Journal of Geophysical Research: BiogeosciencesVolume 116, Issue G2 Free Access Improving canopy processes in the Community Land Model version 4 (CLM4) using global flux fields empirically inferred from FLUXNET data Gordon B. Bonan, Gordon B. Bonan [email protected] National Center for Atmospheric Research, Boulder, Colorado, USASearch for more papers by this authorPeter J. Lawrence, Peter J. Lawrence National Center for Atmospheric Research, Boulder, Colorado, USASearch for more papers by this authorKeith W. Oleson, Keith W. Oleson National Center for Atmospheric Research, Boulder, Colorado, USASearch for more papers by this authorSamuel Levis, Samuel Levis National Center for Atmospheric Research, Boulder, Colorado, USASearch for more papers by this authorMartin Jung, Martin Jung Max Planck Institute for Biogeochemistry, Jena, GermanySearch for more papers by this authorMarkus Reichstein, Markus Reichstein Max Planck Institute for Biogeochemistry, Jena, GermanySearch for more papers by this authorDavid M. Lawrence, David M. Lawrence National Center for Atmospheric Research, Boulder, Colorado, USASearch for more papers by this authorSean C. Swenson, Sean C. Swenson National Center for Atmospheric Research, Boulder, Colorado, USASearch for more papers by this author Gordon B. Bonan, Gordon B. Bonan [email protected] National Center for Atmospheric Research, Boulder, Colorado, USASearch for more papers by this authorPeter J. Lawrence, Peter J. Lawrence National Center for Atmospheric Research, Boulder, Colorado, USASearch for more papers by this authorKeith W. Oleson, Keith W. Oleson National Center for Atmospheric Research, Boulder, Colorado, USASearch for more papers by this authorSamuel Levis, Samuel Levis National Center for Atmospheric Research, Boulder, Colorado, USASearch for more papers by this authorMartin Jung, Martin Jung Max Planck Institute for Biogeochemistry, Jena, GermanySearch for more papers by this authorMarkus Reichstein, Markus Reichstein Max Planck Institute for Biogeochemistry, Jena, GermanySearch for more papers by this authorDavid M. Lawrence, David M. Lawrence National Center for Atmospheric Research, Boulder, Colorado, USASearch for more papers by this authorSean C. Swenson, Sean C. Swenson National Center for Atmospheric Research, Boulder, Colorado, USASearch for more papers by this author First published: 18 May 2011 https://doi.org/10.1029/2010JG001593Citations: 463AboutSectionsPDF ToolsRequest permissionExport citationAdd to favoritesTrack citation ShareShare Give accessShare full text accessShare full-text accessPlease review our Terms and Conditions of Use and check box below to share full-text version of article.I have read and accept the Wiley Online Library Terms and Conditions of UseShareable LinkUse the link below to share a full-text version of this article with your friends and colleagues. Learn more.Copy URL Abstract [1] The Community Land Model version 4 (CLM4) overestimates gross primary production (GPP) compared with data-driven estimates and other process models. We use global, spatially gridded GPP and latent heat flux upscaled from the FLUXNET network of eddy covariance towers to evaluate and improve canopy processes in CLM4. We investigate differences in GPP and latent heat flux arising from model parameterizations (termed model structural error) and from uncertainty in the photosynthetic parameter Vcmax (termed model parameter uncertainty). Model structural errors entail radiative transfer, leaf photosynthesis and stomatal conductance, and canopy scaling of leaf processes. Model structural revisions reduce global GPP over the period 1982–2004 from 165 Pg C yr−1 to 130 Pg C yr−1, and global evapotranspiration decreases from 68,000 km3 yr−1 to 65,000 km3 yr−1, within the uncertainty of FLUXNET-based estimates. Colimitation of photosynthesis is a cause of the improvements, as are revisions to photosynthetic parameters and their temperature dependency. Improvements are seen in all regions and seasonally over the course of the year. Similar improvements occur in latent heat flux. Uncertainty in Vcmax produces effects of comparable magnitude as model structural errors, but of offsetting sign. This suggests that model structural errors can be compensated by parameter adjustment, and this may explain the lack of consensus in values for Vcmax used in terrestrial biosphere models. Our analyses show that despite inherent uncertainties global flux fields empirically inferred from FLUXNET data are a valuable tool to guide terrestrial biosphere model development and evaluation. Key Points FLUXNET diagnostic models guide terrestrial biosphere model development Revisions to the Community Land Model (CLM4) improve gross primary production Model structural errors can be compensated by parameter adjustment 1. Introduction [2] Models of Earth's land surface, including its terrestrial ecosystems, for climate simulation have expanded beyond their hydrometeorological heritage to include biogeochemical cycles (e.g., carbon and nitrogen), land use, and vegetation dynamics [Bonan, 2008]. These models, coupled to their host climate model, are important research tools to study land-atmosphere interactions, climate feedback from ecological processes, and land management practices to mitigate climate change. [3] The development and evaluation of global terrestrial biosphere models for climate simulation have long utilized eddy covariance tower measurements of energy and carbon fluxes. Such analyses typically involve model calibration and evaluation at one or more flux tower sites [Morales et al., 2005; Friend et al., 2007; Stöckli et al., 2008; Mercado et al., 2009a; Randerson et al., 2009; Williams et al., 2009; Zaehle and Friend, 2010; Mahecha et al., 2010] and leave unresolved model evaluation at larger regional to continental scales. However, data-oriented diagnostic techniques to upscale gross primary production (GPP) and latent heat flux from the FLUXNET network of tower sites to global 0.5° gridded data products provide a means to evaluate the models at large scales [Jung et al., 2009, 2010; Beer et al., 2010; M. Jung et al., Global patterns of land-atmosphere fluxes of carbon dioxide, latent heat, and sensible heat derived from eddy covariance, satellite, and meteorological observations, submitted to Journal of Geophysical Research, 2011], notwithstanding potential errors in the data products. Here, we use the FLUXNET upscaled GPP and latent heat flux to evaluate and improve canopy processes in one such model, version 4 of the Community Land Model (CLM4) [Oleson et al., 2010; Lawrence et al., 2011]. [4] CLM4 substantially overestimates carbon uptake during GPP compared with data-driven estimates and with other models, and the model has a particularly high bias in the tropics [Beer et al., 2010]. We investigate biases in GPP, and associated errors in latent heat flux, arising from model parameterizations (termed model structural error) and from uncertainty in a key model photosynthetic parameter (termed model parameter uncertainty). Model structural errors entail radiative transfer, leaf photosynthesis and stomatal conductance, and canopy scaling of leaf processes. In particular, the distribution of absorbed photosynthetically active radiation among sunlit and shaded portions of the canopy as implemented by Thornton and Zimmermann [2007] is theoretically incorrect. CLM4 also simulates high rates of leaf photosynthesis compared with other photosynthesis models, as demonstrated by Chen et al. [2010]. We show that revisions to the model to correct these errors substantially improve simulated GPP and latent heat flux compared with the upscaled FLUXNET data. [5] Chen et al. [2010] suggested that the impact of model structural differences can be compensated by parameter adjustment, particularly the photosynthetic parameter Vcmax. This leaf-level parameter describes the maximum rate of carboxylation by the photosynthetic enzyme Rubisco, and other parameters such as the maximum rate of electron transport and leaf maintenance respiration scale with Vcmax [Farquhar et al., 1980; Collatz et al., 1991, 1992], but estimates of Vcmax vary greatly and the range of possible values is large, even within a plant functional type [Wullschleger, 1993; Beerling and Quick, 1995; Kattge et al., 2009]. For example, Kattge et al. [2009] derived Vcmax from a literature synthesis, and those values are much different than the values used in CLM4. We quantify the effect of this parameter uncertainty on simulated GPP and latent heat flux and use the upscaled FLUXNET data to evaluate Vcmax parameter estimation. 2. Methods Model Description [6] CLM4 continues earlier versions CLM2 [Bonan et al., 2002] and CLM3 [Oleson et al., 2004; Dickinson et al., 2006], and it succeeds CLM3.5 [Oleson et al., 2008; Stöckli et al., 2008] with revised hydrology and snow parameterizations, organic soils, a 50 m deep ground column, and an updated distribution of plant functional types [Oleson et al., 2010; Lawrence et al., 2011]. The model simulates CO2 assimilation by the plant canopy (GPP) as part of its coupled photosynthesis-stomatal conductance parameterization. Leaf area index for each plant functional type is specified by a globally gridded monthly data set derived from satellite data [Lawrence et al., 2011]. CLM4 includes a biogeochemical parameterization of the terrestrial carbon and nitrogen cycles, in which GPP drives prognostic leaf area and vegetation and soil carbon pools and in which the associated nitrogen cycle constrains carbon fluxes. That version of the model (denoted CLM4CN) has large biases in leaf area [Lawrence et al., 2011]. Here, we do not utilize the carbon-nitrogen biogeochemistry and instead use the prescribed satellite-derived monthly leaf area index so that GPP is unaffected by biases in the CLM4CN biogeochemistry and prognostic leaf area. [7] For these simulations, a 57 year (1948–2004) meteorological data set was used to force the model in offline simulations uncoupled from a climate model, as in the works of Oleson et al. [2008] and Lawrence et al. [2011]. Land cover was held constant at values for year 2000, but atmospheric CO2 varied as in the historical record. The spatial resolution of the model is 1.25 degrees in longitude by 0.9375 degrees in latitude. Model Simulations and Test Data [8] We performed four simulations to document biases arising from model structural errors (Table 1): CLM4, a control simulation with CLM4; RAD, a simulation with revisions to the two-stream radiative transfer parameterization to correctly account for sunlit and shaded leaves (section 2.3); RAD-PSN, as in RAD but with revisions to the leaf photosynthesis and stomatal conductance formulation (section 2.4); and RAD-PSN-KN, as in RAD-PSN but with revised canopy scaling to account for exponential decline in foliage nitrogen concentration with depth in the canopy (section 2.5). This latter simulation is denoted CLM4a to distinguish the full model with all parameterization improvements from CLM4. Table 1. Model Simulations Simulation Description Model Structural Error CLM4 control simulation with CLM4 RAD revised two-stream radiative transfer RAD-PSN RAD and revised leaf photosynthesis RAD-PSN-KN RAD-PSN and revised canopy scaling (also denoted CLM4a) Model Parameter Uncertainty CLM4a control simulation (same as RAD-PSN-KN) Vcmax 25 = Vcmax 25opt f(D)f(N) OPT CLM4a with non-nitrogen-limited Vcmax 25, Vcmax 25 = Vcmax 25opt f(D) KAT CLM4a with Kattge et al. [2009]Vcmax 25, Vcmax 25 = Vcmax 25Kattge f(D) DAY CLM4a without day length factor for Vcmax 25, Vcmax 25 = Vcmax 25opt f(N) [9] We also performed simulations to investigate uncertainty in Vcmax. CLM4 uses a potential value Vcmax 25opt (derived from prescribed, time-invariant foliage nitrogen concentration as described by Thornton and Zimmermann [2007]), and this defines the maximum attainable carboxylation rate (adjusted to 25°C). The realized rate (at 25°C) is obtained after adjusting for day length and nitrogen limitation. The expression Vcmax 25 = Vcmax 25opt f(D) provides the potential carboxylation rate in the absence of nitrogen limitation, after reduction for day length using the function f(D) = (D/Dmax)2, where D is day length and Dmax is maximum day length. Seasonal changes in photosynthetic capacity have been observed in trees [e.g., Niinemets et al., 1999; Wilson et al., 2000; Xu and Baldocchi, 2003], and the CLM4 parameterization assigns this variability to day length [Oleson et al., 2010]. [10] When the carbon-nitrogen biogeochemistry is active, the amount of nitrogen required to support the potential growth is diagnosed, and GPP is reduced if nitrogen availability is insufficient to sustain the potential biomass increment. Without carbon-nitrogen biogeochemistry (as in our CLM4 simulations reported here), the same formulation is used but Vcmax 25opt is reduced by a prescribed nitrogen factor so that Vcmax 25 = Vcmax 25opt f(D)f(N) is the realized value. This ensures that leaf photosynthetic rates (and GPP) are adjusted for nitrogen availability. The term f(N) is scaled between zero and one to represent nitrogen constraints on photosynthesis, varies among plant functional types, and is derived from a CLM4CN simulation [Bonan and Levis, 2010]. [11] We performed three simulations (Table 1) to quantify the effects of uncertainty in the CLM4 values of Vcmax 25, as represented by the terms Vcmax 25opt, f(N), and f(D). The first simulation examined the term f(N). The inferred nitrogen limitation factors f(N) yield lower values for Vcmax 25 compared with Vcmax 25opt (Table 2), yet both estimates are within the range of published values obtained from synthesis studies [Wullschleger, 1993]. Therefore, we performed a CLM4a simulation using the maximum values obtained with f(N) = 1 (denoted OPT). The second simulation used the Kattge et al. [2009] estimates of Vcmax 25. Kattge et al. [2009] derived Vcmax 25 based on a synthesis of photosynthetic measurements extrapolated to natural vegetation using observed foliage nitrogen content (Table 2), and we evaluated their performance in CLM4a (denoted KAT). In the third simulation, we removed the CLM4 day length factor so that Vcmax 25 does not vary through the year (denoted DAY). Table 2. Values for Vcmax25opt (μmol m−2 s−1) Plant Functional Type CLM4 Kattge et al. [2009]Vcmax 25Kattge Vcmax 25opt Vcmax 25opt f(N) Needleleaf evergreen tree, temperate 61 55 62 Needleleaf evergreen tree, boreal 54 42 62 Needleleaf deciduous tree, boreal 57 29 39 Broadleaf evergreen tree, tropical 72 66 41a Broadleaf evergreen tree, temperate 72 51 61 Broadleaf deciduous tree, tropical 52 36 41a Broadleaf deciduous tree, temperate 52 30 58 Broadleaf deciduous tree, boreal 52 40 58 Broadleaf evergreen shrub, temperate 72 36 62 Broadleaf deciduous shrub, temperate 52 30 54 Broadleaf deciduous shrub, boreal 52 19 54 C3 grass, arctic 52 21 78 C3 grass 52 26 78 C4 grass 52 25 78b Crop 57 31 101 a Kattge et al. [2009] report a low value of 29 μmol m−2 s−1. b Not reported by Kattge et al. [2009] and assigned a value for C3 grass, as in CLM4. [12] We compared model simulations with observationally based GPP and latent heat flux derived from the FLUXNET network of eddy covariance towers. The global FLUXNET upscaling uses data-oriented diagnostic models trained with the eddy covariance flux data to provide empirically derived, spatially gridded fluxes. For this study, the global FLUXNET upscaling utilized the model tree ensembles (MTE) approach, described by Jung et al. [2009, also submitted manuscript, 2011] and applied to GPP and latent heat flux [Beer et al., 2010; Jung et al., 2010, also submitted manuscript, 2011]. The upscaling relies on remotely sensed estimates of the fraction of absorbed photosynthetically active radiation (fAPAR), climate, and land cover data. The FLUXNET-MTE upscaling provides monthly fluxes at 0.5° spatial resolution. We regridded the data to the CLM4 grid, excluding FLUXNET-MTE grid cells with no data (typically desert and barren land cover). We analyzed the 23 year period 1982–2004. Radiative Transfer [13] CLM4, and its predecessors, utilizes the two-stream approximation [Sellers et al., 1996a] to calculate radiative transfer and surface albedo for direct beam and diffuse radiation and for visible (<0.7 μm) and near-infrared (≥0.7 μm) wave bands. In CLM4, absorbed photosynthetically active radiation (the visible wave band) is partitioned to sunlit and shaded leaves for photosynthesis [Thornton and Zimmermann, 2007]. Dai et al. [2004] developed a sunlit and shaded leaf canopy parameterization for the Common Land Model (CoLM) with analytical solutions to the two-stream approximation (Appendix A). CLM4 does not use this solution and instead diagnoses the radiation absorbed by sunlit and shaded leaves from the total radiation absorbed by the canopy. [14] Thornton and Zimmermann [2007] describe the sunlit and shaded leaf parameterization, and Oleson et al. [2010] provide the numerical implementation. The direct beam radiation absorbed by the canopy is partitioned into unscattered direct beam and scattered direct beam. Sunlit leaves receive all the unscattered direct beam radiation absorbed by the canopy and additionally a fraction of the total diffuse radiation (scattered direct beam radiation and atmospheric diffuse radiation) absorbed by the canopy. Shaded leaves receive only diffuse radiation. The diffuse radiation absorbed by the canopy is apportioned to sunlit and shaded leaves in relation to the sunlit and shaded fractions of the canopy. [15] The CLM4 diagnosis of sunlit and shaded leaf radiation differs markedly from the analytical solution of Dai et al. [2004] (Figure 1). The two parameterizations are similar in their absorption of direct beam photosynthetically active radiation, but not for diffuse radiation. CLM4 apportions the total canopy absorption of diffuse radiation to sunlit and shaded leaves based on the sunlit and shaded fractions of the canopy. The amount of diffuse radiation absorbed by sunlit leaves declines with leaf area index >∼2 m2 m−2 because the sunlit fraction of the canopy declines; and similarly the amount absorbed by shaded leaves increases. The Dai et al. [2004] two-stream solution shows near constant absorption for leaf area index greater than ∼6 m2 m−2, and sunlit leaves absorb more diffuse radiation than do shaded leaves. Figure 1Open in figure viewerPowerPoint Radiative transfer for photosynthetically active radiation in CLM4 compared with other parameterizations. (a) Fraction of the canopy that is sunlit and shaded in relation to leaf area index. (b-e) Fraction of incident direct beam absorbed by the canopy and by the sunlit and shaded leaves. (f-i) Same as Figures 1b–1e, but for incident diffuse radiation. Shown are the CLM4 two-stream solution [Thornton and Zimmermann, 2007]; the Dai et al. [2004] analytical two-stream solution implemented in CoLM; the multilayer theory of Goudriaan [1977] and Goudriaan and van Laar [1994] implemented in the CABLE land surface model [Kowalczyk et al., 2006]; and the multilayer approach of Norman [1979] implemented in the CANOAK plant canopy model [Baldocchi and Wilson, 2001; Baldocchi et al., 2002]. Simulations are for a canopy of leaves with spherical leaf angle orientation, leaf reflectance of 0.10 and transmittance of 0.05, soil albedo of 0.10, and zenith angle of 30°. [16] For comparison, we also considered the multilayer radiative transfer theory of Goudriaan [1977] and Goudriaan and van Laar [1994], implemented in plant canopy models [e.g., de Pury and Farquhar, 1997; Wang and Leuning, 1998] and in the CABLE land surface model [Kowalczyk et al., 2006]; and the multilayer approach of Norman [1979], implemented in the CANOAK plant canopy model [Baldocchi and Wilson, 2001; Baldocchi et al., 2002]. These parameterizations behave similar to the Dai et al. [2004] two-stream solution, though the exact partitioning of radiation between sunlit and shaded leaves varies somewhat among the three approaches (Figure 1). Leaf Photosynthesis and Stomatal Conductance [17] CLM4, and its predecessors, utilizes a coupled leaf photosynthesis and stomatal conductance model that is a variant of the Ball-Berry stomatal conductance model [Ball et al., 1987; Collatz et al., 1991], the Farquhar et al. [1980] C3 photosynthesis model as implemented by Collatz et al. [1991], and the Collatz et al. [1992] C4 photosynthesis model. Bonan [1995] described this parameterization, and the numerical implementation in CLM4 [Oleson et al., 2010] is unchanged from earlier versions of the model [Bonan, 1996; Oleson et al., 2004]. [18] We updated the photosynthesis-conductance parameterization based on literature synthesis and to account for theoretical advances since its original implementation, and we denote the new formulation PSN (Appendix B). In particular, CLM4 has higher rates of leaf photosynthesis than a variant of the Farquhar/Ball-Berry/Collatz model used in CoLM [Chen et al., 2010]. This is related in part to colimitation of photosynthesis in CoLM, used also in the C3 and C4 models of Collatz et al. [1991, 1992] and implemented in the Simple Biosphere model (SiB) [Sellers et al., 1996a, 1996b]. Moreover, the temperature kinetics of Rubisco derived from experimental studies [Bernacchi et al., 2001, 2003; Leuning, 2002] is quite different than that implemented in CLM4 and models such as CoLM and SiB. Key parameterization changes introduced in PSN include: colimitation among Rubisco-, light-, and export-limited rates; revised photosynthetic parameters for Rubisco kinetics and their temperature responses; electron transport rate for light-limited photosynthesis with a maximum rate Jmax; exported-limited photosynthesis based on the rate of triose phosphate utilization; and C4 photosynthesis similar to Collatz et al. [1992] and SiB [Sellers et al., 1996a, 1996b]. [19] The C3 leaf photosynthetic rates are lower for PSN than for CLM4 (Figure 2). This is related in part to the introduction of colimitation in PSN, noted also in a comparison between the CLM4 and CoLM parameterizations [Chen et al., 2010]. Photosynthetic rates are higher without colimitation (compare PSN with colimitation and PSN* without colimitation). The new parameter values for Rubisco kinetics (Kc, Ko, and Γ*) further reduce photosynthetic rates (compare CLM4 and PSN*, both without colimitation). Additionally, the electron transport rate used in PSN adds curvature to the light response curve compared with the linear function used in CLM4. In the CO2 response curve, the reduced export-limited rate used in PSN compared with CLM4 lowers the photosynthetic rate at high CO2 concentration. The temperature functions lower the optimum temperature for PSN by 5–6°C compared with CLM4. Figure 2Open in figure viewerPowerPoint Simulated leaf gross photosynthetic rates for C3 and C4 plants. (a and b) Response to absorbed photosynthetically active radiation. (c and d) Response to ambient CO2 concentration. (e and f) Response to leaf temperature. (g and h) Response to vapor pressure deficit. Shown are the CLM4 solution; the PSN parameterization used in this study; and the PSN parameterization without colimitation (PSN*). As a reference, we show results for CoLM [Chen et al., 2010] with its documented parameterization (CoLM) and with revised electron transport (CoLM*). Reference values are ca = 379 μmol mol−1; oi = 0.209 mol mol−1; Patm = 1013.25 hPa; ϕ = 2000 μmol photon m−2 s−1; Tv = 25°C; air temperature is 25°C and relative humidity is 100%; and gb = 5 cm s−1. In these simulations, Vcmax 25 = 40 μmol m−2 s−1 (C3) and 33 μmol m−2 s−1 (C4), which are representative values used in CLM4 [Oleson et al., 2010]. [20] Colimitation similarly reduces leaf photosynthetic rates in C4 plants (Figure 2). The CO2 response curves differ because of the higher CO2-limited rate (we) for C4 photosynthesis used in PSN compared with CLM4. This causes photosynthesis to saturate at lower CO2 concentration. It also affects the vapor pressure deficit response, because the rate of photosynthesis is not limited by we at the ambient CO2 concentration used in the simulations (379 ppmv) and thus does not depend on intercellular CO2 (ci). Consequently, even though stomata close with greater vapor pressure deficit, photosynthesis is insensitive to vapor pressure deficit. In contrast, the photosynthetic rate is limited by we (and thus depends on ci) at the ambient CO2 in CLM4 and decreases with higher vapor pressure deficit as stomata close. The temperature optimum is shifted about 2°C warmer in PSN compared with CLM4. [21] Chen et al. [2010] previously compared the CLM4 and CoLM photosynthesis-stomatal conductance models. For reference, we compared our results with similar simulations using the CoLM parameterization (Figure 2). The C3 light and CO2 response curves for PSN are similar to CoLM. The PSN temperature response has an optimal temperature about 3°C lower than CoLM. The vapor pressure deficit responses are similar, except when PSN limits the response at high vapor pressure deficit. The CoLM C4 parameterization produces much lower photosynthetic rates than PSN. CoLM limits the electron transport rate to a value less than Jmax/4, used for both C3 and C4 plants. When this limitation is removed (CoLM*), the CoLM photosynthetic rates for C4 plants are similar to PSN. Canopy Integration [22] Sellers et al. [1992] introduced canopy scaling of leaf photosynthesis and stomatal conductance based on gradients of foliage nitrogen in the canopy, and Sellers et al. [1996a, 1996b] implemented this parameterization in SiB. The photosynthetic parameter Vcmax varies with leaf nitrogen concentration. The original theory postulated that plants optimally allocate resources to maximize carbon gain such that area-based leaf nitrogen is distributed through the canopy in relation to the time-mean profile of photosynthetically active radiation, but it is now recognized that the nitrogen gradient is shallower than the light gradient [Hollinger, 1996; Carswell et al., 2000; Meir et al., 2002; Niinemets, 2007; Lloyd et al., 2010]. [23] Many plant canopy models [e.g., de Pury and Farquhar, 1997; Wang and Leuning, 1998] and terrestrial components of climate models including CoLM [Dai et al., 2004; Chen et al., 2010], GISS [Friend and Kiang, 2005], CABLE [Kowalczyk et al., 2006], and O-CN [Zaehle and Friend, 2010] now parameterize canopy scaling using concepts of sunlit and shaded leaves in combination with an exponential profile of foliage nitrogen (defined by the decay coefficient Kn). The canopy is divided into sunlit and shaded fractions, and the photosynthesis-conductance parameterization is solved using canopy-integrated parameters derived from leaf-level parameters. Canopy values for Vcmax 25 are found by integrating leaf nitrogen concentration over the sunlit and shaded fractions of the canopy (Appendix C). Other parameters scale similarly. [24] Values for Kn vary among models, but are generally shallower than the light extinction coefficient. Friend and Kiang [2005] reported Kn = 0.11 for the GISS model, derived from Amazonia rain forest data [Carswell et al., 2000] and used also in O-CN [Zaehle and Friend, 2010]. Alton et al. [2007] used Kn = 0.15, inferred from measurements in a variety of forests, in simulations of boreal needleleaf forest, temperate broadleaf forest, and Amazonian rain forest with JULES. Mercado et al. [2006] derived Kn = 0.18 for Amazonian rain forest [Carswell et al., 2000], and Mercado et al. [2009a] used values of 0.16–0.25 (mean, 0.20) for five rain forest sites in the Brazilian Amazon (L. Mercado, personal communication, 2010). Larger values have been used in some models. CoLM uses Kn = 0.5 for Vcmax scaling and 0.72 for Jmax scaling [Dai et al., 2004]. [25] CLM4 uses a comparable scaling approach, but the canopy gradient in foliage nitrogen is specified through a linear decrease in foliage mass per unit leaf area Ma (g C m−2), or an increase in specific leaf area SLA (m2 g−1 C), with greater cumulative leaf area from the canopy top [Thornton and Zimmermann, 2007]. The gradient is specified such that Ma decreases twofold from canopy top to canopy bottom with a leaf area index of 8 m2 m−2. Mass-based foliage nitrogen concentration Nm (g N g−1 C) is constant with canopy depth, but area-based foliage nitrogen Na (g N m−2) decreases because Ma decreases with depth (Na = Nm Ma). Canopy values for Vcmax 25 are found by integrating Ma over the sunlit and shaded fractions of the canopy to obtain sunlit and shaded Na, from which Vcmax 25 is obtained (Appendix C). [26] While canopy scaling based on gradients of Ma, or conversely SLA, may be a useful conceptual framework, its implementation in CLM4 has several limitations. Observational studies find that Ma decreases exponentially with greater depth in forest canopies [Niinemets and Tenhunen, 1997; Meir et al., 2002; Lloyd et al., 2010]. Indeed, CLM4 has a shallower gradient in Vcmax than seen in observations or used in other models (Figure 3). The CLM4 profile of Vcmax compares favorably with data of Niinemets and Tenhunen [1997] for broadleaf deciduous tree at low leaf area, but declines too gradually at high leaf area. In contrast, an exponential profile with Kn = 0.11 [Friend and Kiang, 2005; Zaehle and Friend, 2010] more closely matches the observations. Kn = 0.15 [Alton et al., 2007] produces a sharper decline, and Kn = 0.50 [Dai et al., 2004] yields a steep decline. Lloyd et al.'s [2010] estimates of Kn for 16 temperate broadleaf f
