全球变化知识资源中心

The net ecosystem production (NEP) of an ecosystem represents its carbon (C) balance at daily to decadal scales. Despite considerable study, the main drivers of NEP are still unclear. Climate^{4, 5}, climatic trends⁶, nitrogen deposition^{7, 8}, disturbance and management^{8, 9} have been suggested to influence NEP. These studies, however, either were unable to explain a substantial percentage of the spatial variability in NEP or collected data in a restricted subset of climatic space, indicating that it is not yet known what factor(s) most strongly govern NEP, one of the critical pathways by which terrestrial ecosystems feedback to climate.

At the ecosystem scale, nitrogen deposition has been suggested to enhance the NEP of forests^{3, 7}. Nutrient availability is indeed a key variable explaining patterns of carbon allocation in forests; nutrient-rich forests exhibit higher biomass production, biomass production efficiency (defined as biomass production/gross primary production (GPP) ratio) and shoot-to-root biomass production ratio². By converting a larger fraction of GPP to woody biomass and thereby increasing the residence time of the assimilated C, forests growing on more fertile soils can be expected to exhibit higher NEP. Carbon-use efficiency at the ecosystem level (CUEe), defined as NEP of an ecosystem per unit of GPP, measures the proficiency of an ecosystem to store C absorbed from the atmosphere. We thus suggest that both NEP and CUEe increase with increasing nutrient availability in forest ecosystems.

To test this hypothesis, we updated and analysed a global forest data set of mean annual carbon flux (GPP, ecosystem respiration (Re) and NEP), stand biomass, stand age and information on management. The resulting data set of 92 forests included scattered data from 1990 to 2010 from boreal, temperate, Mediterranean and tropical forests⁹ (Supplementary Fig. 1). We added all published information on the nutrient status of these forests and we classified them as forests with high nutrient availability (without apparent nutrient limitation) and low nutrient availability (apparently strongly nutrient-limited, in the sense of ref. 2, considering a holistic combination of availability of nutrients and soil characteristics). We based the nutrient availability classification on a multivariate factor analysis containing information about soil type, soil and foliar nutrient concentrations (N, P), soil pH, soil C/N ratio, nitrogen deposition and mineralization, history of the stand, specific reports of nutrient availability and an assessment by the principal investigator of the site (Supplementary Table 1). This analysis clearly separated nutrient-rich from nutrient-poor forests (Supplementary Fig. 2). We also established a medium category that was used for additional testing; it contained forests with information indicating moderate availability of nutrients or with little information about their nutrient status. Mean annual temperature and precipitation (MAT, MAP) from the WorldClim database¹⁰ and water deficit (WD) derived from Moderate Resolution Imaging Spectroradiometer (MODIS) evapotranspiration time series (MOD15A2 product) were used as climatic predictors. We then used generalized linear models to disentangle the effects of climate, management and stand age from those of nutrient availability on NEP and CUEe (see Methods for details on data sets and methodology).

NEP in nutrient-rich forests averaged 33 ± 4% (mean ± s.e.m) of GPP, whereas nutrient-poor forests accumulated only 6 ± 4% of the photosynthesized carbon (CUEe in Fig. 1, difference = 27 ± 7%, analysis of variance P < 0.001). Only nutrient-rich forests showed a clear positive relationship between GPP and NEP (Fig. 1). In contrast, nutrient-poor forests channelled a larger proportion of GPP into Re (Fig. 2), with NEP being almost independent of GPP. Higher nutrient availability thus seems to channel C fixed by GPP towards storage in biomass and soils, rather than being respired back to the atmosphere.

The bar chart inside the main graph shows that carbon-use eciency at the ecosystem level (CUEe) (net ecosystem production (NEP)/gross primary production (GPP) ratio) in nutrient-rich forests (red) is more than five times higher than in nutrient-poor forests (blue). We also present results for forests with GPP < 2,500 gC m⁻² yr⁻¹, because values of GPP >2,500 gC m⁻² yr⁻¹ were available only for nutrient-poor forests. When considering only forests with GPP < 2,500 gC m⁻² yr⁻¹, the Nutrients*GPP (where Nutrients = nutrient availability) interaction (where * indicates an interaction) was significant at the 0.006 level. In the bar chart, error bars indicate the s.e.m. and *** indicates significant differences at the P < 0.001 level.

Sources of data.

We used data of mean annual carbon flux from a global forest database⁹. This data set contains complete measurements of carbon balance and uncertainties of GPP, Re and NEP of forests around the world. The WorldClim database¹⁰ (resolution ˜1 km at the Equator) and MODIS evapotranspiration time series (MOD15A2 product) provided climatic data (MAT and MAP from WorldClim and potential and actual evapotranspiration from MODIS). The reliability of the data from the WorldClim database was tested with the available observed climatic values from the forests (N = 123). Results indicated a strong correlation between observed and WorldClim values for annual temperature and precipitation (R² = 0.96, P < 0.001 and R² = 0.84, P < 0.001 respectively).

All continents were represented in our analyses (Supplementary Fig. 1), although most of the forests were located in Europe and North America. Boreal (N = 31) and especially temperate (N = 68) sites outnumbered Mediterranean (N = 14) and tropical (N = 16) sites. Of the forests included, 61 were coniferous, 57 were broadleaved and 11 were mixed.

Information on nutrient availability.

For each forest, we compiled all available information from the published literature (carbon, nitrogen and phosphorus concentrations of soil and/or leaves, soil type, soil texture, soil C/N ratio, soil pH, measures of nutrients, and so on.) related to nutrient availability. Then we followed the criteria shown in Supplementary Table 3 to code these variables as three-level factors indicating high, medium or low nutrient availability. Next, we transformed these factors into dummy variables and performed a factor analysis. The first factor extracted explained 14.8% of the variance in the data set and was related to nutrient-rich dummy variables whereas the second factor explained 8.7% of the variance and was related to nutrient-poor dummy variables (Supplementary Fig. 2a). Then, on the basis of the aggregations across the two main factors extracted (Supplementary Fig. 2b) we classified the forests as having clearly high or clearly low nutrient availabilities. The remaining forests, for which empirical evidence was insufficient to classify them as nutrient-rich or nutrient-poor or indicated moderate nutrient availability, were classified as medium nutrient availability. To maximize robustness, we included only the forests with clearly high (N = 23) and clearly low (N = 69) nutrient availabilities in the main analysis, discarding data from the 37 remaining forests with medium nutrient availability. We also present the analysis with all of the available data (including the medium nutrient availability category) in Supplementary Fig. 8 and in Supplementary Models.

Statistical analyses.

We ran generalized linear models to test for differences in CUEe, NEP, Re and GPP between forests of high and low nutrient availability, accounting for the possible effects of GPP, mean stand age, management (as a binary variable: managed or unmanaged) and climate (MAT, MAP and WD = 1 − (AET/PET) ∗ 100), where AET and PET represent actual and potential evapotranspiration, respectively. That is, NEP ˜GPP + nutrient availability + Age + Management + MAT + MAP + WD. We tested for interactions up to the second order among GPP, nutrient availability, age and management. The significant variables of the final model (minimum adequate model) were selected using stepwise backward variable selection and the Akaike information criterion of the respective regression models. To evaluate the variance explained by each predictor variable, we used the averaged over orderings method (the lmg metric, similar to hierarchical partitioning) to decompose R² from R (ref. 28) with the package relaimpo (Relative Importance for Linear Regression²⁹). Finally, we tested whether nutrient status, management, age and climatic variables could lead to changes in patterns of biomass allocation with stepwise forward regressions. Model residuals met the assumptions required in all analyses (that is, normality and homoscedasticity).

The robustness of our analyses was tested by five different methods: running weighted models using the inverse of the uncertainty of the estimates as a weighting factor; using only data derived from eddy-covariance towers; restricting comparison of nutrient-rich and nutrient-poor forests to a common rank of GPP (GPP < 2,500 gC m⁻² yr⁻¹ in Figs 1 and 2, thus excluding most of the tropical forests and using forests presenting GPPs above 1,000 and below 2,500 gC m⁻² yr⁻¹ in Supplementary Fig. 10); using an alternative classification of nutrient availability (the second most plausible classification) as an analysis of sensitivity; and using the factors extracted for the classification of nutrients as nutrient richness covariates instead of using the binary factor nutrient availability. Detailed information about the methods used in this paper is presented in the Supplementary Information.

Corrected online 28 May 2014

In the version of this Letter originally published, the following text was omitted from the acknowledgements section: 'We also thank all site investigators, their funding agencies, the various regional flux networks (Afriflux, AmeriFlux, AsiaFlux, CarboAfrica, CarboEurope-IP, ChinaFlux, Fluxnet-Canada, KoFlux, LBA, NECC, OzFlux, TCOS-Siberia, USCCC), the Office of Science (BER) and US Department of Energy (for funding the development of measurement and data submission protocols), and the Fluxnet project, whose work and support is essential for obtaining the measurements without which the type of integrated analyses conducted in this study would not be possible.' This has been corrected in the online versions of the Letter.

Corrected online 13 March 2015

In the version of this Letter originally published, in the final paragraph, the section of text including 'Earth system models should … nutrient cycling)' was misleading and should have been:
"Models simulating the dynamics of the terrestrial biosphere currently consider the effects of nitrogen on vegetation and soils^{25, 26} but they still do not consider the effects of other nutrients such as phosphorus or potassium. Future models should consider the interactions of nitrogen as well as these other nutrients with the entire forest carbon balance. The relationship between GPP and NEP appears to be so strongly controlled by the nutrient status of the forest that terrestrial biosphere models may be unable to accurately predict the carbon balance of forest ecosystems without information on background nutrient availability²⁷—soil nitrogen, phosphorous, potassium, pH—and on changes in soil and plant nutrient cycling resulting from human activities (such as nitrogen deposition, climate change and elevated CO₂)."
To accommodate these changes, the original ref. 27 (W. De Vries and M. Posch, M. Environ. Pollut. 159, 2289-2299; 2011) was removed and the remaining references renumbered. The original references numbered 31–33 have been moved to the Supplementary Information, as they are uncited in the main text.

1. Dixon, R. K. et al. Carbon pools and flux of global forest ecosystems. Science 263, 185–90 (1994).
   • CAS
   • ADS
   • ISI
   • PubMed
   • Article
2. Vicca, S. et al. Fertile forests produce biomass more efficiently. Ecol. Lett. 15, 520–526 (2012).
   • CAS
   • PubMed
   • Article
3. Janssens, I. A. et al. Reduction of forest soil respiration in response to nitrogen deposition. Nature Geosci. 3, 315–322 (2010).
   • CAS
   • ADS
   • ISI
   • Article
4. Valentini, R. et al. Respiration as the main determinant of carbon balance in European forests. Nature 404, 861–865 (2000).
   • CAS
   • ADS
   • ISI
   • PubMed
   • Article
5. Kato, T. & Tang, Y. Spatial variability and major controlling factors of CO₂ sink strength in Asian terrestrial ecosystems: Evidence from eddy covariance data. Glob. Chang. Biol. 14, 2333–2348 (2008).
   • Article
6. Piao, S. et al. Footprint of temperature changes in the temperate and boreal forest carbon balance. Geophys. Res. Lett. 36, L07404 (2009).
   • CAS
   • ADS
   • Article
   URL: