Environmental control of leaf area production: Implications for vegetation and land-surface modeling


  • Sharon A. Cowling,

    1. National Center for Ecological Analysis and Synthesis (NCEAS), University of California (UCSB), Santa Barbara, California, USA
    2. Now at Department of Geography, University College London, London, UK.
    3. Permanently at Department of Geography, University of Toronto, Toronto, Ontario, Canada.
    Search for more papers by this author
  • Chris B. Field

    1. Department of Plant Biology, Carnegie Institution of Washington, Stanford, California, USA
    Search for more papers by this author


[1] Leaf surface area per unit ground cover (leaf area index, LAI) is one of the major controls on plant productivity and biospheric feedbacks on atmospheric energy and water exchanges. Nearly all vegetation and land-surface models include parameterizations of LAI, however not much research currently focuses on the validation of simulated responses of LAI to environmental change. The objective of our research was to quantitatively review the plant science literature to extract information on the response of LAI to variations in soil moisture, soil fertility and atmospheric CO2. Our synthesis confirms that LAI is likely co-limited by a number of resources, including water, nitrogen and light. Atmospheric CO2 influences LAI in much the same manner as other plant resources. When CO2 supply is strongly limiting to gross primary production (i.e., at relatively low CO2 concentrations), LAI is strongly correlated with CO2, whereas when CO2 is abundant, LAI sensitivity to CO2 dramatically decreases. Such a nonlinear relationship between leaf area production and atmospheric CO2 may introduce a potential bias for global change modeling, particularly in the simulation of low-density vegetation that has the potential to significantly increase canopy size without inducing self-shading.

1. Introduction

[2] How much photosynthetic machinery is produced by vegetation is an important determinant of carbon accumulation in ecosystems. The ratio of leaf surface area to unit ground cover called leaf area index (LAI, m2 m−2), is an integrative measure of carbon and water balance in plants because it describes the potential surface area available for leaf gas exchange (CO2, O2 and H2O). LAI is frequently correlated with net primary production (NPP) of both conifers [Korner and Arnone, 1992; Cutini, 1996; Warren and Adams, 2000; Reich et al., 2001] and deciduous species [Bolstad et al., 2001]; therefore, not surprisingly, it is a standard parameter in models estimating terrestrial productivity.

[3] The representation of LAI in global vegetation models differs according to underlying conceptual framework [Cramer et al., 1999]. LAI can be prescribed based on published biome-specific values estimated from field experiments and remotely-sensed data [Law and Waring, 1994; Chen and Cihlar, 1996; Carlson and Ripley, 1997] or can be simulated from first principles using NPP, carbon allocation schemes [Friedlingstein et al., 1999], and resource optimization theory [Bondeau et al., 1999]. Intermodel differences in the parameterization of LAI can often account for discrepancies in the predicted strengths of terrestrial carbon sinks [Schimel et al., 2000], as well as inter-annual variations in NPP and net ecosystem production (NEP) [Bondeau et al., 1999].

[4] Because LAI represents the major surface area for gas exchange between the atmosphere and terrestrial biosphere, it is also an important parameter in interactive models of land surface and atmospheric processes [Sellers et al., 1996; Field and Avissar, 1998; Pielke et al., 1998; Bounoua et al., 1999]. LAI influences surface energy exchange through the partitioning of incoming solar radiation into latent and sensible heat fluxes [Pielke, 2001; Pielke et al., 1998; Buermann et al., 2001; Law et al., 2001], and not only has the potential to influence regional climate [Eastman et al., 2001] but also global atmospheric circulation [Zhou et al., 2001a; Chase et al., 1996].

[5] Common to both terrestrial productivity and land-surface models is an assumption that LAI has been suitably parameterized to capture responses to various environmental constraints. As such, much contemporary research centers on comparisons of simulated LAI among suites of vegetation models [Parton et al., 1996; Bondeau et al., 1999; Kucharik et al., 2000]. Fewer studies focus on validating simulated responses of LAI to environmental change [Osborne and Woodward, 2001]. More research is needed to better understand why LAI simulations differ between models, and to help identify which of the possible LAI parameterization frameworks best captures empirical trends.

[6] Our ability to realistically characterize the degree of LAI response to environmental change is critical for global change research for several reasons. LAI is the major, if not dominant control of gross primary production (GPP) and evapotranspiration (ET). Small increases in leaf area production can substantially expand the potential for whole-plant photosynthetic carbon uptake, particularly in low-density vegetation where self-shading is nonlimiting to canopy development. For research concerning the response of vegetation to future doubled atmospheric CO2 and warmer climate (“2 × CO2” used hereafter), the reliability of predicted LAI becomes even more important. Increases in LAI may counter-balance stomatal responses to CO2 (i.e., canopy-level compensatory responses [Field et al., 1995]), ultimately negating the improvement in water use promoted by decreased stomatal conductance through LAI enhancement of rates of transpiration.

[7] The objectives of our research are threefold: (1) to review the plant science literature on the response of LAI to various environmental changes, with the intent of initiating a database from which broad model-data comparisons can be made; (2) to synthesize LAI data for response trends, including comparisons along lines of high versus low LAI vegetation, vegetation type, and in the case of atmospheric CO2, treatment methodology (i.e. open- versus closed-topped chambers); and (3) to discuss global change modeling activities as they relate to a generalized theory of external and internal controls on leaf area production.

[8] The data we collected describes the response of LAI to three environmental variables: soil moisture, soil nutrients (with an emphasis on nitrogen) and atmospheric CO2. To our knowledge, this is the first attempt to quantify LAI response to environmental change for use in data-model comparisons. We recognize the importance of temperature in modifying a variety of processes affecting leaf surface area [Marshall et al., 1992; Battaglia et al., 1998; Gunn and Farrar, 1999]; however, due to lack of data, we are unable to address them here.

[9] We supplement our data analysis with a data-model comparison of LAI responses to subambient atmospheric CO2 concentrations, using the biogeochemistry–biogeography vegetation model, BIOME3 [Haxeltine and Prentice, 1996]. Model sensitivity experiments of LAI predicted by BIOME3 over lower than ambient CO2 concentrations had been performed previously [Cowling et al., 2001], with conclusions made about the response of paleoecosystems to paleo-CO2 levels. An outstanding question to this research was whether or not the low LAI's predicted by BIOME3 were consistent with the bulk of published physiological data. Data-model comparisons incorporating a suite of different process-based, global vegetation models will be the focus of future research, thus will not be addressed any further than the low CO2 comparison presented in the following paper.

2. Methodology

2.1. Data Selection

[10] We reviewed the plant science literature to extract data quantifying the effects of changes in environmental conditions on LAI (Table 1). Our database contains information from studies on the response of LAI to atmospheric CO2 (n = 30), soil fertility (n > 100) and soil moisture (n > 60). A significant amount of research concentrates on the response of whole-plant or single-leaf surface area to environmental changes (see review by Pritchard et al. [1999]), but much less on responses in LAI. An increase in total plant leaf area does not always represent a change in LAI because LAI is determined by factors other than leaf size (i.e., leaf number, orientation and retention) [Lawlor, 1995]. Each of these factors can be differentially modified by environmental conditions, such that opposing responses might ultimately result in no observable trend in LAI. Only research papers that explicitly measure LAI, therefore, are included in our study. Experimental species, experimental conditions such as closed versus open-top growth chambers, age of experimental plants and duration of the experiment, were recorded for each study.

Table 1. Summary of the Research Used to Compile a Database on the Response of Leaf Area Index to Changes in Atmospheric CO2 (CO2), Soil Moisture Content (Drought), and Soil Nutrition (Nutrients)a
TreatmentVegetation TypeStudiesReference Citations
  • a

    Data were separated according to vegetation types; categories including C3 crops, woody species, native species, C4 plants, and plant communities. “Studies” refers to the number of independent sources for species-based data. Since some articles contain information on several treatments using different species, they will appear more than once within the table.

CO2crops11Allen et al., 1991; Chen et al., 1997a, 1997b; Ewers et al., 1999; Lawson et al., 2001; Lee et al., 1997; Mayeux et al., 1997; Mullholland et al., 1997, 1998; Rowland-Bamford et al., 1991; Rudorff et al., 1996; Sims et al., 1999
woody8Ceulemans et al., 1995, 1996; Idso et al., 1993; Jach et al., 2000; Overdieck and Forstreuter, 1994; Tingey et al., 1996; Tognetti et al., 1999
natives5Cook et al., 1998; Lutze and Gifford, 1998; Manderscheid et al., 1997; Schapendonk et al., 1997; Schenk et al., 1995
C41Rudorff et al., 1996
communities6Arnone and Korner, 1995; Hattenschwiler and Korner, 1998; Korner and Arnone, 1992; Lovelock et al., 1998; Niklaus et al., 1998; Polley et al., 1992
Watercrops47Begum and Paul, 1993; Chapman et al., 1993; De Costa and Shanmugathasan, 1999; De Tafur et al., 1997; Fernandez et al., 1994; Frederick et al., 1991; Gerik et al., 1996; Gill and Narang, 1993; Giunta et al., 1995; Gupta et al., 1996; Haqqani and Pandey, 1994; Hooda and Kalra, 1981; Jefferies and Mackerron, 1993; Joel et al., 1997; Kasturikrishna and Ahlawat, 1999; Lawlor, 1995; Link et al., 1995; Lopez et al., 1997; Lu et al., 2000; Majid and Simpson, 1997; Mandal et al., 1991; Mondal and Paul, 1994; Nandan and Prasad, 1998; Nielsen and Nelson, 1998; Ockerby et al., 1993; Padmani et al., 1994; Pannu and Singh, 1993; Ramesh, 2000; Reddy and Ahlawat, 1998; Robertson and Giunta, 1994; Saeed and El-Nadi, 1997; Sammis et al., 1986; Sepaskhah and Llampour, 1995; Shrotriya and Misra, 1977; Singh and Singh, 1995; Singh et al., 1994; Sirait et al., 1994; Tiwari et al., 1997; Vivek and Chakor, 1992; Yadav et al., 1993, 1994a, 1994b, 1999; Yao and Goue, 1992; Zhang et al., 1998
woody4Albaugh et al., 1998; Cutini, 1996; Ewers et al., 1999; Sampson and Allen, 1998
natives2Daoust and Tayler, 1969; Huang et al., 1998
C44Farre et al., 2000; Otegui et al., 1995; Singh and Singh, 1995; Steduto and Hsiao, 1998
communities4Crombie, 1992; Jana et al., 1995; Jose and Gillespie, 1997; O'Grady et al., 2000
Nutrientscrops75Abbas et al., 1994; Ahmad et al., 1998; Bali et al., 1991; Bednartz et al., 2000; Bisht and Chandel, 1991; Booij et al., 1996; Boonchoo et al., 1998; Cantero-Martinez et al., 1999; Chawla and Narda, 2000; De Tafur et al., 1997; Dingkuhn et al., 1992; Fagade and De datta, 1971; Fernandez et al., 1994; Frederick and Camberato, 1995; Gill and Narang, 1993; Grashoff and d'Antuono, 1997; Gupta et al., 1996; Heitholt et al., 1998; Hooda and Kalra, 1981; Ingram et al., 1991; Joel et al., 1997; Johnston and Fowler, 1992; Kappen et al., 1998; Kasturikrishna and Ahlawat, 1999; Khalifa, 1973; Khan, 1996; Khan et al., 1993, 1996, 2000; Khanpara et al., 1993; Kolar and Grewal, 1994; Kumar et al., 1999; Lawlor, 1995; Liebman et al., 1995; Morris, 1999; Nandan and Prasad, 1998; Nielsen and Halvorson, 1991; Ockerby et al., 1993; Olasantan et al., 1994; Padmani et al., 1994; Pal et al., 1996; Parihar et al., 1997; Patel et al., 1996; Patra et al., 1996; Pellet and El-Sharkawy, 1993; Peltonen-Sainio, 1997; Pettigrew and Meredith, 1997; Pospisil et al., 2000; Ramamurthy and Shivashankar, 1996; Reddy and Ahlawat, 1998; Rodriguez et al., 2000; Saini and Thakur, 1996; Sarkar et al., 1998; Scholberg et al., 2000; Sharif and Eghbal, 1994; Shrotriya and Misra, 1977; Singh et al., 1994, 1996; Stutterheim and Barbier, 1995; Tenebe et al., 1996; Thakur et al., 1999; Tiwari et al., 1997; Tomar et al., 1996; Tomar and Dadhwal, 1999; Vivek and Chakor, 1992; Yadav, 1981; Yadav et al., 1993, 1994a, 1994b; Yahiya et al., 1996; Zaongo et al., 1997.
woody14Albaugh et al., 1998; Balster and Marshall, 2000; Carlyle, 1995, 1998; Colbert et al., 1990; Cromer et al., 1993; Ewers et al., 1999; Fife and Nambiar, 1995, 1997; Heilman and Xie, 1994; Sampson and Allen, 1998; Stoneman et al., 1996; Tingey et al., 1996.
natives9Aerts et al., 1992; Aerts and De Caluwe, 1994; Bailey and Laidlaw, 1998; Daoust and Tayler, 1969; Laidlaw and Withers, 1998; Link et al., 1995; Lutze and Gifford, 1998.
C48Bullock et al., 1993; Olasantan et al., 1994; Pellerin et al., 2000; Plenet et al., 2000; Saini and Shekhar, 1998; Tolk et al., 1999; Vidovic and Pokorny, 1973; Wu et al., 1993.
communities7Chiang et al., 2000; Fahey et al., 1998; Hattenschwiler and Korner, 1998; Herbert and Fownes, 1995; Maurer et al., 1999.

[11] To minimize errors associated with seasonal trends in LAI of annual species, several conventions were adopted. Because peak LAI was frequently reported in studies listing only one LAI value for each experimental condition, we recorded peak LAI in as many studies as possible. If only a few LAI values were recorded for sequential harvests, then LAI was averaged between harvests. If treatments were begun part way through the experiment, then LAI at final harvest was recorded.

[12] Plant density can influence leaf area development, thereby modifying LAI [Cutini, 1996; Kumar et al., 1999; Pospisil et al., 2000]. For surveyed studies containing plant density as an experimental treatment, data from the least dense treatment was chosen for our study. If an experiment was replicated, then replicate LAI's were averaged and if the experiment was replicated on several sites, LAI was averaged between sites. Some studies focused on the response of cultivars (i.e., of the same species but having different phenotypic characteristics) or hybrids to changes in environmental conditions. In such cases, LAI was recorded as an average between cultivars/hybrids. Some of the more recent studies concentrated on plant response to ozone or other environmental stresses in conjunction with water, nitrogen or CO2 treatment. In these cases, we used only the studies incorporating full factorial experiments, so that we could extract data influenced solely by the desired environmental variable.

2.2. Standardization of Data

[13] For studies investigating the effects of atmospheric CO2 on LAI, experimental and control CO2 levels were recorded as they were reported by the authors. Most of the published LAI-CO2 research focused on CO2 enrichment (87%), but a few studies investigated low CO2-LAI interactions (13%), enabling us to present line graphs of LAI versus atmospheric CO2 between the range of 160 and 900 μmol mol−1.

[14] It was more difficult to standardize studies investigating the effects of soil fertilization on LAI because the soil fertility of control treatments varied considerably between studies. This was primarily a result of our incorporating global field data, which meant that LAI data were aggregated from experimental plots having soils of differing texture, composition and nutrient composition. Fertilization regimes also differed between studies; nitrogen (N) was added alone or with small amounts of phosphorus (P) and potassium (K), equal parts N:P:K, or large amounts of macronutrients (N, P, K) with small additions of micronutrients (i.e., S, Zn). LAI-fertilization data, therefore, were grouped into fertilization classes where control represents treatments with no external fertilization regime, mild, moderate and high fertilization classes represent fertilization treatment rates of 1–199, 200–349, and >350 kg nutrient (N, P, or K) per hectare, respectively.

[15] Standardizing data on LAI responses to changes in soil water content was the most difficult because soil moisture of control treatments was frequently not recorded, nor was the degree of water stress quantified in each study. For example, several experimental treatments were described as simply “irrigated” or “rainfed only.” For these reasons it was necessary to present bar graphs according to loosely grouped soil drought categories: a control (no water stress, i.e., the wettest treatment in every study) and two water stress categories (moderate and severe). Therefore, LAI data was somewhat arbitrarily assigned a water stress class depending on how the authors had described their water treatments.

2.3. Data Analysis

[16] The available data did not support a formal metaanalysis, however we were able to maintain the philosophy of synthesizing data to discriminate between subsets of data. In our case, these subsets include control plants having either low or high initial (control) LAI, differences in vegetation type (i.e., annual crops versus woody perennials), and variations in treatment methodologies (i.e., open versus closed environment treatments).

[17] One of the more important distinctions for the data we collected concerned differences in plant functional type. Experimental data were divided according to five vegetation classes: (1) C3 crop species, (2) woody perennials, (3) native plants and herbs, (4) C4 plants, and (5) plant communities (Table 1). Plant communities were defined as having two or more species incorporated into the same experimental plot. Examples of communities include the European spruce forest (Picea, Oxalis, Homogyne, Melampyrum), calcareous grasslands (Bromus, Carex), and European mixed forest (Fagus, Picea). Data from all five classes were pooled and presented in graphs described as “all plant types.”

[18] Leaf area index was compared between studies using a ratio (R-LAI) that we define as the treatment LAI divided by the experimental control LAI. LAI responses to changes in environmental variables can be meaningfully compared between studies because R-LAI is an indication of the relative degree of change associated with a specific environmental condition, because it eliminates interstudy differences in LAI of control plants.

[19] Control treatments were defined as 360 μmol mol−1 for the atmospheric CO2 studies, the most watered experiment for water stress experiments, and the unfertilized plots for the soil fertilization studies. If LAI showed little to no response to environmental treatment, then R-LAI approaches 1.0. Alternatively, an R-LAI greater than 1.0 indicates a positive enhancement of LAI.

2.4. Limitations

[20] Because some of the annual plant studies reported a single LAI value that could have represented either average lifetime or peak LAI, and because LAI measurements of perennial plants did not represent peak nor lifetime averaged LAI, there exists some unavoidable limitations to our study. The objective of this synthesis, however, was to identify broad trends in LAI following changes in environmental conditions; therefore systemic data selection limitations will probably not significantly alter our general conclusions.

[21] An additional source of error in our analysis may arise from the different methods used to measure LAI in each of the surveyed studies. In some studies, LAI was measured during destructive harvests by dividing total leaf area by total ground surface area. Most of our LAI data (particularly crop data) were recorded in this manner. For most of the plant community studies and some of the woody perennial experiments, optical techniques were used to estimate LAI. The amount of solar radiation reaching the ground is measured by a canopy light analyzer and related to LAI via algorithms based on light extinction coefficients. Optical techniques tend to underestimate LAI relative to harvest methodologies [Smith et al., 1993], but in general, there is good agreement between various LAI estimation methods [Cutini, 1996; Chen et al., 1997a, 1997b].

3. Results

[22] Most of the data we collected on LAI responses to soil fertilization were specific to nitrogen, with the most common experimental species being mustard (Brassica), wheat (Triticum), and pine (Pinus). Consequently, we were able to analyze LAI-N interactions in each of these species (Figure 1). Regression analyses (using a SigmaPlot statistical package) indicate that linear curves provide the best fit to data, with regression coefficients (r2) of 0.24, 0.25, and 0.38 for Brassica, Triticum and Pinus, respectively. The low r2 values indicate that although LAI is influence by nitrogen application, N-fertilization cannot solely explain variation in leaf area production. The LAI curve of the woody perennial, Pinus, has a somewhat higher regression coefficient than either of the two crop species, but this difference is likely not indicative of any significant trend between plant functional types. The Pinus plot is based on low sample size (n = 7) and data from juvenile plants (seedlings/saplings). The developmental tendency for early plant growth to be strongly linear is likely responsible for confounding the positive relationship between LAI and N-fertilization.

Figure 1.

Effects of nitrogen fertilization (kg N ha−1) on leaf area index (LAI) in (a) Brassica (mustard), (b) Triticum (wheat), and (c) Pinus (pine). Regression coefficient is denoted by r2. Number of independent studies is denoted by n.

[23] A bar plot of pooled fertilization data reveals that the linear response of LAI to N observed in Brassica, Triticum and Pinus is not uniform for all plant species, soil nutrients, or fertilization rates (Figure 2). The bar graph indicates that when small to moderate amounts of fertilizer (mainly nitrogen, phosphorus and potassium) are added to nutrient impoverished soils, LAI strongly increases. Most of the data points used in Figure 1 fall within this low to moderate fertilization class (i.e., below an application rate of 250 kg nutrient per hectare). However, for higher nutrient application rates, LAI sensitivity decreases (i.e., saturates) indicating that something other than soil fertility is limiting to canopy development.

Figure 2.

Effects of increased fertilization (mostly nitrogen, phosphorus, and potassium) on leaf area index (LAI). Fertilization classes are defined as low (1–199 kg nutrient per hectare), moderate (200–349), and high (>350). LAI is presented as a ratio of treatment LAI over control (R-LAI). Error bars represent standard error about the mean (SEM) (SigmaStat/Plot).

[24] The response of LAI to increasing soil water content (Figure 3) is close to that of soil fertility, in that the shape of the LAI-response curves is similar (Figure 2). Increasing water availability in severely droughted soils causes a significant increase in LAI; however, once soil water stress is alleviated, LAI becomes much less sensitive to variations in soil water content (Figure 3). Ultimately, LAI encounters a saturation ceiling, whereby further resource input has little to no influence on leaf area production.

Figure 3.

Effects of decreasing soil water stress on leaf area index (LAI). Water stress categories are assigned according to descriptions of water treatments given by authors of respective research papers. R-LAI is defined in Figure 2.

[25] Several trends are noted for the response of LAI to variations in atmospheric CO2. As expected, the regression coefficient (r2) for the curve of open-top chamber (OTC) data was lower than for closed-environment chamber (CEC) data owing to the variability inherent in open design experiments (Figure 4). Data from CEC experiments indicate that LAI increases sharply from low (200 μmol mol−1) to ambient levels of CO2 (360 μmol mol−1), begins to saturate at CO2 levels nearing 650–700 μmol mol−1, and then decreases in response to very high CO2 environments (i.e., >800 μmol mol−1).

Figure 4.

Comparison of the effects of atmospheric CO2 on LAI between experiments with open-top chamber (OTC) (dots) and closed-environment chamber (CEC) (triangles) experimental design. R-LAI is defined in Figure 2.

[26] In contrast, OTC data do not provide information on the response of LAI to CO2 concentrations less than current ambient, and shows contradictory responses of LAI to increases in atmospheric CO2 above 360 μmol mol−1. The number of studies showing an increase in LAI with increasing CO2 appears to be nearly compensated by the number of studies showing a decrease, thus contributing to an absence of LAI-CO2 correlation (i.e., r2 = 0.04). Although there exists disparity in CEC and OTC LAI curves, these differences do not preclude the pooling of CO2 data for further analysis.

[27] Regression of independent crop, plant community and native species data indicate non-linear responses of LAI to increases in atmospheric CO2 (Figure 5). LAI curves for crops and plant communities indicate a strong response of LAI to increases in atmospheric CO2 up to current ambient (Figures 5a and 5b). Curves for crops, communities, and native plants (Figure 5d) indicate that CO2 levels above current ambient have a minor influence on LAI. In several cases, a decrease in LAI is associated with rising CO2 (in particular, see Figure 5d), with regression lines barely increasing above an R-LAI value of 1.0.

Figure 5.

Effect of changes in atmospheric CO2 on (a) LAI (crop species), (b) LAI (plant communities), (c) LAI (woody perennials), and (d) LAI (native plants). R-LAI is defined in Figure 2.

[28] Although woody data indicate a positive linear response of LAI with increasing CO2 (Figure 5), this trend is probably not significant. Atmospheric CO2 experiments involving woody perennials are almost always conducted with seedlings or saplings, such that the developmental tendency of linear growth in young trees is likely affecting LAI-CO2 trends. This behavior is apparent in vegetation types other than woody perennials. Research shows that crop species are most responsive to atmospheric CO2 changes early in their growth phase [Allen et al., 1991]. Most of our surveyed crop studies, however, investigate the effects of CO2 over the full life span of the plant.

[29] To test whether or not pretreatment LAI influenced how plants responded to changes in atmospheric CO2, we assumed that LAI-CO2 interactions were linear, to calculate the slope of the LAI-CO2 curve (R-slope defined as the rate of change in LAI relative to change in atmospheric CO2). A plot of R-slope against control LAI will reveal the presence of data clumping, and signal the potential for control LAI to alter LAI sensitivity. If high or low control LAI consistently results in large or small responses, for example, then data will cluster around similar points. Our analysis indicates an absence of data clumping (Figure 6).

Figure 6.

Slope of the response curve (R-slope, defined as the rate of change in LAI relative to change in atmospheric CO2) plotted against pretreatment LAI (m2 m−2).

[30] An aggregation of CO2 data from all vegetation types indicates that response of LAI to CO2 concentrations below current ambient is distinctly different from that above (Figure 7). Although our database contains many fewer studies on the effects of changes in atmospheric CO2 relative to soil nutrients and moisture content, regression analysis of data representing all plant types yields a fairly strong correlation with atmospheric CO2 (r2 = 0.42). LAI response to variations in atmospheric CO2 can be described as strongly linear at levels less than current ambient. In contrast, fairly substantial increases in atmospheric CO2 have an overall minimal influence on LAI. The differential (nonlinear) response of carbon processes to changes in atmospheric above and below current ambient CO2 has also been observed in experimental studies investigating leaf gas exchange above natural grasslands [Mielnick et al., 2001].

Figure 7.

Effect of changes in atmospheric CO2 on LAI (all vegetation types).

4. Discussion

4.1. LAI and Resource Optimization in Plants

[31] Representation of LAI in mechanistic vegetation models is often based on the optimization of LAI according to resource use and availability. In response to low soil moisture or high atmospheric vapor pressure deficit, leaf pores (stomata) partially close to help offset plant desiccation, with a tradeoff arising from the subsequent reduction in rates of carbon uptake. Allocation of carbon to leaf production results in greater surface area for water loss during transpiration, which in turn can be limiting to plant growth. Consequently, several vegetation models estimate LAI in terms of an optimization of leaf area with respect to whole-plant water conservation [Haxeltine and Prentice, 1996; Kergoat, 1998; Foley et al., 1996].

[32] Alternatively, an optimal LAI representing maximal canopy photosynthesis can exist for a given total amount of nitrogen in the canopy [Anten et al., 1995; Sands, 1995; Hirose et al., 1997; Williams and Rastetter, 1999; Hartz-Rubin and De Lucia, 2001; Moorcroft et al., 2001]. Vegetation models, therefore, have also been developed on the basis of LAI optimization for canopy nitrogen content [Yin et al., 2000; Dickinson et al., 2002].

[33] Our synthesis confirms that LAI is likely colimited by a number of resources, including water and nitrogen. Nitrogen alone cannot explain LAI trends in selected crop and woody species (i.e., low r2 values found in Figure 1), nor does soil moisture uniformly influence LAI across a range of soil water deficits (Figure 3). A recent inter-model data comparison by Bondeau et al. [1999] highlight the tendency of vegetation models to generally overestimate LAI in physiogeographic regions because models usually simulate only one of several colimiting constraints on LAI. For example, if a vegetation model optimizes LAI according to only water conservation, then LAI of northern boreal forests might be overestimated in regions where LAI is strongly constrained by soil nitrogen availability.

[34] At some resource threshold, however, addition of greater and greater amounts of fertilizers and water will have no further influence on LAI (Figures 2 and 3). The saturation of LAI with high fertilizer and water application is likely indicative of a light (photon) limitation. Dense canopies attenuate light, causing self-shading to promote negative carbon balances in low-canopy leaves. Above the saturation threshold, whole-plant carbon balance equalizes, causing LAI to stabilize.

[35] Because resources such as water and soil nutrients are often more limiting than carbon, atmospheric CO2 is not often viewed as a plant resource per se. However, our synthesis indicates that LAI is considerably lower when carbon is limiting as a critical resource for plant growth (i.e., at CO2 concentrations less than ∼ambient) (Figure 7). The strongly positive linear relationship between LAI and CO2 is observed up to current levels of CO2, after which the sensitivity of LAI to CO2 substantially decreases. It is difficult to define the precise CO2 level at which LAI saturates, although data indicate that the slope of the LAI response curve begins to level with steadily increasing levels of CO2.

[36] Due to the similarity in the influence of atmospheric CO2 on LAI relative to other plant resources, we propose a general model of the response of canopy LAI to changes in resource availability (Figure 8). When resources are low, canopy development is strongly influenced by changes in gross primary production (GPP). Alleviation of environmental or resource stress will result in higher rates of GPP and greater investment of carbon into leaf area production. A good proportion of natural ecosystems likely lie within this part of the LAI response curve, evidenced by the bulk of field studies that indicate a strong correlation between resource availability, GPP and maximum LAI [Schimel et al., 1991; Sala et al., 1994; Fassnacht and Gower, 1997; Vogel and Gower, 1998; Ares and Fownes, 1999; Balster and Marshall, 2000; Lane et al., 2000; Le Dantec et al., 2000; Bolstad et al., 2001].

Figure 8.

Hypothetical LAI-resource response curve. Dashed line indicates threshold where sensitivity of LAI to resource availability begins to decrease and eventually saturates.

[37] At low resource availability, the slope of the LAI response curve should be determined by the balance between resource investment into extra activity per unit of leaf area (i.e., increasing photosynthetic efficiency) and into increasing total leaf area (i.e., maximizing surface area available for light capture). LAI saturation (Figure 8; dotted line) can be defined as the threshold after which any further increase in LAI is compensated by negative carbon balances in shaded lower canopy leaves.

[38] Simulations of the response of LAI to future 2 × CO2 are typically based on the assumption that the LAI-CO2 relationship is linear across a wide range of CO2 concentrations. For vegetation whose LAI is not currently limited by self-shading, elevated CO2 should cause a reduction in leaf light compensation point (LCP), the minimum amount of energy required to overcome negative carbon balances resulting from plant respiration [Long and Drake, 1991]. In support, experimental data show the potential for elevated CO2 to lower LCP and promote leaf extension and retention in lower canopy leaves [Long and Drake, 1991].

[39] Our synthesis, however, indicates that LAI will likely saturate at CO2 levels lower than have been previously assumed. Increase in the quantity of yellowing and senescing leaves is often greater with elevated CO2 treatments [Korner and Arnone, 1992; Arnone and Korner, 1995; Gielen et al., 2001; Hartz-Rubin and De Lucia, 2001]. Thus, higher leaf turnover may account for the maintenance of stable LAI through time, despite observed higher rates of carbon uptake. If high CO2 decreases LCP at the leaf-level, then why does LAI saturate at the level of canopy?

[40] The explanation for why canopy LAI saturates with high CO2 remains uncertain, but may be related to a canopy-scale down-regulatory mechanism promoted in response to an abundance of CO2 [Lutze and Gifford, 1998; Hattenschwiler and Korner, 1998]. Plastic response of LAI to changing environments (i.e., evolutionary influences) may also be constraining the sensitivity of modern-day vegetation canopies to environmental change [Field et al., 1992]. Alternatively, some other essential plant nutrient (like phosphorus) may be playing a larger role in modifying canopy-scale LAI than we currently suspect.

4.2. Implications for Global Change Research

[41] Having confidence that LAI is simulated realistically in current vegetation and land-surface models is important because many of the biological and physical consequences of 2 × CO2 are based on modeled trends in LAI. The realistic simulation of LAI in response to environmental controls is critical because in many cases LAI is the only variable used to describe vegetation and ecosystem structure. Consequently, more confidence can be placed on LAI simulations if data support simulated trends.

[42] One example of the utility of LAI data-model comparisons in global change research involves the simulation (with BIOME3; [Haxeltine and Prentice, 1996]) of the effects of last glacial maximum (LGM) climate on vegetation in the Amazon Basin. Modeling results highlight the importance of low atmospheric CO2 in lowering rain forest LAI, and have been extrapolated to help explain high species diversity in the neotropics [Cowling et al., 2001] as well as the “openness” of Pleistocene vegetation in general [Mercader et al., 2000]. In support, other LGM vegetation simulations predict decreases in LAI with low atmospheric CO2 [Levis et al., 1999].

[43] Because modeling sensitivity experiments indicate that LAI is more influenced by LGM CO2 level (200 μmol mol−1) than climate, one might wonder whether the strong LAI × CO2 relationship simulated by BIOME3 might be no more than a modeling artifact. We conducted a data-model comparison using the low CO2 data collected in our synthesis and the BIOME3-simulated LAI values obtained from Cowling et al. [2001] (Figure 9). Modeled LAI values represent forest LAI averaged across the entire lowland Amazon Basin, and were predicted based on today's climate (temperature and precipitation) simulated over a range of atmospheric CO2 levels. We observe a good fit between modeled and observed responses of LAI to changes in CO2, despite the fact that the observed data represent individual crop responses to CO2, and modeled results represent LAI trends averaged across a large geographical region. Not only does this comparison indicate that BIOME3 realistically predicts the response of LAI to low CO2, but that our database captures the general response of LAI to CO2, independent of spatial scale. Data-model comparisons such as above have become (and will continue to remain) a mainstay of global change research [Cramer and Field, 1999].

Figure 9.

Observed (triangles) versus modeled (dots) responses of LAI to decreases in atmospheric CO2 from 360 to 150 μmol mol−1.

[44] Recent (NDVI) satellite data on vegetation “greenness” indicate that LAI has increased in the Mediterranean basin [Osborne et al., 2000] and Eurasian continent [Zhou et al., 2001b; Lucht et al., 2002] over the past 2 decades. Interacting biotic and abiotic factors have been cited as possible underlying mechanisms. Zhou et al. [2001b] propose greater precipitation as the catalyst for increased productivity in Eurasia. Lucht et al. [2002] conclude that northern greening can mostly be attributed to biogeochemical responses of vegetation to increasing temperature. Osborne and Woodward [2001], however, suggest that a significant portion of the 7% increase in Mediterranean LAI is due to the recent 10% increase in atmospheric CO2 since the start of the Industrial Revolution. Their conclusions are drawn from vegetation modeling simulations of the response of LAI to the past two decades of recorded increases in atmospheric CO2. Our data synthesis cannot exclude CO2 as a possible catalyst, however we caution that any observed CO2-LAI response will not continue indefinitely.

[45] Our ability to predict how the terrestrial biosphere might respond to future changes in climate and CO2 [Friend and Cox, 1995; Schimel et al., 2000; Pan et al., 1998; Kittel et al., 1995] depends on the sophistication of our current modeling framework and variable parameterizations. The simulation of leaf area development remains one of the weakest features in crop models, despite the vast quantity of research focusing on understanding crop growth [Marcelis et al., 1998]. Simulation of LAI in global vegetation models is often based on linear scaling of leaf-level processes to the canopy [Field et al., 1995]. If a model incorporates linear LAI-CO2 responses, then for low-density vegetation (when self-shading is nonlimiting to canopy structure) an increase in CO2 should result in substantial increases in LAI, following the effect of high CO2 in increasing water-use efficiency and photosynthetic carbon uptake. Our synthesis, however, indicates that the LAI-CO2 relationship is likely nonlinear, thus for vegetation that is nonlimited by self-shading, additional CO2 will not enhance whole-plant carbon gain. Such sentiments have been expressed previously by others [Lockwood, 1999, and references therein].

[46] Several recent papers highlight the importance of vegetation feedbacks in modifying climate under 2 × CO2. Such simulations predict increases in global LAI as a result of changes in vegetation cover, for example, a transition from tundra to boreal forest in high latitudes or a replacement of grasses by trees in the tropics [Levis et al., 2000; Bergengren et al., 2001]. Widespread enhancement of LAI in vegetation types is also simulated as a result of CO2 fertilization. In the latter case, simulation results reveal that LAI-CO2 responses contribute to an offsetting effect on stomatal responses [Betts et al., 1997]. Our synthesis indicates that the general assumption that the sensitivity of LAI to increases in CO2 is solely limited by self-shading, is likely not valid, and therefore future climate change simulations should be considered with this caveat in mind. In the absence of CO2-fertilization, for example, modeling experiments conducted by Friend and Cox [1995] demonstrate that elevations in CO2 above ambient can reduce LAI by 21%, because of the respiratory costs associated with higher surface temperatures.

5. Concluding Statements

[47] Cooperation between vegetation modelers, physiologists and field ecologists will encourage the flow of information needed to test critical model assumptions [Luo et al., 1999]. This synthesis should help to emphasize the importance of empirical LAI data for modeling research, as the assumption that leaf surface responses to environmental change can be scaled to the ecosystem, may not necessarily apply to canopy development. Heightened attention toward incorporating empirical trends in state-of-the-art vegetation models will ensure greater confidence attached to predictions of how the terrestrial biosphere might change in the near and distant future. Inclusion of nonlinear biotic-abiotic relationships like the one highlighted here will go a long way in satisfying this objective.


[48] We would like to thank Jack Williams and two anonymous reviewers for providing useful comments on earlier drafts of this manuscript. This work was conducted while SAC was a Postdoctoral Associate at the National Center for Ecological Analysis and Synthesis (NCEAS), a Center funded by NSF (grant DEB-0072909), the University of California, and the Santa Barbara campus.