Too early to infer a global NPP decline since 2000



[1] The global terrestrial carbon cycle plays a pivotal role in regulating the atmospheric composition of greenhouse gases. It has recently been suggested that the upward trend in net primary production (NPP) seen during the 1980's and 90's has been replaced by a negative trend since 2000 induced by severe droughts mainly on the southern hemisphere. Here we compare results from an individual-based global vegetation model to satellite-based estimates of NPP and top-down reconstructions of net biome production (NBP) based on inverse modelling of observed CO2 concentrations and CO2 growth rates. We find that simulated NBP exhibits considerable covariation on a global scale with interannual fluctuations in atmospheric CO2. Our simulations also suggest that droughts in the southern hemisphere may have been a major driver of NPP variations during the past decade. The results, however, do not support conjecture that global terrestrial NPP has entered a period of drought-induced decline.

1. Introduction

[2] A considerable part of the uncertainty surrounding the magnitude of a future climate change is related to the response of the terrestrial carbon cycle. The terrestrial carbon cycle can be divided into two components or fluxes of opposite sign. Net primary production (NPP) constitutes the uptake flux, in which carbon from the atmosphere is sequestrated by plants through the balance between photosynthesis and plant respiration. The opposing release flux is dominated by heterotrophic respiration, in which carbon is released to the atmosphere by microbes and other organisms that consume dead and decaying biomass in the litter and organic layers of soils. Biomass burning through wildfires constitutes a smaller but significant component of the release flux. The resulting net land-atmosphere flux, net biome production (NBP) is the small difference between these two large fluxes. Today the terrestrial carbon cycle acts as a sink for carbon dioxide from the atmosphere, sequestrating around 30% of the global anthropogenic emissions, along with a similarly sized sink in the oceans [Le Quéré et al., 2009].

[3] Le Quéré et al. [2009] showed that the fraction of the total CO2 emissions remaining in the atmosphere, the airborne fraction, is likely to have increased since 1959, suggesting that CO2uptake by the global land and ocean sinks has lagged behind the average increase in emissions. Alternatively, the increased airborne fraction could be a result of a climate change-induced decrease in the sink capacity of the ocean and/or land sinks. However,Knorr [2009] found no significant trend in the airborne fraction when including CO2 measurement and emission data uncertainties in a statistical model, inferring that the sink has not yet (up to 2007) experienced a significant decrease. Dynamic global vegetation models (DGVMs) account for the main ecosystem processes responsible for interannual changes in terrestrial ecosystem carbon balance [Cramer et al., 2001; Prentice et al., 2001]. Intercomparison studies have shown that DGVMs disagree as to the fate of the global terrestrial carbon sink during the coming century, for example when forced by a “business as usual” climate scenario [Cramer et al., 2001; Sitch et al., 2008]. The magnitude and overall sign of changes in terrestrial carbon balance simulated over the coming century vary depending on differences in the regional distributions of precipitation and temperature projected by different GCMs [Berthelot et al., 2005; Schaphoff et al., 2006; Scholze et al., 2006]. Studies using DGVMs coupled to general circulation models (GCMs) to account for the feedback of climate-driven carbon balance shifts on radiative forcing tend to indicate that the land sink (NBP) is likely to become weaker in a changing future climate [Friedlingstein et al., 2006; Sitch et al., 2008]. Whether a weakening sink will be mainly a result of reduced NPP or increased respiration is still uncertain and depends on model assumptions [Friedlingstein et al., 2006]. Depending on the time scale, different processes will likely dominate in controlling the land sink. On shorter time scales, the changes in the land sink will likely be dominated by a near-instantaneous response of NPP to changes in climate. On longer time scales, decomposition of soil organic matter, which includes fractions with mean residence times from decades to centuries, will increasingly dominate changes in NBP. In a longer-term perspective, however, NPP, which provides the substrate for heterotrophic activity, must be the controlling process.

[4] The regional distribution of future changes in climate will be decisive for the future evolution of the land-sink. The response of NPP to a given absolute change in climate can vary markedly between regions. Increased temperatures leading to a longer growing season may tend to benefit NPP in colder regions while in warmer climate zones, increased maintenance respiration and the influence of higher evaporative demand on plant available water can reduce NPP [Beer et al., 2010; Nemani et al., 2003]. Studies using process models to explain a remotely-sensed increase in land surface greenness observed by satellites during the 1980's and 90's have inferred a positive trend in NPP during this period, attributed mainly to temperature increases at mid to high northern latitudes [Lucht et al., 2002; Nemani et al., 2003].

[5] In a recent study, Zhao and Running [2010]used a satellite data-constrained model to estimate spatial and temporal variability in global NPP for the past decade, reporting that the positive NPP trend seen in the 1980's and 90's (0.19 PgC/yr) [Nemani et al., 2003] might already have been disrupted and replaced by a negative trend (−0.055 PgC/yr) since 2000. Their main explanation for this trend shift was the repeated incidence of large-scale drought events mainly in the tropics and subtropics – areas susceptible to increased temperature and decreased precipitation. Zhao and Running's reported trend is inferred from a time series (reproduced inFigure 1aof this paper) of only 10 years' of global NPP estimates, too short for inferring a trend shift in a chaotic-dynamic system characterised by climate oscillations on time scales longer than a decade. Additionally the NPP estimates exhibit marked interannual variability relative to the weak negative trend.

Figure 1.

Carbon flux timeseries. (a) Global NPP anomalies (2000–2009) from LPJ-GUESS andZhao and Running [2010] with atmospheric CO2growth rate anomalies from NOAA/ESRL. (b) Global NBP, NPP, heterotrophic respiration + biomass burning (HRFI) from LPJ-GUESS with CO2 growth rate and CO2 inversion range anomalies. A positive NBP signifies an uptake of C by the terrestrial biosphere. NB: scale differs from other figure frames. (c) Southern hemisphere NPP and water deficit index (ω) from LPJ-GUESS and NPP from Zhao and Running (ω ranges from 0–1, 1 signifies no water stress). (d) Northern hemisphere NPP and ωfrom LPJ-GUESS and NPP from Zhao and Running. NB: for consistency, all carbon fluxes are plotted downward for sinks and upward for sources relative to the terrestrial biosphere.

[6] In this paper we evaluate and further explore the findings of Zhao and Running [2010], using an alternative, independent method of estimating NPP variations. We employ a bottom-up approach, forcing a detailed, individual-based DGVM with data on historical climate and atmospheric CO2concentrations, employing a time-invariant land use mask, but no satellite data. The model predicts NPP and NBP variations in space and time in response to this forcing. We compare the NBP simulated by our model to top-down reconstructions based on inverse modelling of observed atmospheric CO2 concentrations and global annual CO2 growth rate anomalies, which are dominated by the variability of NBP [Baker et al., 2006; Bousquet et al., 2000; Keeling et al., 1996; Keeling et al., 2005; Le Quéré et al., 2003].

2. Model Description

[7] Global patterns and interannual variation in NPP and NBP were simulated with LPJ-GUESS, a generalized, process-based model of vegetation dynamics and biogeochemistry designed for regional or global applications. It combines features of the Lund-Potsdam-Jena (LPJ) DGVM [Gerten et al., 2004; Sitch et al., 2003] with a more detailed treatment of plant resource competition and demography, based on the interactions of plant individuals at the neighbourhood or plot scale [Smith et al., 2001]. The individual-based dynamics are computationally expensive, but have been demonstrated to improve the realism of the model in simulating transient shifts and spatial patterns of vegetation and carbon balance [Smith et al., 2001]. An explicit representation of size structure and tree demographics, as included in our model, may be important for the accurate simulation of large-scale forest carbon stocks and fluxes [see, e.g.,Fisher et al., 2010; Purves and Pacala, 2008; Wolf et al., 2011]. Vegetation is represented as a mixture of 11 plant functional types (PFTs) differentiated by bioclimatic limits, growth form, phenology and life history strategy. Individuals of different age/size classes (for trees) and grassy ground-layer vegetation interact in competition for light and water within a number (20 in this study) of replicate patches in each simulated 0.5 × 0.5 degree gridcell. Photosynthesis, respiration, stomatal conductance and phenology are simulated on a daily time step. Growth of plant individuals is effected annually, accounting for tissue turnover, sapwood-to-heartwood conversion, and allocation of the accrued NPP to leaves, sapwood and fine roots in conformity with a set of allometric rules [Sitch et al., 2003].

[8] Plants experience water-deficit stress when the supply of water from the root zone, given as a soil water-dependent proportion of a prescribed maximum transpiration rate, cannot meet the transpirative demand. The latter depends on temperature and radiation, while the influence of humidity is accounted for by a boundary-layer parameterisation that relates potential evapotranspiration to aggregate conductance at canopy scale [Huntingford and Monteith, 1998]. Under water deficit, plants down-regulate their stomatal conductance, reducing water loss but lowering the intercellular CO2 concentration and thereby photosynthesis [Haxeltine and Prentice, 1996]. In addition, water-deficit results in an increased allocation of NPP to roots at the expense of leaves and stems, further reducing photosynthesis the following growing season [Sitch et al., 2003].

[9] Demographic changes within and among PFTs are the outcome of annual establishment, mortality and disturbance in each patch. The model also simulates soil hydrology (two-layer bucket with baseflow and percolation between layers) [Gerten et al., 2004] and disturbance by wildfires [Thonicke et al., 2001]. Decomposition of plant litter and two soil organic matter pools follows first-order kinetics with dependency on soil temperature and moisture. A detailed description of LPJ-GUESS is available in [Smith et al., 2001]. This study uses Version 2.1 which includes the updates described by Hickler et al. [2012]. The PFT set employed in this study, and their prescribed parameters, are given in Tables S1 and S2 in Text S1 in the auxiliary material.

[10] LPJ-GUESS has been evaluated extensively and exhibits comparable skill to other approaches and models in reproducing observed temporal and spatial variation in large-scale vegetation patterns and biogeochemistry. Simulated carbon and evaporative fluxes have been compared to ecosystem flux measurements [Morales et al., 2005; Wramneby et al., 2008], forest inventory data and site-based measurements of NPP, leaf area index and biomass, spanning many of the world's biomes, including drought-affected biomes such as savannah, steppe and warm-climate grasslands [Hickler et al., 2006; Smith et al., 2011, 2008; Tang et al., 2012, 2010; Zaehle et al., 2006]. As an example, Hickler et alshowed that the model reproduced precipitation-driven, satellite-observed inter-annual variations in LAI in the African Sahel. Comparisons of these variables to satellite estimates have also been performed [Smith et al., 2011, 2008; Tang et al., 2010]. In a detailed sensitivity-uncertainty analysis,Wramneby et al. [2008] demonstrate that simulated carbon fluxes remain robust within the parameter uncertainty of the model. Figure S1 in Text S1shows a zonal comparison between NPP from this study and inventory data from the Ecosystem Model-Data Intercomparison (EMDI) database [Olson et al., 2001]. Global aggregated fluxes and carbon pool sizes are given in Table S3.

3. CO2 Inversion Data

[11] CO2 flux inversion is a method in which a priori knowledge of fluxes, observations and transport models is applied to infer the magnitude and geographic location of CO2 sources and sinks [Baker et al., 2006; Ciais et al., 2010; Gurney et al., 2002]. The inversions facilitate separation of anthropogenic emissions from natural fluxes and land fluxes from ocean fluxes. Thus the results can be compared to those from bottom-up methods [Bousquet et al., 2000]. We obtained filtered CO2 inversion products as yearly sums for the northern and southern hemisphere from CarboScope ( We selected the time period covered by three or more inversion products (1996–2008), and available time series for the natural land-atmosphere flux from the LSCE AN v2.1 [Peylin et al., 2005] (available time period: 1996–2004), LSCE VAR v1.0 [Chevallier et al., 2005] (1988–2008), Jena S96 v3.3 [Rödenbeck, 2005] (1996–2009), Jena S99 v3.3 [Rödenbeck, 2005] (1999–2009) and CarbonTracker CTE2008 [Peters et al., 2007] (2000–2007) inversion products. From the flux time series we calculated the inter-model range of anomalies relative to the time series average.

4. Methods

[12] We forced LPJ-GUESS with monthly precipitation, temperature, short wave radiation and number of rain-days (precipitation exceeding 1 mm) per month at 0.5 × 0.5° resolution based on the CRU-NCEP (v2.0) dataset (N. Viovy and P. Ciais, unpublished data, 2010). The CRU-NCEP dataset is a combination of the CRU TS 2.1 dataset [Mitchell and Jones, 2005] and NCEP reanalysis I [Kalnay et al., 1996] extending from 1901 through 2009.

[13] Two runs were realized, one using only grass (C4 and C3) PFTs, to represent croplands and pastures, the second also including woody PFTs, representing forests and potential natural vegetation. Results from these two simulations were merged into a single data set, weighing the resultant ecosystem-atmosphere carbon fluxes (see below) for each year and grid cell by a time-invariant land use mask based on the HYDE 3.1 dataset [Klein Goldewijk et al., 2010] for the year 2005. The simulations were initialized with a 500 year “spin-up” from “bare ground” to attain vegetation and carbon pools in each grid cell in approximate equilibrium with the early-20th century climate. The spin-up was forced by a detrended version of the climate driver data for the period 1901–1930, replicated every 30 years. Atmospheric CO2 concentrations ([CO2]) were set at the 1861 level of 287 ppmv until the simulation year corresponding to 1861, thereafter annual [CO2] from atmospheric or ice core measurements.

[14] Results were analysed in terms of the simulated component fluxes of the carbon exchange between ecosystems and the atmosphere, i.e. NBP, NPP and the net release flux comprising the sum of heterotrophic respiration and biomass burning. The water deficiency stress experienced by plants in the simulations was quantified by the water deficiency index, ω, in the range 0–1 where 1 signifies no water stress [Sitch et al., 2003, equation 13]. In our analysis, ωwas weighed with the grid-cell average simulated NPP for the period 1980–2009 to better capture the relative effect of water stress on NPP in different global regions.

5. Comparison on a Global Scale

[15] Time series of NPP from our simulations show marked similarities to those of Zhao and Running [2010] (Figure 1a), but in contrast to the latter study, they exhibit a positive overall trend for 2000 through 2009 (0.33 PgC/yr, < 0.01; Mann-Kendall test). Note that for consistency with the other fluxes shown, NPP increases downward inFigure 1. Similar to Zhao and Running's study, the time series exhibits marked (s.d. ∼1.2 PgC/yr) interannual variations; including some agreement with interannual variation in the atmospheric CO2 growth rate (= −0.43, n.s) (NOAA/ESRL, (NB: the CO2 growth rate anomaly data for 2009 (∼−0.6) differ slightly from the value given by Zhao and Running (∼−0.2), due to an update to the value reported in the NOAA/ESRL database.)

[16] The above results concern simulated NPP from our model. When we instead consider NBP, which accounts also for heterotrophic respiration and emissions from biomass burning, the correlation with CO2 growth rate anomalies rises (= −0.86; < 0.005) (Figure 1b). Simulated NBP also shows similarities with the average of the CO2 inversions (r = −0.70; < 0.05) (Figure 1b). The simulated and inversion-based time series diverge after 2004, the simulated results indicating a moderate reduction in sinks, while the inversions suggest an increased sink. In the simulations, an increase in NPP is offset by increased heterotrophic respiration and biomass burning, resulting in a weakened sink.

6. Regional Patterns

[17] NPP variations simulated for the southern hemisphere (SH) are similar to the results of Zhao and Running and reflect the strong influence of drought dynamics they find using their method (Figure 1c). The decline in NPP over the study period is weaker (−0.007 PgC/yr) in our simulations compared to Zhao and Running's results (−0.183 PgC/yr). The relationship between the water deficiency index ω and NPP is strong, indicating a strong influence of droughts on simulated NPP interannual variability over the SH. In 2007, simulated NPP increases in response to ameliorated water stress (increased ω), in contrast to Zhao and Running's results, which suggest a decrease in NPP for this year. Interannual variability over the northern hemisphere (NH) is lower (Figure 1d), a finding common to Zhao and Running's results and our own. Like Zhao and Running, our model simulated a positive trend in NH-NPP (0.33 PgC/yr; Zhao and Running 0.128 PgC/yr). The relationship betweenω and simulated NPP is also weaker and ω shows less variability, suggesting that factors (temperature, insolation) other than available water exerted a stronger relative control on ecosystem production in the NH in the simulations. See Figure S2a in Text S1 for the simulated NPP trend, also see Auxiliary Material section 4, and Figure S3 in Text S1 for comparison of NBP and CO2 inversion products at the hemispherical scale.

7. Meteorological Data

[18] The CRU-NCEP dataset is a combination of CRU TS 2.1 and NCEP reanalysis I. To ensure that the differences in our results compared with Zhao and Running could not occur because of differences in the climate forcing data used, we repeated the same analysis using only the NCEP reanalysis data employed by Zhao and Running as forcing to our model (the forcing consisted of NCEP reanalysis I [Kalnay et al., 1996] for the period 1948 through 1978 and NCEP-DOE reanalysis II [Kanamitsu et al., 2002] for the period 1979 through 2009). The resulting fluxes and their trends are very similar to those reported above with the exception that they exhibit a larger amplitude (an effect of increased shortwave radiation in the NCEP dataset), and do not affect the conclusions drawn in our study.

8. Discussion and Conclusion

[19] Variations in the simulated carbon balance in our model depend only on variations in climate and atmospheric [CO2], yet our results show many common features with those attained by Zhao and Running [2010] using satellite data to estimate NPP. Similarly to Zhao and Running's findings our results also suggest that terrestrial ecosystems during the first decade of this century have exhibited pronounced interannual fluctuations in carbon balance, and that these fluctuations are dominated by the influence of drought episodes in various regions of the southern hemisphere in particular years.

[20] A method that employs direct spectral observations from space of the world's vegetation might be expected to give more reliable results than a model forced only by meteorological data and atmospheric [CO2]. However, the MODIS NPP products also rely on meteorological data and their quality [Zhao et al., 2006], model assumptions and parameters to translate the reflectance measurements into estimates of NPP [Zhao and Running, 2010]. For example, the method is dependent on biome specific parameters, where assumptions as to the strongest limiting factor for plant production (water, radiation or temperature) are an important source of uncertainty [Churkina et al., 1999; Nemani et al., 2003]. In separate critiques of Zhao and Running's study, Medlyn [2011] and Samanta et al. [2011] suggest that the apparent overall decline in NPP after 2000 could be an artifact caused by the strong temperature dependence of the NPP model used, persistent warming over 2000–2009 tending to exacerbate plant respiration costs, thereby reducing NPP. Consistent with such an interpretation, Samanta et al. [2011]note that some 85% of the land south of 70°N exhibits no significant decline in the MODIS-derived NDVI and EVI vegetation indices over the same decade.

[21] Due to the marked influence of terrestrial ecosystem functioning on interannual variations in atmospheric [CO2], simulated global NBP may be verified against independent data by comparing it with atmospheric [CO2] and, at the hemispheric scale, by comparison with results from CO2 flux inversions. NBP in our study shows persuasive agreement with the CO2 growth rate anomalies and CO2 inversions on the global scale (Figure 1b). In the absence of dependable, independent estimates of large-scale NPP it is not possible to ascertain whether the results from our model more accurately replicate actual variation in NPP compared withZhao and Running's [2010] study.

[22] The results of this study lend support to the finding of Zhao and Running [2010] that SH ecosystems have shown considerable sensitivity, in terms of their carbon balance, to droughts over the past decade. CO2inversion studies paint a similar picture. The drought-related fluctuations in NPP may be seen as a warning sign of the threats posed to ecosystems and their services by a continuation of current warming trends, and as a reminder that the present-day land carbon sink is a transient anomaly that can not persist in the long term. Heterotrophic respiration and biomass burning exhibit positive trends in our simulations, but they are overwhelmed by a growing-season related increase in NH NPP. In their headline conclusion, Zhao and Running suggested that global terrestrial NPP has entered a period of drought-induced decline. Our results suggest that such a conclusion may be premature.


[23] This study was financially supported by the Foundation for Strategic Environment Research (MISTRA) through the Mistra-SWECIA programme, and the Swedish Research Council FORMAS. The study is a contribution to Lund University's Strategic Research Areas Modelling the Regional and Global Earth System (MERGE) and Biodiversity and Ecosystem Services in a Changing Climate (BECC). We acknowledge CarboScope and its contributors for sharing the CO2inversion data used in the study. We also acknowledge N. Viovy and P. Ciais for providing the CRU-NCEP dataset. The NCEP Reanalysis 2 data was provided by the NOAA/OAR/ESRL PSD, Boulder, Colorado, USA, from their Web site at

[24] The Editor thanks two anonymous reviewers for assisting in the evaluation of this paper.