Seasonal and interannual variability in 13C composition of ecosystem carbon fluxes in the U.S. Southern Great Plains

The δ13C value of terrestrial CO2 fluxes (δbio) provides important information for inverse models of CO2 sources and sinks as well as for studies of vegetation physiology, C3 and C4 vegetation fluxes, and ecosystem carbon residence times. From 2002.2009, we measured atmospheric CO2 concentration and δ13C-CO2 at four heights (2 to 60 m) in the U.S. Southern Great Plains (SGP) and computed δbio weekly. This region has a fine-scale mix of crops (primarily C3 winter wheat) and C4 pasture grasses. δbio had a large and consistent seasonal cycle of 6.8‰. Ensemble monthly mean δbio ranged from -25.8 ± 0.4 ‰ ( ± SE) in March to -20.1 ± 0.4‰ in July. Thus, C3 vegetation contributed about 80% of ecosystem fluxes in winter-spring and 50% in summer-fall. In contrast, prairie-soil δ13C values were about -15‰, indicating that historically the region was dominated by C4 vegetation and had more positive δbio values. Based on a land-surface model, isofluxes (δbio ± NEE) in this region have large seasonal amplitude because δbio and net ecosystem exchange (NEE) covary. Interannual variability in isoflux was driven by variability in NEE. The large seasonal amplitude in δbio and isoflux imply that carbon inverse analyses require accurate estimates of land cover and temporally resolved 13CO2 and CO2 fluxes.


Abstract
The δ 13 C signature of terrestrial carbon fluxes (δ bio ) provides an important constraint for inverse models of CO 2 sources and sinks, insight into vegetation physiology, C 3 and C 4 vegetation productivity, and ecosystem carbon residence times. From 2002-2009, we measured atmospheric CO 2 concentration and δ 13 C-CO 2 at four heights (2 to 60 m) in the U.S. Southern Great Plains (SGP) and computed δ bio weekly. This region has a fine-scale mix of crops (primarily C 3 winter wheat) and C 4 pasture grasses. δ bio had a large and consistent seasonal cycle of 6-8‰. Ensemble monthly mean δ bio ranged from -25.8±0.4‰ (±SE) in March to -20.1±0.4‰ in July. Thus, C 3 vegetation contributed about 80% of ecosystem fluxes in winter-spring and 50% in summer-fall. In contrast, prairie-soil δ 13 C values were about -15‰, indicating that historically the region was dominated by C 4 vegetation and had more positive δ bio values. Based on a land-surface model, isofluxes (δ bio × NEE) in this region have large seasonal amplitude because δ bio and net ecosystem exchange (NEE) covary. Interannual variability in isoflux was driven by variability in NEE. The large seasonal amplitude in δ bio and isoflux imply that carbon inverse analyses require accurate estimates of land cover and temporally resolved 13 CO 2 and CO 2 fluxes.

Introduction
Atmospheric δ 13 C-CO 2 values (δ atm ) provide insight into ecosystem carbon fluxes and the plant physiological processes that drive them. For example, δ atm is used by inverse and forward models to quantify oceanic, terrestrial, and fossil carbon sources and sinks (Ciais et al. 1995;Fung et al. 1997;Rayner et al. 1999Rayner et al. , 2008. Because air-sea gas exchanges do not strongly affect δ atm whereas terrestrial CO 2 exchanges do, δ atm has been used to distinguish land versus ocean carbon fluxes (Keeling et al. 1989;Tans et al. 1993;Enting et al. 1995;Francey et al. 1995;Battle et al. 2000) and to attribute interannual variability in the global carbon sinks (Keeling et al. 1995;Francey et al. 1995;Langenfelds et al. 2002;Le Quéré et al. 2003;Randerson et al. 2002). These applications require prior information about the magnitude and isotopic signature of the ecosystem carbon exchanges (δ bio ; Bakwin et al. 1998) that force these atmospheric changes. Likewise, coal, petroleum products, and natural gas have different δ 13 C signatures, nearly all distinct from terrestrial ecosystem exchange. Finally, δ atm and δ bio are also useful for predicting biosphere responses to CO 2 fertilization (e.g., Randerson et al. 1999).
The δ 13 C signature of land surface-atmospheric carbon exchange (photosynthetic uptake and respiration) is also useful for distinguishing the proportional contribution to total flux by plants using different photosynthetic pathways: C 4 plants discriminate less than C 3 plants against 13 CO 2 during photosynthesis, as described below. Being able to distinguish C 3 and C 4 exchange is useful for studying C 3 and C 4 plant phenology, their different responses to changes in CO 2 and climate, and their competitive coexistence (Niu et al. 2008), as well as the ability to correctly simulate these in carbon cycle models is critical to the de-convolution problem.
Currently, the application of isotopic tracers at ecosystem and global scales is limited by insufficient measurements of 13 CO 2 fluxes (isofluxes) and limited confidence in models that predict them. In fact, a number of studies have concluded that the largest uncertainties associated with land:ocean partitioning are in modeling spatial and temporal variations in terrestrial 13 C fractionation and isotopic disequilibria between gross primary productivity and respiration (Heimann and Meier-Raimer 1996;Rayner et al. 1999Rayner et al. , 2008Townsend et al. 2002;Randerson et al. 2005;Scholze et al. 2008). Global inversion models are very sensitive to the magnitude and isotopic signature of photosynthesis and respiration. For example, a 3‰ difference in photosynthetic discrimination globally can yield a 0.7 Pg y -1 difference in the estimated terrestrial carbon sink and a 0.2‰ underestimate in the GPP-ecosystem respiration disequilibrium can yield about a 0.5 Pg y -1 difference (Ciais et al. 1995(Ciais et al. , 1999Fung et al. 1997).
One of the largest effects on land-surface isotopic fluxes is the difference between C 3 and C 4 plant types. Because of biochemical and anatomical differences, C 3 and C 4 plants fractionate against 13 CO 2 to different degrees during photosynthesis (Farquhar et al. 1989). Typically,C 4 and C 3 photosynthetic assimilation have isotope fractionation of 4 and 20‰, respectively, and tissue of these plant types has δ 13 C values of roughly -12‰ and -28‰. As a result, changes in land cover, such as conversion of tropical forests to pastures or sugar cane, can have large effects on δ atm and therefore its interpretation in atmospheric carbon budgets (Ciais et al. 1999;Townsend et al. 2002;Still et al. 2003b). Considering the isotopic disequilibria caused by land use changes to C 4 pastures and crops or interannual C 3 /C 4 variations driven by climate, Scholze et al. (2008) concluded that not accounting for C 4 disequilibria in a 13 C deconvolution calculation can alter the land:ocean partitioning by 1 Pg C yr -1 . Better understanding of spatial and temporal variations in ecosystem C 3 /C 4 fractions is therefore critical to accurately assess variation in terrestrial 13 C fractionation as well as the resulting isotope disequilibria for 3-D inversion studies employing both CO 2 and 13 CO 2 (Still et al. 2003b).
Grasslands with a mixture of C 3 and C 4 species are found in temperate North America, South America, Asia, and Australia (Collatz et al. 1998). Historically, the dominant vegetation cover of the U.S. Great Plains was shortgrass and tallgrass prairie that contained a mixture of C 4 grasses and C 3 grasses and forbs, and riparian forests (McNab and Avers 1994). Much of the native prairie has been converted to cropland, grazed grasslands, and pastures, with the latter two subject to fairly frequent burning (Knapp et al. 1998). In mixed grasslands, C 3 grasses generally grow in spring and early summer, while C 4 grasses dominate during mid-and late summer (Kemp and Williams 1980;Freeman 1998;Knapp et al. 1998). This seasonal partitioning results from differences in the biochemistry of each photosynthetic pathway (Ehleringer et al. 1997;Collatz et al. 1998) and their largely temperate (C 3 ) and tropical (C 4 ) origins (Edwards and Still 2008). The C 4 fraction of prairie grasses increases from the northwest to the southeast of the Southern Great Plains, similar to the trend across the Great Plains as a whole (Epstein et al. 1997;Tieszen et al. 1997;Knapp et al. 1998).
The Southern Great Plains is an excellent location for studying the δ 13 C value of terrestrial carbon fluxes (δ bio ) because there is comprehensive meteorological and carbon cycle infrastructure, simple topography, and fairly consistent wind direction. Furthermore, land cover is similar to other temperate mixed agricultural and C 3 /C 4 grasslands and the region is located in the domain of many forward and inverse model studies of CO 2 sources and sinks.
The objectives of this study were to characterize the δ 13 C values of ecosystem respiration and net ecosystem exchange from a multi-field area of the SGP over eight years and use this record to investigate the (1) seasonal and interannual variation in 13 C signature of these fluxes; (2) relative contribution of C 3 versus C 4 productivity to CO 2 fluxes observed at a centrallylocated 60 m tower; and (3) seasonal and interannual variation in sub-regional 13 C isofluxes. We generated weekly estimates of δ bio at the SGP 60 m tower for 8 years, by collecting 16 flasks per week from four tower heights and calculating the δ 13 C value of nighttime respiratory and daytime net ecosystem (NEE) carbon fluxes using the Keeling plot approach (Keeling 1959 and1961). To estimate isofluxes, we combined output from a spatially explicit, isotope-enabled ecosystem model (ISOLSM; Riley et al. 2002Riley et al. , 2009Aranibar et al. 2006) with a simple footprint estimation.

Site Description
Our study region is the U.S. Southern Great Plains (SGP), centered in north-central Oklahoma ( Figure 1) and extending to southern Kansas. The climate is continental to semi-arid ( Figure 2). Mean annual precipitation in the region ranges from 1270 mm in the southeast to 380 mm in the NW. Oklahoma state mean annual temperature is about 15ºC, but mean daily temperatures frequently exceed 20 ºC during summer months. There is strong horizontal advection, predominantly from the south except during some spring periods ( Figure 3).
Land cover in SGP is dominated by pasture, grazing lands, and annual crops, with winter wheat covering more than 40% of cropland ( Figure 1). Other crops, such as sorghum (C 4 ), soybean (C 3 ), oats (C 3 ), barley (C 3 ), and corn (C 4 ) are commonly grown in summer. The typical field size is a quarter section, or 64.7 ha. Because land holdings are small and the crops are annuals, land cover has fine scale spatial heterogeneity and individual fields can change cover type from year to year. The U.S. Department of Energy Atmospheric Radiation Management Program (ARM) Climate Research Facility (ACRF) supports a large testbed, roughly 300×300 km, for measurements and modeling in the Southern Great Plains (Ackerman et al. 2004). All atmospheric and climatic variables measured in the ACRF are available from the ARM Data Archives (www.archive.arm.gov). The Oklahoma and Kansas Mesonet networks of meteorological stations also contribute data to our analysis. The main site for our measurements is the ACRF Central Facility (36.61°N, 97.49°W), located in the center of the testbed in the midst of farmland in Lamont, Oklahoma.

Atmospheric gas sampling for CO 2 and 13 CO 2
Atmospheric samples were collected from a 60 m tower at the Central Facility.
Instruments and flask samplers were installed in a climate-controlled shed at the base of the tower. Air was pulled from inlets mounted on the tower at 2, 4, 25, and 60 m through 3/8" Dekaron tubing to the shed. The stainless steel inlet assemblies contain a quartz filter and fine metal frit. In the shed, air was dried by passage through a condensing unit and collected in glass flasks.
We used an automated system for trapping the air into glass flasks. Sampling tubes from each of the four heights on the tower led to a 4-port manifold with automated valves. A datalogger controlled which sampling height was routed to the flask box. Incoming sample air was pumped at more than 500 ml min -1 through a multi-port valve (Valco ST configuration E 16 position valve) to 100 ml glass flasks (Kontes Custom Glass Shop). Before sampling, the flasks were filled with dry air from a cylinder, closed, and shipped to the Central Facility. On the collection day, flasks were placed in the 16-flask unit and opened. They were sealed from outside air by the multi-port valve. To collect a sample, the valve position changed to allow air from the tower inlet to flow through the flask. After 3 minutes, the automated valve changed position, sealing the flasks. At the end of the diel cycle, the flask-valves were closed manually and the flasks were shipped to California for analysis.
Flask samples were analyzed at the Carnegie Institution of Washington (Stanford, California) with an integrated system that measures the CO 2 concentration and isotope ratios (δ 13 C and δ 18 O) of small air samples, as described by Ribas-Carbo et al. (2002). Five repeated measurements on the air in each flask yields high-precision data, with standard errors of the measurement around 0.5 ppm for CO 2 and 0.03‰ for δ 13 C for each data point. Gases from NOAA-ESRL are used for calibration and span of the instrument. All values are reported using the delta (δ) notation with per mil (‰) variations relative to VPDB (Vienna Pee Dee Belemnite):

Flask sampling protocol
To estimate δ bio , we collected 16 flask samples over 1-2 days every week, with the heights and timing of sampling following either a "nighttime" or "diel" protocol. We alternated protocols. Every other week, we collected a flask sample at each of the four heights (2, 4, 25, 60 m) at each of four time points in the dark (22:00, 00:00, 2:00, and 4:00 local time). This is referred to as the "nighttime" protocol. On the alternating weeks, we collected a flask sample at two heights (2, 60 m) at each of 8 time points, starting in the late afternoon and continuing to mid-afternoon of the next day (18:00, 21:00, 00:00, 3:00, 6:00, 9:00, 12:00, and 15:00 local time). This is referred to as the "diel" protocol. We split the data from the diel collection, using sunrise and sunset as boundaries, to look at carbon exchange when it was dark (21:00-6:00) and light (9:00-15:00); the 18:00 flasks were used for the diel-dark keeling plots when sunset occurred before 18:00 local time (for a total of 8-10 flasks for dark and 6-8 for daylight). These δ bio values are referred to as diel-dark and diel-light, respectively.

NOAA flasks
We collected two flasks at 60 m at roughly 2 pm local time each week on the day that the Keeling plot collection began. The samples were collected using a flask package built by NOAA and the samples were analyzed by NOAA-ESRL. These data are used in the ESRL GLOBALVIEW product and were used in this study to provide validation for the atmospheric δ atm data used in the Keeling plot mixing model. We also collected flasks in the free troposphere by aircraft each week that were analyzed by NOAA.

Keeling plot analysis
To estimate the 13 C isotopic signature of ecosystem CO 2 exchange, some studies have used the mass-weighted-average plant biomass 13 C as a proxy, but it is preferable to measure the isotopic signature of the CO 2 flux directly because this gives the flux-weighted physiological activity of C 3 and C 4 plants rather than simply their relative biomass abundance (e.g., Tieszen et al. 1997;Dawson et al. 2002). One established method for estimating δ bio is a mixing model called the 'Keeling plot', based on measurements of atmospheric CO 2 mixing ratios and 13 C values (see Methods; Keeling 1959 and1961).
We estimated δ bio as the intercept of a 'Keeling plot' in which δ 13 C values were regressed against the reciprocal of CO 2 concentration (i.e., [CO 2 ] -1 ) for a series of flask measurements (Keeling 1958(Keeling , 1961Flanagan and Ehleringer 1998;Yakir and Sternberg 2000;Pataki et al. 2003); see Figure S1 in online supporting information.
We used a linear regression model (type I) to calculate intercept values. Some studies have advocated a model II, or geometric mean, regression to incorporate errors in both the concentration and isotope values (Friedli et al. 1987;Flanagan et al. 1996;Zar 1996;Sokal and Rohlf 1995;Bowling et al. 1999;Harwood et al. 1999;Pataki et al. 2003). It has been shown, however, that there is almost no difference between these two regression models at high correlation coefficient values such as those in the Keeling plots reported in this study (Zobitz et al. 2006) and it is more straightforward to estimate errors with a type I model (Sokal and Rohlf 1995;Laws 1997;Zar 1996).
We applied quality standards to our regression intercept values by using only those collections for which the intercept standard error (SE) for the regression of δ 13 C on 1/CO 2 was less than 1.5‰. We evaluated the potential bias introduced by using standard error (SE), R (correlation coefficient), and CO 2 mixing ratio ranges as cutoff criteria and found no discernable bias in intercept values created by the SE cutoff. Keeling plots were analyzed with fewer than 16 flasks if some flasks did not pass laboratory analytical checks.

Fractional contribution of C 3 and C 4 plant types to ecosystem carbon exchange
To calculate the fractional contribution of C 3 and C 4 plants to ecosystem carbon exchanges (f C3 and f C4 , respectively), we used a two-member mixing model based on the assumption that NEE variations in these plant types drives the observed variations in δ bio as: where δ is the δ 13 C value, and subscripts indicate the measured Keeling plot result (δ bio ) or the C 3 and C 4 contributions. As a two member mixing model, we set f C3 + f C4 = 1. Substituting f C3 = 1-f C4 and rearranging gives: Given measurements of CO 2 and δ atm in the nocturnal boundary layer and 13 C end members for C 3 and C 4 plant types, Keeling plot analysis has been used to infer the C 4 respiration fraction at ecosystem to regional scales (Miranda et al. 1997;Still et al. 2003a;Lai et al. 2003;). The partitioning relies on the large and fairly consistent δ 13 C offset between C 3 and C 4 plants. The offset can be affected by variable C 3 isotope fractionation resulting from moisture stress, with much less variation in C 4 grass isotopic composition (Farquhar et al. 1989;Henderson et al. 1992;Mole et al. 1994;Buchmann et al. 1996;Still et al. 2003a).

Isotopic end members for C 3 and C 4 plant types
We estimated the isotopic end members for each plant type from leaf δ 13 C values published by Still et al. (2003a) and of leaf samples that we collected. These are a proxy for average δ 13 C values of leaf photosynthetic uptake and whole plant respiration. More accurate estimates would be generated by direct measurements of these fluxes, for C 3 and C 4 plant types, over the region, for each time point. We lack such measurements but note that they would do little to advance our main goal which is to define the influence of this region on the atmosphere and to test our ability to simulate this forcing.
We Soils were sampled to 1 m depth in two fields of lightly grazed prairie at GRL, using a Giddings brand hydraulic corer (n=10 per field; 7 depths per core). One field was subject to a controlled burn in March 2005 and the other had not been burned for more than 10 years.
Collection dates were March 2005 (right after the burn) and March 2007. The soil core was split length-wise. One half was sieved at 2 mm and roots were removed by hand picking for five minutes. The sieved soil was dried at 50ºC and ground to produce a homogenized bulk sample.
Organic carbon content and δ 13 C were determined on soil and leaf samples with an elemental analyzer connected to an isotope ratio mass spectrometer at the University of suggests these are representative average C 3 and C 4 leaf values for this region.

Isoflux
Isoflux (I, µmol m -2 s -1 ‰) is defined as the CO 2 flux multiplied by its δ 13 C value (Bowling et al. 2008;Riley et al. 2003). As a result of the sign convention for fluxes, in which negative flux is a transfer of carbon from atmosphere to ecosystem, a positive isoflux implies transfer of 13 C-depleted CO 2 from the atmosphere, thereby enriching δ atm .
We estimated weekly average NEE and ecosystem respiration with a spatially-explicit, isotope-enabled land-surface model (ISOLSM; Riley et al. 2009) that has been calibrated and tested in this region against eddy flux tower observations (Billesbach et al. 2004;Fischer et al. 2008) and is driven by Mesonet and ARM Extended Facilities climate observations and MODISderived land cover and LAI. For the simulations used here, we have imposed a time varying C 3 /C 4 fraction based on the work by Still et al. (2003a); we note that the partitioning in that work is consistent with that calculated in the current study. Using methods described in Aranibar et al. (2006) and based on those described by Farquhar et al. (1989), ISOLSM predicts photosynthetic discrimination and respiratory 13 C fluxes as a function of plant type, phenology, water status, and atmospheric conditions.
The predicted NEE and δ bio used here are weekly average values in the upwind direction of the tower. We did not estimate a proper footprint for the Keeling plot intercept. Rather, we used the land surface of SGP in the weekly-average wind direction (denoted here as α). If α was more than 5° E or W of South, we used the land surface of the SW or SE quadrant of SGP, respectively. If α was less than 5º off south, we used the southern half of SGP as the source area.
NEE and δ bio were modeled at 10 km resolution in the selected area. The use of weekly average NEE implicitly treats the weekly δ 13 C value of photosynthetic uptake and respiration as the same. This appears to be a reasonable simplification for SGP, since the difference between δ bio for light and dark periods was small, especially compared to the seasonal signal (see Results section). We predicted isoflux by multiplying ISOLSM predictions of δ bio and NEE.

Atmospheric CO 2 and 13 CO 2
Atmospheric CO 2 mixing ratio and δ atm in the Southern Great Plains display a distinct δ atm values also have a large diurnal range because δ bio of respired CO 2 is much more negative than the δ value of the background atmosphere. The stable nocturnal boundary layer acts somewhat like a chamber and the buildup of CO 2 changes δ atm on a given night by as much as 3‰. Including day and nighttime data, the range in CO 2 concentration within any year was 75 ppm and in δ atm was more than 3.5‰ ( Figure 4).

δ 13 C values of ecosystem carbon exchanges (δ bio )
We used the diurnal cycle in CO 2 and δ atm and the boundary layer's chamber-like effect to analyze the flasks collected between 2 and 60 m with a Keeling plot regression analysis ( Figure S1). The flask-collection protocols sampled a vertical and temporal CO 2 gradient created by ecosystem respiration into a stable nocturnal boundary layer ("nighttime" and "diel-dark" protocols) or by net photosynthetic drawdown during the day ("diel-light" protocol) ( Keeling plots with a CO 2 span below 5 ppm) and were excluded from this analysis, leaving 231 weekly δ bio signatures in the time series reported here. A comparison among SE, R-values, and CO 2 range of all Keeling plots is presented in Figure S2. Large temporal and vertical 13 CO 2 gradients and high isotope-measurement precision yielded high regression coefficients and low SE for flask sets with CO 2 ranges as low as 5-10 ppm.
We computed separate Keeling plot intercepts for the light and dark periods of the diel sampling. There was no difference between the diel-dark δ bio and the full diel value (Figure 5a) for all weeks when both intercepts were significant. We conclude that the nighttime respiration observations dominate the diurnal Keeling plot.
There were 33 weeks with diel-light intercepts with SE < 1.5‰, distributed across spring, summer, and fall, and in every year. Although it is more difficult to obtain significant Keeling plots in the day time, 14 weeks had SE < 0.5‰ and 26 had SE < 1‰.
There were 30 weeks for which both diel-light and diel-dark intercepts had SE<1.5‰, allowing us to compare periods influenced by photosynthetic uptake and respiration versus just respiration ( Figure 5b). Of these, 16 weeks showed no significant difference between light and dark Keeling plot intercepts, with a mean difference of only 0.03‰. (The mean of the absolute value of the differences was 0.77 ± 0.15‰.) Considering all 30 pairs, the diel-dark values were slightly more negative than the diel-light values (mean dark-light difference -0.49 ±0.27‰ and mean of absolute differences 1.27±0.16‰). There was no seasonal pattern to the dark-light differences. Unless noted, subsequent results refer to respiration δ bio , calculated from the nighttime (n=97 weeks) and diel-dark (n= 116 weeks) protocols.
Weekly respiration δ bio measured at the tower ranged between -16‰ (summer) to -32‰ (winter), with only 3 values were below -30‰. There was a consistent seasonal cycle, with a minimum in winter and a maximum in summer and a seasonal amplitude of about 8‰ ( Figure   6). The ensemble monthly means (n=5-8 months) ranged from a high of -20.1±0.4‰ (±SE) in July to a minimum of -25.8±0.4‰ in March (Table 2). There was no apparent secular annual trend in the peak, minimum, or timing of the seasonal cycles ( Figure 6). The values generated by the nighttime and diel-dark protocols, although slightly different, had the same seasonal patterns and magnitudes ( Figure 6; see Figure S3 for a plot of nighttime vs. diel δ bio ).
To look for anomalies in weekly δ bio , we computed the difference between monthly average Keeling plot values for each month and the ensemble mean for that month (Table 2), and compared these to means and anomalies in monthly average precipitation, temperature, and vapor pressure deficit. The study period had significant inter-annual variability in monthly rainfall but overall was slightly drier than the climatological average (  Table 2), we found no correlation between monthly precipitation anomalies and δ bio anomalies, but considering the summer months only (July, August, September) there was a positive correlation with a small but significant slope.
Specifically, higher rainfall in summer months was correlated with more positive δ bio values. For July and September, precipitation anomalies explained >70% of the variance in monthly δ bio anomalies, with a slope of 0.03‰ per mm additional rain (p < 0.04; n = 6 for each month except 2004 and 2009). Precipitation anomalies in July and September ranged from +40 mm to -63 mm, associated, according to this slope, with anomalies in the mean monthly δ bio of +1.2‰ and -1.9‰, respectively. (The regression for August months was not significant and had a different slope, resulting in a summer (July-August-September) regression with R 2 = 0.2, slope = 0.01 ‰ mm -1 p < 0.08; n = 19.)

Fractional contribution of C 3 and C 4 plant types to ecosystem carbon exchange
δ bio can be used to estimate the relative contribution of C 3 versus C 4 fluxes in the measurement footprint (Still et al. 2003a;Lai et al. 2003). Using the two-member mixing model and leaf-tissue end members given in the Methods and Table 1, we estimated that ecosystem carbon exchange was 85% C 3 in March to 50% C 3 in summer (Table 2; Figure S4). The fraction of total ecosystem exchange ascribed to C 3 plants peaked in winter because there was almost no C 4 uptake or respiration. In absolute terms, C 3 uptake fluxes peaked in April and May, according to eddy flux data and land cover maps (Billesbach et al. 2004;Fischer et al. 2008;Riley et al. 2009). Note that the influence of ecosystem fluxes on atmospheric δ 13 C depends both on the isotopic ratio and magnitude of NEE.
The uncertainty in C 4 /C 3 ratio due to uncertainty in the intercept, propagated as in Phillips and Gregg (2001), was less than two percentage points. For end members, the variation in leaf δ 13 C values, among species, sites, seasons, and years that we sampled, was less than 0.2‰. A sensitivity analysis using this range of uncertainty in end members gives an uncertainty in f C3 and f C4 of 1.4%. Thus, the uncertainty in using leaf tissues to represent leaf fluxes (Bowling et al. 2008;Cernusak et al. 2009) is much smaller than the magnitude of temporal patterns observed.

Isofluxes
To investigate the influence of our observed patterns in δ bio on the atmosphere, we used ISOLSM to predict ecosystem carbon exchange and calculate isofluxes for 2006-2008. The NEE predicted for the SGP area upwind of the tower is shown in Figure 7a. ISOLSM predicted net CO 2 uptake (regional average uptake > 2 µmol C m -2 s -1 ) from March to October. The variability in NEE reflects mainly seasonal phenology and climate, but also wind direction. The latter is not a dominant effect, as the NEE in Figure 7a is almost the same as the whole-region average NEE for the same time (Riley et al. 2009). Using the ISOLSM 13 C routines and the same upwind subregions, we predicted δ bio at the Central Facility tower. These predictions match the observed values for δ bio within 1-2‰ most of the time, both in magnitude and in weekly and seasonal variability, but there were some weeks with larger differences (Figure 7b). Figure 7c shows 2006-2008 isoflux calculated as the product of predicted NEE and predicted δ bio .
Four patterns are evident in the isoflux results. First, the timing and broad seasonal pattern of isoflux mirrors that of observed δ bio , with late-winter and mid-summer extremes.
Second, the relative amplitude of variation is much larger for isofluxes than for δ bio due to covariance in δ bio and NEE seasonality. Peak uptake occurs in April and May, when the winter wheat is at peak productivity and growth of C 3 grassland species has begun. In other words, the most negative NEE coincides with relatively negative δ bio , and thus the isoflux also peaks in this period.
Third, inter-annual variability in predicted isoflux was large and the dominant cause was variation in NEE rather than in δ bio (Figure 7a and 7b). Specifically, 2006 had much lower peak and cumulative isoflux than did 2007 or 2008. The cumulative annual isoflux upwind of the tower was 0.6, 1.4, and 1.8 MgC ‰ m -2 y -1 in , 2007respectively. Similarly, 2008 had the largest summertime isoflux of the three years, even though it had almost the same δ bio as in 2007, highlighting the importance of the inter-annual variability in NEE. Finally, in all three years, cumulative modeled NEE was negative (a net carbon sink) leading to a positive isoflux. In other words, because the land surface in this region is predicted to be a net carbon sink (due at least in part to export of agricultural products out of the region, and therefore not being returned to the soil), the annual isoflux caused a net enrichment of δ atm in this region.

Seasonal and inter-annual patterns in δ atm and δ bio
While atmospheric δ atm values in SGP followed the global seasonal and annual patterns for this latitude, the δ bio in SGP had much larger seasonality and was strongly influenced by local land use and land cover change. The measured weekly δ bio varied by 8‰ seasonally, from spring minimum to summer maximum, with little interannual variation in this cycle across the 8-year record.
Keeling plot intercepts generated by nighttime and daylight sampling are affected by different processes. The nighttime intercept captures only ecosystem respiration, and may reflect small changes in footprint throughout the night. The diel intercept is influenced by a growing boundary layer, isotopic composition of both photosynthesis and heterotrophic respiration, and changing footprints (Pataki et al. 2003;Griffis et al. 2007;Shimoda et al. 2009). Nevertheless, when comparing light and dark plots from the same week, or diel and nighttime plots from consecutive weeks, we observed similar δ bio values ( Figure 6; Figure 5). Although the larger range of CO 2 values and unidirectionality of nighttime CO 2 fluxes tended to yield lower SE values, about one third of the day time samplings at SGP resulted in good Keeling plot intercepts.
We can explore potential isotopic disequilibrium between photosynthesis and respiration by comparing the diel-light and diel-dark Keeling intercepts from the same 24-h diel sampling.
The diel-dark values were slightly more negative on average (-0.49 ±0.27‰, n=30) than the paired diel-light values. As noted above, the difference could be due to isotopic disequilibrium as well as atmospheric mixing and transport. The fact that half of the light-dark comparisons, distributed in spring, summer, and fall, showed no difference (-0.03 ±0.25‰, n=16 of 30) suggests there was no difference between the respiratory and NEE δ bio values during those times; the alternative hypothesis is that a complex coincidence of atmospheric conditions and landsurface exchanges compensated for a real difference. In any case, the day-night differences in any given diurnal cycle were much lower than the seasonal variations associated with shifts in C 3 -C 4 fractions.
There are a few properties of SGP that could create large disequilibrium between the isotopic signatures of heterotrophic respiration and net photosynthetic fluxes. First, C 3 and C 4 plants tend to have different phenology, whether growing together in grasslands or in different fields. If ecosystem respiration lags productivity on order of months, there would be significant C 3 carbon respired during peak C 4 productivity, or vice-versa. Second, farmers may plant a C 3 crop in fields that had C 4 plants the previous year, or vice-versa. We could deduce disequilibrium between respiration and photosynthesis if we observed differences in δ bio between the light and dark segments of each diel sampling or from a relatively positive δ bio in months when there was no active C 4 growth, i.e., October-April. Since we did not observe a large day-night difference, and did not see more than 15% C 4 fraction in the winter, our observations suggest that in this region most respiration is of recently photosynthesized material.

Fractional contribution of C 3 and C 4 plant types to ecosystem carbon exchange
Using a two-member mixing model with leaf-δ 13 C end members for C 3 and C 4 vegetation (Table 1), we estimated that ecosystem respiration and NEE in the study region varied from 80% C 3 in winter-spring to 50% C 3 in summer-fall ( Table 2). The spring shift from C 3 dominance to balanced C 3 -C 4 carbon exchange coincided with the timing of wheat senescence and harvest. We observed fairly consistent seasonal cycles in the C 3 :C 4 ratio of ecosystem carbon fluxes despite large variation in weather conditions and model estimates of regional NEE. There was only a weak positive correlation between anomalies in precipitation and δ bio in the summer months.
The observed seasonal cycle of δ bio was too large to explain by water stress, phenology, or disequilibrium between photosynthesis and respiration alone. Rather, the large seasonal changes reflect changes in the relative contribution of C 3 -versus C 4 -derived fluxes to the air sampled at the Central Facility. We cannot say, however, that the observed changes in δ bio reflect a change in C 3 and C 4 exchange from a particular set of fields (i.e., a static footprint). It is possible that the footprint influencing the tower air changes over time in such a way that it samples regions of different C 3 and C 4 activity. Both explanations are plausible at this site. For the static footprint case, previous studies have shown that the phenology of the dominant C 3 and C 4 crops and grasses in the study region follow the δ bio pattern we observed (Fischer et al. 2008;Billesbach et al. 2004;Suyker et al. 2003;Still et al. 2003b). For the second case, the SGP quadrant southeast of the tower has more C 4 productivity compared to southwest of the tower, and the prevailing wind typically shifts slightly from southwest to southeast in late spring, coincident with the observed increase in δ bio . Both phenomena-seasonal shifts in C 3 and C 4 productivity and seasonal shifts in wind direction-may be convolved in the patterns reported here.
Prior studies of nighttime δ bio in prairie and pastures report large seasonal cycles in δ bio due to changing proportions of C 4 and C 3 activity (Still et al. 2003a;Lai et al. 2003Lai et al. , 2006. Our summer results are consistent with the δ bio values and seasonal pattern of previous studies. But the annual cycle we report has important differences, notably a larger seasonal δ bio amplitude, because we observed a larger region that included wheat and other C 3 crops in addition to grasslands. Our sampling footprint was larger than that of these earlier studies, which were conducted using shorter towers (<5 m) and in some cases shorter duration for Keeling plot sampling.
Those prior studies also found that higher-than-average spring precipitation increased the C 3 contribution to spring respiration (Still et al. 2003a) and less spring rainfall corresponded to reduced C 4 contribution to summer ecosystem respiration via impacts of spring soil water on the C 3 -to-C 4 transition (Lai et al. 2003(Lai et al. , 2006. The positive effect of precipitation on summer δ bio within mixed C 3 -C 4 grasslands arises because C 4 grasses can outcompete C 3 vegetation during the warmer months. We saw evidence of a weak, positive precipitation response during July and September. We did not see as strong a precipitation response in C 3 :C 4 ratio because multiple fields influenced our observations, including summer C 3 monocultures that are also stimulated by higher rainfall.
One caveat to our use of a two-member mixing model to estimate C 3 and C 4 fractions is that fossil emissions and seasonal burning are sources of CO 2 in SGP. Three winter Keeling plotintercepts gave δ bio less than -30‰, indicating that there was a source of very depleted CO 2 , such as fossil fuel-derived CO 2 , that was observable when plant and soil fluxes were very small-and thus should be included as a third end member in the mixing model. However, for most of the year ecosystem fluxes are much larger than fossil CO 2 emissions, and therefore likely to have a much greater influence on atmospheric CO 2 concentrations and δ atm .

Historic δ bio
The observed pattern of C 3 -dominated fluxes is very different from that of the native vegetation in this region. We have measured the δ 13 C of soil organic carbon (SOC) in native prairie and pasture as well as in fields that were converted from prairie to winter wheat many years ago. The native prairie soil at GRL had δ 13 C values of -14.9‰ (SE=0.2, n=48). This indicates a historical ecosystem with about 85% C 4 net primary productivity. In contrast, the surface soil carbon under a field at the Central Facility that had been cultivated with winter wheat for at least the prior decade had δ 13 C values around -22‰ in 2004. Thus, as a result of land cover conversion, both the phenology of plant uptake and its isotopic signature have changed significantly in the Southern Great Plains. In addition, during the period when C 4derived SOC was being replaced with C 3 -derived SOC under long-term wheat cultivation, the region must have been a source of 13 C to the atmosphere, because soil carbon in the vast prairie areas converted to C 3 crops was becoming lighter, perhaps 7-12‰ lighter.
Observations of δ bio are useful for inverse models that utilize δ atm measurements as a constraint (e.g., Rayner et al. 2008). For example, parameterizing the region with the historic vegetation would result in large errors in estimated δ bio . Likewise, assuming a constant yearround δ bio value, be it C 4 or C 3 , would create large errors. For inversions and other purposes, such agricultural regions cannot be simply categorized as C 3 or C 4 , or represented by historical vegetation type (Lai et al. 2003;Shimodo et al. 2009).

Isotopic Disequilibria
While SGP does not appear to have a large 13 C disequilibrium on diurnal timescales, it has very large seasonal variation, with a δ 13 C amplitude of ~8‰. This variation is due mainly to the seasonal transition from C 3 to C 4 photosynthesis, and the fact that most respired CO 2 was from recently photosynthesized carbon. On annual timescales, we see a net transfer of 13 C between atmosphere and ecosystem because the landscape is a net carbon sink, primarily due to the export of NPP in plant and animal products. Finally, historically this region shows a decadal to centennial disequilibrium due to historical land cover change from C 4 to C 3 . The large decrease in δ bio due to land cover change increases the 13 C signal between ocean uptake and land uptake in this region.

Isofluxes
The impact of δ bio on atmospheric 13 C/ 12 C composition depends on the flux rate and the flux δ 13 C values. As a result of the covariance in seasonal cycles in δ bio and predicted NEE and, the seasonal amplitude in isoflux was larger proportionally than variation in δ bio (Figure 7). In addition, there was substantial inter-annual variability in isoflux between 2006 and 2008. The seasonal cycle in δ bio did not vary appreciably from year to year. Rather, interannual variability in isoflux in this region mainly results from interannual variability in NEE, which is sensitive to rainfall, other climate conditions, and land use.
Because the land surface in this region is typically a net carbon sink, the annual isoflux causes a net enrichment of atmospheric 13 CO 2 in this region. However, this effect is diminished by the significant contribution to NEE by C 4 plants in this region. For example, the isotopic enrichment in the boundary layer would be roughly 0.016‰ less for each ppm of CO 2 drawdown in mid-summer than it would be if the ecosystem was populated by only C 3 vegetation. This difference may seem small, but given the large CO 2 draw-downs that occur over the midcontinent in summer (Desai et al. 2005), the resulting downwind isotopic signal would be clearly measurable by current technology, and such measurements could assist in identifying the source of the observed CO 2 signal. This difference is also critical for understanding the cumulative change in the concentration and isotopic compostion of CO 2 in the atmosphere over annual and interannual time frames.

Research Needed: Footprints
Interpretation of δ bio values is limited because there is no published theory for estimating the precise footprint of tower-based observations of δ bio . Some of the change in δ 13 C between flasks in a single collection period may come from changes in background air advection (i.e., footprint) rather than in ecosystem fluxes. While there are fairly established protocols for estimating the footprint influence function for atmospheric concentration measurements and for fluxes estimated from high frequency observations by eddy covariance analysis, it is more difficult to determine the footprint for a Keeling plot intercept. The measurements in the Keeling plot regression may each have a different footprint because they were collected at different heights and/or times (Griffis et al. 2007). Studies using low towers (< 5 m above the ground) have assumed that the footprint is field scale (100's of m) (Pataki et al. 2003;Still et al. 2003a), but the use of different sampling heights can complicate interpretation even over low-statured vegetation (Pataki et al. 2003). Analyses of δ 13 C data from tall towers, like WLEF in Park Falls, Wisconsin, have assumed a large, regional footprint, but to our knowledge these footprints have not been quantified. In one comparison, the estimated footprint for Keeling plots (based on the individual concentration footprints) was up to 20 times greater than the footprint of the colocated eddy covariance measurements (Griffis et al. 2007). More footprint quantification tools are needed to take advantage of opportunities for combining CO 2 eddy flux measurements with Keeling plot or 13 C eddy flux results.

Summary and Implications
We report here one of the longest time series available of the δ 13 CO 2 value (δ bio ) of ecosystem carbon fluxes. The data are available from the ARM archive (www.arm.gov) or the authors. In the heterogeneous, agricultural Southern Great Plains, there was a consistent seasonal cycle in δ bio , with large amplitude of almost 8‰ due to C 3 dominance in spring and C 4 dominance in summer. Because of land cover change to C 3 crops, the regional isotopic signature of plant uptake in SGP is currently much more negative (roughly -20 to -27‰) compared to historic soil carbon values (roughly -15‰). Although the seasonal cycle in δ bio was fairly constant from year to year, there were large seasonal and inter-annual variations in predicted isoflux, due to seasonal covariance of NEE and δ bio and inter-annual variation in NEE.
Carbon sources and sinks inferred by atmospheric carbon budgets can depend strongly on the assumed δ 13 C value of ecosystem CO 2 exchanges (Ciais et al. 1995(Ciais et al. , 1999Fung et al. 1997;Scholze et al. 2008). For the Southern Great Plains, assuming historic or potential vegetation would result in large errors in estimated δ bio and thus in regional to global carbon flux estimates.
Likewise, assuming a constant δ 13 C value of NEE year-round could create large errors in inferred fluxes. As a result, grasslands and savannah, as well as agricultural regions should not be categorized as purely C 3 or C 4 , or represented by historical vegetation types, because land use and land cover change on seasonal to decadal timescales can have large impacts on the isotopic fluxes between the land surface and atmosphere.    (N=42 for dark (2); N=30 for dark and light (3)). The symmetry of scatter around the dashed 1:1 line shows the lack of bias between either the whole diel vs. diel-dark periods or the diel-dark vs.

Tables
diel-light periods. Flasks were collected every three hours at 2 and 60 m starting at 18:00 local time. Most intercepts generated from a range of CO2 concentrations > 5 ppm had SE < 1.5‰. The ability to generate significant intercepts with low CO2 ranges was likely due to the large isotopic gradients at our site and high precision of the IRMS used to analyze the samples (Zobitz et al. 2006;Ribas-Carbo et al. 2002).