141, https://doi.org/10.5194/acp- C., MacDougall, A., Rinke, A., Saito, K., Zhang, W., Alkama, 14-133-2014, 2014. R., Bohn, T. J., Delire, C., Hajima, T., Ji, D., Lettenmaier, D. P., Seneviratne, S. I., Corti, T., Davin, E. L., Hirschi, M., Jaeger, E. B., Miller, P. A., Moore, J. C., Smith, B., and Sueyoshi, T.: Simulated Lehner, I., Orlowsky, B., and Teuling, A. J.: Investigating soil high-latitude soil thermal dynamics during the past 4 decades, moisture–climate interactions in a changing climate: A review, The Cryosphere, 10, 179–192, https://doi.org/10.5194/tc-10- Earth-Sci. Rev., 99, 125–161, 2010. 179-2016, 2016. Seneviratne, S. I., Nicholls, N., Easterling, D., Goodess, C., Kanae, Peñuelas, J., Ciais, P., Canadell, J. G., Janssens, I. A., Fernández- S., Kossin, J., Luo, Y., Marengo, J., McInnes, K., and Rahimi, Martínez, M., Carnicer, J., Obersteiner, M., Piao, S., Vautard, M.: Changes in climate extremes and their impacts on the natu- R., and Sardans, J.: Shifting from a fertilization-dominated to a ral physical environment, Managing the risks of extreme events warming-dominated period, Nature Ecology Evolution, 1, 1438– and disasters to advance climate change adaptation, in: Manag- 1445, 2017. ing the risks of extreme events and disasters to advance climate Peylin, P., Law, R. M., Gurney, K. R., Chevallier, F., Jacobson, change adaptation: Special report of the Intergovernmental Panel A. R., Maki, T., Niwa, Y., Patra, P. K., Peters, W., Rayner, P. on Climate Change, 109–230, Cambridge University Press, pp. J., Rödenbeck, C., van der Laan-Luijkx, I. T., and Zhang, X.: 109–230, 2012. Atmos. Chem. Phys., 19, 12361–12375, 2019 www.atmos-chem-phys.net/19/12361/2019/ A. Bastos et al.: Seasonal-cycle amplitude change 12375 Sitch, S., Friedlingstein, P., Gruber, N., Jones, S. D., Murray- Welp, L. R., Patra, P. K., Rödenbeck, C., Nemani, R., Bi, J., Piper, S. Tortarolo, G., Ahlström, A., Doney, S. C., Graven, H., Heinze, C., and Keeling, R. F.: Increasing summer net CO2 uptake in high C., Huntingford, C., Levis, S., Levy, P. E., Lomas, M., Poul- northern ecosystems inferred from atmospheric inversions and ter, B., Viovy, N., Zaehle, S., Zeng, N., Arneth, A., Bonan, comparisons to remote-sensing NDVI, Atmos. Chem. Phys., 16, G., Bopp, L., Canadell, J. G., Chevallier, F., Ciais, P., Ellis, 9047–9066, https://doi.org/10.5194/acp-16-9047-2016, 2016. R., Gloor, M., Peylin, P., Piao, S. L., Le Quéré, C., Smith, B., Wenzel, S., Cox, P. M., Eyring, V., and Friedlingstein, P.: Projected Zhu, Z., and Myneni, R.: Recent trends and drivers of regional land photosynthesis constrained by changes in the seasonal cycle sources and sinks of carbon dioxide, Biogeosciences, 12, 653– of atmospheric CO2 , Nature, 538, 499–501, 2016. 679, https://doi.org/10.5194/bg-12-653-2015, 2015. Yin, Y., Ciais, P., Chevallier, F., Li, W., Bastos, A., Piao, S., Wang, Smith, W. K., Cleveland, C. C., Reed, S. C., and Running, S. W.: T., and Liu, H.: Changes in the response of the Northern Hemi- Agricultural conversion without external water and nutrient in- sphere carbon uptake to temperature over the last three decades, puts reduces terrestrial vegetation productivity, Geophys. Res. Geophys. Res. Lett., 45, 4371–4380, 2018. Lett., 41, 449–455, 2014. Yu, Z., Wang, J., Liu, S., Piao, S., Ciais, P., Running, S. W., Poul- Smith, W. K., Reed, S. C., Cleveland, C. C., Ballantyne, A. P., An- ter, B., Rentch, J. S., and Sun, P.: Decrease in winter respiration deregg, W. R., Wieder, W. R., Liu, Y. Y., and Running, S. W.: explains 25% of the annual northern forest carbon sink enhance- Large divergence of satellite and Earth system model estimates ment over the last 30 years, Global Ecol. Biogeogr., 25, 586–595, of global terrestrial CO2 fertilization, Nat. Clim. Change, 6, 306– 2016. 310, 2016. Zeng, N., Zhao, F., Collatz, G. J., Kalnay, E., Salawitch, R. J., West, Thomas, R. T., Prentice, I. C., Graven, H., Ciais, P., Fisher, J. B., T. O., and Guanter, L.: Agricultural Green Revolution as a driver Hayes, D. J., Huang, M., Huntzinger, D. N., Ito, A., Jain, A., of increasing atmospheric CO2 seasonal amplitude, Nature, 515, and Mao, J.: Increased light-use efficiency in northern terrestrial 394–397, https://doi.org/10.1038/nature13893, 2014. ecosystems indicated by CO2 and greening observations, Geo- Zhu, D., Peng, S. S., Ciais, P., Viovy, N., Druel, A., Kageyama, phys. Res. Lett., 43, 11339–11349, 2016. M., Krinner, G., Peylin, P., Ottlé, C., Piao, S. L., Poulter, B., Verburg, P. H., Crossman, N., Ellis, E. C., Heinimann, A., Hostert, Schepaschenko, D., and Shvidenko, A.: Improving the dynam- P., Mertz, O., Nagendra, H., Sikor, T., Erb, K. H., Golubiewski, ics of Northern Hemisphere high-latitude vegetation in the OR- N., and Grau, R.: Land system science and sustainable devel- CHIDEE ecosystem model, Geosci. Model Dev., 8, 2263–2283, opment of the earth system: A global land project perspective, https://doi.org/10.5194/gmd-8-2263-2015, 2015. Anthropocene, 12, 29–41, 2015. Zhu, Z., Piao, S., Myneni, R. B., Huang, M., Zeng, Z., Canadell, Viovy, N.: CRUNCEP data set, available at: ftp://nacp.ornl.gov/ J. G., Ciais, P., Sitch, S., Friedlingstein, P., Arneth, A., and Cao, synthesis/2009/frescati/temp/land_use_change/original/readme. C.: Greening of the Earth and its drivers, Nat. Clim. Change, 6, htm (last access: 27 July 2017), 2016. 791–795, 2016. Walker, D. A., Raynolds, M. K., Daniëls, F. J., Einarsson, E., Elve- bakk, A., Gould, W. A., Katenin, A. E., Kholod, S. S., Markon, C. J., Melnikov, E. S., and Moskalenko, N. G.: The circumpolar Arctic vegetation map, J. Veg. Sci., 16, 267–282, 2005. www.atmos-chem-phys.net/19/12361/2019/ Atmos. Chem. Phys., 19, 12361–12375, 2019