SEARCH

SEARCH BY CITATION

Keywords:

  • eddy covariance;
  • water use efficiency;
  • isotope discrimination;
  • carbon dioxide;
  • multilayer model

Abstract

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Methods
  5. 3. Results
  6. 4. Discussion
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

[1] Forest ecosystems across the globe show an increase in ecosystem carbon uptake efficiency under conditions with high fraction of diffuse radiation. Here, we combine eddy covariance flux measurements at a deciduous temperate forest in central Germany with canopy-scale modeling using the biophysical multilayer model CANVEG to investigate the impact of diffuse radiation on various canopy gas exchange processes and to elucidate the underlying mechanisms. Increasing diffuse radiation enhances canopy photosynthesis by redistributing the solar radiation load from light saturated sunlit leaves to nonsaturated shade leaves. Interactions with atmospheric vapor pressure deficit and reduced leaf respiration are only of minor importance to canopy photosynthesis. The response strength of carbon uptake to diffuse radiation depends on canopy characteristics such as leaf area index and leaf optical properties. Our model computations shows that both canopy photosynthesis and transpiration increase initially with diffuse fraction, but decrease after an optimum at a diffuse fraction of 0.45 due to reduction in global radiation. The initial increase in canopy photosynthesis exceeds the increase in transpiration, leading to a rise in water-use-efficiency. Our model predicts an increase in carbon isotope discrimination with water-use-efficiency resulting from differences in the leaf-to-air vapor pressure gradient and atmospheric vapor pressure deficit. This finding is in contrast to those predicted with simple big-leaf models that do not explicitly calculate leaf energy balance. At an annual scale, we estimate a decrease in annual carbon uptake for a potential increase in diffuse fraction, since diffuse fraction was beyond the optimum for 61% of the data.

1. Introduction

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Methods
  5. 3. Results
  6. 4. Discussion
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

[2] Solar radiation is one of the major drivers for ecosystem gas exchange processes such as transpiration, photosynthesis and production of sensible heat. Over the last several decades, however, changes in solar radiation received at the earths' surface have been observed. From the 1950s to the 1990s, global radiation (Rs) i.e., the incoming short wave radiation has decreased across the globe by about 0.27% per year (0.51 ± 0.05 W m−2 per year), called global dimming [Qian et al., 2006; Stanhill and Cohen, 2001]. In sum, this dimming has resulted in a total reduction of global radiation equaling 20 W m−2 between the years 1952 and 1992 [Stanhill and Cohen, 2001]. Since the 1990s however global radiation (Rs) has increased at many sites, particular in Europe and the United States described as global brightening [Wild et al., 2005]. Although the exact reasons for these changes are still unclear, a line of evidence suggests that anthropogenically emitted aerosols as well as an increase in cloud formation have caused this reduction; the recovery followed the reduction in air pollution in the industrialized countries [Liepert, 2002; Ohmura, 2006; Ramanathan et al., 2001; Wild et al., 2005]. The strongest reduction in global radiation was observed in northern latitudes, where the strongest increase in fossil fuel emission occurred during this period. Also, the reduction in global radiation at individual sites seems to be link to the degree of urbanization [Alpert et al., 2005]. Concurrent with the change in global radiation is an change in diffuse radiation as observed at several sites across the globe [Abakumova et al., 1996; Russak, 1990].

[3] Direct measurements of canopy carbon dioxide (CO2) exchange have shown that canopy photosynthesis is enhanced under conditions with a high proportion of diffuse radiation compared to conditions with the same global radiation but with a lower proportion of diffuse radiation [Baldocchi et al., 1997; Gu et al., 2003; Hollinger et al., 1994; Jenkins et al., 2007; Niyogi et al., 2004; Young and Smith, 1983]. The response to diffuse light, however, varies among ecosystems and seems to be related to canopy properties (R. Zhang et al., The response of eight contrasting ecosystems to direct versus diffuse radiation, submitted to Global Change Biology, 2008). Similarly, whole tree chambers which produced a higher diffuse radiation fraction as a side effect have produced an enhancement of radiation use efficiency [Denmead et al., 1993]. This phenomenon seems to be an emergent property at ecosystem scale, but not observed at leaf level. While the response of individual leaves to radiation and other environmental drivers is well understood and can be modeled quite accurately [Farquhar et al., 1980], the response of canopy photosynthesis to light is quite complex because phenomena emerge at the canopy scale that are quite different than at the leaf scale [Baldocchi and Amthor, 2001]. Most prominent is the increase in radiation use efficiency with diffuse radiation [Jarvis et al., 1985; Jenkins et al., 2007; Niyogi et al., 2004; Roderick et al., 2001] (J. Still et al., Influence of clouds and diffuse radiation on ecosystem-atmosphere CO2 and C18O exchanges, submitted to Journal of Geophysical Research, 2008). Thus, ecosystem scale observations such as eddy covariance measurements and ecosystem scale modeling are essential to understand these phenomena.

[4] Roderick et al. [2001] argue that under clear sky condition sunlit leaves are typically radiation saturated while shaded leaves have very low radiation load and hence are sensitive to radiation changes as they are on the linear part of the light response curve. When conditions become cloudier sunlit leaves receive less direct radiation but still remain on the radiation-saturated part of the photosynthesis-light response curve and thus maintain high photosynthesis. On the other hand, shaded leaves receive more diffuse radiation; diffuse radiation is omni-directional and it penetrates through the canopy better. Since shaded leaves are typically located on the linear part of the photosynthesis-light response curve they benefit strongly from small increases in radiation load resulting in a net increase in canopy photosynthesis. In addition to the radiative effect, an indirect thermal effect may play a role, too. The reduction in global radiation due to aerosols or cloudiness might lead to a decrease in leaf and surface temperature. Such a decrease would result in an increase in leaf photosynthesis under conditions where air temperature exceeds the optimum temperature for photosynthesis [Baldocchi and Harley, 1995; Steiner and Chameides, 2005] and may lower leaf, stem or soil respiration, too. Steiner and Chameides [2005] argue, from their modeling exercise, that the thermal effect might be larger than a radiative effect. Finally, a third potential mechanism would involve an interaction between diffuse fraction and air humidity under cloudy conditions resulting in more favorable conditions for leaf photosynthesis [Gu et al., 1999]. This is because stomatal conductance increases with reduced vapor pressure deficits. A potential forth mechanism would be the stimulation of photosynthesis and stomata opening due to an increase in the blue/red light ratio under conditions of high diffuse fraction [Urban et al., 2007].

[5] The volcanic eruption of Mount Pinatubo in 1991 offered the opportunity to test some of the hypotheses about the diffuse radiation effect on canopy photosynthesis. The eruption led to a global increase in atmospheric aerosols, a decrease in global radiation and an increase in diffuse fraction, without any first-order changes in vapor pressure deficit. Roderick et al. [2001] and Gu et al. [2003] argued that the increase in diffuse radiation had a stronger impact on photosynthesis at forest sites than the decrease in global radiation and hence increased the terrestrial carbon sink. Yet, other - more indirect - studies based on tree rings [Krakauer and Randerson, 2003] and atmospheric measurements [Angert et al., 2004] did not find support for an increase in forest productivity across wide regions. The reasons behind the discrepancy among those studies - besides technical issues - remain unclear. In the case of the tree rings one could argue that stand properties and antecendent physiological conditions may play an important role. Roderick et al. [2001] suggest that diffuse radiation growth enhancement would be expected to be most pronounced for closed canopy forests with high leaf area indices. Also the effect would be especially significant for understory plants with more leaves shaded from direct radiation [Krakauer and Randerson, 2003]. Tree ring analysis, however, is often based on dominant trees where diffuse radiation growth enhancement would be less pronounced. Also, a study from an old growth boreal forest in Canada suggest that tree ring width and carbon uptake are not well correlated [Rocha et al., 2006]. Furthermore, the level of aerosols also seems to play a role. Yamasoe et al. [2006] and Oliveira et al. [2007] showed that photosynthesis in a tropical rain forest increased at intermediate levels of aerosol optical depth but decreased at high levels of aerosol optical depth. The direct link to aerosols was shown by a study from the Sierra Nevada in California, USA where air pollutions from the nearby Central Valley led to an increase in aerosol load in the afternoon and consequently to higher carbon uptake rates in the afternoon compared to the morning [Misson et al., 2005]. Finally, a multisite study with direct observation of aerosol optical depth and direct measurements of carbon dioxide exchange with the eddy covariance flux method showed that the diffuse radiation effect is ubiquitous across North America, but its magnitude varies between forest, grassland and agricultural land suggesting a dependency on canopy structure [Niyogi et al., 2004].

[6] In this study, we aim to address some of the unresolved aspects of the diffuse radiation effect by analyzing eddy covariance flux data from two mixed beech forests that differ in their canopy structure and by employing the multilayer biophysical canopy model CANVEG to diagnose the mechanisms and explanations for the observations. We hypothesize that the diffuse radiation effect on canopy photosynthesis increases with leaf area index and increases with a multi layer canopy structure (Hypothesis 1). We further hypothesize that the diffuse radiation effect is more pronounced in canopies where more radiation penetrates into deeper layers and less where high leaf reflectance and transmittance lead to production of diffuse radiation within the canopy (Hypothesis 2). We then extend this analysis to other trace gases and hypothesize that the diffuse radiation effect is stronger for photosynthesis and less pronounced for transpiration (Hypothesis 3). Consequently we expect to see an increase in water use efficiency with increased diffuse fraction as observed by Min [2005]. This combination of effects would result in a decrease in carbon isotope discrimination according to knowledge from a simple big leaf model. Finally, we hypothesize that with an increased diffuse fraction leaf temperature would decline (Hypothesis 4).

[7] To achieve our goal of testing the articulated hypotheses, we demonstrate how well the CANVEG model reproduces carbon, water, and energy fluxes at a deciduous forest site in Germany and test whether the CANVEG model accurately reproduces the observed diffuse radiation effect. Second, we use the model to investigate the relevance of leaf inclination angle, leaf optical properties, leaf clumping, and leaf area index for the diffuse radiation effect. Third, we derive from radiation measurements the local relationship between increase in diffuse fraction and decrease in global radiation. We then use this relationship to simulate the impact of a concurrent increase in diffuse fraction and decrease in global radiation on photosynthesis, transpiration, leaf temperature, water use efficiency as well as on carbon isotope discrimination and isoprene fluxes. We show that the diffuse radiation effect is largely caused by a redistribution of radiation under conditions with increased diffuse fraction and an associated increase in shade leaf photosynthesis. Furthermore, the response of canopy photosynthesis exceeds the response in transpiration leading to an increase in water-use-efficiency. This increase results in an increase in carbon isotope discrimination, which is in contrast to commonly used simple big-leaf models. Finally, we show that at our site atmospheric conditions were mostly beyond the optimum of the diffuse radiation response function resulting in a decrease in annual carbon uptake if diffuse fraction was to increase.

2. Methods

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Methods
  5. 3. Results
  6. 4. Discussion
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

2.1. Site Description

[8] Environmental and scalar flux measurements were carried out at two beech forests in Thuringia, in Central Germany. The Hainich tower site is an old unmanaged mixed beech forest located within the Hainich National Park, near the city of Eisenach [Knohl et al., 2003]. The Hainich National Park was established in 1997 to protect one of the largest beech forests in Central Europe and covers an area of about 7600 ha. Due to its history as a military base, a large part of the forest had been taken out of regular management and developed relatively undisturbed. None of the area has been clear felled. Consequently, the forest displays characteristics typical of an unmanaged old-growth forest; for example it maintains a wide range of age classes (from one to 250 years), comparatively large dead wood pools, canopy gaps and a vertically structured canopy. The forest is dominated by beech (Fagus sylvatica, 65%), is codominated by ash (Fraxinus excelsior, 25%) and maple (Acer pseudoplantanus and Acer plantanoides, 7%), and sustains some European hornbean (Carpinus betulus), and elm (Ulmus glabra). Maximum effective leaf area index (PCA-LAI 2000 measurements not corrected for clumping effects or stems and branches) of the woody vegetation (excluding ground vegetation) was ∼5–6 m2 m−2 with a multilayered leaf canopy structure. Carbon dioxide, water vapor and energy flux measurements began at the site in October 1999. Flux data of this site have been investigated in detail with regard to site quality in combination with footprint modeling. Overall the measurements reveal good representation of the surrounding landscape by the flux measurement and high data quality [Rebmann et al., 2005; Reithmaier et al., 2006].

[9] The Leinefelde tower site is a managed and pure-stand of beech trees (Fagus sylvatica) located in the Geney forest district, near Leinefelde [Anthoni et al., 2004]. The forest has been managed as a shelterwood system since 1838. The beech forest around the tower is characterized by a sequence of relatively homogenous, even-aged stands at different stand age which were naturally regenerated under the shelter of a few remaining old trees of the previous stand. Regular thinning has occurred on a 10–20-year-cycle, with the last two thinning in 1982 and 1999. Maximum effective leaf area index in this stand was ∼5–6 m2 m−2 with a generally top weighted leaf canopy structure. Measurements at the site started in April 2002.

2.2. Flux and Meteorological Measurements

[10] Fluxes of carbon dioxide, water vapor and heat were continuously measured since October 1999 at the Hainich tower site and since April 2002 at the Leinefelde tower site using the eddy covariance technique. The flux system consisted of a triaxial sonic anemomenter (Gill Solent R3, Gill Instruments, Lymington, UK) and a fast response CO2/H2O infrared gas analyzer in absolute mode (LiCor 6262-3, LiCor Inc. Lincoln, NE, USA). The methodology to calculate fluxes followed standard procedures and included de-spiking, calculation of the time lag associated with the transport of air drawn through the tube, block averaging, spectral correction for dampening in the tube, planar-fit coordinate rotation fluxes, correction for CO2 storage in the air column underneath the eddy covariance sensors, and u*-correction for nighttime data with low turbulence [Aubinet et al., 2000]. Water vapor fluxes and sensible heat fluxes were corrected for nonclosure of the energy balance. Independent measurements of CO2 fluxes using an open path sensor during a four week period showed agreement within 7% of our measurements [Knohl et al., 2003]. In addition, the tower was equipped with instruments to measure net radiation (Schulze-Däke LXG055, Dr. Bruno Lange GmbH, Berlin, Germany), albedo and shortwave (global) radiation (CM14, Kipp & Zonen, Delft, NL), diffuse shortwave radiation (CM11, Kipp & Zonen, Delft, NL), photosynthetic photon flux density (LI-190SA, LiCor Inc., Lincoln, NE, USA), air humidity and air temperature (HMP35D, Vaisala, Helsinki, Finland) and air pressure (PTB101B, Vaisala, Helsinki, Finland). Precipitation was continuously collected inside and outside the forest (RainGauge, Young, Traverse City, MI, USA). Soil moisture and soil temperature were measured in vertical soil profiles in the surrounding of the tower (ML-2x, DeltaT, Cambridge, UK, PT-100-temperature sensors). Measured net ecosystem exchange (NEE) was decomposed in ecosystem respiration and gross primary productivity (GPP) based on an extrapolation of a nighttime CO2 flux-temperature regression to daytime respiration [Knohl et al., 2003]. Full details on instrumentation and calculation can be found in the work of Knohl et al. [2003] and Anthoni et al. [2004].

2.3. Model

[11] For the modeling exercise we used the one-dimensional multilayer biophysical canopy model CANVEG [Baldocchi, 1997; Baldocchi and Wilson, 2001]. CANVEG computes the biosphere-atmosphere exchange of water vapor, carbon dioxide and sensible heat flux densities and the microclimate within and above the forest at 1-h time steps. The model consists of coupled micrometeorological and eco-physiological modules (see the appendix in Baldocchi et al. [1999] for details). The micrometeorological modules compute leaf and soil energy exchange, turbulent diffusion, scalar concentration profiles and radiative transfer through the canopy. Environmental variables, computed with the micrometeorological module, in turn, drive the physiological modules that compute leaf photosynthesis, stomatal conductance, transpiration and leaf, bole and soil/root respiration. The newest version of the model, as used here, includes a soil water module, but it is most applicable for well water conditions at present. The model has been tested and validated a number of times with direct flux measurements for deciduous forests for well water conditions [Baldocchi, 1997; Baldocchi and Harley, 1995] and has performed favorably when compared with the performance of other ecosystem models [Hanson et al., 2004].

[12] For the transfer of radiation through the canopy the model computes individually flux densities of photosynthetically active (PAR), near infrared (NIR) and longwave radiation (IR). For each layer the probability of sunlit and shaded leaves is calculated. The radiative transfer model was derived from probabilistic theory and assumes that foliage is randomly distributed in space and the sun is a point source [Norman, 1979]. The probability of beam penetration is based on a Markov model, an extended Poisson distribution [Myneni et al., 1989] and the probability of sunflecks is calculated according to [Gutschick, 1991]. Penumbra is not included in the radiative transfer calculations as it is typically considered to be only of secondary importance and to affect canopy photosynthesis by less than 5–10% [Denholm, 1981]. Scattering of radiation was computed for the visible and near infrared wave bands using the slab, ‘adding’ approach of [Norman, 1979]. The model includes transmission and reflectance of radiation by leaves with reflectance (ρ) and transmittance (τ) of beech leaves taken as ρ = 0.057 and τ = 0.048 (Francois, personal communication), updated on Dufrene et al. [2005]. Leaf inclination angle distribution is calculated from mean leaf inclination angle and its standard deviation using a beta distribution [Goel and Strebel, 1984]. Leaf inclination angle is the angle between the leaf surface plane and the horizontal plane and influences the amount of direct radiation absorbed by the leaf at a given sun elevation angle. The beta distribution has shown to best represent leaf inclination angle distribution observed in the field [Wang et al., 2007]. Mean leaf inclination angle is assumed to be high (45°) at canopy top, decreases in the middle canopy and then reaches a fairly low value at the lower canopy (25°) [Planchais and Sinoquet, 1998; Utsugi et al., 2006]. Photosynthesis parameters are taken from measurements at nearby beech trees (Kutsch, personal communication). Photosynthetic capacity (Vcmax = 55 μmol m−2 s−1) is assumed to decline linearly within the canopy [Baldocchi and Harley, 1995].

2.4. Model Validation

[13] We used quasi-continuous measured flux and meteorological data (1 h time resolution) from year 2002 to calibrate and validate the model for the Hainich site. We chose year 2002 for this study since 2002 was a sufficiently wet year (986 mm precipitation) without water limitation during the summer [Knohl et al., 2008]. This way we were able to reduce the influence of soil water as a confounding factor. Also, we focused on summertime growing period only where phenology was almost constant reducing further the influence of leaf development. First, we used data from day 160 to 200 and from 251 to 290 in 2002 to calibrate the model. Details on model parameters are given in Table 1. Second, we validated the model with data from our target period from day 200 to 250 in 2002 which was then also used for all subsequent analysis. We use the following measures to report quality of our validation: (a) BIAS: slope of a linear regression line of model versus measurement. (b) R2: correlation coefficient of a linear regression line of model versus measurement showing the fraction of explained variance. (c) NSEE: normalized standard error estimate as a dimensionless estimate of relative uncertainty

  • equation image
Table 1. Key Parameters in the Model CANVEG Applied for the Hainich Site
ParameterValue
  • a

    Vcmax is the maximum carboxylation rate at canopy top.

  • b

    Jmax is the maximum rate of electron transport.

Leaf area index6 m2 m−2
Canopy height33 m
Vcmaxaat 25°C55 μmol m−2 s−1
Ratio of Jmaxb to Vcmax2
Ratio of dark respiration to Vcmax0.015
Stomatal conductance factor11
Quantum yield0.29
Leaf clumping factor0.79
Leaf reflectance0.057
Leaf transmittance0.048

[14] The model predicted latent and sensible heat fluxes and carbon dioxide fluxes well (R2 > 0.8, BIAS ≈ 1, see Table 2). Also the model predicted canopy albedo well (within 0.02) and canopy net radiation.

Table 2. Model Validationa
 BIASR2NSEE
  • a

    BIAS reflects the slope of a 1:1 plot (model versus measurement), R2 the correlation coefficient, NSEE the normalized standard error estimate, a relative measure of uncertainty.

Sensible heat flux1.0210.8690.397
Latent heat flux0.9260.8250.257
NEE0.9970.8570.185
GPP1.0040.8590.175
Net radiation1.0470.9780.152

3. Results

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Methods
  5. 3. Results
  6. 4. Discussion
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

3.1. Diffuse Radiation Effect at Two Forest Sites

[15] In order to quantify the effect of diffuse radiation (Rd) on carbon uptake we analyzed flux measurements at two nearby located forest sites that were exposed to almost identical weather conditions. Both sites had similar leaf area index (around 5–6 m2 m−2), but differed in canopy structure. The Hainich site had an uneven-aged, multilayer canopy, while the Leinefelde site had an even-aged, single layer canopy. In order to illustrate the effect of diffuse radiation on net ecosystem exchange and to reduce confounding effects of changes in global radiation (Rs), Niyogi et al. [2004] normalized the flux measurements by global radiation. Since in our case air humidity (expressed as vapor pressure deficit) and temperature influenced net ecosystem fluxes, too, we normalized net ecosystem carbon exchange fluxes (NEEmeas) by a statistical model (NEEstat) based on an extended nonrectangular light response regression that included photosynthetically active radiation (PAR), vapor pressure deficit (VPD), and air temperature (Ta).

  • equation image

where quantum yield (α), maximal assimilation (Amax), curvature (θ), and dark respiration (R) are fitting parameters. In order to account for the influence of air temperature and vapor pressure deficit we replaced quantum yield (α = a + b · Ta) and Amax (Amax = c + d · VPD) with linear regressions. This regression model resulted from tests with various regression models showing the lowest Akaike information criterion (AIC) indicating that there is no redundancy in fitting parameters [Crawley, 2002]. For our target period from day 200 to day 250 in 2002 we obtained regression parameters by nonlinear regression (SPlus, Insightful, WA, USA). The regression model explained 82% and 79% of the variance at the Hainich and Leinefelde site, respectively. We then normalized measured NEE (NEEmeas) by NEE from the regression (NEEstat).

[16] The normalized NEE fluxes at both two flux sites showed a strong and significant (p < 0.001) positive correlation with the diffuse fraction in short wave radiation, indicating increased carbon uptake with increasing diffuse radiation (Figure 1a, slope of 0.54 ± 0.06 for Hainich, 0.45 ± 0.11 for Leinefelde). The slope of this correlation reflects the strength of the diffuse radiation effect and will be used as a proxy in the subsequent analysis. Normalized NEE fluxes increase by about 50% during a rise in diffuse fraction from Rd/Rs = 0.15 to Rd/Rs = 0.85. In contrast to our hypothesis 1 (second part) the effect of diffuse light on canopy CO2 exchange at the multilayer canopy (Hainich) and at the single layer canopy (Leinefelde) did not show a significant difference as reflected by the regression slopes and their 1-σ standard deviation in Figure 1. Thus canopy structure (at similar leaf area index) seems not to play a role for the diffuse radiation effect based on direct carbon flux measurements at our two sites.

image

Figure 1. Response of normalized net ecosystem exchange NEE [dimensionless] to diffuse fraction Rd/Rs [dimensionless] in short wave radiation for measurements at the (a) Hainich and Leinefelde site and (b) for measurements and model at the Hainich site. Slope and ± standard deviation are given.

Download figure to PowerPoint

[17] Our model, parameterized for the Hainich site, reproduced the diffuse radiation effect observed in the flux measurements (Figure 1b, slope of 0.51 ± 0.05). Good performance was achieved with our one-dimensional radiative transfer model even though it did not represent penumbra or a 3-dimensional radiation transfer [Alton and North, 2007; Cescatti, 1997a, 1997b]. The validity of using one-dimensional radiative transfer schemes has been demonstrated in previous studies, where we compared the radiative transfer scheme, used in CANVEG, with measurements of light transfer through a deciduous forest [Baldocchi et al., 1985]. We found that the model computations of light transfer matched field measurements well, as long as clumping was considered. Overall, the match with observations gives confidence in the application of the model at this site. Further confidence in the biophysical model comes from an analysis based on gross primary productivity (GPP) calculated from the eddy covariance data instead of net ecosystem exchange (NEE). Results reveal an identical pattern (data not shown) with the only difference being that the slope is slightly smaller (0.43 ± 0.04); GPP is larger than NEE and therefore reduces the relative contribution of the diffuse light on the normalized fluxes.

3.2. Ecosystem Properties Influencing the Diffuse Radiation Effect

[18] In order to test whether the diffuse radiation effect on canopy photosynthesis (here we use the slope of normalized gross ecosystem productivity versus diffuse fraction in order to focus on canopy processes alone) increases with increasing leaf area index (LAI) (first part of our hypothesis 1) we ran the model over a range of leaf area indices (LAI = 1.5 m2 m−2 to 9 m2 m−2). With increasing LAI the model predicts an increase in the diffuse radiation response slope, from 0.29 at LAI = 1.5 m2 m−2 to a slope of 0.48 at a LAI = 9 m2 m−2 (Figure 2). These results support the argument of Roderick et al. [2001] that the effect of diffuse radiation is most prominent in dense closed canopies with high leaf area index. It is also consistent with field studies, where the predominant number of studies showing a diffuse light enhancement of photosynthesis come from measurements over dense forests [Baldocchi, 1997; Gu et al., 1999; Hollinger et al., 1994; Jenkins et al., 2007; Niyogi et al., 2004]. Furthermore these computations indicate that individual dominant trees may not record a positive response to diffuse radiation in their tree rings as found by Krakauer and Randerson [2003].

image

Figure 2. Diffuse radiation effect (slope of normalized gross primary productivity versus diffuse fraction) [dimensionless] in the model increases with increasing leaf area index (LAI).

Download figure to PowerPoint

[19] Next we examine the impact of other canopy properties on the diffuse radiation effect, e.g., leaf clumping factor, leaf inclination angle and leaf optical properties such as leaf reflectance and transmittance. Our results show that the diffuse radiation effect decreases with leaf clumping factor, i.e., it is highest for clumped canopies (leaf clumping factor small) and lowest for canopies with randomly distributed leaves (leaf clumping factor close to one) (Figure 3, top). Also, the diffuse radiation effect increases with increasing leaf inclination angle (Figure 3, middle). Both canopy properties control how deeply radiation penetrates into the canopy. A small leaf clumping factor, for example, indicates that leaves are clumped and form aggregates. This effect results in a higher transfer of radiation into lower canopy layers compared to nonclumped canopies (data from model, not shown). Similarly, high leaf angles result in a deeper penetration of radiation inside the canopy (data from model, not shown). Our results suggest that the diffuse radiation response is stronger in canopies that allow a stronger radiation transfer into the canopy, supporting hypothesis 2. The strongest influence, however, was observed for variations in leaf transmittance and, to a lesser degree, in leaf reflectance. Increasing transmittance and reflectance lead to a decline in the diffuse radiation effect (Figure 3). Leaf transmittance and reflectance control the transformation of direct radiation into diffuse radiation within the canopy through scattering effects. Canopies with low transmittance or reflectance scatter radiation less strongly and produce less diffuse radiation themselves. Consequently they are more sensitive to diffuse radiation imposed by cloudy conditions or aerosols.

image

Figure 3. Sensitivity of the diffuse radiation effect [dimensionless] in the model for changes in (a) leaf clumping factor [dimensionless], (b) leaf inclination angle at canopy top [°], and (c) leaf optical properties [dimensionless]. Default values for leaf clumping factor = 0.79, for leaf inclination angle = 55°, for transmittance = 0.048, and for reflectance = 0.057.

Download figure to PowerPoint

3.3. Changes in Radiation With Diffuse Fraction

[20] In order to test how ecosystem fluxes and processes change with an increase in diffuse fraction we performed model experiments over a range of diffuse fraction. If diffuse fraction increases, the total amount of radiation received at the surface decreases due to absorption and reflection of radiation by clouds or aerosols [Spitters et al., 1986]. Roderick [1999] shows that diffuse fraction and atmospheric transmission (Rs/R0) are strongly and linearly correlated over a large range. For our study we derived a local relationship between atmospheric transmission and diffuse fraction (Rd/Rs) from surface measurements and use this relationship to vary global radiation along with diffuse fraction (Figure 4a). Atmospheric transmission is calculated from surface measurements of global radiation (Rs) and estimated irradiance at the top of the atmosphere (R0) based on solar-earth geometry according to Spitters et al. [1986] and Gu et al. [2003]. Please note that we inverted axes in our Figure 4a compared to Roderick [1999], as we wanted to express transmission in dependence of diffuse fraction. Also, please note that at high diffuse ratio (Rd/Rs > 0.85) and low atmospheric transmission (Rs/R0 < 0.25) the linear relationship breaks down [Roderick, 1999]. We therefore performed our model experiments only over a range from Rd/Rs = 0.1 to Rd/Rs = 0.8.

image

Figure 4. (a) Relationship between atmospheric transmission (Rs/R0) and the diffuse fraction (Rd/Rs) (Rd/Rs < 0.85: Rs/R0 = −0.7657 Rd/Rs + 0.8842, R2 = 0.855, P < 0.001, n = 1125) and (b) between normalized atmospheric long wave radiation (Ls/Lclear sky(Ta)) and the diffuse fraction (Rd/Rs) (Ls/Lclear ky(Ta) = 0.4225 Rd/Rs + 0.9059, R2 = 0.773, P < 0.001, n = 1125) where Rd, Rs and R0 are the diffuse, global and top of the atmosphere solar irradiance. Ls is long wave irradiance at the surface and Lclear sky(Ta) is calculated long wave radiation from the atmosphere based on air temperature (Ta) and Stefan-Boltzmann law with emissivity according to Swinbank [1963]. All data are 30 min values measured at the top of the tower at the Hainich site.

Download figure to PowerPoint

[21] With an increase in diffuse radiation we also observed an increase in long wave irradiance from the atmosphere (Figure 4b). In order to include this observation in our model experiment we used the linear relationship of normalized atmospheric long wave radiation (Ls/Lclear sky(Ta)) and the diffuse fraction (Rd/Rs). We normalized measured long wave irradiance at the surface with a clear sky long wave radiation as calculated from the Stefan-Boltzmann law (based on air temperature) and a clear sky emissivity after Swinbank [1963] which includes the dependency of atmospheric emissivity on air temperature (directly) and atmospheric water content (indirectly). This consideration and inclusion of longwave radiation allows us to evaluate net radiation and surface energy balance correctly.

[22] Under constant solar radiation (R0) at the top of the canopy, increasing diffuse fraction (Rd/Rs) leads to a decrease in beam (direct) radiation (Rb) and an initial increase in diffuse radiation (Rd) (Figure 5). Due to absorption and reflection processes in the atmosphere (at clouds or aerosols) the overall amount of solar radiation, i.e., global radiation, reaching the earth surface (Rs, the sum of Rb and Rd) decreases with diffuse fraction. As a consequence, the flux density of diffuse radiation decreases at high diffuse fraction (Rd/Rs > 0.6). It is important to note that the exact relationship between diffuse fraction (Rd/Rs) and radiation at the surface (Rs) depends on scattering and absorption processes taken place in the atmosphere and are expected to be different for clouds and aerosols. A multisite study across the United States shows that increasing aerosols loading will increase the diffuse fraction of the radiation without significantly reducing the total radiation itself [Niyogi et al., 2004]. Consequently our approach to reduce global radiation (Rs) along with increasing diffuse fraction leads to a conservative estimate of the effect of diffuse radiation on canopy processes.

image

Figure 5. Change in solar radiation at the surface (Rs), beam radiation (Rb) and diffuse radiation (Rd) and in long wave irradiance from the atmosphere (Ls) with changes in diffuse fraction (Rd/Rs) under constant solar radiation at the top of the atmosphere (R0) based on the radiation measurements at the Hainich site (regression in Figure 4).

Download figure to PowerPoint

3.4. Response of Ecosystems to Diffuse Fraction

[23] For the model experiments we selected five sunny clear sky days (Days 169, 177, 190, 210, 230 of 2002) which had a diffuse fraction of Rd/Rs ≈ 0.1 at noon and increased stepwise the diffuse fraction of each 1-h data point by 0.1 steps up to Rd/Rs ≈ 0.7. Along with diffuse fraction we modified global radiation and long wave irradiance according to the regression obtained in Figure 4 and as shown in Figure 5. All other environmental drivers were kept as measured on those days. As a result we separated confounding effects of air temperature and vapor pressure deficit allowing us to study the effect of radiation changes alone. In the following we present daily values weighted by photosynthetically active radiation and then averaged across all five days (±s.e., n = 5). Since we use radiation-weighted daily averages, our diffuse fraction varies between Rd/Rs = 0.16 (measured) to Rd/Rs = 0.76 (model experiment). We apply the appropriate error propagation where derived variables were calculated, e.g., water use efficiency. The diffuse radiation effect (0.56 ± 0.11, slope of normalized NEE fluxes versus diffuse fraction) over those five days with the modified diffuse fraction matches well the diffuse radiation effect from model (0.51 ± 0.05) and measurements (0.54 ± 0.06) over our target period from day 200 to 250 in 2002. This finding indicates that our artificially changed radiation regime reflects natural conditions well.

[24] Canopy photosynthesis and net ecosystem exchange (includes leaf, stem and soil respiration), show nearly identical increases with diffuse fraction for Rd/Rs ≤ 0.45 (Figure 6). At high diffuse fraction (Rd/Rs > 0.45) canopy photosynthesis and net ecosystem exchange (NEE) decrease, reflecting the overall reduction in radiation (see Figure 5). It is important to note that the absolute change in NEE is almost identical to the absolute change in canopy photosynthesis. This observation suggests that the impact of changes in diffuse fraction on respiration processes in soil, leaf, and stem are minor compared to changes in photosynthesis. Canopy transpiration also increases with diffuse fraction. When evaluated in relative terms the impact of diffuse light on transpiration is smaller than for canopy photosynthesis. When carbon and water fluxes are viewed in tandem, we find an initial increase in water use efficiency (photosynthesis/transpiration) supporting our hypothesis 3. Only at very high diffuse fraction does water use efficiency decline due to a higher reduction in photosynthesis compared to transpiration. Our model experiments do not include surface-atmosphere feedbacks and hence they may underestimate the effect of surface cooling under cloudy conditions. To incorporate these complex feedbacks a fully coupled surface layer-boundary layer model including cloud generation and longwave radiation changes would have been required [McNaughton and Spriggs, 1986], but this was beyond the scope of our analysis. In order to approximate the impact of such an surface cooling (including surface warming through increased longwave radiation from increased cloud cover), we derived a linear regression between air temperature (Tair) and diffuse fraction (Rd/Rs) from local measurements at the Hainich site. This regression (Tair = −7.669 Rd/Rs + 22.139, R2 = 0.238, n = 1419) was significant (p < 0.0001) but contained large scatter (NSEE = 0.31) and implied that air temperature varied from about 16°C to 21°C as the diffuse fraction ranged from 0.8 to 0.1. If we apply this regression to modify air temperature according to diffuse fraction and rerun the model we find a stronger increase in canopy photosynthesis and net ecosystem uptake with diffuse fraction since leaf, stem and soil respiration are reduced due to lower temperatures (data not shown). This effect is not more than 25% of the response without such a feedback. Shape and optima of the curves in Figure 6, however, remain identical. Transpiration also responds to surface air cooling due to the associated decrease in vapor pressure deficit; it declines by 25% at high diffuse fraction and forces an incremental increase in water use efficiency.

image

Figure 6. (a) Changes in canopy photosynthesis and net ecosystem productivity and (b) canopy transpiration and water-use-efficiency with diffuse fraction (Rd/Rs). Values are averages (±standard error) of five days (each hour of each day weighted by photosynthetically active radiation).

Download figure to PowerPoint

3.5. Mechanisms of the Diffuse Radiation Effect

[25] Since Figure 6 indicates that canopy photosynthesis might be the dominant factor controlling the diffuse radiation effect, and not respiration related processes, we investigated the impact of diffuse and direct radiation on individual canopy layers. For the same model runs and days, as in Figure 6, we see that the diffuse fraction has little effect on leaf photosynthesis of sunlit leaves (Figure 7, top right). Only at high diffuse fraction does a small reduction in leaf photosynthesis occur (mainly in the middle canopy layers). In contrast, there is a strong increase in leaf photosynthesis of the shaded fraction of canopy leaves (Figure 7, bottom right). At high diffuse fraction, however, shade leaf photosynthesis decreases as well. The impact of diffuse fraction on leaf respiration is comparatively small both for sunlit as well as shaded leaves (Figure 7, left) even though leaf temperature decreases with increased transpiration (see below) and does not significantly contribute to the overall observed increase in net ecosystem productivity with diffuse fraction. Leaf respiration is fairly small as we considered photoinhibition of respiration under light by 50% in our current model implementation [Amthor et al., 1994]. Since the process and exact value of photoinhibition is not yet resolved [Tcherkez et al., 2005], we performed a sensitivity analysis and found that including photoinhibition or not did not change the relevance of respiration for the diffuse radiation effect.

image

Figure 7. Leaf gross photosynthesis and leaf respiration per m2 ground area (daily averages weighted by PAR at canopy top, n = 5 d) in each layer for different diffuse fractions.

Download figure to PowerPoint

[26] In order to understand why the decrease in direct radiation (following Figure 5) had a minor impact on leaf photosynthesis while the increase in diffuse radiation led to a strong increase in leaf photosynthesis we compared the distribution of leaf area receiving radiation under different diffuse fraction scenarios. Under sunny conditions with low diffuse fraction (Rd/Rs = 0.16) a large part of leaf area received a high load of photosynthetically active radiation (PAR) (Figure 8). Under conditions with high diffuse fraction, the leaf area distribution is shifted, i.e., sunlit leaves receive less radiation, and shaded leaves receive more radiation. Since leaf photosynthesis saturates at high PAR values (with our model parameters at around PAR > 600 μmol m−2 s−1) an overall reduction in direct PAR from 1200 to 600 μmol m−2 s−1 reduces leaf photosynthesis marginally. On the other hand, an increase in the low range of PAR results in a steep increase in leaf photosynthesis.

image

Figure 8. (a) Changes in leaf area distribution receiving photosynthetically active radiation for conditions with different diffuse fraction (Rd/Rs) and radiation response curve of leaf photosynthesis. (b) Corresponding change in canopy photosynthesis with diffuse fraction and histogram of diffuse fraction during daytime (Rs > 50 W m−2) measured at the Hainich site in 2002.

Download figure to PowerPoint

3.6. Effect of Diffuse Fraction on Other Ecosystem Processes

[27] Since the response of ecosystem to diffuse radiation leads to an increase in water use efficiency we would expect to see changes in carbon isotope discrimination as water use efficiency and carbon isotope discrimination are typically closely coupled. If we consider a simple big leaf model we can describe water use efficiency (WUE) at the leaf level as follows:

  • equation image

where GPP denotes canopy photosynthesis or gross primary productivity [μmol m−2s−1], T transpiration [mmol m−2s−1], gs stomatal conductance [μmol m−2s−1], pa atmospheric pressure [kPa], ca atmospheric CO2 mixing ratio [ppm], ci the CO2 mixing ratio inside the stomata cavity, ea atmospheric water vapor pressure [kPa]. ei the water vapor pressure inside the stomata [kPa]. Equation (3) can be rearranged and solved for ci/ca resulting in:

  • equation image

[28] Carbon isotope discrimination Δ [‰] is linearly related to ci/ca in a first approximation as follows [Farquhar et al., 1982]:

  • equation image

[29] Often (eiea) is assumed to be equal to the atmospheric vapor pressure deficit as the pore space underneath the leaf stomata is typically water saturated. Following this approach we would expect to see a decline in ci/ca, and hence a decline in carbon isotope discrimination with the observed increase in water use efficiency. Our calculations with CANVEG, however, show the opposite, i.e., an increase in ci/ca and Δ with increasing Rd/Rs and increasing water use efficiency (Figure 9). The two differences between the calculation in CANVEG and with the simple big leaf model (equation (4)) are: (a) that in CANVEG we do not assume that (eiea) is equal to the atmospheric vapor pressure deficit; the model calculates ei directly based on leaf surface temperature and (b) that in CANVEG we include boundary layer conductance which are slightly different for water vapor and CO2. If we include the correct (eiea) based on leaf temperature we get the same response in ci/ca to water use efficiency (Figure 9, bottom). The offset is largely explained by not considering the different boundary layer conductances in the simple big leaf model. The zig zag at high water use efficiency results from different leaf surface temperature and hence different (eiea) due to different diffuse fraction at a similar water-use efficiency.

image

Figure 9. (a) Changes in canopy discrimination and ratio of CO2 concentration inside the stomata and in the atmosphere (ci/ca) with diffuse fraction. (b) Relationship between ci/ca as well as canopy discrimination and water use efficiency calculated based on the multilayer model CANVEG, a simple big leaf model with eiea = atmospheric vapor pressure deficit and a simple big leaf model with eiea = saturation vapor pressure at leaf temperature minus atmospheric vapor pressure.

Download figure to PowerPoint

[30] The concurrent decline in global radiation (Rs) with increasing diffuse fraction also leads to a reduction in leaf temperature by about 1.5°C over a range from Rd/Rs = 0.16 to Rd/Rs = 0.76 supporting hypothesis 4. As a consequence other processes related to leaf temperature are affected as well (Figure 10). Most prominently our model suggests that isoprene emitting tree species, such as oak trees, would show a decline in isoprene emissions from about 65 nmol m−2 s−1 to 32 nmol m−2 s−1. This decline, however, is mostly driven by changes in leaf temperature due to reduced global radiation. Instead, if we keep global radiation constant and only increase diffuse fraction, the isoprene emission would rise slightly caused by a somewhat larger leaf area contributing to the isoprene emissions.

image

Figure 10. Changes in isoprene flux and leaf temperature with diffuse fraction (Rd/Rs). Values are averages (±standard error) of five days (each day weighted by photosynthetically active radiation).

Download figure to PowerPoint

3.7. Influence of Diffuse Fraction on Annual Fluxes

[31] To address the ecological implications of diffuse light on ecosystem metabolism we must evaluate its effect of diffuse light on canopy photosynthesis and net ecosystem exchange on annual timescales. In preliminary computations we found that net annual canopy carbon dioxide exchange increased by 78 g C m−2 y−1 at a deciduous forests in Tennessee, USA when we reduced direct radiation by 20% and converted this energy to diffuse radiation; but we did not change global radiation [Baldocchi et al., 2002]. Since changes in cloud cover and in aerosol loading might results in a different relationship between diffuse radiation fraction and global radiation received at the earth surface [Niyogi et al., 2004], we test two new cases at annual time integrals: (A) global radiation reduces and long wave incoming radiation increases with increased diffuse radiation fraction according to our local measurements (see Figures 4 and 5) and (B) global radiation and long wave incoming radiation remain constant even if diffuse radiation fraction is increasing (Table 3). Our model simulations give two contrasting results: In A, GPP decreases at the annual timescale by 3 to 8% with an increase in diffuse radiation of 5 to 15%, while in B, GPP increases by 2 to 7%. Other processes such as latent heat fluxes, sensible heat fluxes, net radiation, and isoprene fluxes also show a decrease in scenario A indicating that the decline in global radiation overrules any benefit from an increase in diffuse fraction. Scenario B, where direct radiation is not changed with diffuse fraction, results in increases in the respective fluxes. The only exception is sensible heat flux which declines with increased diffuse fraction since latent heat fluxes increase stronger with diffuse fraction than net radiation causing sensible heat flux to compensate for the missing energy.

Table 3. Impact of Increases in Diffuse Fraction on Ecosystem Processes at an Annual Scalea
 Annual Flux (Simulated)Rs Declines With Rd/Rs IncreaseRs Independent of Rd/Rs Increase
+5%+10%+15%+5%+10%+15%
  • a

    Estimates are given for scenario where global radiation (Rs) reduces (long wave incoming radiation increases) concurrently with increases in diffuse fraction according to our local measurements and scenario where global radiation (and long wave incoming radiation) is not influenced by increases in diffuse fraction. All other input variables are kept constant.

ΔGPP1579 g C m−2 y−1−3%−6%−8%+2%+5%+7%
ΔLE1106 MJ m−2 y−1−3%−5%−7%+2%+3%+4%
ΔH406 MJ m−2 y−1−12%−24%−34%−2%−3%−5%
ΔRn1453 MJ m−2 y−1−5%−11%−15%+1%+1%+2%
ΔFisoprene56 mol m−2 y−1−3%−6%−9%+1%+2%+3%

4. Discussion

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Methods
  5. 3. Results
  6. 4. Discussion
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

4.1. What Explains the Diffuse Radiation Effect?

[32] Our model simulations show an initial increase in GPP and NEE with diffuse fraction but a decline in fluxes after an optimum, around Rd/Rs = 0.45 (Figure 6). The response function reflects a superposition of an increase in GPP and NEE with diffuse fraction and a decrease associated with the concurrent decline in global radiation. One may argue that fluxes beyond the optimum are still larger than if radiation was reduced but with no additional diffuse radiation, e.g., clear sky radiation in the morning or evening hours. Thus, radiation use efficiency is expected to remain larger even beyond the optimum compared to clear sky conditions. This observation was made independently at two mixed deciduous forests in Eastern United States [Jenkins et al., 2007; Min, 2005]. Our observed optimum of about Rd/Rs = 0.45 is higher than the 0.31 estimated by Roderick et al. [2001] based on a quadratic function representing a range of crop and forest sites, but lower than the 0.6 to 0.7 as suggest by Alton et al. [2007b]. Measurements in a Nothofagus forest in New Zealand also suggest a higher optimum [Hollinger et al., 1994]. Our model results in Figures 2 and 3 suggest that the exact location of the optimum depends on canopy properties such as leaf area index, leaf optical parameters, leaf clumping and leaf angle and will therefore vary among ecosystems.

[33] Overall, our model results suggest that the enhanced penetration of radiation into the canopy under diffuse radiation conditions and the more even distribution of radiation among shade and sun leaves explain the diffuse radiation effect. Conceptually, a higher diffuse fraction leads to a decline in radiation load on sunlit leaves and an increase in radiation load on shade leaves (Figure 11a). The radiation reduction on sunlit leaves, however, does not initially result in a substantial decrease in leaf level photosynthesis as they remain radiation saturated. Shade leaves, however, operate on the linear part of the light response curve and therefore respond sensitively to more radiation resulting in an increase in shade leaf photosynthesis. As a result overall canopy photosynthesis rises even when sum of diffuse and direct radiation decreases (Figure 11b). Thus, canopy photosynthesis has a steeper response to incoming photosynthetically active radiation (PAR) under diffuse conditions than under clear sky conditions resulting in a higher radiation use efficiency (Figure 11b). Only at high diffuse fraction (Rd/Rs > 0.45) when a large fraction of sunlit leaves are not anymore radiation saturated, canopy photosynthesis starts to decline. This mechanism seems to dominate compared to other potential mechanisms such as the interaction with vapor pressure deficit (keep constant in our model analysis) as well as a reduction in leaf, stem or soil respiration (see Figures 6 and 7). A reduction in leaf, stem or soil respiration might gain more relevance if surface-atmosphere feedbacks were included via a coupled surface layer-planetary boundary layer model. Our first guess approximation of such a feedback, based on a local air temperature versus diffuse fraction regression, indicated that the temperature effect would account for less than 25% of the overall effect. A model experiment, though, cannot prove that the suggested canopy radiation distribution mechanism is correct, as we could model the right results for the wrong reason. It does, however, show that this mechanism is sufficient to explain the diffuse light effect as observed from eddy covariance measurements. At other sites or times, e.g., with a different canopy structure or where soil water limitations become relevant, other mechanisms such as interaction between diffuse fraction and precipitation may become important as well. Our findings, though, confirm and provide mechanistical details for the hypothesis proposed by Roderick et al. [2001], Farquhar and Roderick [2003], and Alton et al. [2007b] stating that the redistribution of radiation load among sun and shade leaves under diffuse conditions causes the observed diffuse radiation effect. Similar results were found in a Czech spruce stand where NEE was increased by up to 150% under cloudy conditions and where the more efficient penetration of radiation inside the canopy was considered to be the most relevant process [Urban et al., 2007].

image

Figure 11. Conceptual figure showing (a) the response of sunlit and shade leaves to changes in photosynthetically active radiation (PAR) under increased diffuse radiation and decreased direct radiation, (b) the response of canopy photosynthesis to a decrease in PAR and increase in leaf photosynthesis.

Download figure to PowerPoint

4.2. How Does Diffuse Fraction Affect Other Ecosystem Processes?

[34] Our model results show water use efficiency responds to diffuse light because canopy photosynthesis responds more strongly to diffuse radiation than transpiration (Figure 6). This confirms observation from a deciduous forest in Massachusetts (USA) where water use efficiency was enhanced under cloudy conditions [Min, 2005]. Furthermore, our results exhibit an increase in carbon isotope discrimination with increasing water use efficiency (Figure 9). This is counterintuitive based on common knowledge produced by the simpler big leaf model, which suggests a decline in carbon isotope discrimination with increasing water use efficiency, but it is consistent with direct measurements of leaf stable carbon isotopes in shading experiments [Israeli et al., 1996; Yakir and Israeli, 1995]. Our observation indicates that the relationship between water use efficiency and carbon discrimination is overruled by a thermal disequilibrium between air and leaves. As a consequence, the leaf to air vapor pressure gradient (eiea) does not equal atmospheric vapor pressure deficit. This result confirms findings by Baldocchi and Bowling [2003] and indicates that simple big leaf models that are used to calculate water use efficiency or ci/ca may lead to erroneous values if they assume that (eiea) equals the atmospheric vapor pressure deficit. Our finding is a typical example of a nonlinear feedback in the processes controlling biosphere-atmosphere exchange: water use efficiency depends on leaf surface temperature which itself is influenced by leaf transpiration, a major component of water use efficiency. Simple big leaf models that do not include such feedbacks might therefore be limited in their usefulness for understanding processes in biosphere-atmosphere exchange.

[35] Isoprene is the dominant nonmethane hydrocarbon emitted from temperate broadleaved forest canopies and is produced by many different tree species during the carbon fixation pathway [Fuentes et al., 2000]. Isoprene is of interest in atmospheric chemistry because it contributes to the formation of the radiatively active oxidant gas, ozone, which it a strong tropospheric pollutant. Our results suggest that the reduced global radiation under cloudy conditions leads to surface cooling and hence to a decline in isoprene emission from isoprene-producing tree species. This effect might have been even more pronounced if we had included a planetary boundary layer model that captures surface-atmosphere feedbacks resulting in a decline in air temperature under cloudy conditions. Our first guess approximation of such a feedback based on a local air temperature versus diffuse fraction regression indicated that the decline in isoprene emissions would be enhanced by additional 30%. Our canopy model combines a leaf level model of isoprene emission [Guenther et al., 1993] with the multilayer canopy model CANVEG [Baldocchi et al., 1999]. There is a growing body of literature related to isoprene emissions from plants suggesting that other factors such as available photosynthates control isotope emissions as well (e.g., [Arneth et al., 2007; Niinemets et al., 1999; Zimmer et al., 2000] and references reviewed therein). We, therefore, consider our results as a first approximation. A more biochemical-based model would be needed to fully understand the physical-biochemical feedbacks underlying the emission of isoprene and its response to changes in diffuse fraction.

4.3. How Do Changes in Diffuse Fraction Influence Annual Fluxes?

[36] Overall, we are not only interested in understanding the mechanisms of the diffuse radiation effect on ecosystem processes during episodes, but also in quantifying the relevance of potential changes in diffuse radiation on ecosystem processes at an annual timescale. When we run the model over an entire year we find contrasting results depending on whether global radiation reduces with diffuse fraction (scenario A) or remains unaffected by diffuse fraction (scenario B). In scenario A, GPP, LE, H, net radiation as well as isoprene fluxes decrease, while they increase under scenario B (except for H). One may wonder why GPP is decreasing in scenario A even for a small increase in diffuse radiation indicating that the ecosystem is not benefiting from an increase in diffuse fraction at an annual scale. Looking closer at the distribution of diffuse fraction during daytime (Figure 8) it becomes apparent that more than 61% of daytime data are already beyond the maximum of the diffuse radiation response (approx. at a diffuse fraction of 0.45, Figure 5). As a consequence an increase in diffuse fraction across all conditions will overall lead to a decrease in ecosystem GPP even though there are individual times (days with low diffuse fraction) when the ecosystem profits from more clouds and hence more diffuse fraction. If cloudiness were to increase only during midday or previously very sunny conditions then we would expect to see a positive response in scenario A. These findings are supported by a recent study in a boreal forest in Siberia where a decline in NEE by 9% is estimated under conditions with increased cloudiness even though there is a general positive response to diffuse radiation [Alton et al., 2007b]. Additionally, potential interaction between diffuse fraction and precipitation may become relevant when we investigate dry year or dry sites. Similarly, surface-atmosphere feedbacks resulting in surface air cooling maybe also reduce respiration fluxes and hence change the overall response of NEE. One important question that remains open is whether forest ecosystems are adapted with their canopy structure, leaf area index and distribution of Rubisco within the canopy to optimally use the locally existing diffuse radiation regime, i.e., adapting the optimum of the diffuse radiation response to the distribution of diffuse fraction at the site. Our data are inconclusive as our model does not yet include biosphere feedbacks on the planetary boundary layer and acclimation affects. A fully coupled biosphere atmosphere model including planetary boundary layer processes and acclimation or a synthesis across a multitude of sites may help to elucidate this point. A recent study using the land-surface model JULES suggests that most of the forest stands are not fully light-acclimated and hence could benefit by 8–13% in gross primary productivity if they optimize the distribution of leaf nitrogen to acclimate to the local light environment [Alton and North, 2007].

5. Conclusions

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Methods
  5. 3. Results
  6. 4. Discussion
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

[37] Forest ecosystems across the globe show an increase in ecosystem carbon uptake efficiency under conditions with high fraction of diffuse radiation. This phenomenon seems to be an emergent ecosystem property that is not observed at the leaf level. Here, we combine eddy covariance flux measurements and canopy-scale modeling in order to investigate the impact of diffuse radiation on various canopy gas exchange processes such as canopy photosynthesis, water use, carbon isotope discrimination and isoprene emission and to elucidate the underlying mechanisms.

[38] At our two study sites in central Germany, the response to diffuse radiation was almost identical - even though canopy structure was different (multilayer versus single-layer) - and was well reproduced by our multilayer biophysical model CANVEG. Our model computations suggest that the diffuse radiation effect increases with leaf area index and with leaf inclination angle, but decreases with leaf clumping, leaf reflectance and leaf transmittance indicating the important role of radiative transfer through the canopy for the diffuse radiation effect.

[39] In order to simulate the impact of a potential change in diffuse radiation we derived from measurements local relationships of atmospheric transmission and sky long wave radiation with diffuse fraction and varied incoming radiation along with diffuse fraction. Our model experiments showed that both canopy photosynthesis as well as transpiration initially increase with diffuse fraction, but decrease after the optimum at a diffuse fraction of 0.45 due to the reduction in global radiation. Our results show that the increased contribution of shaded leaf area to canopy photosynthesis is sufficient to explain the diffuse radiation effect observed with eddy covariance flux measurements. Interaction with vapor pressure deficit and reduced leaf respiration seems to be of minor importance (only up to 25% of the overall effect). Since our results suggest that the layered structure of a canopy is essential to explain the diffuse light effect, it seems likely that global scale models would benefit in their capacity of modeling the response to diffuse light from a multilayer canopy. Alton et al. [2007a] argue that current global models that do not include multilayer canopies overestimate global gross primary productivity by about 10% and do not capture the response of ecosystems to diffuse light. Consequently, global models will remain limited in their ability to assess the impact of air pollution and increased cloudiness on the global carbon cycle.

[40] The initial increase in canopy photosynthesis exceeds the increase in transpiration leading to an associated rise in water-use-efficiency. In contrast to commonly used simple big-leaf models our multilayer canopy model predicts an increase in carbon isotope discrimination with water-use-efficiency and hence diffuse fraction. This resulted from thermal disequilibria between air and leaf causing a discrepancy between atmospheric vapor pressure deficit and the leaf to air water vapor gradient. This aspect becomes essential when stable isotopes of carbon are used to derive water use efficiency or vice versa.

[41] At annual timescale, the overall effect of a potential increase in diffuse radiation due to higher cloud cover or higher aerosol load seems to depend on the associated effect on global radiation. Only under conditions when global radiation is not or only to a minor degree affected by changes in diffuse fraction an overall increase carbon uptake seems to occur. Under conditions when global radiation is decreasing annual carbon uptake seems to reduce even though there is a positive response to diffuse radiation at individual times. At annual scales, however, the distribution of diffuse fraction is already beyond the optimum for 61% of the data resulting in an overall decline in carbon uptake. The interaction between diffuse fraction and atmospheric transmission seems to determine the overall effect of a change in diffuse fraction on ecosystem carbon uptake.

Acknowledgments

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Methods
  5. 3. Results
  6. 4. Discussion
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

[42] This research was financed by Marie Curie fellowships to A.K. under contract MOIF-CT-2004-002543 and contract MEXT-CT-2006-042268. Model development by D.B. was supported in part by the Office of Science (BER), U.S. Department of Energy, grant DE-FG02-03ER63638. We thank the administration of the National Park Hainich for the possibility to conduct our research at this site, and we thank Detlef Schulze (MPI Biogeochemistry) and Olaf Kolle (MPI Biogeochemistry) with his team for the support with the flux measurements. We thank Matthias Cuntz (MPI Biogeochemistry) for valuable discussions on CANVEG.

References

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Methods
  5. 3. Results
  6. 4. Discussion
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Methods
  5. 3. Results
  6. 4. Discussion
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information
FilenameFormatSizeDescription
jgrg378-sup-0001-t01.txtplain text document0KTab-delimited Table 1.
jgrg378-sup-0002-t02.txtplain text document0KTab-delimited Table 2.
jgrg378-sup-0003-t03.txtplain text document1KTab-delimited Table 3.

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.