Journal of Geophysical Research: Biogeosciences

Nitrogen controls plant canopy light-use efficiency in temperate and boreal ecosystems



[1] Optimum daily light-use efficiency (LUE) and normalized canopy photosynthesis (GEE*) rate, a proxy for LUE, have been derived from eddy covariance CO2 flux measurements obtained at a range of sites located in the mid to high latitudes. These two variables were analyzed with respect to environmental conditions, plant functional types (PFT) and leaf nitrogen concentration, in an attempt to characterize their variability and their potential drivers. LUE averaged 0.0182 mol/mol with a coefficient of variation of 37% (42% for GEE*). Foliar nitrogen N of the dominant plant species was found to explain 71% of LUE (n = 26) and 62% of GEE* (n = 44) variance, across all PFTs and sites. Mean Annual Temperature, MAT, explained 27% of LUE variance, and the two factors (MAT and N) combined in a simple linear model explain 80% of LUE and 76% GEE* variance. These results showed that plant canopies in the temperate, boreal and arctic zones fit into a general scheme closely related to the one, which had been established for plant leaves worldwide. The N-MAT-LUE relationships offer perspectives for LUE-based models of terrestrial photosynthesis based on remote sensing. On a continental scale, the decrease of LUE from the temperate to the arctic zone found in the data derived from flux measurements is not in line with LUE resulting from inversion of atmospheric CO2.

1. Introduction

[2] Canopy light-use efficiency (LUE), defined as either the ratio of gross (GPP) or net primary productivity (NPP) to absorbed light has received increasing attention over the last decades, primarily because the combination of remotely-sensed absorbed photosynthetically active radiation (APAR) and estimates of LUE allows investigations of GPP and NPP over large areas. Since the launch of the NOAA-TIROS satellites in 1981 and the consequent development of algorithms to derive APAR from space, LUE-based approaches have become a widely applied tool [Prince, 1991; Potter et al., 1993; Ruimy et al., 1994; Field et al., 1995; Running et al., 2004] and LUE-based productivity models have greatly contributed to the characterization of the temporal variability of global-scale terrestrial productivity [e.g., Nemani et al., 2003]. At a smaller scale, productivity models based on LUE, often called Production Efficiency Models (PEM), have been developed for a range of different ecosystems. Such models take advantage of field data sets of productivity and biomass when details of physiology or ecology are not known [e.g., Prince, 1991; Medlyn, 1998; Mäkelä et al., 2008]. Both, global and ecosystem models depend on accurate estimates of LUE.

[3] Because it is the ratio of two key physiological properties (light capture and photosynthesis), LUE subsumes a broad range of processes and has also been applied as an integrative diagnostic tool. As such it has been used, for instance, to analyze and intercompare output from ecosystem models that differ in their complexity, their parameterizations and/or their representation of processes. An example was provided by the intercomparison of global models of NPP [Ruimy et al., 1999], which demonstrated that grid-cell light-absorption and NPP were highly correlated for ten out of twelve global models, the two “exceptions” being models that predicted NPP on the basis of the nitrogen cycle. Ruimy et al. [1999] also drew attention to the large differences among average LUE values from the ten LUE-like global models, highlighting the need for accurate estimates of large-scale LUE. In addition, their results indicated that the relationship between LUE and the nitrogen cycle still has to be clarified at the global scale. The importance of accurate LUE values is also exemplified by atmospheric inversion studies. Such studies infer the surface sinks and sources of CO2 from atmospheric measurements and transport models. LUE-based estimates of productivity are often used as a constraint in or as an end-product of the inversion process [Randerson et al., 2002; Kaminski et al., 2002; Still et al., 2004; Chevallier et al., 2005, among others].

[4] Originally, field studies suggested LUE being rather invariable among different, well-watered crops [Monteith, 1977] but later reviews by Prince [1991], Ruimy et al. [1994], and Medlyn [1998] have demonstrated significant variation among vegetation types at least for LUE derived from NPP (LUENPP). Part of this variation may be related to measurement aspects as neither NPP nor APAR are easy to capture precisely, especially across sites and across Plant Functional Types (PFT) [Gower et al., 1999], but there is little doubt overall that the assumption of a constant LUENPP does not provide an accurate description of terrestrial ecosystems [Binkley et al., 2004; Bradford et al., 2005].

[5] From a physiological perspective some authors have argued that LUE derived from GPP (LUEGPP) should be less variable than LUENPP, mainly because differences in carbon allocation and respiration estimates are responsible for some of the variability in LUENPP and should not affect LUEGPP [Ruimy et al., 1996a; Goetz and Prince, 1999]. The first analyses of LUEGPP led however to a somewhat contradictory picture with the ratio either being more or less constant across ecosystems (ca. 0.02 mol CO2/mol APAR, Ruimy et al. [1995]) or varying widely [Turner et al., 2003, 2005]. Intuitively a certain fluctuation of LUEGPP would be expected as GPP varies not only with APAR but also with other factors, e.g., soil water and nutrient availability, the ratio of direct to diffuse radiation, canopy age or site history [Alton et al., 2007; DeLucia et al., 2007]. Yet, when investigating LUE over the course of one to several years, water and nutrient supply will be reflected to some extent in the canopy leaf area index, and therefore in the absorbed PAR. How LUE and the fraction of absorbed PAR (fPAR) are related to environmental constraints is largely unknown. Whether light capture and light-use efficiency show coordinated responses to environmental constraints has received theoretical interest [Field et al., 1995; Goetz and Prince, 1999], but little observational support so far, and the opposite view of light-use efficiency increasing with decreasing light availability has also been considered [Binkley et al., 2004].

[6] In addition to the question on the range and variability of LUE there is also a debate on its global patterns. When plotted against latitude LUE increased toward the north for some global models, but decreased for others [Ruimy et al., 1999], illustrating the lack of consensus on the underlying processes. Kaminski et al. [2002] and Still et al. [2004] showed that large-scale CO2 inversion studies tend to impose a large increase of LUE from temperate to arctic ecosystems. In terms of modeling, Haxeltine and Prentice [1996] suggested that a pole-ward increasing trend in plant nitrogen content could support higher LUE at higher latitudes. Conversely, on the basis of CO2 flux data over boreal sites, Lafont et al. [2002] found a correlation between mean annual temperature and LUE, which leads to a decrease in LUE toward high northern latitudes. This result was supported by the analysis of Schwalm et al. [2006] who, in addition, did not detect any significant correlation with foliar nitrogen based on 11 flux measurement sites.

[7] Clearly, for increased confidence in satellite-derived productivity estimates as well as to offer better diagnostics to large-scale ecology, it is important to reduce the uncertainties affecting large-scale LUE patterns and to identify the relevant drivers for its variation. The large number of net CO2 flux data that are by now becoming available give an unprecedented access to estimates of ecosystem GPP, owing to effort in collecting and processing data in networks like FLUXNET [Baldocchi et al., 2001; Friend et al., 2007; Owen et al., 2007]. For this study, we derive LUE from CO2 flux time-series to estimate LUEGPP over a variety of sites spanning the temperate, boreal and arctic ecosystems. The questions we address are: (1) Is optimum LUEGPP (when GPP achieves seasonal maxima) variable among these ecosystems? (2) If yes, what are the large-scale and local-scale patterns behind LUE variability? (3) What are the major controls of spatial LUE variation, and how can it be parameterized?

2. Data and Methods

2.1. Sites Characteristics

[8] CO2 fluxes based on the eddy covariance technique have been compiled from the FLUXNET, EUROFLUX, AMERIFLUX, BOREAS and EUROSIBERIAN CARBONFLUX databases [Baldocchi et al., 2001; Heimann, 2002; Sellers et al., 1995; Valentini et al., 2003] as well as from studies that had been conducted independently of these data sets. The emphasis has been put on mid to high latitude sites. Site descriptions and references are summarized in Table 1. As a result of the considerable efforts of the participants to these projects, there is a relatively high degree of homogeneity in the methods and algorithms used at different sites. To take advantage of this effort, we considered only fluxes measured with eddy covariance methods, and did not retain for example, fluxes measured and up-scaled from chamber techniques.

Table 1. List of Sites and References and Derived Informationa
ReferencesNameSpecies or Canopy TypePFTLatMAT (°C)LAIFPARN (%)LUE (mol/mol)GEE* (μmol m−2 s−1)
  • a

    The first 42 rows contain type I data (see text) sorted by latitude, the 35 rows below are for type II data. PFT is for plant functional type, Lat is latitude, MAT is mean annual temperature, LAI is leaf area index, fPAR is the fraction of absorbed PAR, N is for leaf nitrogen on a mass basis, LUE is optimum daily light-use efficiency, and GEE* is normalized canopy photosynthesis. PFT are indicated by the following symbols: equation image, evergreen needleleaf; △, deciduous needleleaf; •, evergreen broadleaf; ○, deciduous broadleaf; equation image, mixed forest, ▪ tundra or boreal wetland, *, C4 grass or crop; +, C3 grass or crop.

  • b

    Data from FLUXNET web site.

  • c

    Data from CARBODATA CD.

  • d

    Data from EUROSIB project database.

  • e

    Values from Lindroth et al. [1998].

  • f

    Data from BOREAS CD (revised 2004).

Harazono et al. [2003]BarrowbWet sedge tundra70.3−
Oechel et al. [2000]UpadbWet sedge tundra70.3−
Vourlitis and Oechel [1999]Happy valleybMoist tussock tundra69.0−
Wallin et al. [2001]FlakalidencNorway spruceequation image64.
Markkanen et al. [2001]HyytialacScots pinesequation image61.
Lloyd et al. [2002]ZotinodScots pineequation image61.0−1.5-0.700.990.012917.7
Lohila et al. [2004]Joikonnen BarleyBarley+60.93.950.913.620.02635.2
Lindroth et al. [1998]Norundac,eSpruce pine60.
Milyukova et al. [2002]FyodorovskoyedNorway spruce56.
Goulden et al. [1997]BN-OBSfOld black spruce56.0−2.9-0.860.660.012815.6
McCaughey et al. [1997]BN-YJPfYoung jack pine56.0−2.9-0.650.930.011418.4
Moore et al. [2000]BN-OJPfOld jack pine56.0−2.9-0.760.900.006815.0
Pilegaard et al. [2001]SoroecBeech forest55.
Yang et al. [1999]BS-OAfOld aspen54.01.0-0.902.500.018930.3
Jarvis et al. [1997]BS-OBSfOld black spruce54.01.0-0.850.700.013119.5
Baldocchi et al. [1997]BS-OJPfOld jack pine54.01.0-0.781.000.010621.8
Dolman et al. [2002]LooboscScots pine52.
Kowalski et al. [2000]BraschaatcScots pineequation image51.310.23.10.80-0.017026.7
Grünwald and Bernhofer [2007]TharandtcNorway spruce50.
Aubinet et al. [2001]VielsalmcFir spruce pine beechequation image50.
Klemm and Mangold [2001]WeidenbrunnencNorway spruceequation image50.
Granier et al. [2000]HessecBeech48.
Davis et al. [2003]Park FallbMaple aspen pineequation image45.
Hollinger et al. [1999]HowlandbSpruce dominatedequation image45.
Berbigier et al. [2001]LandescMaritime pine44.712.53.10.851.100.017627.2
Law et al. [2001]Metolius-oldbPonderosa pineequation image44.
Nakai et al. [2003]Japan forestbBirch oak43.
Valentini et al. [1996]CollelongocBeech42.
Wofsy et al. [1993]Harvard forestbOak maple hemlock42.
Reichstein et al. [2002]CastelporzianocQuercus ilex42.
Turner et al. [2003]BondvillebCorn*
Turner et al. [2003]BondvillebSoybean+
Monson et al. [2002]NiwotbFir spruce pine40.
Goldstein et al. [2000]BlodgettbPonderosa pine39.
Suyker and Verma [2001]ShidlerbTallgrass prairie*
Hanan et al. [2002]Ponca CitybWheat+
Wilson and Baldocchi [2000]Walker BranchbOak maple36.
Katul et al. [2001]DukebPine36.
Meyers [2001]Little WashitabGrassland*
Stylinski et al. [2002]Sky Oaks-youngbChaparral33.
Stylinski et al. [2002]Sky Oaks-oldbChaparral33.
Campbell et al. [2001]RiceRice+
Lloyd [2001]Ny Alesund,Tundra, semi-desert80.0−
Nordstoem et al. [2001]Zackenberg fenFen74.5−
Soegaard et al. [2000]Zackenberg heathHeath tundra74.5−
Laurila et al. [2001]Zackenberg willowWillow tundra74.5−
Williams et al. [2000]AFS 1Coastal wet tundra70.5−8.2-0.571.14-6.7
Williams et al. [2000]AFS 13Alder tussock tundra70.5−8.2-0.551.0-12.7
Williams et al. [2000]AFS 14Acidic tussock tundra70.5−8.2-0.501.0-12.5
Williams et al. [2000]AFS 3Non-acidic tundra70.5−8.2-0.750.87-6.2
Williams et al. [2000]AFS 4Acidic tussock tundra70.5−8.2-0.830.78-8.0
Williams et al. [2000]AFS 6Acidic tussock tundra70.5−8.2-0.591.0-16.5
Williams et al. [2000]AFS 9Shrub tundra70.5−8.2-0.522.05-30.8
Williams et al. [2000]AFS heathHeath tundra70.5−8.2-0.650.94-6.6
Laurila et al. [2001]KevoFen69.8−
Aurela et al. [2001]PetsikkoMountain birch69.5−
Laurila et al. [2001]KaamanenFen69.1−
Aradóttir et al. [1997]GunnarsholtcBlack cottonwood63.
Röser et al. [2002]Zotino-birchdBirch61.0−
Röser et al. [2002]Zotino-mixeddFir spruce birchequation image61.0−
Röser et al. [2002]Zotino-poledFir61.0−
Lohila et al. [2004]Joikonnen GrassGrass+
Hollinger et al. [1998]YakutskLarch60.0−
Griffis and Rouse [2001]ChurchillSubarctic fen58.8−6.90.6 --29.1
Clement et al. [2003]AberfeldycSitka spruce56.58.080.94--30.9
Suyker et al. [1997]BS-FENfBoreal fen54.
Anderson et al. [1995]BS-YJPfYoung jack pine54.01.0-0.691.10-20.8
Fan et al. [1995]LabradorBlack spruceequation image53.8−4.9-0.650.59-21.5
Flanagan et al. [2002]LethbridgebGrassland*
Chen et al. [2002]Wind cranebFir hemlockequation image46.08.710.00.95--20.0
Lafleur et al. [2001]Mer bleueOmbrotrophic bog45.
Lee et al. [1999]Camp BordenMaple aspen ash44.
Hirano et al. [2003]Japan LarchLarch42.
Schmid et al. [2000]MorganMaple tulip poplar oak39.311.13.40.79--40.7
Hollinger et al. [1994]MaruiaNothofagus−
Arneth et al. [1998]New ZealandPinus radiataequation image−42.810.87.00.951.30-31.0
Hunt et al. [2002]NZ tussockTussock grassland+−

[9] For each site, canopy leaf area index (LAI; projected leaf area basis and usually including understorey vegetation) or fPAR (the fraction of PAR absorbed), and mean annual temperature (MAT) data were compiled (Table 1). MAT was either provided by the database and reference articles or derived from climatology, using the gridpoint closest to the site [Leemans and Cramer, 1991, updated 1995]. The vegetation at the sites was classified into the following Plant Functional Types: Evergreen needleleaf, evergreen broadleaf, deciduous needleleaf, deciduous broadleaf, mixed forest, tundra and boreal wetlands, C4 grasses and crops, and C3 grasses and crops. Databases and literature were screened for site leaf or needle nitrogen content expressed on a mass basis (gN/g dry matter, hereafter N). For most sites, only the dominant species have been sampled for N, with a few exceptions in herbaceous canopies, which provide canopy-average nitrogen. For evergreen plants, most studies provided an average over the different needle or leaf age-classes. When seasonal course of nitrogen content was available, we retained the values closest to the date of maximum CO2 flux. Leaf nitrogen was not corrected for sugar content.

2.2. Derivation of Light-Use Efficiency and Normalized Photosynthetic Rate

[10] From the CO2 flux time series two variables were derived: Optimum daily photosynthetic light-use efficiency (hereafter simply noted LUE), and a proxy for LUE which is the normalized maximum photosynthesis rate (or Gross Ecosystem Exchange) noted GEE*. The CO2 flux data compiled in this study are of two types (noted I and II), for which different methods had to be used.

2.2.1. Type I Data Set

[11] For 42 data sets, typically one year or longer, for which flux data were available, time-series of half-hourly GEE were derived from NEE and an estimate of ecosystem respiration Reco. As used here, GEE is considered positive, whereas NEE and Reco follow the classical micrometeorological conventions, being positive when the CO2 flux is upward.

equation image

[12] For the long-term comprehensive data it was possible to estimate Reco using two different methods, including a simple one, which can also be applied for the less comprehensive type II data (see below). The first method estimates Reco from soil temperature using an Arrhenius-type relationship with parameters that may vary seasonally. Nighttime flux data were selected above wind speed and/or friction velocity thresholds before fitting equation (2) to soil temperature (5 to 10 cm) for consecutive two-month periods of time [e.g., Falge et al., 2001]. This allows to account for seasonal variations in plant phenology, water budget and microbial processes.

equation image

where T0 is the reference soil temperature (283.16 K), F0 is the fitted respiration rate at reference temperature, Ea is the fitted activation energy and R is the gas constant.

[13] To obtain daily totals, Reco was extrapolated during daytime periods based on soil temperature and GEE computed accordingly.

[14] A second, simpler method estimates ecosystem respiration rate as the average of nighttime fluxes (i.e., period when global radiation Rg < 10 W/m2 over a 24 h period of time, equation (3))

equation image

where brackets indicate averaging. Half-hourly GEE were computed using a constant value for half-hourly Reco over a given day. This simple method assumes that the differences between nighttime-average and mid-day ecosystem respiration are small compared to seasonal and inter-sites differences, which is applicable in ecosystems where respiration does not respond strongly to rainfall events.

[15] Once time-series of GEE had been derived, the maximum canopy photosynthesis, GEEmax, was computed as the average of the upper 98.5–99.5% bin of the half-hourly GEE histogram. These 98.5 and 99.5 thresholds were defined to retain photosynthesis rates typical of optimal environmental conditions, while discarding outliers. All days with at least one half-hourly GEE value falling in the 98.5–99.5% interval were considered “optimal” in terms of environmental conditions. For these days, 24h totals (GEE24h, PAR24h) were derived from half-hourly GEE and PAR. An optimum daily light-use factor (LUF), on the basis of incident PAR, was derived from the slope of the linear relationship between GEE24h and PAR24h. To avoid circular analysis no gap filled data sets were used since light is used to fill gaps in NEE and GEE time series.

equation image

To account for differences in PAR absorption due to canopy openness, daily LUE was computed as:

equation image

Depending on the site, fPAR was either obtained from literature and database sources or derived from projected LAI using:

equation image

with k = 0.5 and Θs being the sun elevation at solstice.

2.2.2. Type II Data

[16] For the additional 35 type II data sets, which either were short-term studies or not included in the above databases, maximum rates of canopy photosynthesis GEEmax were derived by averaging 3 consecutive half-hourly CO2 flux values during optimal environmental conditions. Averaging consecutive data prevents overestimation of GEEmax created by statistical variation sometimes present in the half-hourly eddy flux data. A few studies directly provide GEE time series either from temperature driven Reco, or from the intercept of light/NEE curves, otherwise, Reco was estimated with equation (3).

[17] For comparison among sites with different leaf area index, GEEmax was normalized by fPAR, using a reference fPARc of 0.95 corresponding to a closed canopy. Additionally, GEEmax was normalized by the cosine of Θs, to compensate for difference in incident PAR caused by latitude, assuming optimum conditions for CO2 flux occurring near the solstice.

equation image

GEE* is expected to be a good proxy for LUE in type II data sets since both variables share the same normalization by absorbed PAR, although simplified in the case of GEE*, providing that daily integrated GPP and daily maximum GPP values are related.

2.2.3. Sensitivity Tests

[18] To evaluate the robustness of equations (5) and (7) to fPAR (equation (6)), we tested the sensitivity of LUE and GEE* to three fPAR estimates. Firstly, we assumed that the period of maximum GEE departs from solstice depending on latitude, ranging from day of year 180 at 45° to 220 at 80° [Falge et al., 2002]. Θs is then replaced by the sun elevation at 12 h (local solar time) for the corresponding day of year. Secondly, fPAR was assumed to be a linear mixture of fPAR for direct (equation (6)) and diffuse irradiance:

equation image

assuming a fraction of direct PAR, λ, of 0.65. Lastly, we tested the hypothesis that the extinction coefficient k is PFT dependent [e.g., Yuan et al., 2007].

equation image

assuming kpft = 0.45 for needleleaf, 0.55 for broadleaf, and 0.5 for other canopies instead of using k = 0.5 for all sites.

3. Results and Discussion

[19] The consistency of the GEE* estimation methods was first evaluated using the 42 type I data set (Table 1). GEE* derived from either equation (2) or (3) were similar (r2 = 0.93, slope 1.02, intercept −0.3 μmol s−1 m−2). A similar picture was found for LUE data based on equations (2) and (3) (r2 = 0.87, slope 0.93, intercept −0.0007 mol/mol). Therefore, in the following, we choose to use equation (3), which allowed us to merge type I and type II into a homogeneous GEE* data set. For consistency, we consequently used equation (3) also for derivation of LUE.

[20] GEEmax computed with equation (3) were well correlated (r2 = 0.81) with the maximum GEE derived from 25 FLUXNET sites by Falge et al. [2002]. Our values were slightly higher since Falge et al. [2002] used a smoothing over 15 days to compute the seasonal course of “all weather” GEE whereas we are interested here in optimum half-hourly values. The general agreement between our GEEmax and the one given by Falge et al. [2002], on the basis of a detailed examination of the data and methods, supports our simplified data processing.

3.1. Analysis of LUE and GEE* Variability

[21] As expected LUE and GEE* from type I data were highly correlated (Figure 1) with a r2 of 0.88 (n = 42). Or in other words, GEE* can be used as a proxy for LUE for type II data set. Since the full data set for GEE* spans a larger range of ecosystems, latitudes, soil and environmental conditions, especially in the harsh arctic environment, it provides more robust statistics than an analysis restricted to long term comprehensive flux data. Therefore, in the following, results are shown for both LUE and GEE*. For the purpose of clarity, plots showing GEE* have a second y-axis (on the right) that maps GEE* into LUE units based on the linear relationship in Figure 1b (LUE = 0.0006 GEE* + 0.0023).

Figure 1.

(a) Histogram of optimum daily light-use efficiencies (LUE) for the 42 sites with type I data. (b) LUE versus canopy normalized photosynthesis GEE* for type I data (n = 42).

[22] The histogram of LUE (Figure 1a) reveals a significant variability in among-site LUE, with a mean of 0.0182 mol/mol, and a standard deviation of 0.0067 resulting in a coefficient of variation of 37% (n = 42, Table 2). Considering the range of ecosystems included, it is not surprising that GEE*, derived for the entire data set in Table 1 was also highly variable, averaged at 26.2 μmol m−2 s−1, with a standard deviation of 11 μmol m−2 s−1 and a coefficient of variation of 42% (n = 77).

Table 2. Optimum Daily Light-Use Efficiency (LUE, mol/mol) and Canopy Normalized Photosynthesis (GEE*, μmol m−2 s−1): Average, Coefficient of Variation (CV) and Number of Data, Grouped by Plant Functional Types
Deciduous broadleaved0.022510 (n = 7)34.514 (n = 13)
Evergreen broadleaved0.014443 (n = 3)22.524 (n = 4)
Mixed forests0.018717 (n = 3)27.210 (n = 4)
Evergreen needleleaved0.015532 (n = 19)23.624 (n = 25)
Deciduous needleleaved--21.771 (n = 2)
Tundra, wetlands0.011641 (n = 3)18.048 (n = 19)
C3 grasses and crops0.027042 (n = 3)35.851 (n = 5)
C4 grasses and crops0.024528 (n = 4)43.627 (n = 5)
All grasses and crops0.025632 (n = 7)39.738 (n = 10)
All plant types0.018237 (n = 42)26.242 (n = 77)

[23] The canopy LUE values derived here are naturally lower than commonly measured leaf-level quantum yields which may be of the order of 0.06 under normal environmental conditions [e.g., Ehleringer and Pearcy, 1983]. Several reasons underlie the leaf vs. canopy difference: Photosynthesis of leaves saturates in high irradiance, thereby decreasing canopy daily LUE, which is integrated over the course of a day and includes periods of high and low irradiance. Moreover, although we select the optimum seasonal LUE value, environmental conditions still impose some stress on the plants. Highest canopy LUE were obtained for two crops in the data set, rice and corn (Table 1), which show almost no saturation of leaf level GEE/PAR curves at the half hourly time step, despite rice being a C3 plant. The average value of 0.018 is close to but lower than the value of 0.02 proposed by Ruimy et al. [1995, 1996b].

[24] For further illustration, Figure 2a shows LUE and GEE* as functions of MAT at the site separated per PFT. MAT correlated weakly but significantly with LUE (Figure 2a, r2 = 0.27, n = 42, P = 0.0004) and GEE* (Figure 2b, r2 = 0.34, n = 77, P < 10e-7). The sensitivity tests performed with the different formulations of fPAR and APAR (equations (8) and (9)) did not affect these results (not shown). Since PAR irradiance tends to decrease from temperate to arctic latitudes, the trend of maximum GPP versus MAT is obviously stronger than GEE* (not shown).

Figure 2a.

Optimum daily light-use efficiency (LUE, mol/mol) versus annual mean temperature (MAT) for type I data. Symbols are for ecosystem type (Table 1) as: Evergreen needleleaf (▴), evergreen broadleaf (•), deciduous needleleaf (△), deciduous broadleaf (○), mixed forest (equation image), tundra and boreal wetlands (▪), C4 grasses and crops (*), and C3 grasses and crops (+).

Figure 2b.

Same as Figure 2a, but for canopy normalized photosynthesis (GEE*, μmol s−1 m−2) for type I and II data. Right y-axis is a linear mapping to LUE units (mol/mol) (see Figure 1).

[25] LUE was not to be expected to be a simple function of MAT because of a range of additional factors. In continental high latitude ecosystems photosynthesis takes place over a short (2–3 months) and well defined period with sometimes quite warm temperatures [e.g., Lloyd et al., 2002; Arneth et al., 2002]. However, the correlation almost disappears when MAT is replaced by the temperature corresponding to the period of CO2 flux data retained to compute LUE and GEE* (r2 = 0.12, P = 0.06 and r2 = 0.19, P = 0.01 respectively). What is more, effects of physiology may override those of temperature with, for instance, LUE of a well-fertilized, C4 crop is expected to be higher than that of a “natural” system at similar MAT. Nevertheless, LUE and GEE* tended to be organized along a MAT gradient (Figures 2a and 2b), with a clear tendency to form clusters for some of the plant types like deciduous broadleaved forests, whereas for most of the other types a large variability was found (e.g., needleleaved forests, grasses and crops, and tundra/wetlands). The better correlation with GEE* is caused by a broader sampling of the possible range of MAT, including a number of sites with low annual temperatures (MAT < 0). The correlation of LUE and GEE* with MAT we find here is lower than the values reported by Lafont et al. [2002] for 18 boreal sites and closer to results reported by Schwalm et al. [2006] for yearly mean LUE at 17 Canadian forest and wetland sites. The broader range of PFT included here, especially inclusion of warmer broadleaved evergreen, grasses, crops on the one hand and cold climate fens and deciduous forests on the other hand explains the difference with Lafont et al. [2002]. These PFT add to the scattering of the GEE*/MAT relationship (Figure 2b and Table 2). Averaged per PFT, LUE ranges from 0.0116 for tundra and wetlands to 0.0270 for C3 grasses and crops (Table 2).

3.2. Role of Leaf Nitrogen Content

[26] At the leaf level, numerous studies have demonstrated a strong link between nitrogen content and photosynthesis [Field and Mooney, 1986; Wullschleger, 1993]. A general framework for leaf structure and function, relating leaf assimilation rate, leaf nitrogen content and leaf mass per area (LMA) has been proposed by Reich et al. [1997] and extended worldwide by Wright et al. [2004]: Leaf photosynthesis (on a mass basis) correlates positively with leaf nitrogen (on a mass basis) and negatively with LMA. At the canopy level, several studies point toward a similarly strong incidence of nitrogen on photosynthesis, productivity and even net CO2 flux [e.g., Choudhury, 2001; Williams et al., 2000; Smith et al., 2002]. However, Schwalm et al. [2006] did not find any significant correlation between foliar nitrogen and canopy LUE across 9 forest sites. How leaf-level relations translate to the canopy level therefore remained elusive so far.

[27] When plotted against leaf nitrogen content expressed on a mass basis, both LUE and GEE* significantly increased with N (Figures 3a and 3b). The variance explained by N, in a single variable 1:1 linear relationship, reaches values of r2 = 0.71 (n = 26, all type I sites with nitrogen data) and r2 = 0.62 (n = 44, all sites with nitrogen data) for LUE and GEE* respectively. It can be seen from Figure 3a that the increase of LUE with N becomes less well defined at leaf N levels > ca. 2% for deciduous broadleaf forest sites, suggesting a curvilinear relationship might exist for some biomes. The C4 crops and grasses achieved high LUE and GEE* at relatively low levels of leaf N.

Figure 3a.

Optimum daily LUE versus leaf nitrogen content N, mass basis.

Figure 3b.

Same as Figure 3a, but for canopy normalized photosynthesis (GEE*).

[28] Table 1 and Figures 3a and 3b show the large differences in leaf N that exist among, and also within, plant types. Still, LUE and GEE* tended to group in well-defined PFT-clusters. In that respect, canopy-level data behave like the leaf-level data presented by Reich et al. [1997], which also show such PFT clusters.

3.3. Combination of MAT and N

[29] MAT and leaf N were not related in the data set analyzed here (Figure 4), except for a weak tendency of evergreen tree N to increase with temperature and for deciduous trees and the two crop sites to have higher leaf N across the entire range of MAT when compared with other PFTs.

Figure 4.

Leaf nitrogen N versus annual mean temperature.

[30] However, the residuals of the linear regression between LUE and N (Figure 3a) were weakly correlated with MAT (not shown) which implies that the N:LUE relationships were not completely independent of climate. At same levels of leaf nitrogen, highest LUE were thus observed at warmest temperature. This was true also for the N vs. GEE* residuals. Consequently the combination of N and MAT explained a larger part of LUE and GEE* variability and a simple linear model LUE = a MAT + b N + c was therefore fitted to the data (Figure 5a, same for GEE* Figure 5b).

Figure 5a.

Optimum daily LUE versus linear combination of annual mean temperature and leaf nitrogen N.

Figure 5b.

GEE* versus linear combination of annual mean temperature and leaf nitrogen N.

[31] This simple model explained 80% of LUE variance (LUE = 0.0063 N + 0.00036 MAT + 0.0064, n = 26) and 76% of GEE* variance (GEE* = 10.85 N + 0.66 MAT + 8.41, n = 44). These regression coefficients are quite high, considering that the derivation of LUE and GEE* are affected by several approximations and uncertainties (fPAR estimates at low LAI for instance).

[32] Differences in leaf nitrogen reflect site quality combined with plant type. As opposed to foliar N, which is central to plant photosynthesis and physiology, MAT is a surrogate for a number of variables and processes in interaction, like the length of the growing season, the nutrient cycle, the water budget, to mention only three. Therefore the relationship of LUE with MAT is best viewed as a large-scale pattern that subsumes these effects and that is locally supplanted by the role of PFT, leaf nitrogen, and other factors. A good example comes from the BOREAS data, where stands of aspen, Jack pines and black spruce co-exist at the landscape scale but show dramatically different LUE and leaf N, sometimes even within the same stand type [O'Connell et al., 2003]. At the regional scale, averages of the LUE, weighted by the relative surface of the different PFT in BOREAS Northern and Southern Sites, tend to fall into the general large-scale MAT/LUE gradient (not shown). Large-scale patterns of LUE therefore depend on the relative surfaces occupied by different PFT or plant differing in N content, in line with the findings of Still et al. [2004] of higher LUE in Eurasia than North America due to higher deciduous trees cover, combined to the overall environmental conditions correlated to MAT.

3.4. Canopy Index and Function

[33] To make use of the explanatory power of LMA, in addition to foliar nitrogen, it is tempting to scale the leaf level relationship of Reich et al. [1997] up to the canopy level. This was done by Green et al. [2003], who proposed a “Canopy Index”, as a combination of canopy nitrogen, LMA and fPAR. Such a canopy index is theoretically related to canopy LUE (equation (10), see Appendix A). Green et al. [2003] further expressed LUE in terms of LAI, fPAR and leaf nitrogen (see Appendix A for equations and suggestions on alternative index derivation)

equation image

[34] When applied to our data set, the canopy index (hereafter noted IG) was found to explain a similar variance of LUE than foliar nitrogen alone (see Figures 6a and 6b). This was the case for both the percentage of variance explained as well as for the complementarities between IG and MAT. The comparison of Figures 6 and 3 confirms the analysis by Green et al. [2003] such that a canopy index increases the linearity of LUE prediction compared to leaf nitrogen alone. However, the scatter also increased in our data and the overall predictive power did not improve. Particularly the C4 sites stand out as outliers in the IG index/GEE* relationship. Alternative formulations were developed and tested, that accounted for both, a non-linear relationship between leaf photosynthesis and leaf N, as well as for the differences in incoming PAR to refine the formulation of Green et al. [2003]. This revised index (N0.77 LAI/fPAR.1/Ipar; see Appendix A) still did not improve the overall r2 of the relationship shown in Figure 6, suggesting that the scattering of canopy index/LUE or GEE* may come from assumptions in scaling leaf level to canopy level, possibly in averaging LMA or leaf properties over the canopy. Given the limited number of sites and the approximations in fPAR derivation, it is at this stage difficult to further evaluate the appropriateness of the different mass-based canopy indices. Indeed, canopy structure, foliage clumping, non-leaf tissues can make the derivation of the factor LAI/fPAR particularly difficult, especially at the scale of a flux tower footprint.

Figure 6a.

LUE versus the canopy index IG of Green et al. [2003].

Figure 6b.

Normalized canopy photosynthesis (GEE*) versus Green et al. [2003] canopy index IG.

[35] Whether a unique nitrogen-based relationship can be applied to estimate LUE for different plant types is central to both application of remote sensing data and understanding and interpreting the observed LUE patterns. Leaf N concentration may be a better predictor of productivity than total N integrated over the canopy as soon as different plant types are considered. For instance, Smith et al. [2002] were able to relate aboveground NPP to canopy averaged N concentration for deciduous and evergreen tree species using the same relationship. Conversely, different relationships for deciduous and evergreen trees emerged when canopy total N was used. Considering trees, grasses and crops, Green et al. [2003] drew a similar conclusion, favoring N concentration rather than canopy total N. Similarly, expressing leaf nitrogen on a mass basis rather than on a surface basis may also seem counter-intuitive (see also discussion by Smith et al. [2002] and Hikosaka [2004]). From a practical point of view, foliar nitrogen content expressed on a mass basis is far less variable within the canopy than area basis nitrogen, and therefore easier to measure. As a consequence, mass basis N or indices based on mass basis N prove more convenient to evaluate in the field and can provide general framework for canopy LUE variability when addressing global or regional issues. So far our own, and other published evidence indicate that both canopy N concentration or mass-basis leaf N may be robust predictors of canopy functioning in terms of light use.

3.5. Idealized Canopy Models

[36] Considerations about the optimum functioning of canopies have suggested that the capture of different resources, like PAR, water, and nutrients could be regarded in an integrated fashion. Such an integrated perspective would ensure coordinated resource acquisition resulting in a Balanced Canopy Functioning [Field et al., 1995], an appealing theoretical concept. As a result, plant canopies should not invest in PAR acquisition if PAR cannot be transformed into NPP because of other resource limitations or environmental conditions. Leaf N content may reflect nutrient limitation, and therefore may co-vary with APAR. Other studies have suggested that resource-use could be either constant over a wide range of conditions [Goetz and Prince, 1999] or related to resource-capture. Opposing views postulate either that resource use efficiency increases when resources availability decreases, or that resource use efficiency increases with the availability of a resource, as a result of improved overall functioning [Binkley et al., 2004]. Development of such theoretical frameworks and their evaluation with observations are necessary to develop and test plant productivity models [see Field et al., 1995].

[37] Our compilation of data allows to test some aspects of these concepts. It becomes apparent (Figure 7), for instance, that LUE and GEE*, although highly variable, are not related to light resource capture (i.e., fPAR) in a simple way. Likewise, leaf nitrogen and fPAR are not related (Figure 8). Arguably, there are few high LUE associated with very low fPAR sites in our data set, suggesting that high light-use efficiency is generally accompanied by high fPAR at least in the absence of opposing management practices, which is in line with Binkley et al. [2004]. The same is true for higher leaf N (e.g., >1.5%) being associated with relatively larger fPAR. However, other factors are required to explain the occurrence of LUE and N variability at similarly high fPAR level. For instance, the possible role of dense evergreen foliage as a nitrogen reservoir in nutrient poor environment (resulting in high mass of foliage with relatively low N) plays against a scaling of maximum LUE with fPAR because it increases fPAR and not LUE. It must be kept in mind though, that we investigate optimum LUE only. Using time-integrated variables may result in a different picture [Field et al., 1995] but based on our analysis there is little evidence for theoretical schemes relating of resource use to resource capture.

Figure 7a.

Optimum daily LUE versus fraction of absorbed PAR (fPAR).

Figure 7b.

Same as Figure 7a, but for GEE*.

Figure 8.

Canopy fPAR versus leaf nitrogen N.

[38] At the leaf level, both empirical evidence and mechanistic analyses have established the strong relationship between N and photosynthesis. Leaf nitrogen is recognized as critical for the photosynthesis apparatus, but extensive discussions have addressed the observed departures from a single inter-specific relationship. Allocation of N to Rubisco, activity of Rubisco, C3 or C4 metabolisms, diffusion of CO2, respiration are some factors, which change the N-photosynthesis relationship [Hikosaka, 2004, and references therein]. Differences between evergreen and deciduous trees have been reviewed by Warren and Adams [2004], who pointed CO2 diffusion and overinvestment in Rubisco as important factors. At the canopy level however, mechanisms are still lacking and it is not possible to associate the relationship between LUE and nitrogen of the dominant plant to patterns of allocation of N to Rubisco, chlorophyll or other forms. The factors explaining inter-specific differences in leaf-level data potentially drive canopy-level differences. The question is complicated by the existence of gradients of N allocation within canopies. Different studies have found a trade-off between N allocation to Rubisco and chlorophyll according to light availability, but such allocation pattern has often been shown to be sub-optimal.

[39] The empirical evidence in our data set either implies that there is a scaling between leaf N of the dominant plants and the whole canopy functioning, or implies that the variability within canopies is of second order compared to the explanatory power of leaf N of the dominant plants. The lack of relation between LUE or N and fPAR, as well as the relatively poor results of idealized canopy models, show that more pluri-specific studies are needed.

3.6. Concluding Remarks: Latitudinal Patterns of LUE and Remote Sensing-Based Models

[40] A simple relationship of N with either, fPAR, MAT or latitude would facilitate the development and use of global LUE models greatly but the existence of such large-scale patterns is a matter of debate. Some studies point toward an increase of foliar N with latitude, MAT, and altitude, whereas others point toward an opposite pattern [Reich and Oleskyn, 2004, and references therein]. Reich and Oleskyn [2004] found the highest foliar N values for mid-latitude and a weak decrease toward the coldest climate zones. In the data analyzed in our study the only significant trend in N versus latitude was for the evergreen needleleaf PFT.

[41] It has been argued that LUE may increase with latitude because of a parallel increase in leaf nitrogen [Haxeltine and Prentice, 1996]. LUE derived from atmospheric inversions [e.g., Knorr and Heimann, 1995; Kaminski et al., 2002; Randerson et al., 2002; Still et al., 2004] tend to indicate increasing LUE values from temperate to arctic latitudes, together with an increase for highly continental zones [Kaminski et al., 2002]. Interestingly, Still et al. [2004] found higher LUE for continental Eurasia than for North America and suggested that the distribution of deciduous trees with higher needle N could explain this pattern, which is inline with our study.

[42] Overall, our estimate of optimum LUE and GEE* along latitude gradients (e.g., Figures 9a and 9b) indicates a decreasing trend toward high latitude, supporting the statement of Kaminski et al. [2002], who considered obtaining the highest LUE for tundra as unrealistic.

Figure 9a.

Optimum daily LUE as a function of latitude for ∣lat∣ >45°.

Figure 9b.

GEE* as a function of latitude for ∣lat∣ >45°.

[43] A variety of reasons may help to explain the discrepancy between our findings and LUE derived from inversion analyses: atmospheric inversion studies rely on an estimate of the heterotrophic respiration and other surface processes like snow-related processes (insulation effect, impact on water availability), whose errors can impair LUE estimates [Randerson et al., 2002]. However, inaccuracy of the atmospheric vertical mixing or meridional transport may also cause such a discrepancy. If this were to be the case, the consequences on the estimates and localization of the carbon sinks/sources might be significant, as it has been shown by Stephens et al. [2007], and deserve further examination.

[44] Overall, our results strongly support the view that LUE varies significantly both across and within biomes, and Plant Functional Types. Our data do not support the view that LUEGPP might be less variable that LUENPP and might therefore span a small range of values [Goetz and Prince, 1999; Ruimy et al., 1996a]. Our extensive use of the flux measurement network strengthens and extends the results of Turner et al. [2003] and Yuan et al. [2007], who compared LUEGPP from 4 and 28 flux measurements sites respectively, as well as Choudhury [2001], who used data and canopy modeling and suggested a significant variability of LUEGPP. Convergent conclusions were drawn for aboveground LUENPP in the past [Gower et al., 1999, and references therein]. Given the high diversity of measurement sites, which encompass managed and unmanaged stands, mono- or pluri-specific canopies, leaf nitrogen emerges as a strong organization factor of optimum canopy LUE and canopy photosynthesis rate.

[45] As far as remote-sensing models are concerned LUE-based GPP models therefore have to account for variability in optimum LUE within an ecosystem and on biome scale. This has important consequences for the validation of global LUE-models with local data. Such models often combine an optimum LUE with different stress factors. Temporal variability in stress factors, incoming and absorbed PAR usually result in favorable model/data comparison, especially if site quality is accounted for [e.g., Yuan et al., 2007; Mäkelä et al., 2008]. Remote sensing of the xanthophylls cycle (PRI), surface radiative temperature or fluorescence may help capturing some the short to seasonal variability of LUE [Grace et al., 2007] and may be included in the next generation of LUE-based models. Improvements of optimum LUE estimates will be achieved if information on leaf nitrogen can be obtained in addition to fPAR, which may be done using ground-based or airborne sensor (e.g., AVIRIS, CASI). For instance, Smith et al. [2002] showed that forest productivity could be assessed through estimation of leaf nitrogen content mapped with AVIRIS. Boegh et al. [2002] were able to simulate canopy photosynthesis for different crops combining nitrogen-based modeling and hyperspectral data from CASI. The robustness of inversion methods for chlorophyll and nitrogen content, on the basis of satellite-borne sensors like MERIS, CHRIS-PROBA or HYPERION, has to be investigated in the perspective of LUE-modeling.

Appendix A

[46] One way to derive the canopy index proposed by Green et al. [2003] is to state that, for different canopies, daily GEE is proportional to GEEmax, the maximum rate of canopy photosynthesis. Thus LUE defined as the ratio of daily GEE to daily absorbed PAR writes:

equation image

[47] Note that the use of GEE* as a proxy for LUE is based on the same assumption (GEE* being close to equation (A1) right-hand-side). Assuming the maximum canopy photosynthesis rate is proportional to the integral of leaf level maximum photosynthesis rate (on a surface basis, Asmax), we have

equation image

Using the leaf-level equation of Reich et al. [1997], (equation (A3)), and expressing leaf photosynthesis on a mass basis (Ammax), we have:

equation image
equation image

[48] Assuming LUE values are measured at similar incident PAR level, LUE is proportional to the index of Green et al. [2003]. To avoid assumptions of equations (A3) and (A4), we also use the following index

equation image


[49] This work would not have been possible without the considerable efforts of a large group of people and many funding agencies (see acknowledgments for each site in the corresponding reference). Each flux measurement site usually requires several scientists and students to be run on the long term. Sites in remote places, in the arctic for instance, are among the most challenging places to run flux systems. The authors would like to thank all these people for sharing their data under “fair-use policy”, as well as providing unpublished information. Special thanks to Dennis Baldocchi, who provided energy, ideas, and enthusiasm to the FLUXNET community.