The terrestrial carbon
fluxes show the largest variability among the components of the global carbon
cycle and drive most of the temporal variations in the growth rate of
atmospheric
Atmospheric
It has been long debated whether it is GPP or TER that controls the spatial and temporal variability in NEE. Several studies have ascribed inter-annual variability in NEE to variability in either GPP (Ahlström et al., 2015; Janssens et al., 2001; Jung et al., 2011, 2017; Stoy et al., 2009; Urbanski et al., 2007) or TER (Morgenstern et al., 2004; Valentini et al., 2000) or both (Ma et al., 2007; Wohlfahrt et al., 2008b). GPP and TER show comparable ranges of IAV, typically larger in absolute terms than that observed for NEE due to the temporal correlation between the two gross fluxes (Richardson et al., 2007). Given that photosynthesis and respiration may respond differently to environmental drivers (Luyssaert et al., 2007; Polley et al., 2008), the interpretation of climate impacts on the variability in NEE requires an understanding of the relation between the IAV of NEE and that of GPP and TER (Polley et al., 2010).
The environmental factors driving the IAV of NEE (IAV
The temporal dynamic of NEE has been addressed in numerous studies, based on either “top–down” approaches, which primarily focus on aircraft atmospheric budgets (Leuning et al., 2004), tower-based boundary layer observations (Bakwin et al., 2004) and tracer transport inversion (Baker et al., 2006; Gurney et al., 2002; Rödenbeck et al., 2003), or on “bottom–up” methods that rely on data-driven gridded products derived from the upscaling of flux data (Jung et al., 2011, 2017; Papale et al., 2015; Papale and Valentini, 2003) or process-based biogeochemical models that simulate regional carbon budgets (Desai et al., 2007, 2008; Mahadevan et al., 2008).
Despite the broad literature on the subject, very few examples of IAV
analysis based on multiple data streams are available in the literature
(Desai et al., 2010; Pacala, 2001; Poulter et al., 2014). In the present
study, patterns and controls of the inter-annual variability in NEE have been
analysed using three different data streams: ecosystem-level data from the
FLUXNET database, the MPI-MTE (model tree ensemble) bottom–up product
resulting from the statistical upscaling of in situ flux data (le Maire et
al., 2010) and the Jena CarboScope Inversion top–down product, which
estimates land (and ocean) fluxes from atmospheric
Data on an ecosystem scale were retrieved from two releases of the FLUXNET
dataset, namely La Thuile (
On a global scale, two sources of gridded data were used: a bottom–up data
product, namely the MPI-MTE product (Jung et al., 2009), and, as a top–down
product, the Jena CarboScope
To derive surface fluxes, the Jena CarboScope Inversion combines modelled
atmospheric transport with high-precision measurements of atmospheric
As the inversion estimates the total land flux, calculated as the difference
between the total surface flux and prescribed anthropogenic emissions, it
also includes
In order to analyse the role of climatic drivers in the inter-annual
variability, global maps of temperature and precipitation were used. Gridded
air temperatures were obtained from the Climatic Research Unit (CRU) at the
University of East Anglia on a monthly timescale and
The inter-annual variability in NEE was estimated as the standard deviation
of annual NEE values generated by trend and residuals, computed on time
windows of 12 months shifted with a monthly time step (Luyssaert et
al., 2007; Shao et al., 2015; Yuan et al., 2009) and calculated with the same
methodology for the three data streams used in the analysis. Average values
of IAV for plant functional type (PFT) were determined using the PFT
classification of FLUXNET sites and the MCD12C1 product (aggregated at the
appropriate spatial resolution using the dominant PFT) for the MPI-MTE and
the Jena Inversion. Map grid cells were also classified according to mean
annual temperature and precipitation, and the mean value of
IAV
For the two gridded products, which provide a 30-year-long time series (1982–2011), the IAV was partitioned into two components, namely the variance explained by the temporal trend and that due to annual anomalies (Ahlström et al., 2015). For this purpose a linear model was fitted to the time series at each pixel, and the determination coefficient of the regression was used to measure the fraction of variance explained by the trend, whereas its complement to 1 was the fraction of variance due to anomalies.
The spatial correlation between IAV and climatic drivers (air temperature and
precipitation) was analysed on a global scale for the MPI-MTE by calculating
the spatial correlation coefficient between the temporal standard deviation
(IAV amplitude) of NEE and the average annual temperature or precipitation in
moving spatial windows of
Finally, in order to identify which process between photosynthesis and
respiration drives IAV
The spatial pattern of inter-annual variability for the three datasets is
shown in Fig. 1. The IAV of NEE at individual FLUXNET sites ranges from 15 to 400
Spatial distribution of NEE standard deviation used as a measure of
inter-annual variability (IAV
In terms of IAV, the two global products show a reasonable qualitative
correspondence for North America and Eurasia, whereas they disagree for South
America, with MPI-MTE showing a minimum of IAV in the humid tropics, where
the inversion product shows large variability. MPI-MTE in particular shows
maximum values along the eastern coast of South America while the Jena
Inversion shows an almost opposite pattern. A similar behaviour is also
observed in Africa, where the top–down product shows a maximum in central
Africa, while MPI-MTE shows a minimum in the Congo Basin and higher values in
arid zones like the Sahel and southern Africa. These discrepancies could, on the one hand, be ascribed to the
limits of the bottom–up approach in dealing with the low seasonality of the
fraction of absorbed radiation (FaPAR) in evergreen broadleaf forests, given
the relevance of this predictor in the MPI-MTE estimates. A second reason for
the discrepancy could be the
As far as the Northern Hemisphere is concerned, a good correspondence is observed in western Eurasia, while some discrepancies are observed in other zones; for example MPI-MTE shows a large IAV in India, probably driven by the changes in FaPAR related to agricultural intensification, which does not emerge from the inversion product that has little observational constraint in this area. To summarize, the spatial pattern of IAV in the two products better agrees in the Northern Hemisphere for temperate and cold temperate zones, whereas for the Southern Hemisphere, and in particular for the humid evergreen forests, the two products show a poor match. In general, it has to be considered that both the MPI-MTE product and the Jena Inversion are driven by data from surface networks that are very limited in the tropics and the Southern Hemisphere, and, therefore, these observation-driven estimates are under-constrained in those areas. These results highlight that for achieving more robust and consistent estimates of the terrestrial carbon fluxes, it is of key importance to increase the availability of atmospheric and ecosystem flux observations in the tropical region, either by establishing new sites where the network is sparse or improving the sharing of data where the monitoring stations are available but not connected to global networks (e.g. flux stations in Amazonia).
IAV
Dependency of NEE interannual variability on GPP and
NEE
Dependence of IAV
The results presented in the maps of Fig. 1 are summarized in the climate space in Fig. 2. The left panels show that peak values of IAV are located in different climate regions for the two gridded products (temperate humid for MPI-MTE and tropical humid for the Jena Inversion). These results highlight that top–down and bottom–up estimates do not agree on the main sources of temporal variability in the terrestrial carbon budget and call for more investigation to pin down the reasons for these large discrepancies. Given that the IAV of NEE increases with the primary productivity at the FLUXNET sites (Fig. 3), in Fig. 2 (right column) we normalized the IAV of both MPI-MTE and the Jena Inversion by the average GPP of the specific climate bin from the MPI-MTE. Normalization using GPP (which is always positive) offers a more robust metric of relative IAV if compared to normalization with NEE (that spans 0). Figure 2 shows either the mean IAV (left column) or the ratio of the mean IAV and GPP (right column) in each climate bin, since this metric is less sensitive to outliers than the mean of ratios and gives more weight to points with larger fluxes. The normalized IAV shows a consistent pattern between the three different data products, with a clear decreasing trend with increasing temperature and precipitation (i.e. increasing productivity). Ultimately arid regions seems to have the higher relative variation in land carbon fluxes, in accordance with previous findings (Ahlström et al., 2015). Interestingly, the two gridded products show slightly different climatic location for the peak in relative IAV, with MPI-MTE pointing to warm arid regions, whereas the Jena Inversion points to cold arid systems.
The dependency of IAV
The importance of the spatial scale of analysis for the IAV
Maps of the fraction of NEE variance explained by temporal trends
and anomalies for MPI-MTE NEE and the Jena Inversion; latitudinal band
(15
Boxplot of NEE inter-annual variability averaged in PFT classes for
MPI-MTE NEE and the Jena Inversion; green dots represent observations at FLUXNET
sites (different
Climatic drivers of the spatial variability in NEE interannual
variability. The panels
The fractions of IAV
For the two gridded products the analysis of IAV (either in terms of absolute
IAV
The climatic dependence of the spatial variability in IAV
The climate dependencies of IAV are further separated between the variability
due to trends and anomalies (Fig. 7b, d). The two components of
IAV
An improvement in the mechanistic understanding of IAV
To investigate how good these proxies are, the ratios TER
Bar plot of the TER
Control on IAV by GPP-TER and CUP-CRP, expressed as the difference
of the determination coefficients (
Control on IAV by GPP-TER and CUP-CRP, expressed as the difference of the determination coefficients plotted in a temperature–precipitation space. The two top panels refer to MPI-MTE, while the bottom panel refers to the Jena Inversion; dots refer to FLUXNET sites.
Comparison of the annual NEE anomalies between the Jena Inversion,
MPI-MTE and the Global Carbon Project data. Time series of annual anomalies
are shown in
In order to identify which of the gross fluxes controls the variability in
the net land flux, we assessed the fraction of variance (
Finally, in order to place our analysis in a broader context, global annual
values of the gridded products used in the present analysis have been
compared with the estimates of the Global Carbon Project (Fig. 11). On an
annual timescale, the Jena Inversion shows excellent agreement with the GCP,
and this is not surprising since GCP land fluxes are estimated as a residual
term from the atmospheric
In conclusion, this study assessed the temporal variability in the terrestrial C budget with three different datasets to diagnose common patterns and emerging features. Some discrepancies between data products have emerged, in particular in the tropics, where a chronic deficiency of atmospheric and ecosystem observations severely limits the accuracy of large-scale assessments. On the other hand, several important global features have been identified and confirmed by the different products, like (i) the dominant role played by photosynthesis in the short-term variability in the land carbon budget, (ii) the high relative IAV in water-limited ecosystems, and (iii) the dependence of IAV on spatial scales and ecosystem productivity. Ultimately, the variability in the land fluxes observed in the recent decades proved to be extremely valuable to investigate the controlling mechanisms and the sensitivity and vulnerability of the terrestrial C balance to climate drivers.
All the datasets used for this analysis are available for
download via the following links. Ecosystem carbon fluxes: FLUXNET La
Thuile dataset (Baldocchi et al., 2001;
The authors declare that they have no conflict of interest.
This study was supported by the JRC project AgForCC no. 442. The MCD12C1 product was retrieved from the online data pool, courtesy of the NASA EOSDIS Land Processes Distributed Active Archive Center (LP DAAC), USGS/Earth Resources Observation and Science (EROS) Center, Sioux Falls, South Dakota. This work used eddy covariance data acquired and shared by the FLUXNET community, including these networks: AmeriFlux, AfriFlux, AsiaFlux, CarboAfrica, CarboEuropeIP, CarboItaly, CarboMont, ChinaFlux, FLUXNET-Canada, GreenGrass, ICOS, KoFlux, LBA, NECC, OzFlux-TERN, TCOS-Siberia, and USCCC. The FLUXNET eddy covariance data processing and harmonization was carried out by the ICOS Ecosystem Thematic Center, AmeriFlux Management Project and Fluxdata project of FLUXNET, with the support of CDIAC, and the OzFlux, ChinaFlux and AsiaFlux offices. Edited by: Alexey V. Eliseev Reviewed by: Nir Krakauer and two anonymous referees