1. Introduction
Results from a GCM are strongly influenced by the exchange of radiation, momentum, sensible heat, and latent heat between the atmosphere and the land surface. The Simple Biosphere Model (SiB) (Sellers et al. 1986) is one of the many land surface parameterizations implemented for a GCM (Garratt 1993). It is a soil–vegetation–atmosphere transfer scheme (SVATS) that uses site-specific biophysical and physiological vegetation characteristics coupled to the meteorology of the adjacent atmosphere to calculate albedo, drag, conductance, and energy partitioning of a vegetated land surface. Recently, Sellers et al. (1992a, hereafter S92a) updated SiB to include a new single-layer canopy integration scheme incorporating a leaf model based on a mechanistic description of photosynthetic CO2 uptake and a semiempirical parameterization of stomatal conductance (Collatz et al. 1991, 1992; hereafter C91 and C92). Notably, CO2 exchanges associated with C3 and C4 photosynthesis and respiration by the canopy and soil are included in this SVATS. The new version, SiB2, is more robust, has fewer adjustable parameters, and provides an improved theoretical basis for relating biophysical properties of the canopy to spectral vegetation indices. The use of SiB2 within the Colorado State University (CSU) GCM is described in the series of papers by Sellers et al. (1996a, hereafter S96), Sellers et al. (1996b), and Randall et al. (1996). These simulation studies have led to new insights into the coupling of the biosphere and the atmosphere. For example, Sellers et al. (1996c) examined the significance of physiological mechanisms in the response of continental climates to increased CO2 concentration, and Denning et al. (1995; 1996a,b) examined the role of covariation of atmospheric convection with diurnal and seasonal cycles of CO2 exchange in determining the distribution of CO2 within the atmosphere. The present study is part of an ongoing effort to evaluate the accuracy of land surface processing in the CSU GCM by independent validation of SiB2.
GCMs have been validated using historical averages of regional surface observations of air temperature, pressure, wind speed, precipitation, and soil water, and satellite observations of albedo and outgoing longwave (Gates et al. 1990); and seasonal cycles of atmospheric CO2 concentration (Denning et al. 1996b). A SVATS may be evaluated directly by intercomparison of a suite of flux measurements (e.g., net radiation, evapotranspiration, sensible heat, soil heat) with simulations obtained when the model is driven with meteorological observations (Garratt et al. 1993). However, flux measurements and surface meteorology are inherently local in scale. The “footprint” of an eddy-correlation experiment is typically a few square kilometers, whereas a typical GCM grid box is thousands of square kilometers.
While there may be important differences in the behavior of the land surface on the scale of a GCM grid box from that of a single eddy-correlation site (see Avissar and Pielke 1989), the fundamental mechanisms that control ecosystem exchanges of energy, water, and CO2 are not expected to be scale dependent. Furthermore, measurements at the scale of eddy-correlation experiments are presently the largest scale at which it has been possible to obtain measurements that close the water, carbon, and energy budgets of the land surface for time intervals of weeks to years. Area averaging of meteorological variables and flux observations from multiple sites located in intensively studied areas such as FIFE (Betts and Ball 1993, 1998) have been used to construct regional datasets for validation studies; for example, Viterbo and Beljaars (1995) evaluated two alternative versions of the European Centre for Medium-Range Weather Forecasts (ECMWF) land surface parameterization, run in stand-alone mode, using sitewide averages of energy flux and soil water content of the FIFE site for 140 days. Useful checks for scale-dependent effects in extrapolating from a single site to larger scales can be obtained by comparing such regional-scale simulations with direct flux measurements by aircraft (MacPherson et al. 1992); with integrative measures of flux such as soil moisture runoff (Famiglietti et al. 1992; Wood and Lakshmi 1993; Liang et al. 1994); or to remote sensing indicators of flux (Sellers et al. 1992b, hereafter S92b; 1995). However, in the context of testing the fundamental biophysical and physiological parameterizations of land surface schemes, it is not clear whether regional-scale datasets offer any substantial advantage over datasets acquired at a single site. These considerations lead us to suggest that measurements at the eddy-correlation scale presently provide the best opportunity for rigorous testing of surface schemes in land surface models.
Recent technological advances have made it possible to conduct eddy-correlation studies over long time spans. The primary motivation has been to assess whether particular ecosystems are acting as sources or sinks for atmospheric CO2 (Wofsy et al. 1993); but these data also provide an opportunity to compare SVATS models in a stand-alone mode with observed seasonal behavior of ecosystem–atmosphere exchanges. That CO2 is now commonly measured is a significant advantage because CO2 is a sensitive diagnostic for physiological processes related to the control of canopy conductance (C91) and (as shown here) for soil temperature and moisture content. Some of the first work on seasonal changes in net ecosystem CO2 flux was initiated with FIFE (Verma et al. 1992), and similar work is now in progress at dozens of sites around the world.
These advances provide a stimulus for conducting longer term tests of SVATS and for including CO2 flux as a diagnostic in these tests. The PILPS (Project for Intercomparison of Land-surface Parameterizations Schemes) study (Henderson-Sellers et al. 1995) is making use of long-term datasets from the Cabauw grassland in the Netherlands and elsewhere to compare the outputs of SVATS models to observations of net radiation, sensible and latent heat flux, ground heat storage, and surface temperature using year-round simulations. However, CO2 flux is not included in the PILPS study. A number of studies have attempted to simulate CO2 flux, but most of these have been for a few days at most (Grant and Baldocchi 1992; Amthor et al. 1994; Baldocci 1994). The pioneering study of Saugier and Ripley (1975) is particularly noteworthy for simulations of CO2 exchange and carbon balance in a Canadian grassland for two 140-day intervals. Other studies extending several months have been published by Grant et al. (1993), Chen and Coughenour (1994), and Gao (1994). McMurtrie et al. (1992) have conducted a five-year study of an Australian pine plantation. However, most of these CO2 modeling studies have used models that are not appropriate for use as SVATS in climate models.
Our objective in this study was to exercise several features of SiB2 (particularly the capacity to simulate net CO2 exchange, physiological response to water stress, and detailed soil water balance) over the growing season of 1987 at FIFE. The design of the simulationexperiments is outlined in Fig. 1. We made extensive use of process-level studies of photosynthesis, soil respiration, and soil evaporation to test and calibrate component models included in SiB2. For most of this work SiB2 was run in a stand-alone mode driven by meteorological observations. Its output was compared, using statistical analysis of scatter and bias, to observed fluxes of heat, net radiation, and CO2 measured by eddy correlation at a single site (Verma et al. 1992, hereafter Site 16) conducted as part of FIFE (Sellers et al. 1988, 1992c; Sellers and Hall 1992). The FIFE experimental area is located near Manhattan, Kansas, in an area including the Konza Prairie Long-Term Ecological Research (LTER) area. The area is a tallgrass prairie ecosystem composed primarily of a mixed stand of C3 and C4 grasses whose abundances change during the growing season (S92b). Additional observations were made at Site 16 that were ancillary to the eddy flux experiment, including canopy biophysics, soil properties, surface temperature, soil temperature, surface reflectance, surface irradiance, and soil water. All observations used in this study, except site CO2 flux, are available on the FIFE CD-ROM (Strebel et al. 1994). Previous papershave described the flux measurements at Site 16 (Kim and Verma 1990a, b, 1991a; Verma et al. 1992), correlated these to changes in the available soil water (Kim and Verma 1991b), examined the ability of satellite-determined vegetation indices to predict the latent heat flux from this system (Verma et al. 1993), and estimated a site carbon budget (Kim et al. 1992). These studies have highlighted strong modulation of the fluxes by physiological mechanisms presumably related to seasonal changes in the availability of soil water at this site (Stewart and Verma 1992).
We performed seasonlong (142 day) runs with Tuned, Control, and Calibrated versions of the model (Fig. 1) driven by continuous micrometeorological observations. The Tuned version produced a good match to the fluxes and soil measurements taken over the growing season. The Control and Calibrated runs illustrate that substantial differences in water and carbon balance of the ecosystem can accrue over multimonth runs with relative small differences in parameter values or model structure. The process of developing the Tuned version of the model is used to highlight areas where SiB2 needs improvement.
2. Materials and methods
a. Notation and terminology
All notations used in this paper are defined in Table A1. They follow as closely as possible those of S96. We use the term prognostic to refer to the SiB2 variables whose values are retained from one time step to the next (Tc, Tg, Td, W1, W2, W3, Mc, Mg, and gc).
b. Site description
What we refer to in this study as “Site 16” is equivalent to FIFE grid 4439 (collocated 1987 station IDs 16, 11, 18, 73, and 106). The site has an elevation of 443 m and was centered at 39°03′06" lat, −96°32′28" long. It was described as ungrazed, burned on 15 April, mostly level, but with at least 7% north-facing steep slopes. Its species composition was listed in detail by Kim and Verma (1991a). They described it as being dominated by the C4 grasses Andropogon geradii, Sorghastrum nutans, and Panicum virgatum. The soil column is about 140 cm in depth changing from silty–clay–loam to clay to gravel to impermeable bedrock.
c. Short-term calibration runs
Thirteen short-term simulations of Site 16 were conducted ranging in duration from 4 to 58 simulated hours. These periods corresponded to times when the eddy flux station was operational at Site 16. Here gc was initialized to a steady-state value consistent with the conditions of the first time step [see S96, Eq. (C.16)]; Mc and Mg were intialized to zero; W1, W2, W3 were set to the same water content using an average of the start-day’s 0–140 cm water content from Verma et al. (1993); and Tc, Tg, and Td were initialized using field observations. These short-term simulations were conducted as part of the development of the Tuned version described below.
d. Growing season continuous runs
Three seasonlong SiB2 simulations of Site 16 were conducted for the 142-day interval, 28 May through 16 October 1987. They are identified by the version names Control, Calibrated, and Tuned (Fig. 1) and correspond to lab run numbers 18.46, 18.89, and 18.45, respectively. The Tuned version included several code modifications to SiB2v1.0 (see the Appendix). There were also version differences in parameter values, which are listed as separate columns in the parameter tables presented below. All versions were driven by the same meteorological observations consisting of S↓, L↓, em, Tm, P, and um (section 2f). Linear interpolation of the 30-min driver data into 6-min intervals was performed at run time. This allowed a more stable solution of the SiB2 energy and soil water budgets. The variable gc was initialized to a steady-state value consistent with the conditions of the first time step; Mc and Mg were intialized to zero;and W1, W2, W3, Tc, Tg, and Td were initialized using field observations.
e. Statistical comparisons of simulations with observations
We used a set of statistical techniques similar to Baldocchi (1992), which included a consideration of flux uncertainties. Fifteen simulation output variables were compared to corresponding observations (λEm, Hm, G, Rn, Am, W1, W2, W3, S↑, L↑, Tskin, T0–5cm, T10cm, T50cm, and fw). The following comparison statistics were calculated where observations are oi, simulations are si, and the total number of observations are n
- Standard error of the estimate (SEE, units are the same as oi):
- “Normalized” SEE, an estimate of relative uncertainty (unitless, equivalent to the rms deviation divided by the rms observation):
“1:1” Linear regression: The slope (BIAS), intercept (INTER), and their standard errors (SE) were calculated for the unconstrained linear regression line of all (oi, si) data pairs. Ideally the BIAS should be 1.0, and the INTER should be 0.0. The BIAS and INTER were tested for significant differences from these ideal values using a t test.
f. Meteorological dataset
To produce continuous meteorological driver-data for Site 16 simulations, we started with a FIFE areawide dataset (Betts and Ball 1993). The dataset was generated by averaging observations from all available FIFE automated meteorological stations (AMS), which measured conditions at 30-min intervals for days 145–288 of the 1987 growing season. Some short gaps had to be filled by linear interpolation and with data from the nearby LTER site and FIFE flux stations. Time was corrected from UTC to solar time by subtracting 6 h 26 min. We were concerned that the areawide-average data might be inconsistent with the micrometeorology of Site 16 since there was significant variability in the micrometeorology across the entire FIFE site. Therefore, discontinuous micrometeorological observations taken at Site 16 (section 2g) and at a site with similar aspect (E. Smith, Station 2, FIFE-CD-ROM) were used to checkfor inconsistencies between the areawide averages and the individual sites. In general the data were consistent, but we found that it was possible to improve the agreement by applying the correction factors 0.95, 0.93, and 0.75 to the areawide S↓, L↓, and um, respectively. The comparison of um should have taken into account the differing heights of the AMS (5.4 m) and eddy flux (2.5 m) stations, which we only became aware of after the simulations had been performed; however, the 0.75 correction factor used was probably adequate. Vapor pressure (em) was calculated from areawide Tm and Twet. Also extracted from the areawide dataset was Tskin, T10cm, T50cm, and S↑ for comparison with SiB2 simulation output. We found that L↑ (used for comparison with SiB2 simulation output) could be much more accurately estimated as a function of areawide Tskin using the standard blackbody radiation equation, than as a residual of areawide Rn, S↓, L↓, and S↑.
g. Eddy flux dataset
This dataset was constructed from eddy flux station data collected by S. Verma during 1987 at station 16 and were used for comparison with SiB2 simulation output. Here λEm, Hm, G, Rn, and T0–5cm were obtained from the FIFE CD-ROM, and Am was obtained from S. Verma. The method of data collection was described in Kim and Verma (1991b). The dataset covers the breadth of the growing season, but it contains several large gaps. Thirteen representative periods were selected from the dataset for comparison with the short-term SiB2 runs. For the seasonlong SiB2 runs all available eddy flux data were used for comparison with SiB2 output. Time was corrected from local daylight savings time to solar time by subtracting 1 h 26 min. For the Tuned version of SiB2 we calibrated S↓PAR calculations against the meteorological observations taken at Site 16, by adjusting S↓PARfac (Table 1).
h. Soil water dataset
These data are all shown in Fig. 2. Neutron probe measurements of soil water content centered at 20 cm (10–25 cm), 30 cm (25–35 cm), 40 cm (35–45 cm), 50 cm (45–55 cm), 60 cm (55–70 cm), 80 cm (70–90 cm), 100 cm (90–110 cm), 120 cm (110–130 cm), and 140 cm (130–140 cm) were conducted several times during the season by the FIFE staff. They supplemented these data near the surface by gravimetric measurements of soil water centered at 2.5 cm (0–5 cm) and 7.5 cm (5–10 cm), which were converted to volumetric soil water content using soil density measurements. The vertical bars mark the estimated range of soil water content available to the canopy (field capacity to wilting point). At different times during the growing season, the soil water profile shows very dynamic changes near the soil surface in response to the timing and quantity of precipitation received at the site. On the other hand, waterin the subsoil (>1 m) was constant and apparently not drawn down by the plants even during periods of severe drought. To check on the consistency of the site hydrology measurements, we compared changes in the total moisture stored in the soil (ΔWsoil) to the measured soil inputs (precipitation, 249.9 mm) and outputs (latent heat by eddy flux) integrated over the interval from yearday (DOY) 183 to 289 (322.3 mm). This analysis was not exact since we had to interpolate through intervals with no evaporation measurements, and we assumed that runoff and drainage (not measured) were negligible. Nevertheless, the changes in soil water measured at nine times during the interval matched (
i. Time-invariant miscellaneous site parameters
These can all be found in Table 1. With a few minor exceptions Control and Calibrated parameter valueswere all obtained from Sellers et al. (1996b). Latitude and longitude were obtained from Verma et al. (1993). Initial values of Tuned leaf reflectance and transmittance were estimated from Walter-Shea et al. (1992). These were modified by experimentation during SiB2 calibration runs to produce better matches to the following light observations: S↑ (section 2f); S↑PAR and S↑NIR (Site 16 surface MMR data, FIFE-CD-ROM); and S↑PAR and S↓PAR (Site 16 UNL light-bar data, FIFE-CD-ROM). Tuned soil reflectance values were set as in S92b. Parameters Csoilfac,
j. Time-varying biophysical site parameters
The biophysical parameters LT, N, fc3, and fc4 (S92b) and z2 (FIFE CD-ROM) are listed in Table 1 with their range of reported values. Daily values used in the simulations (available from author J. Berry) were derived by interpolation of the reported intermittant values. S92b derived their fc3 and fc4 data from species abundance data. Here LT was calculated by dividing LG by N, both from S92b. Their LG data were derived from Verma et al. (1992), and their N data from dried green and total biomass data measured by the FIFE staff. Unfortunately, a comparison of LG data from S92b with LG data measured by the FIFE staff revealed major differences possibly due to reported leaf curling complications. This variance in Site 16 leaf area measurements has been analyzed in detail by Kim et al. (1989). To be consistent with the other biophysical data we used only LG values from S92b. Intermittant measurements of z2 at Site 16 were made by the FIFE staff. Here, zm was obtained from Verma et al. (1993). The ll, lw, G1, G4, and χl aerodynamic constants were obtained from Dr. P. Sellers. The zc, z1, and zs (not measured in the field) were given daily values as a function of z2, using the equations in Table 1, which were obtained from P. Sellers. Note in Table 1 the special case, where Control and Calibrated fc3 and fc4 were set to constant (time invariant) values; otherwise all versions shared the same biophysical values.
k. Time-varying aerodynamic site parameters
Values of the SiB2 run-time aerodynamic parameters (z0, d, C1, C2, RB1, RB2, ha, G2, and G3) were calculated off-line for each day of the simulations, using the site surface characteristics LT, ll, lw, χl, V, zm, z2, zc, z1, zs, G1, and G4 (Table 1), and a special preprocessing program called MOMOPT (P. Sellers). The theory behind MOMOPT (a first-order approximation to a second-order closure scheme) is presented in Sellers et al. (1989, appendix A). Note the Tuned value of V was used in MOMOPT, so there would be no difference in aerodynamics between the versions.
l. Time-varying solar angle term (the Π factor)
m. Time-invariant soil hydraulic site constants
Table 2 lists the values of our Site 16 SiB2 soil hydrology parameters. Control and calibrated parameter values were obtained from the indicated references. Tuned values were set using soil survey horizon data and Clapp and Hornberger (1978) hydraulic parameters (CH hereafter). Average values of percent clay, percent silt, and percent sand for the SiB2 soil layers corresponding to D1, D2, and D3 were calculated from data on the soil profile weighted by thickness. SiB2 B, Ψs, Ks, and θs were set using the CH
n. Time-invariant plant physiology site parameters
Table 3 lists the plant physiology parameters. Control and Calibrated C4 and Tuned C3 parameter values were obtained from the indicated references. Tuned C4 values were set using a SiB2 leaf gas exchange chamber simulation mode to simulate leaf-level observations. Theenergy and water budget calculations of the main SiB2 model were by-passed, and parameter values for temperature Ta, humidity ea, incident par SPAR↓, CO2 concentration ca, and boundary layer conductance gb were prescribed using a special input file. To construct leaf response curves, one of these parameters was stepped over a range of values, while the other parameters were held constant. To collapse the canopy to a single leaf, canopy light extinction parameters were set to 1.0 (N, LT, Π). For each new value of the stepped parameter the photosynthesis submodel was iterated until photosynthesis and stomatal conductance came to steady state.
3. Results and discussion
a. Calibration from process-level studies
SiB2 uses the C3 and C4 leaf models for photosynthesis and stomatal regulation (C91 and C92) and a canopy integration scheme, which relates the response of the canopy to that of single leaves at the top of the canopy (S92a), to calculate canopy net photosynthesis and canopy resistance to water vapor transport. Extensive measurements of C4 photosynthesis at the leaf scale were conducted during the FIFE experiment (Polley et al. 1992) and as part of the LTER project (Knapp 1985). These data provided an opportunity to calibrate the physiological constants used by SiB2. The C4 photosynthesis parameters were fit to measurements of leaf photosynthesis, which were digitized from the published curves. Figures 3a–c illustrate leaf responses of the C4 grass Panicum virgatum to variations in light and intercellular CO2 (from Polley et al. 1992) and leaf temperatures (from Knapp 1985). Simulations (solid lines), using SiB2 in its leaf chamber mode (section 2n), are shown compared to observations (symbols). Similar results were obtained with two other C4 species (Sorgastrum nutans and Andropogon gerardii). The parameters that were fit included the dark respiration coefficients (Rd,
Soil respiration was estimated using the process model of Norman et al. (1992). Equation (A7) predicts CO2 flux at the soil surface as a function of canopy development, soil moisture, and soil temperature.
b. Calibration of canopy capacity from site-level measurements
Figure 4 shows the results of simulations of site-scale energy and CO2 fluxes for three consecutive days (dashed lines) compared with observations (points). Note that net site CO2 flux simulations and observations include soil respiration. These simulations cover a period (4–7 June, yeardays 155–158) early in the growing season following heavy rains and mild evaporative conditions. The availability of soil water was near its highest point of the season but still lower than field capacity (Fig. 2). Temperatures were mild, the humidity was moderate, and we assume there was no water stress.
The value of the major adjustable parameter of the physiological model, C4–Vmax0, was adjusted to fit the observations. A value of 30 μmol m−2 s−1 for a leaf at the top of the canopy resulted in good agreement of the canopy model with the observations. The simulations shown in Fig. 4 are for a mixed canopy. Lacking any measurements of photosynthesis of C3 species at FIFE, the C3–Vmax0 of rubisco was set to a plausible value of 120 μmol m−2 s−1 (C91). Similar values of C4–Vmax0 were obtained whether the canopy was modeled as a pure stand of C4 species or the real mixture (39% C3 and 61% C4 species). This compares with a weighted-average leaf-level calibration value of 23 μmol m−2 s−1 [weighted by the Kim and Verma (1991b) reported abundance percentages], to the 35 μmol m−2 s−1 reported for corn (C92), and is identical to the value selected for GCM simulations of C4–grassland ecosystems.
One caveat with respect to the Vmax0 calibrations is that any error in the SiB2 canopy integration factor Π [see S96, Eqs. (C.19), (C.21), (C.22)] will appear in the apparent Vmax0 in the canopy-scale calibrations. We prescribed Π from field measurements of canopy biophysics [Eq. (3)]. These Π values appeared reasonably consistent with a few test values obtained from local satellite observations (data not shown). However, given the overall uncertainty, a discrepancy of 30% between Vmax0 values obtained by calibration at the leaf and canopy scales provide support for the hypothesis presented by S92a that process-level studies of leaf physiology can be used to calibrate models intended for simulation of canopy processes. It is also encouraging and noteworthy that the model, using the calibrated value of C4–Vmax0, correctly simulated water and energy fluxes at the site without further tuning (Fig. 4).
c. Calibration of the soil water stress parameter from site-level measurements
SiB2 contains a variable for soil water stress, fw, which attenuates the value of Vmax0 (of both C3 and C4 species) as soil water availability falls [see S96, Eq. (C.17)]. This is intended to reproduce the effect but does not represent the mechanism of water stress. Earlierstudies have highlighted strong modulation of the fluxes by physiological mechanisms presumably related to seasonal changes in the availability of soil water at this site (Stewart and Verma 1992; Verma et al. 1993). No direct physiological studies of water stress responses were reported from FIFE. We made use of measurements of soil moisture at the site and seasonal changes in net site CO2 flux to calibrate the water-stress responses of SiB2.
Figure 5 presents diurnal plots of observed and simulated net site CO2 flux (Am) for six representative periods from a total of 13 periods used for calibrating the response to water stress. Figure 6 presents plots of observed and simulated energy fluxes (Rn, λEm, Hm, and G) corresponding in time to Fig. 5. At the beginning of the season (panel a), climate, canopy, and soil water conditions were optimal; Am was at peak levels, and surface energy exchange was dominated by λEm (Hm was very low). By 10 October (panel f) virtually all photosynthetic activity was gone. The soil was dry, temperatures were low, the canopy was in senescence, and Hm dominated surface energy exchange. We assume here that water stress accounts for all variation in canopy and soil CO2 flux over the season that could not otherwise be accounted for by SiB2 using the above calibration under unstressed conditions. For each short-term run, the model was initialized and driven from observed soil moisture, canopy biophysics and meteorology. The water stress parameter (fw) was adjusted for each of the 13 short-term simulation periods (while keeping Vmax0 constant) to fit the net CO2 flux. The resulting values of fw for all 13 short-term periods are plotted in Fig. 7 as a function of the yearday (DOY). Note that in SiB2 fw is normally calculated as a function of prognostic soil moisture. The “fitted” values of fw give a set of empirical “observations” that were used to calibrate SiB2’s functional water stress response. Note that only fw was tuned.
The fitted simulations (shown as lines on Figs. 5 and6) generally matched the energy fluxes and the net site CO2 flux fairly well. This grassland apparently responded quickly to both onset and relief of drought stress. When water stress was severe in August (panels c and d), modeled CO2 flux tended to be too low in the morning and too high in the afternoon. This same pattern was observed in trial simulations using August data from FIFE86 (Verma et al. 1989). The basis of this pattern is not known, but one possible explanation is the phenomenon known as hydraulic lift (Field and Goulden 1988), which we did not attempt to model.
Substantial seasonal changes in soil respiration (indicated by Am at night) may also be seen in Fig. 5. This was presumably a function of changes in soil moisture and temperature. Our short-term simulations, using (A7) and observed values for soil moisture and temperature, accurately matched the observed net site CO2 flux at night. Note that in these simulations both the soil water content and the soil temperature were initialized from field observations and changed insignificantly over the simulation period.
We next turn to the relationship between the apparent level of water stress and soil water status. Using linear interpolation of the Site 16 soil moisture measurements from Fig. 2, we produced 13 detailed soil water profiles (with 11 soil layers each) corresponding in time to each of the 13 empirical estimates of fw. Each of these 11-layer profiles was recast to the corresponding layers of SiB2 using the
Figure 8 shows the fw “observations” (solid symbols) plotted against volumetric water content of the the rooting zone (2–90 cm) of SiB2. These data, which spanseveral dry-down and recovery cycles, show a very strong and consistent relationship between the apparent level of physiological drought stress and soil water content.
d. Calibration of the soil in tuned version
After many iterations of seasonlong runs with adjustments to the physiology and soil hydrology, we came to the conclusion that the structure of the soil system in SiB2 was resulting in unrealistic behavior of the soil–plant system, making it difficult to arrive at a satisfactory calibration. For example, the top layer of soil in the model contained no roots. Thus, precipitation entering the soil did not affect the plants until there was sufficient accumulated precipitation to penetrate to the rooting layer, yet with abundant roots near the surface, plants appeared to obtain some water even from light rainfall events. The rooting layer in the Control version of SiB2 (2–90 cm) spans a depth interval with very dynamic variations in moisture content (see Fig. 2), yet as noted above the model only “knows” the averagewater content of the entire layer. We found that there could be very large errors in seasonlong simulations of net CO2 flux—even if W2 and the canopy water stress parameter were correctly simulated. This was traced to the strong effect of moisture on respiration in the organic-rich soil near the surface. The structure of the soil layers in the Control version of SiB2 (Table 1) were not appropriate to obtain the accurate estimates of moisture and temperature needed by the soil respiration model of Norman et al. (1992). The resulting large errors in soil respiration in seasonlong runs made it difficult to use net CO2 flux as a diagnostic in the calibration studies.
For these reasons, we modified the Control version of SiB2 with the goal of obtaining a more realistic simulation of the location of water in the soil profile relative to the sinks for water (uptake by roots, direct evaporation from the surface, and drainage) and the physiological elements that respond to water. This is referred to as the Tuned version. This version included the mixed C3/C4 canopy submodel noted above, but the most fundamental change in this version was that the canopy could extract water from all three layers of the soil and the depth of these layers was adjusted (Table 1). The canopy water stress factor was thus no longer a property of a single layer of soil, instead it was calculated by weighting the water availability in each soil layer (a local value of fw) by the density of roots in that layer (A5). For simplicity we used a linear relationship (see Fig. 8) between the water content of each layer and its layer specific fw (A4). The constants of this equation, Wisp and Wcsp, were assumed to be the same for all layers. To set the root density profile to time-invariant values, it was assumed to be correlated with the soil profile of organic matter. Extraction of water from each layer followed the product of the fractions of available water and roots in each layer (A6). These changes permitted the model to generate more realistic simulations of the soil water during seasonlong runs.
e. Analysis of season-long runs
Figure 9 shows simulations of five output parameters (dotted lines) from three different 142-day runs of SiB2, beginning on DOY 148 and ending on DOY 289, all compared to observations (solid circles). The runs are labeled Control, Calibrated, and Tuned. In the Am row of panels each simulated day is horizontally compressed so that its normal diurnal cycle (such as in Fig. 4) appears as a vertical trace, where the peak indicates the maximum midday flux of CO2 uptake, and the nadir indicates the maximum rate of respiratory CO2 release at night (principally from the soil). The fw row of panels shows the simulated and observed seasonal course of the water stress factor, and the next three rows of panels show the corresponding moisture content of the soil layers (W1, W2, and W3).
These three runs illustrate some of the complex interactions that occur in seasonlong runs. All runs were started with prognostic initialization values from day 148 observations and were driven by the same precipitation and micrometeorolgical observations. All of the models contained the soil respiration parameterization of Norman et al. (1992), but simulated soil respiration differed from run to run, because of differences in simulated soil moisture and temperature. In the following we will attempt to explain the differences between these simulation results. However, we acknowledge that any“insights” gained will be somewhat artificial since they are based partly on modeled behavior.
The Control simulation of Am shown in Fig. 9 (upper left) match observations very well early in the season, but after day 200, the simulations of Am at midday are consistently too high (overestimating photosynthesis) and generally underestimate respiration at night (except for “spikes,” which follow rainfall events). The Calibrated version tracks the seasonal pattern better than the Control, and the Tuned version is better yet. A noticable difference between the Calibrated and Tuned Am is the lack of recovery from water stress in the Calibrated version after a major rain event on DOY 217 (which did not penetrate to the root zone, 5–95 cm, of this model version).
Inspection of the second row of panels shows that the soil water stress parameter (fw) calculated from the prognostic soil moisture was above the observations for most of the season in the Control run, but it matched the observations much better in the Tuned and Calibrated runs. As shown in Fig. 8, the Control soil texture and water stress parameterization were not appropriate for this site. The clay soil at this site holds more water (and a larger fraction of that water is unavailable to plants) than the loam soil specified in the Control. The Tuned and Calibrated runs used water stress parameterizations that were fit to local site observations. However, this is only a partial explanation for the absence of water stress in the Control simulation. As layer 2 dryed down due to the extraction of excessively transpired water, water was “wicked up” inappropriately from the third layer (Fig. 9).
Table 4a provides a summary of the simulated hydrologic balance of the three runs. It is interesting how differently the hydrological balance was solved in these runs. For heuristic purposes, we will assume here that the Tuned run is a reasonable approximation of the“real” response of the site. Note, however, that there were no direct measurements at Site 16 of infiltration, runoff, or the partitioning of evaporation between canopy and soil, so these model predictions cannot be verified. Over the season, net water loss from the Control soil was 273 mm, whereas the corresponding drawdown of the soil of the Tuned run was only 86 mm. Several factors apparently contributed to this difference of 187 mm. The Control compared to Tuned produced 60 mm less infiltration of precipitation into the soil (
The net water balance of the Calibrated version was much closer to the Tuned. It lost only 18 mm more water over the season. However, this balance was achieved somewhat differently than in the Tuned run: 85 mm less water infiltrated the soil, but this was offset by canopy transpiration being 87 mm lower, and direct soil evaporation being 22 mm higher. One notable difference between the Calibrated and the Control run is in the amount of direct soil evaporation. An improved soil evaporation parameterization developed from process-level studies at FIFE by S92b reduced Egs from 184 to 151 mm.
The larger runoff from the Calibrated run relative to the Control is related to the lower saturated conductivity (Ks) of a clay soil compared to a loam (Table 2). Empirical studies of the soil water balance (section 2h) were not compatible with such a large runoff. We did not attempt to correct this in the Calibrated version, but we did in the Tuned version. To get more water to infiltrate into the soil of the Tuned version (which also had a clay soil), we experimented with increasing Ks. However, this had the undesirable side effect of inappropriately increasing unsaturated flow between layers. To avoid this we introduced a separate constant (Kinfil) that applied only when there was free water in the above-ground“puddle” (Mg). We choose a value for Kinfil about 10 times larger than Ks. This implies that the infiltrationcapacity for the field soil should, in fact, be different than the saturated flow of a homogeneous soil column. One possible rationalization is that there are “conduits” for free water in the soil structure such as those produced by roots, gopher and ant holes, or cracks (Germann and Beven 1985). We also arbitrarily reduced direct evaporation of precipitation from the canopy (Eci) by making the interception capacity (
The near absence of water stress in the Control run caused it to transpire the largest total quantity of water (Table 4a). It also had correspondingly higher gross photosynthesis (Table 4b). At a finer level, photosynthesis and transpiration of the Tuned run were greater than the Calibrated run. This was partly due to differences in the soil parameterization that permitted 85 mm more water to infiltrate into the soil over the course of the Tuned run. Both runs appear to have similar levels of water stress (Fig. 9, fw), but overall stress was slightly higher in the calibrated version. Water use efficiency (WUE) calculated on the basis of net photosynthesis was similar in all versions. For the C4 component, WUE = 5.56–6.02 mmol CO2 per mol H2O. For the C3 component (only in the Tuned version), WUE = 3.73. These compared to 7 and 3 mmol CO2 per mol H2O expected for C4 and C3 species respectively (Berry and Downton 1982). The presence of C3 species in the Tuned version decreased its WUE relative to the Control and Calibrated runs.
Net ecosystem carbon balance (Am) of these runs is remarkably different (Table 4b). The Tuned run resulted in a slight net accumulation of carbon over the course of the run of 53 gC m−2. The Calibrated run accumulated about 3.4 fold more and the Control run 18.2 fold morecarbon over the season. These differences stem largely from the simulation of soil respiration (Rsoil). Gross soil respiration in the Control run was less than half of that in the Tuned run. Yet, the same soil respiration parameterization was used in each model run. Inspection of the model output showed that this was occurring because the simulated water content at 10 cm (W10cm) required by the process model of Norman et al. (1992) was not realistic in the Control run. Note that W10cm was not explicitly simulated by SiB2. Instead it was obtained by weighted interpolation of the SiB2 soil water profile (W1, W2, and W3) using (A9). To address these errors we modified the Tuned version to have a more realistic treatment of the soil profile with layers of 0–15 cm, 15–90 cm, and 90–140 cm, and the model was changed to permit the canopy to extract water from all layers of the soil depending on the root density and water availability of the layers. As shown in Fig. 9, the Tuned version reproduced the soil water content of all three layers very accurately. The prediction of W1 was greatly improved by allowing D1 to be thicker than the Control version, resulting in an improved estimation of W10cm and consequently Rsoil. This is illustrated by the values of Am at night closely matching observations throughout the season. In SiB2 canopy respiration (RD) represents only leaf dark respiration (typically 2%–3% of Vmax0). SiB2 currently has no method of calculating canopy maintainence or growth respiration; RD was 4–5 times higher in the Tuned version compared to Control and Calibrated. This was a direct result of the higher value of the Tuned dark respiration rate parameter (Rd) as fit to local leaf-level measurements. It is possible that unaccounted canopy respiration quantities in the Control and Calibrated versions, which used a standard value of Rd, are somehow reflected in the higher Tuned Rd, but we have no way to evaluate this. These problems illustrate the importance of accurate prediction of site respiration as well as photosynthesis.
f. Statistical analysis of the simulation results
To provide a more quantitative basis for the analysis of different version runs, we turned to a statistical approach. Figure 10 shows “one-to-one” plots of eight output variables (y axis) compared to corresponding observations (x axis) for the Tuned version. Table 5 presents the data for three statistical indices SEE, NSEE, and BIAS (defined in section 2e). We have used these indices to assess the match of paired simulations and observations for the Control, Calibrated, and Tuned runs. The BIAS statistic indicated any systematic bias in the relationship between simulations and observations. For example, there is visible bias in the plot of Tuned Hm (Fig. 10), which has a regression slope of 1.42 (Table 5). The scatter about the one-to-one line is indicated by the SEE statistic. For example, Tuned Am, which has a BIAS of 1.00 (no bias), does have significant scatter with an SEE of 4 μmol m−2 s−1. The NSEEstatistic indicates the relative errors in these estimates. For example, Tuned Am has an NSEE of 30%, while Tuned Hm has an NSEE of 62%, indicating (relative to its own scale) that Am was simulated more accurately than Hm.
Bar charts of the NSEE statistic illustrate the relative accuracies of the Tuned, Calibrated, and Control simulations (Fig. 11). Reported uncertainties in flux measurements are indicated by horizontal dashed lines. This is intended as a nonrigorous statistical technique to evaluate model deviations compared to observational uncertainties. An analogous technique is discussed in Baldocchi (1992).
In this work net CO2 flux (Am) was used as the primary diagnostic for calibration and tuning of the model, and large improvements were realized in the accuracy of Am simulation as shown in Fig. 11. Latent heat flux, soil water, and to a lesser extent, sensible heat flux estimates were also improved as Am was calibrated. Net radiation (Rn) and upwelling longwave (L↑) matched very well to the observations and were not affected by the changes that improved Am. The improvement of the upwelling solar radiation (S↑) in the Tuned version resulted from adjustments to the model leaf reflectances and transmittances (Table 1) guided by Site 16 observations of canopy solar reflectance and transmittance. However, since Rn was unaffected, we presume the improvement in S↑ was inconsequential.
4. Conclusions
The experiments reported here have examined SiB2’s ability to predict changes in temperature, hydrologic state, energy flux, and CO2 flux over the 1987 growing season at FIFE Site 16. The results reported here demonstrate that SiB2, when properly calibrated to this site, can accurately simulate the observed responses of this ecosystem, during the growing season, forced only by an above-canopy atmosphere and a prescribed canopy phenology. In this model, stomatal regulation, which controls transpiration and energy partitioning, is linked to photosynthesis, a process that is universal and mechanistically well understood. Another advantage of basing the parameterization used in SiB2 on photosynthesis is that it can, in theory, be calibrated from leaf-scale measurements of photosynthesis and stomatal conductance. Indeed, the calibration used for general simulation of C4–grasslands in the CSU GCM is based on studies with leaves of Zea mays. In the present study, we compared two independent ways to fit C4–Vmax0, the major adjustable parameter of the photosynthesis model. First, we analyzed published measurements of gas exchange of individual leaves of the dominant species at FIFE. Second, we fit this parameter by tuning the canopy model to fit the observed net CO2 flux measured by eddy correlation above the grassland. These estimates agreed well with each other and were quite similar to the general calibration used for this ecosystem in the GCM.We note that this may be fortuitous, and that this agreement does depend on knowledge of the canopy integration factor (calculated from the leaf area index) and soil respiration. Nevertheless, this study lends support to this calibration scheme.
The present study provided an excellent opportunity to test SiB2’s ability to model seasonal changes in soil moisture stress. During the season, as photosynthetic activity went up and down through several drying cycles, there was a direct correspondence between photosynthetic activity and soil water content, with maximum soil water stress evident during yeardays 200–225 (Fig. 9, Tuned). When the model correctly simulated changes in CO2 flux associated with drought stress, it also corectly simulated the corresponding changes in latent and sensible heat fluxes. These results illustrate that net CO2 flux is a very sensitive diagnostic of the physiological processes that control surface energy exchange.
The mechanisms of water stress effects are still an active area of research (Sharkey and Badger 1982; Kaiser 1987; Bjorkman 1989; Chaves 1991; Quick et al. 1992; Pereira and Chaves 1993). The phyto–hormone abscisic acid (ABA) is most likely involved (Johnson et al. 1991; Tardieu et al. 1991, 1993; Tardieu and Davies 1993a,b). We should note that modelers differ on whether to use soil moisture or leaf water potential as the key state variable, and whether this has a direct oran indirect effect on stomatal conductance. Earlier versions of SiB calculated water stress through leaf water potential (LWP), as a function of soil water potential and the rate of transpiration (Sellers et al. 1986). In SiB2 the calculation of LWP was eliminated. In this study we found canopy water stress could be adequately predicted from soil water content or soil water potential, and by assuming that stomatal conductance was indirectly affected by water stress through its effect on photosynthesis. Other modeling studies have addressed water stress at FIFE. Hope (1992) and Gao (1994) used a direct effect of water stress on stomatal conductance, while Chen and Coughenour (1994) used a method similar to ours. However, neither of these studies specifically calibrated their models for water stress, and their simulations do not appear to reproduce the episodes of water stress at this site. Modeling studies by Kim and Verma (1991b), Kim and Verma (1991a), and Kim et al. (1992) specifically address water stress at this site and show that alternative formulations can be made to work equally well. All of the above studies used observed soil moisture content as an input parameter. Our study differs in that we predict soil water content and the level of water stress.
This study identified some areas of concern. Sparse and noisy data on leaf area index (LAI) are a significant source of concern in this study, and continue to be, as we extend this work into the years 1988–89. We estimatethat simulations of site respiration were the largest source of error in our site CO2 flux modeling. Soil respiration estimation [Eq. (A7)] requires accurate prediction of soil temperature and moisture [note that Kim et al. (1992), Gao (1994), and Chen and Coughenour (1994) avoided this problem by supplying observed 10-cm values]. In our simulations, soil moisture was predicted as output of the model, and respiration was accurately predicted only when soil conditions were correctly simulated. This required a more realistic representation of the profile of roots, moisture, and temperature in the soil. Error analysis revealed a substantial bias in our simulations of Hm (Figs. 10 and 11). Other investigators have discussed problems at FIFE in the context of modeling Hm from a SVATS or remotely sensed Tskin (Vining and Blad 1992; Hall et al. 1992; Cooper et al. 1995), suggesting that the coupling of different surface thermal radiators (e.g., litter, bare soil, rocks, or sun and shade leaves) to aerodynamic conductances is a likely source of error.
The result that SiB2 does an adequate job of simulating the seasonal course of energy, water, and CO2 exchange at a single well-characterized site provides a rigorous test of the way that the biophysical and physiological mechanisms controlling these processes are represented in SiB2. We suggest that this is a fundamental requirement for simulation of these processes at the scale of a GCM. However, we acknowledge that scale-dependent problems (e.g., aggregation of sub-grid-cell processes, and averaging of nonlinear properties) must also be considered in extrapolating from these local-scale calibration studies to the scale of a GCM grid box.
Acknowledgments
We note the passing of our friend and colleague, Dr. Cyril Grivet, who was among those lost in the tragic crash of TWA flight 800 the evening of 17 July 1996. Cyril’s way with instruments, his gift for mathematics, and his generosity are missed. This research was supported by NASA under Grants NAGW-3699 (FIFE) and NAS 531731 (EOS). We are particularly grateful to Alan Betts for the FIFE sitewide average meteorology dataset; to Shashi Verma for the site 16 eddy flux dataset; to the staff of FIFE who assembled the FIFE CD-ROM; and to Jim Collatz, Scott Denning, and Dave Randall for sharing their wisdom.
REFERENCES
Amthor, J. S., M. L. Goulden, J. W. Munger, and S. C. Wofsy, 1994:Testing a mechanistic model of forest-canopy mass and energy exchange using eddy correlation: Carbon dioxide and ozone uptake by a mixed oak–maple stand. Aust. J. Plant Physiol.,21, 623–651.
Avissar, R., and R. A. Pielke, 1989: A parameterization of heterogeneous land surfaces for atmospheric numerical models and its impact on regional meteorology. Mon. Wea. Rev.,117, 2113–2136.
Baldocchi, D., 1992: A Lagrangian random-walk model for simulating water vapor, CO2 and sensible heat flux densities and scalar profiles over and within a soybean canopy. Bound.-Layer Meteor.,61, 113–144.
——, 1994: A comparative study of mass and energy exchange rates over a closed C3 (wheat) and an open C4 (corn) crop. II: CO2 exchange and water use efficiency. Agric. For. Meteor.,67, 291–321.
Berry, J. A., and W. J. S. Downton, 1982: Environmental regulation of photosynthesis. Photosynthesis. Vol II: Development, Carbon Metabolism and Plant Productivity, Academic Press, 263–343.
Betts, A. K., and J. H. Ball, 1998: FIFE surface climate and site-average dataset 1987–89. J. Atmos. Sci.,55, 1091–1108.
——, ——, and A. C. M. Beljaars, 1993: Comparison between the land surface response of the European Centre model and the FIFE-1987 data. Quart. J. Roy. Meteor. Soc.,119, 975–1001.
Bjorkman, O., 1989: Some viewpoints on photosynthetic response and adaptation to environmental stress. Photosynthesis, W. R. Briggs, Ed., Alan R. Liss, 45–58.
Chaves, M. M., 1991: Effects of water deficits on carbon assimilation. J. Exp. Bot.,42, 1–16.
Chen, D. X., and M. B. Coughenour, 1994: GEMTM: A general model for energy and mass transfer of land surfaces and its application at the FIFE sites. Agric. For. Meteor.,68, 145–171.
Clapp, R. B., and G. M. Hornberger, 1978: Empirical equations for some soil hydraulic properties. Water Resour. Res.,14, 601–604.
Collatz, G. J., J. T. Ball, C. Grivet, and J. A. Berry, 1991: Physiological and environmental regulation of stomatal conductance, photosynthesis and transpiration: A model that includes a laminar boundary layer. Agric. For. Meteor.,54, 107–136.
——, M. Ribas-Carbo, and J. A. Berry, 1992: Coupled photosynthesis stomatal conductance model for leaves of C4 plants. Aust. J. Plant Physiol.,19, 519–538.
Cooper, H. J., E. A. Smith, and W. L. Crosson, 1995: Limitations in estimating surface sensible heat fluxes from satellite radiometric skin temperatures. J. Geophys. Res.,100, 25419–25427.
Denning, A. S., I. Y. Fung, and D. A. Randall, 1995: Latitudinal gradient of atmospheric CO2 due to seasonal exchange with land biota. Nature,376, 240–243.
——, G. J. Collatz, C. Zhang, D. A. Randall, J. A. Berry, P. J. Sellers, G. D. Colello, and D. A. Dazlich, 1996a. Simulations of terrestrial carbon metabolism and atmospheric CO2 in a general circulation model. Part 1: Surface carbon fluxes. Tellus,48B, 521–542.
——, D. A. Randall, G. J. Collatz, and P. J. Sellers, 1996b: Simulations of terrestrial carbon metabolism and atmospheric CO2 in a general circulation model. Part 2: Spatial and temporal variations of atmospheric CO2. Tellus,48B, 543–567.
Famiglietti, J. S., E. F. Wood, M. Sivapalan and D. J. Thongs, 1992:A catchment scale water balance model for FIFE. J. Geophys. Res.,97, 18997–19007.
Field, C. B., and M. G. Goulden, 1988: Hydraulic lift: Broadening the sphere of plant-environment interactions. Tree,3, 189–190.
Gao, W., 1994: Atmosphere–biosphere exchange flux of carbon dioxide in a tallgrass pairie modeled with satellite spectral data. J. Geophys. Res.,99, 1317–1327.
Garratt, J. R., 1993: Sensitivity of climate simulations to land–surface and atmospheric boundary-layer treatments—A review. J. Climate,6, 419–449.
——, P. B. Krummel, and E. A. Kowalczyk, 1993: The surface energy balance at local and regional scales—A comparison of general circulation model results with observations. J. Climate,6, 1090–1109.
Gates, W. L., P. R. Rowntree, and Q. C. Zeng, 1990: Validation of climate models. Climate Change: The IPCC Impacts Assessments, W. J. McGegart, G. W. Sheldon, and D. C. Griffiths, Eds., Australian Govt. Publ. Service, 97–130.
Germann, P. F., and K. Beven, 1985: Kinematic wave approximation to infiltration into soils with sorbing macropores. Water Resour. Res.,21, 990–996.
Grant, R. F., and D. D. Baldocchi, 1992: Energy transfer over crop canopies: Simulation and experimental verification. Agric. For. Meteor.,61, 129–149.
——, P. Rochette, and R. L. Desjardins, 1993: Energy exchange and water use efficiency of field crops: Validation of a simulation model. Agron. J.,85, 916–928.
Hall, F. G., K. F. Huemmrich, S. J. Goetz, P. J. Sellers, and J. E. Nickeson, 1992: Satellite remote sensing of surface energy balance: Success, failures, and unresolved issues in FIFE. J. Geophys. Res.,97, 19061–19089.
Henderson-Sellers, A., A. J. Pitman, P. K. Love, P. Irannejad, and T. H. Chen, 1995: The project for intercomparison of land surface parameterization schemes (PILPS): Phases 2 and 3. Bull. Amer. Meteor. Soc.,76, 489–503.
Hope, A. S., 1992: Estimating the daily course of konza prairie latent heat fluxes using a modified terga model. J. Geophys. Res.,97, 19023–19031.
Johnson, I. R., J. J. Melkonian, J. H. M. Thornley, and S. J. Riha, 1991: A model of water flow through plants incorporating shoot/root “message” control of stomatal conductance. Plant Cell Environ.,14, 531–544.
Kaiser, W. M., 1987: Effects of water deficit on photosynthetic capacity. Physiol. Plant.,71, 142–149.
Kim, J., and S. B. Verma, 1990a: Components of surface energy balance in a temperate grassland ecosystem. Bound.-Layer Meteor.,51, 401–417.
——, and ——, 1990b: Carbon dioxide exchange in a temperate grassland ecosystem. Bound.-Layer Meteor.,52, 135–150.
——, and ——, 1991a: Modeling canopy photosynthesis: Scaling up from a leaf to canopy in a temperate grassland ecosystem. Agric. For. Meteor.,57, 187–208.
——, and ——, 1991b: Modeling canopy stomatal conductance in a temperate grassland ecosystem. Agric. For. Meteor.,55, 149–166.
——, C. Hays, S. Verma, and B. Blad, 1989: A preliminary report on LAI values obtained during FIFE by various methods. 27 pp. [Available from Center for Agricultural Meteorology and Climatology, UNL, P.O. Box 830728, Lincoln, NE 68583-0728.].
——, S. B. Verma, and R. J. Clement, 1992: Carbon dioxide budget in a temperate grassland ecosystem. J. Geophys. Res.,97, 6057–6063.
Knapp, A. K., 1985: Effect of fire and drought on the ecophysiology of andropogon geradii and panicum vergatum in a tallgrass prairie. Ecology,66, 1309–1320.
Liang, X., D. P. Lettenmaier, E. F. Wood, and S. J. Burges, 1994: A simple hydrologically based model of land surface water and energy fluxes for general circulation models. J. Geophys. Res.,99, 14415–14428.
MacPherson, J. I., R. L. Grossman, and R. D. Kelly, 1992: Intercomparison results for FIFE flux aircraft. J. Geophys. Res.,97, 18499–18514.
McMurtrie, R. E., R. Leuning, W. A. Thompson, and A. M. Wheeler, 1992: A model of canopy photosynthesis and water use incorporating a mechanistic formulation of leaf CO2 exchange. For. Ecol. Manag.,52, 261–278.
Norman, J. M., R. Garcia, and S. B. Verma, 1992: Soil surface fluxes and the carbon budget of a grassland. J. Geophys. Res.,97, 18845–18854.
Pereira, J. S., and M. M. Chaves, 1993: Plant water deficits in mediterranean ecosystems. Water Deficits Plant Responses from Cell to Community, J. A. C. Smith and H. Griffiths, Eds., BIOS Scientific, 237–251.
Polley, H. W., J. M. Norman, T. J. Arkebauer, E. A. Walter-Shea, D. H. Greegor Jr., and B. Bramer, 1992: Leaf gas exchange of Andropogon gerardii vitman, Panicum virgatum 1., and Sorghastrum nutans (1.) Nash in a tallgrass pairie. J. Geophys. Res.,97, 18837–18844.
Quick, W. P., M. M. Chaves, R. Wendler, M. M. David, M. L. Rodrigues, J. A. Passaharinho, J. S. Pereira, M. D. Adcock, R. C. Leegood, and M. Stitt, 1992: The effect of water stress on photosynthetic carbon metabolism in four species grown under field conditions. Plant Cell Environ.,15, 25–35.
Randall, D. A., P. J. Sellers, J. A. Berry, D. A. Dazlich, C. Zhang, G. J. Collatz, A. S. Denning, S. O. Los, C. B. Field, I. Y. Fung, C. O. Justice, and C. J. Tucker, 1996: A revised land-surface parameterization (SiB2) for GCMs. Part III: The greening of the Colorado State University general circulation model. J. Climate,9, 738–763.
Saugier, B., and E. A. Ripley, 1975: A model of growth and water use for a natural grassland. Proc. Summer Computer Simulation Conf., 945–953. [Available from Simulation Councils, 4838 Ronson Ct., Suite L, San Diego, CA 92111.].
Sellers, P. J., and F. G. Hall, 1992: FIFE in 1992: Results, scientific gains, and future research directions. J. Geophys. Res.,97, 19091–19109.
——, Y. Mintz, Y. C. Sud, and A. Dalcher, 1986: A simple biosphere model (SiB) for use within general circulation models. J. Atmos. Sci.,43, 305–331.
——, F. G. Hall, G. Asrar, D. E. Strebel, and R. E. Murphy, 1988: The First ISLSCP Field Experiment (FIFE). Bull. Amer. Meteor. Soc.,69, 22–27.
——, W. J. Shuttleworth, J. L. Dorman, A. Dalcher, and J. M. Roberts, 1989: Calibrating the simple biosphere model for Amazonian tropical forest using field and remote sensing data. Part I: Average calibration with field data. J. Appl. Meteor.,28, 727–759.
——, F. G. Hall, D. E. Strebel, R. D. Kelley, S. B. Verma, B. L. Markhame, B. L. Blad, D. S. Schimel, J. R. Wang, and E. Kanemasu, 1990: FIFE interim report February 1990. NASA/Goddard, 220 pp.
——, J. A. Berry, G. J. Collatz, C. B. Field, and F. G. Hall, 1992a:Canopy reflectance, photosynthesis and transpiration, III. Remote Sens. Environ.,42, 1–20.
——, M. D. Heiser, and F. G. Hall, 1992b: Relations between surface conductance and spectral vegetation indices at intermediate (100 m−2 to 15 km−2) length scales. J. Geophys. Res.,97, 19033–19059.
——, F. G. Hall, G. Asrar, D. E. Strebel, and R. E. Murphy, 1992c:An overview of the First International Satellite Land Surface Climatology Project (ISLSCP) Field Experiment (FIFE). J. Geophys. Res.,97, 18345–18371.
——, M. D. Heiser, F. G. Hall, S. J. Goetz, D. E. Strebel, S. B. Verma, R. L. Desjardins, P. M. Schuepp, and J. I. MacPherson, 1995: Effects of spatial variability in topography, vegetation cover and soil moisture on area-averaged surface fluxes: A case study using the FIFE 1989 data. J. Geophys. Res.,100, 25607–25629.
——, D. A. Randall, G. J. Collatz, J. A. Berry, C. B. Field, D. A. Dazlich, C. Zhang, and G. D. Colello, 1996a: A revised land surface parameterization (SiB2) for atmospheric GCMs. Part I:Model formulation. J. Climate,9, 676–705.
——, S. O. Los, C. J. Tucker, C. O. Justice, D. A. Dazlich, G. J. Collatz, and D. A. Randall, 1996b: A revised land surface parameterization (SiB2) for atmospheric GCMs. Part II: The generation of global fields of terrestrial biophysical parameters from satellite data. J. Climate,9, 706–737.
——, L. Bounoua, G. J. Collatz, D. A. Randall, D. A. Dazlich, and S. Los, 1996c: Comparison of radiative and physiological effects of doubled atmospheric CO2 on climate. Science,271, 1402–1406.
Sharkey, T. D., and M. R. Badger, 1982: Effects of water stress on photosynthetic electron transport, photophosphorylation, and metabolite levels of Xanthium strumarium mesophyll cells. Planta,156, 199–206.
Stewart, J. B., and S. B. Verma, 1992: Comparison of surface fluxes and conductances at two contrasting sites within the FIFE area. J. Geophys. Res.,97, 18623–18628.
Strebel, D. E., D. R. Landis, K. F. Huemmrich, and B. W. Meeson, 1994: Collected Data of the First ISLSCP Field Experiment, Volume 1: Surface Observations and Non-Image Data Sets. NASA, CD-ROM. [Available from PLDS, Code 923, NASA/GSFC, Greenbelt, MD 20771.].
Tardieu, F., and W. J. Davies, 1993a: Integration of hydraulic and chemical signalling in the control of stomatal conductance and water status of droughted plants. Plant Cell Environ.,16, 341–349.
——, and ——, 1993b: Root-shoot communication and whole-plant regulation of water flux. Water Deficits Plant Responses from cell to Community, J. A. C. Smith and H. Griffiths, Eds., BIOS Scientific, 237–251.
Tardieu, F., N. Katerji, O. Bethenod, J. Zhang, and W. J. Davies, 1991: Maize stomatal conductance in the field: Its relationship with soil and plant water potentials, mechanical constraints and ABA concentration in the xylem sap. Plant Cell Environ.,14, 121–126.
——, J. Zhang, and D. J. G. Gowing, 1993: Stomatal control by both [ABA] in the xylem sap and leaf water status: A test of a model for droughted or ABA-fed field-grown maize. Plant Cell Environ.,16, 413–420.
Verma, S. B., J. Kim, and R. J. Clement, 1989: Carbon dioxide, water vapor and sensible heat fluxes over a tallgrass prairie. Bound.-Layer Meteor.,46, 53–67.
——, ——, and ——, 1992: Momentum, water vapor, and carbon dioxide exchange at a centrally located prairie site during FIFE. J. Geophys. Res.,97, 18629–18639.
——, P. J. Sellers, C. L. Walthall, F. G. Hall, J. Kim, and S. J. Goetz, 1993: Photosynthesis and stomatal conductance related to reflectance on the canopy scale. Remote Sens. Environ.,44, 103–116.
Vining, R. C., and B. L. Blad, 1992: Estimation of sensible heat flux from remotely sensed canopy temperatures. J. Geophys. Res.,97, 18951–18954.
Viterbo, P., and C. M. Beljaars, 1995: An improved land surface parameterization scheme in the ECMWF model and its validation. J. Climate,8, 2716–2748.
Walter-Shea, E. A., B. L. Blad, C. J. Hays, M. A. Mesarch, D. W. Deering, and E. M. Middleton 1992: Biophysical properties affecting vegetative canopy reflectance and absorbed photosynthetically active radiation at the FIFE site. J. Geophys. Res.,97, 18925–18934.
Wofsy, S. C., M. L. Goulden, J. W. Munger, S.-M. Fan, P. S. Bakwin, B. C. Daube, S. L. Bassow and F. A. Bazzaz, 1993: Net exchange of carbon dioxide in a mid-latitude forest. Science,260, 1314–1317.
Wood, E. F., and V. Lakshmi, 1993: Scaling water and energy fluxes in climate systems: Three land–atmosphere modeling experiments. J. Climate,6, 839–857.
APPENDIX
Description of Differences from theStandard SiB2 Model
A copy of version 1.0 of standard SiB2 (SiB2v1.0) FORTRAN source code and documentation can be obtained from J. Collatz (Code 923, Goddard Space Flight Center/NASA, Greenbelt, MD 20771). Copies of the current study’s modified SiB2v1.0 source code and accompanying sets for a Unix workstation with a FORTRAN compiler can be obtained from author J. Berry. Figure A1 presents a schematic diagram of the structure and key parameters of what we call our Tuned version of SiB2 (Fig. 1); see Table A1 for definition of symbols. The differences from SiB2v1.0 are described below.
Mixed C3/C4 canopy
Water stress parameterization
Hydrology
In SiB2v1.0 the maximum capacity of canopy interception and ground surface pools are set as constant values in the source code, and the depth of the top soil layer, D1, is set to 2 cm. In the Calibrated and Tunedversions these were converted into variables allowing them to be assigned different values (see Table 1).
Table 2 lists the soil physics parameters. In SiB2v1.0 a single set of soil physics constants (θs, Ks, Ψs, and B) are used for all soil layers. In the Tuned version these constants were given separate values for each soil layer (i = 1, 3): θs(i), Ks(i), Ψs(i), and B(i). In the control and calibrated versions the maximum rate of infiltration of water into the soil was controlled by Ks. In the Tuned version infiltration was given a separate maximum conductivity constant, Kinfil.
Soil respiration
Soil temperature at arbitrary depths
Soil water at arbitrary depths
Soil evaporation
The experimental design for these simulation studies. The Control configuration (run 46) used“vanilla” SiB2v1.0 with parameter values from look-up tables used to run the CSU GCM for the grid square containing FIFE. The Tuned configuration (run 45) represents our best fit of SiB2 to Site 16 allowing both code and parameter changes to SiB2v1.0. Sensitivity analysis was used to identify a few key changes resulting in the Calibrated configuration (run 89). This was similar to the Control—with minimal tuning.
Citation: Journal of the Atmospheric Sciences 55, 7; 10.1175/1520-0469(1998)055<1141:MOEWAC>2.0.CO;2
Profiles of soil water content at Site 16 in 1987: (a)–(c) volumetric soil water profiles covering the dates 28 May–15 October. Separate profiles are identified by symbols as well as the yearday (DOY). Left and right vertical dark lines represent the soil water contents at the permanent wilting point and at field capacity, respectively (Verma et al. 1989).
Citation: Journal of the Atmospheric Sciences 55, 7; 10.1175/1520-0469(1998)055<1141:MOEWAC>2.0.CO;2
Leaf-level studies of gas exchange responses (data points) and simulations using the Tuned version’s leaf chamber simulation mode (solid lines). P. virgatum is used as an example in panels (a)–(c): (a) net photosynthesis as a function of intercellular CO2 at four different absorbed light levels (Q) in μmol m−2 s−1 [data from Polley et al.(1992), Fig. 1]; (b) net photosynthesis as a function of absorbed quantum flux at saturating intercellular CO2 [data from Polley et al. (1992), Fig. 2]; (c) net photosynthesis as a function of leaf temperature at rate-saturating values of Q [data from Knapp (1985), Fig. 2]. Corresponding plots were done for A. gerardii and S. nutans also (not shown), and the data were pooled into panel (d) a one-to-one plot of modeled versus observed net photosynthesis. The pluses are from plots of net photosynthesis as a function of intercellular CO2. The solid circles are from plots of net photosynthesis as a function of Q. The open circles are from plots of net photosynthesis as a function of temperature. Different values of Vmax0 were used for each species (Table 3). The slope of a linear regression line fitted to the data points and passing through the origin is 0.96 with an r2 of 0.98.
Citation: Journal of the Atmospheric Sciences 55, 7; 10.1175/1520-0469(1998)055<1141:MOEWAC>2.0.CO;2
Short-term simulations (dashed lines) and eddy correlation observations (solid circles) of site fluxes from the period 4–7 Jun 1987 (DOY 155–158). The simulations were performed using a preliminary form of the Tuned version. The canopy C4–Vmax0 was set to 30 μmol m−2 s−1.
Citation: Journal of the Atmospheric Sciences 55, 7; 10.1175/1520-0469(1998)055<1141:MOEWAC>2.0.CO;2
Short-term simulations (dashed lines) and eddy-correlation observations (solid circles) of net site CO2 flux (Am). The simulations were performed using the same value of the C4–Vmax0 as in Fig. 4. The soil water stress factor (fw) was adjusted to fit simulations to observations. Six examples are shown from a total of 13 intervals: (a) 6–8 Jul (DOY 187–189), nearly full green canopy (LG = 2.2) and no soil moisture stress (fw = 1.0); (b) 30 Jul (DOY 211), still nearly full green canopy (LG = 2.1) but with very high soil moisture stress following the first dry down period (fw = 0.20); (c) 10 Aug (DOY 222), a slightly senescent green canopy (LG = 1.8) but with some recovery from the first dry down period following a series of rain events (fw = 0.50); (d) 11–12 Aug (DOY 223–224), still a slightly senescent green canopy (LG = 1.8) but with very high soil moisture stress following the second dry down period (fw = 0.27); (e) 15–16 Aug (DOY 227–228), still a slightly senescent green canopy (LG = 1.7) but with nearly complete recovery from the second dry down period following a major series of rain events (fw = 0.72); (f) 8 Oct (DOY 281), nearly complete canopy senescence (LG = 0.7) coupled with very high soil moisture stress following the third dry down period (fw = 0.30).
Citation: Journal of the Atmospheric Sciences 55, 7; 10.1175/1520-0469(1998)055<1141:MOEWAC>2.0.CO;2
Energy budget simulations and observations associated with the short-term model runs shown in Fig. 5: Rn simulations (solid lines); Rn observations (solid circles); λEm simulations (dashed lines); λEm observations (open squares); Hm simulations (dash–dot lines); Hm observations (solid triangles); G simulations (dotted lines); and G observations (open circles).
Citation: Journal of the Atmospheric Sciences 55, 7; 10.1175/1520-0469(1998)055<1141:MOEWAC>2.0.CO;2
“Observations” of the soil water stress factor (fw) obtained by the best fit of 13 short-term simulations to observed fluxes (cf. Figs. 5 and 6) plotted against the DOY. Note that the lines connecting measurements may not represent the true time course of changes in water stress during these intervals.
Citation: Journal of the Atmospheric Sciences 55, 7; 10.1175/1520-0469(1998)055<1141:MOEWAC>2.0.CO;2
The soil water stress factor from Fig. 7 and the corresponding average volumetric soil water content in the 2–90-cm-depth interval. Interpolations in time between the profiles shown in Fig. 2 were used as necessary to match the times of the data points in Fig. 7. Rainfall and latent heat data were used to correct obvious water status interpolation errors in two cases. The solid line is an empirical fit to the data, using (A4), characterized by the water content at“incipient water stress” (Wisp = 0.27) and that at “complete water stress” (Wcsp = 0.17). A multilayered form of this expression (A5) was used in the Tuned version, which resulted in slightly different values of Wisp and Wcsp (Table 1). The broken lines were obtained from the thermodynamic parameterization of water stress [S96, Eq. (C.17)] with different soil texture–dependent parameters. The dash–dot line was generated using the Control version’s Ψc (from Table 1) and loam soil texture (from Table 2). The dashed line shows Control loam soil texture replaced by the Calibrated version’s clay soil texture. The dotted line is the same as the dashed line with Control Ψc replaced by Calibrated Ψc.
Citation: Journal of the Atmospheric Sciences 55, 7; 10.1175/1520-0469(1998)055<1141:MOEWAC>2.0.CO;2
Continuous long-term SiB2 simulations (dashed lines) of net site CO2 flux (Am), soil water stress (fw), and soil water (W1, W2, and W3) compared to available observations (solid circles) from Site 16, 28 May–16 October 1987 (142 days, DOY 148–289) for the Control, Calibrated, and Tuned versions. Note that at this timescale in the Am panels diel curves (like those in Fig. 4) are compressed to vertical lines. The fw observations are from Fig. 7. The soil water observations were processed from Fig. 2 using Eq. (A9).
Citation: Journal of the Atmospheric Sciences 55, 7; 10.1175/1520-0469(1998)055<1141:MOEWAC>2.0.CO;2
One-to-one plots of Tuned version simulations (y axis) against available observations (x axis) for net site CO2 flux (Am), net radiation (Rn), latent heat (λEm), upwelling shortwave radiation (S↑), sensible heat (Hm), site skin temperature (Tskin), soil heat (G), and deep soil temperature (Td). These data are from the same Tuned run as in Fig. 9. The statistics corresponding to these one-to-one plots are listed in Table 5.
Citation: Journal of the Atmospheric Sciences 55, 7; 10.1175/1520-0469(1998)055<1141:MOEWAC>2.0.CO;2
Bar graphs of the NSEE statistic for the Control, Calibrated, and Tuned versions: net site CO2 flux (Am), net radiation (Rn), latent heat (λEm), soil heat (G), soil water (as an average of the NSEE’s for W1, W2, and W3), upwelling longwave radiation (L↑), sensible heat (Hm), and upwelling shortwave radiation (S↑). These data are from the same runs as in Fig. 9. The NSEE values used in these plots were taken directly from Table 5. Each horizontal dashed line indicates the ±% overall uncertainty assigned by FIFE investigators to that flux measurement (Sellers et al. 1990, 4–22).
Citation: Journal of the Atmospheric Sciences 55, 7; 10.1175/1520-0469(1998)055<1141:MOEWAC>2.0.CO;2
Fig. A1. A schematic diagram of the Tuned version of SiB2. The symbols and units are defined in Table A1. Across the top of the diagram are the SiB2 micrometeorological driving parameters and the resulting net fluxes of radiation, sensible heat, water, and CO2 shown in yellow, red, blue, and green, respectively. The insert in the center of the canopy depicts the biochemical transformations that occur in a chloroplast. The photosynthesis model does not explicitly predict all of these reactions, but it does predict the steady-state limitations of the overall process. Note added in proof: Inside the box rules for Canopy Air Space the symbols A3 and A4 should be omitted.
Citation: Journal of the Atmospheric Sciences 55, 7; 10.1175/1520-0469(1998)055<1141:MOEWAC>2.0.CO;2
Miscellaneous site parameter values. The symbols and units are defined in Table A1.
Soil hydraulic site parameter values. The symbols and units are defined in Table A1.
Plant physiology site parameter values. The symbols and units are defined in Table A1.
Table 4a. Integration of the hydrological budget for the Control, Calibrated, and Tuned SiB2 runs. The symbols and units are defined in Table A1. All values are in mm. Both runs were for 142 continuous days: 28 May–16 Oct 1987. ΔM is always zero, because total surface standing water was unchanged over the run. Soil gain = (ppt − surface losses) = P − (Eci + Ro1 + ΔM). Net soil = (soil gain − soil losses) = PW1 − (Ect + Egs + Q3). Eci is negative when there was more condensation to than evaporation from canopy interception stores. Interlayer flux is negative when net flow was upward. This is not a term in the overall water budget calculations.
Table 4b. Integration of CO2 fluxes and water use efficiency for the Control, Calibrated, and Tuned SiB2 runs. Symbols and units are defined in Table A1. All CO2 flux values are in gC m−2 . Water use efficiency (WUE) units are [(mmol CO2) (mol H2O)−1]. All runs were for 142 continuous days; 28 May–16 Oct 1987. Dashes mean not applicable. The mathematical relationships between the symbols are defined (A1–A3).
Simulation versus observation statistics for Control, Calibrated Control, and Tuned SiB2 runs. The symbols and units are defined in Table A1. Each of the runs was for 142 continuous days: 28 May–16 October 1987. NSEE is given as a percent.
Table A1. Notation.
Table A1. (Continued)
Table A1. (Continued)
Table A1. (Continued)
