Portal de Periódicos da CAPES

Transient ecosystem responses to free‐air CO 2 enrichment (FACE): experimental evidence and methods of analysis

2001; Wiley; Volume: 152; Issue: 1 Linguagem: Inglês

10.1046/j.0028-646x.2001.00247.x

1469-8137

Yiqi Luo,

Atmospheric chemistry and aerosols

How is it possible to extrapolate results from field CO2 experiments to allow the prediction of large-scale, long-term responses of ecosystems to global change? There are two key points of view on this long-standing issue. The first considers that currently running experiments using free-air CO2 enrichment (FACE) facilities or open-top chambers mimic a future atmospheric CO2 concentration (Ca), so that observations from the experiments represent ecosystem responses to the designated CO2 level. However, here both theoretical (Luo & Reynolds, 1999) arguments and experimental (from the Duke Forest FACE experiment) evidence are put forward to support a second viewpoint. In this, the current experiments are considered to exert an ecosystem perturbation, achieved primarily by altering carbon (C) influx. Observation of ecosystem responses to the perturbation helps probe the processes and mechanisms. If this second viewpoint is correct, then we must also ask how we can use results from FACE experiments, where ecological responses to a rather abrupt perturbation are measured, in order ultimately to understand and predict long-term ecosystem responses to a very gradual CO2 change in the real world. FACE facilities have become a premier approach for conducting ecosystem CO2 experiments (Hendrey et al., 1999; Miglietta et al., 2001; Okada et al., 2001). They have been implemented in numerous ecosystems, including an agricultural field in Arizona (Kimball et al., 1994), a loblolly pine forest at Duke University in North Carolina (DeLucia et al., 1999), a pasture in Switzerland (Van Kessel et al., 2000), the Mojave Desert in Nevada (Jordan et al., 1999), a sweetgum forest in Tennessee (Norby et al., 2001), a grassland in Minnesota (Reich et al., 2001), a poplar plantation in Italy (Miglietta et al., 2001), and a rice field in Japan (Okada et al., 2001). There is tremendous enthusiasm for FACE technology among ecologists around the world because it provides an unprecedented opportunity to conduct manipulative CO2 experiments in intact ecosystems with minimal environmental disturbance. The FACE experiments have already yielded a great deal of useful information on ecosystem function and are one of the most effective ways to gain short-term insights into long-term issues. The FACE experiments are inspired primarily by the phenomenon of rising Ca in the real world. Ultimately, we have to extrapolate experimental results beyond the FACE plots to infer the capacity of terrestrial ecosystems to sequester C. The extrapolation is, however, complicated by the fact that the FACE experiments impose a step increase in CO2 concentration whereas ecosystems in the real world are experiencing a gradual increase in Ca. Therefore, the first issue we need to address regarding extrapolation is how comparable the ecosystem responses to a gradually increasing Ca are to the responses to the step CO2 increase. Luo & Reynolds (1999) conducted a modelling study to address that issue. Their modelling study indicates that when CO2 increases gradually as in the real world, the ecosystem C sequestration rate increases gradually. In response to a step CO2 increase, ecosystem C sequestration rate is predicted to increase abruptly by approximately 10-fold and then decline gradually. Similarly, the step and gradual CO2 increases give contrasting patterns of N cycling. The predicted transient responses to a step CO2 increase are consistent with other modelling results (Comins & McMurtrie, 1993; Rastetter et al., 1997; Cannell & Thornley, 1998). In addition to physiological transition (e.g. bud primordia set under ambient CO2) to elevated CO2 (Norby et al., 2001), the transient responses to the step CO2 increase are primarily structural consequences of ecosystem C and N processes. Ecosystem C sequestration rate at a given time is the difference between ecosystem C influx via photosynthesis and efflux via respiration. In response to a step CO2 increase, photosynthesis immediately increases due to its fast response time. Respiration is generally proportional to C pool sizes in plants, litter, and soil organic matter (SOM). In response to the step CO2 increase, changes in pool sizes are cumulative, resulting in a gradual increase in ecosystem respiration. The difference between the step increase in photosynthesis and the gradual increase in respiration translates to a transient response in C sequestration rate, which abruptly increases immediately after the CO2 fumigation and then gradually declines. In response to a gradual Ca increase, both photosynthesis and respiration gradually increase with a time lag in respiration. The time lag creates a difference between photosynthesis and respiration at any given time, resulting in a gradual increase in the ecosystem C sequestration rate over time (Luo & Reynolds, 1999). While it is not feasible to conduct an experiment lasting hundreds of years in the real world to observe ecosystem responses to a gradual Ca increase, we can use experimental data to verify the processes causing the transient responses. The processes are the step increase in photosynthetic C influx and the gradual increase in respiratory C release in response to a step CO2 increase as in FACE experiments. Here, I present experimental evidence from the Duke Forest FACE experiment to demonstrate an immediate, step increase in ecosystem C influx accompanied by a gradual increase in ecosystem C efflux via respiration. The Duke FACE experiment is an on-going project – started in August 1996 – and has been well described in terms of experimental design and facility (Hendrey et al., 1999), climate (Luo et al., 2001a), and soil properties (Andrews & Schlesinger, 2001). Leaf-level photosynthesis at the Duke Forest FACE increased by approx. 40–60% in the past 5 yr (Ellsworth, 2000). In order to provide a whole-ecosystem estimate of C influx, we used the MAESTRA model to estimate photosynthetic CO2 assimilation of the loblolly pine canopy at the FACE site (Luo et al., 2001a). MAESTRA is a three-dimensional model of forest canopy radiation absorption, photosynthesis, and transpiration. It incorporates standard formulations of the mechanistic C3 photosynthesis model of Farquhar et al. (1980) and a more empirical formulation of stomatal conductance described by Ball et al. (1987). The canopy is represented by an array of semiellipsoidal tree crowns. Each crown is divided into six horizontal layers with each layer divided into 12 gridpoints of equal volume. Each layer is specified by a number of physical and physiological properties, including radiation, temperature, leaf area, and leaf N content. Environmental variables that drive model simulations are radiation, air temperature, air humidity, wind speed and Ca above the canopy. The MAESTRA model was validated against measurements of leaf-level photosynthetic rates and canopy photosynthesis derived from eddy-covariance measurements at the FACE site (Luo et al., 2001a). The validated model predicts that daily canopy C uptake ranged from 0 g C m−2 d−1 in the winter on exceptionally cold days to 8 g C m−2 d−1 in the summer in the ambient CO2 plots. In elevated CO2 plots it ranged from 0 g C m−2 d−1 in the winter to nearly 12 g C m−2 d−1 in the summer. Accordingly, elevated [CO2] increased average ecosystem C influx by approx. 40% following the CO2 fumigation in August 1996. The CO2 enhancement in canopy C influx displays a step function, indicating an abrupt, large increase in C influx into the forest ecosystem immediately after the CO2 fumigation (Fig. 1). Results from the Duke Forest FACE experiment demonstrate contrasting patterns of ecosystem photosynthetic (open circles connected with line) vs respiratory (line) responses to the step CO2 increase. The ratio of modelled daily canopy C fluxes at elevated relative to that at ambient CO2 from August 1996 to December 1998 displays a step CO2 stimulation of C influx (Luo et al., 2001a). However, the step increase in the C influx did not lead to a step increase in respiratory C release. Rather the ratio of modelled soil respiration shows little change in the first year but a gradual increase in the second and third years under the elevated CO2 treatment compared to that under the ambient CO2 treatment (Luo et al., 2001b). The step increase in C influx into the elevated CO2 plots did not lead to a step increase in C efflux via respiration. Soil respiration rates (including both root and soil microbial respiration) were measured approximately monthly at the Duke FACE site (Andrews & Schlesinger, 2001). Measured midday values of soil respiration at the site displayed a strong seasonal variation, 0.05 g C m−2 h−1 in winter and 0.4 g C m−2 h−1 in summer. Elevation of CO2 concentration did not result in a statistically significant difference in the soil respiration in 1996 but led to significant increases of 23.2% and 35.5% in 1997 and 1998. The observed gradual increase in soil respiration is qualitatively consistent with the model predictions (Luo & Reynolds, 1999). In order to improve the mechanistic representation of soil C processes as affected by environmental variables, Luo et al. (2001b) modified the Terrestrial C Sequestration (TCS) model, which simulates soil respiration. The model uses the ecosystem C influx as an input, which is partitioned into leaf, wood, and fine roots. Litter from those biomass pools is decomposed by microbes, resulting partly in release of CO2 and partly in formation of SOM. SOM is mineralized to release CO2. Since various rhizosphere C processes possess different response times (or residence time, the mean from C entry via photosynthetic C fixation to exit via respiratory C release), photosynthetically fixed C is released back to the atmosphere via respiration with a range of time lags behind photosynthesis itself. The respiratory C release from the soil surface is the convolution of root respiration and microbial decomposition of litter and SOM from various pools with different time constants (Luo et al., 2001b). Thus, the step increase in photosynthetic C influx is transformed into a gradual increase in respiratory C release. Preliminary results also suggest a gradual increase in plant respiration under elevated CO2 at the Duke FACE site. Measured specific respiration rates in trunks and needles showed little CO2 effect (J. Hamilton & E. DeLucia, pers. comm.) whereas measured plant biomass in elevated CO2 showed a gradual increase since the onset of the FACE experiment compared with that in ambient CO2 (DeLucia et al., 1999), resulting in a gradual increase in plant respiratory C release. Integration of the three components of C processes (i.e. canopy photosynthesis, plant and soil respiration) yields a large C sequestration in the first year of the FACE experiment, followed by smaller net C storage in the second and third years (Fig. 1). This dynamic pattern of C sequestration is qualitatively consistent with that predicted by Luo & Reynolds (1999). Both model simulations and experimental data support the argument that the FACE experiments generate a perturbation to ecosystems, primarily by altering input of C flux. This type of experiment requires that a decent size of perturbation be imposed on ecosystems in order to trace responses in a reasonable time frame. As in most system research, we must analyse results from the perturbation experiments in such a way that we can characterize ecosystem structure and estimate parameter values before prediction of future changes. The structure of ecosystem C processes can conceptually be characterized by compartmentalization, donor-controlled transfer, and sequential linearity (Luo & Reynolds, 1999). The C processes are highly compartmentalized due to the fact that photosynthetically fixed C goes to distinctive compartments, such as plant, litter, and SOM. Donor-controlled transfer is reflected by the fact that C release from each compartment through plant and microbial respiration is controlled by sizes of donor pools and hardly by products of respiration. In addition, the majority of photosynthetically fixed C sequentially transfers from one compartment to another, following a first-order linear function (Bolker et al., 1998). Only a small fraction of C can be recycled between SOM and soil microbes. The three properties have been incorporated in virtually all biogeochemical models (Jenkinson & Rayner, 1977; Parton et al., 1987; Comins & McMurtrie, 1993; Rastetter et al., 1997; Thompson & Randerson, 1999). Although the three properties are derived from experimental evidence, the model structure of C processes is yet to be rigorously tested (Luo et al., 2001b). The parameters we need to know in order to predict ecosystem responses to gradually rising Ca in the real world include ecosystem C influx, C partitioning coefficients among pools, and transfer coefficients from donor pools. It is technically difficult to measure ecosystem-scale C influx, particularly in the elevated CO2 plots even though leaf photosynthesis can be easily measured. An alternative method for estimating ecosystem C influx is modelling synthesis of experimental data, which also allows us to quantify the stimulation of canopy C uptake under elevated CO2. While extensive leaf-level measurements have been made in almost all FACE experiments and a variety of canopy models are available, synthesis of data from FACE experiments with canopy models is not only technically feasible but also has the potential to make a critical contribution to predicting C sequestration. The challenge is how to organize resources to realize the potential. Parameter estimation for C partitioning and transfer coefficients becomes more challenging than that for ecosystem C influx due to lack of experimental data and methodological difficulties. According to the degree of difficulty, derivation of parameter values from data can be divided into four cases. First, experimental data can be directly converted to parameter values. For example, specific rates of litter decomposition can be derived directly from laboratory and field studies on litter decomposition. Second, measured values represent results of two or more simultaneous but counteracting processes. For example, fine root biomass at a given time is determined by the balance between root growth and death, which occur simultaneously but counteractively. Parameter estimation for such processes depends on ancillary information. Third, parameter values are not measurable in experiments due to limited technology. For example, root exudation, which is suspected to be an important pathway of C flow to the rhizosphere (Hu et al., 1999), is not readily measurable in natural ecosystems. Parameterization is largely based on an educated guess. Fourth, a measurable quantity is a convolution of several processes with distinguishable characteristics. For example, soil respiration is regulated by multiple processes, including root exudation, root respiration, root turnover, and decomposition of litter and SOM. Those processes have distinctive response times to C perturbation. For this kind of data, deconvolution is an effective approach to parameter estimation (Luo et al., 2001b). A quantitatively rigorous approach to estimating parameter values is inverse analysis, which has proved to be a very powerful tool for model-data integration in other scientific disciplines. The inverse analysis is an approach that fundamentally focuses on data analysis for tests of model structure and parameter estimation. It is often used interactively with forward analysis, which is usually implemented using simulation models. The latter predicts ecosystem responses to global change with a given model structure and a set of prescribed parameter values. Generally speaking, the forward analysis asks what a model can tell us about the ecosystems whereas the inverse analysis asks what the data can tell us about the same system. Combination of the two approaches allows us to probe mechanisms underlying ecosystem responses to elevated CO2. Since simulation modelling is familiar to most researchers, here I will outline the general procedure of the inverse modelling and relevant data requirements. Inverse modelling usually starts with collection of experimental data (step 1) with an attempt to ask what the observed responses to a perturbation can tell us about the system in question (2, 1). By combining prior knowledge about the system, we try to identify processes underlying the observations (step 2). With major processes identified, we can develop a model to link these processes according to ecological mechanisms (step 3). The model can be used in the forward analysis to predict ecosystem responses to global change and can be used in the inverse analysis as well (step 4). The latter is usually implemented with mathematical algorithms for comparison of model predictions with the observed responses (White & Luo, 2001). By comparing model predictions with observations, step 4 attempts to challenge model structures against and/or to derive parameter values from data so that predictions of the system's behaviour are improved. Steps 1–4 are generally iterated several times until we find a model structure that adequately represents the system and then estimate parameter values that quantify interactions and feedbacks. The inverse analysis eventually results in improvement of our ability to predict future changes of the system (step 5). Although the inverse analysis has been hardly discussed in the literature of ecology, it was applied to photoacoustic signals (Tabrizi et al., 1998), population dynamics (Wood, 1997), seed dispersal (Clark et al., 1999), and rhizosphere C processes (Luo et al., 2001b). General procedure of inverse analysis to test model structure and derive parameter values from experimental data. Inverse analysis requires accurate data, which may derive from proper experimental design and effective measurement plans. Most FACE experiments set two CO2 levels, for example one at ambient CO2 and the other at ambient +200 ppm, a Ca level 40–60 yr from now. In such a perturbation experiment, generation of a perturbation size that is effective to probe mechanisms shall be one of the primary goals of experimental design. The perturbation sizes in a FACE experiment can be measured as the amount of additional C influx into an ecosystem at elevated CO2 in comparison to that at ambient CO2. The amount of additional C influx under elevated CO2 is related to productivity of an ecosystem and the CO2 treatment level. In order to obtain most useful experimental results from FACE experiments, it may even become desirable to estimate the perturbation size before we implement a FACE experiment. In the past, a substantial effort has been made on measurements of leaf physiology. Without integration of leaf-level data into estimation of canopy C influx, measurements are not useful for predicting ecosystem C sequestration. Probably because of difficulties in methodology, data on soil C and N processes are less available. Characterization of ecosystem C processes requires parameter estimation of partitioning coefficients among C pools, pool sizes, and C transfer rates from the pools. In order to quantify those parameters, it is imperative to observe time courses of fluxes and pool sizes following the perturbation of step CO2 changes. Uncertainties in experimental data make parameter estimation difficult. Quantification of C fluxes through different pathways by inverse analysis relies heavily on accuracy of data. All the FACE experiments are implemented in field with natural variability in environmental and biological factors. The field experiments have inherent complications, such as successional changes in ecosystem processes and episodic events of precipitation in deserts. In addition, FACE experiments starting with saplings (Miglietta et al., 2001) carry in developmental variation. CO2 experiments in agricultural fields may be accompanied with other disturbances, such as fertilization, fallow, and ploughing. Those factors all potentially obscure actual effects of elevated CO2 and cause tremendous difficulties in data analysis. How to reduce background variability and increase accuracy of data should be one of the major considerations in designing future FACE experiments. Ecosystem responses to the perturbation generated by a step CO2 increase in field CO2 experiments are different from those to a gradual Ca increase as in the real world. In order to develop our ability to predict ecosystem responses to a gradual Ca increase, we need to analyse data from FACE experiments using an inversion approach to challenge the structure of existing models and derive parameter values. In order to effectively conduct the inverse analysis, we need to collect highly accurate, informative data by improving experimental design and measurement plans for the FACE studies. The author thanks Dr R. Norby and an anonymous reviewer for constructive comments on the manuscript. This study was supported by the NSF/DOE/NASA/USDA/EPA/NOAA Interagency Program on Terrestrial Ecology and Global Change (TECO) by DOE under DE-FG03–99ER62800. This research is also part of the Forest-Atmosphere Carbon Transfer and Storage (FACTS-1) project at Duke Forest. The FACTS-1 project is supported by the US Department of Energy, Office of Biological and Environmental Research, under DOE contract DE-FG05–95ER62083 at Duke University and contract DE-AC02–98CH10886 at Brookhaven National Laboratory.