Interacting effects of temperature, soil moisture and plant biomass production on ecosystem respiration in a northern temperate grassland

Transcription

1 Agricultural and Forest Meteorology 130 (2005) Interacting effects of temperature, soil moisture and plant biomass production on ecosystem respiration in a northern temperate grassland Lawrence B. Flanagan *, Bruce G. Johnson Department of Biological Sciences, University of Lethbridge, 4401 University Drive, Lethbridge, Alta., Canada T1K 3M4 Received 2 December 2004 Abstract Chamber measurements of total ecosystem respiration (TER) in a native Canadian grassland ecosystem were made during two study years with different precipitation. The growing season (April September) precipitation during 2001 was less than onehalf of the 30-year mean ( ), while 2002 received almost double the normal growing season precipitation. As a consequence soil moisture remained higher in 2002 than 2001 during most of the growing season and peak aboveground biomass production (253.9 g m 2 ) in 2002 was 60% higher than in Maximum respiration rates were approximately 9 mmol m 2 s 1 in 2002 while only approximately 5 mmol m 2 s 1 in Large diurnal variation in TER, which occurred during times of peak biomass and adequate soil moisture, was primarily controlled by changes in temperature. The temperature sensitivity coefficient (Q 10 ) for ecosystem respiration was on average , and it declined in association with reductions in soil moisture. Approximately 94% of the seasonal and interannual variation in R 10 (standardized rate of respiration at 10 8C) data was explained by the interaction of changes in soil moisture and aboveground biomass, which suggested that plant aboveground biomass was good proxy for accounting for variations in both autotrophic and heterotrophic capacity for respiration. Soil moisture was the dominant environmental factor that controlled seasonal and interannual variation in TER in this grassland, when variation in temperature was held constant. We compared respiration rates measured with chambers and that determined from nighttime eddy covariance (EC) measurements. Respiration rates measured by both techniques showed very similar seasonal patterns of variation in both years. When TER was integrated over the entire growing season period, the chamber method produced slightly higher values than the EC method by approximately 4.5% and 13.6% during 2001 and 2002, respectively, much less than the estimated uncertainty for both measurement techniques. The two methods for calculating respiration had only minor effects on the seasonal-integrated estimates of net ecosystem CO 2 exchange and ecosystem gross photosynthesis. # 2005 Elsevier B.V. All rights reserved. Keywords: Ecosystem respiration; Temperature response; Soil moisture; Grassland; Eddy covariance * Corresponding author. Tel.: ; fax: address: larry.flanagan@uleth.ca (L.B. Flanagan) /$ see front matter # 2005 Elsevier B.V. All rights reserved. doi: /j.agrformet

2 238 L.B. Flanagan, B.G. Johnson / Agricultural and Forest Meteorology 130 (2005) Introduction Net ecosystem carbon exchange (NEE) is controlled by the balance between carbon uptake during photosynthesis and carbon loss during respiration. Recent studies have indicated that shifts in respiration can be the dominant control on interannual variation in the net carbon sink source status of an ecosystem (Valentini et al., 2000; Saleska et al., 2003; Griffis et al., 2004). Therefore, accurate predictions of future changes in carbon exchange between ecosystems and the atmosphere require a detailed understanding of the controls on ecosystem respiration. Ecosystem respiration is controlled by the complex interaction of environmental and biotic factors. Temperature is well known to be a dominant environmental control on respiration rates (Raich and Schlesinger, 1992; Lloyd and Taylor, 1994). There is mounting evidence that the temperature sensitivity of respiration declines with increasing temperature and decreasing soil moisture (Kirschbaum, 1995; Xu and Qi, 2001; Reichstein et al., 2002a,b; Curiel Yuste et al., 2003; Janssens and Pilegaard, 2003; Lavigne et al., 2004; Xu and Baldocchi, 2004). Ecosystem respiration rates are also positively correlated with photosynthesis rates or site productivity, illustrating some important biotic controls on respiration (Craine et al., 1999; Janssens et al., 2001). The correlation between respiration and photosynthesis rates presumably reflects the role of available carbon substrates in affecting plant and soil microbe respiration. The complexity and interaction of factors controlling respiration have hindered the development of mechanistic models similar to that for predicting leaf photosynthesis (Farquhar et al., 1980). In addition, Xu et al., (2004) have argued that our empirical understanding of the controls on respiration is limited by insufficient field measurements that span across a wide enough range of environmental and biological conditions in order to predict how respiration may shift in response to future environmental change. Eddy covariance measurements can be used to study whole ecosystem net CO 2 exchange, where nighttime measurements represent total ecosystem respiration (TER) (Baldocchi, 2003). However, nighttime eddy covariance measurements can underestimate ecosystem respiration because of lack of turbulent exchange necessary for the application of the technique and because under some conditions advection can result in movement of CO 2 horizontally away from the eddy covariance system so that it is not measured (Massman and Lee, 2002). A common method to deal with these problems is to reject eddy covariance data recorded under calm conditions, as determined by a threshold level of friction velocity. Nighttime eddy covariance data measured under turbulent conditions are then used to construct regression equations based on temperature (and sometimes soil moisture) so that ecosystem respiration can be calculated and used to replace rejected data from calm periods and to estimate daytime respiration rates (Goulden et al., 1996; Baldocchi, 2003). Despite these precautions to prevent underestimation of respiration, several studies have observed that scaled chamber measurements of respiration are higher than eddy covariance estimates of respiration in forest ecosystems (Goulden et al., 1996; Lavigne et al., 1997; Law et al., 2000; Drewitt et al., 2002; Griffis et al., 2004). There remains, therefore, uncertainty about the application of eddy covariance data in calculating annual sums and determining the carbon sink source status of an ecosystem (Field and Kaduk, 2004). In this paper, we report chamber measurements of respiration in a native Canadian grassland ecosystem during two study years (2001 and 2002). Our measurements were used to determine the temperature sensitivity of ecosystem respiration and the interacting controls of soil moisture and changes in plant productivity on the capacity for ecosystem respiration. A second major objective of this study was to compare respiration rates measured with chambers and that determined from eddy covariance measurements. In particular, we were interested to determine the consequences of the two respiration methods for calculations of seasonal-integrated net ecosystem exchange and ecosystem gross photosynthesis. Grasslands offer special opportunities for the study of ecosystem respiration because of the wide range of environmental conditions that normally occur in these ecosystems (Xu and Baldocchi, 2004). In addition, the largest interannual changes in primary production of the major ecosystem types in the continental USA occurs in grasslands, primarily caused by variation in precipitation (Knapp and Smith, 2001). The large correlated changes in biomass production and

3 L.B. Flanagan, B.G. Johnson / Agricultural and Forest Meteorology 130 (2005) environmental conditions provide an excellent model system to study these interacting effects on ecosystem respiration. The low stature of grassland vegetation also allows the entire ecosystem to be easily enclosed in chambers. So independent chamber and eddy covariance measurements of respiration can be made in order to evaluate some of the uncertainties associated with the measurement techniques. 2. Materials and methods 2.1. Study site description The study site is located just west of the city limits of Lethbridge, Alberta, Canada (Lat. N:49.438; Long. W: , 951 m above sea level). The site is classified as mixed grassland and occurs in the northern portion of the Great Plains, which is the second largest eco-zone in North America, covering approximately 2.6 million square kilometers (Ostlie et al., 1997). The physiography of the area is described as morainal and gently undulating (Kocaoglu and Pettapiece, 1980). The study area is relatively flat with slopes equal to or less than 2% grade. The soils are Orthic Dark-Brown Chernozems (Agriculture Canada, 1987) with a clay loam to clay texture and a bulk density of 1.24 g cm 3 (Carlson, 2000). The plant community consisted of the dominant grasses Agropyron dasystachyum [(Hook.) Scrib.] and A. smithii (Rydb.) Other major plant species represented to a lesser extent include: Vicia americana (Nutt.), Artemesia frigida (Willd.), Carex filifolia (Nutt.), Stipa comata (Trin. And Rupr.), Stipa viridula (Trin.) and Bouteloua gracilis [(H.B.K.) Lag.] (Carlson, 2000). The grassland has not been grazed for at least 20 years and there is a substantial amount of dead plant material (litter) on the ground surface. The average (S.E., n = 21) amount of dead plant biomass on the ground surface in 2001 and 2002 was g m 2. Average (S.D.) canopy height during 2001 was cm and during 2002 was cm. Aboveground carbon stock in live and dead vegetation was 265 g C m 2 in 2003, and the below ground carbon stock was approximately 8053 g C m 2 over a 30 cm depth interval. Mean ( ) daily temperatures for January and July are 8.6 8C and C, respectively (Agriculture and Agri-food Canada (AAFC), 2000). The mean annual precipitation ( ) for the area is mm with 32% falling in May and June (AAFC, 2000). During the summer months, average pan evaporation (Class A) exceeds the average precipitation by at least 200% and often 300% (AAFC, 2000). The mean annual wind velocity ( ) is 5.2 m s 1 out of the west (AAFC, 2000) Chamber respiration measurements Total ecosystem respiration was measured with a portable gas exchange system (LI-6200, LI-COR, Lincoln, Nebraska) and a dynamic, closed chamber (LI Soil respiration chamber, LI-COR, Lincoln, Nebraska) that was vented to the atmosphere to maintain pressure equilibrium. The chamber was attached to polyvinyl chloride collars (15 cm tall) that were inserted into the soil approximately 6 cm. The ground area enclosed by a collar was 71.6 cm 2. The height above ground of the inside of the chamber plus the collar was approximately 20.5 cm. The live vegetation in the collar was left intact so that the measurement represented total ecosystem respiration (soil plus above-ground vegetation). Measurements were made at six different collar locations during several intervals throughout a day in order to calculate a temperature dependence of respiration during that day. A set of diurnal measurements was taken on 15 days during the season in 2001 and 10 days during Data for each measurement day were fitted to the following equation: TER ¼ R 10 Q ðt soil T ref =10Þ 10 (1) where TER is total ecosystem respiration, R 10 is the standardized rate of respiration (mmol m 2 s 1 ) at the reference temperature, Q 10 is the temperature sensitivity coefficient (dimensionless) that describes the magnitude of change in respiration rate for a 10 8C change in temperature, T soil is the soil temperature (8C) measured at 4 cm depth, T ref is the reference temperature of 10 8C. Non-linear least squares regression was used to obtain estimates of the parameters R 10 and Q 10 using the Gauss Newton method in the Systat10 (2000) software package of SPSS Inc. based on diurnal measurements of ecosystem respiration and soil temperature on a given day.

4 240 L.B. Flanagan, B.G. Johnson / Agricultural and Forest Meteorology 130 (2005) Eddy covariance and meteorological measurements The eddy covariance technique was used to measure NEE and the fluxes of water vapor and sensible heat on a continuous basis (Baldocchi et al., 1988; Moncrieff et al., 2000). A detailed description of our methods and data processing procedures was provided by Flanagan et al. (2002) and Wever et al. (2002). Briefly, a three dimensional ultrasonic anemometer (Solent 1012, Gill Instruments Ltd., Lymington, England) was mounted on a one meter boom placed on top of a 6 m tall tower and oriented in the prevailing wind direction (west) to measure wind speed, direction and air temperature. The mid-point of the sonic head was located approximately 6 m above ground. Changes in CO 2 concentration (along with water vapor) were measured with a closed path, fast response infrared gas analyzer (LI-6262, LI-COR Inc., Lincoln, Nebraska) housed in an insulated and airconditioned instrument hut. Air for CO 2 and H 2 O analysis was drawn through 15 m of 3 mm inner diameter tubing (Bev-A-Line IV Tubing, LABCOR, Concord, Ontario) by a diaphragm pump (KNF UN828 KNI, KNF Neuberger Inc., Trenton, New Jersey) placed downstream from the infrared gas analyzer (IRGA). The flow rate was 8 L min 1. Calibration of the IRGA for CO 2 was done at weekly intervals for CO 2 using a CO 2 cylinder traceable to the WHMO CO 2 standard at CMDL NOAA. The water vapor channel of the IRGA was calibrated using a dew point generator (Li-610, LI- COR Inc., Lincoln, Nebraska). Fluxes of CO 2 (NEE), water vapor and sensible heat were computed using the University of Edinburgh EdiSol software as described previously (Moncrieff et al., 1997; Flanagan et al., 2002; Wever et al., 2002). Along with the eddy flux instrumentation, a weather station was established to provide meteorological data as described in detail previously (Flanagan et al., 2002; Wever et al., 2002). Measurements and equipment relevant to this study are briefly described below. A LI-COR Quantum Sensor (LI- 190SA, LI-COR, Lincoln, Nebraska) was used to measure photosynthetic photon flux density (PPFD, nm wave band). Two soil heat flux transducers (REBS HFT-3.1, Radiation and Energy Balance System, Seattle, Washington, USA) were placed about 2 cm below the soil surface, in order to calculate a mean soil heat flux. Net radiation was measured by a net radiometer (REBS Q*7.1, Radiation and Energy Balance System, Seattle, Washington, USA), mounted on a 3 m tall tower. Relative humidity and air temperature were measured using a shielded temperature and humidity probe (Vaisala HMP45C, Campbell Scientific Ltd., Edmonton, Alberta) placed 2 m above the ground. Copper-constantan thermocouples were used to measure soil temperatures at soil depths of 2, 4, 8 and 16 cm. Volumetric soil moisture content was measured (over a 0 15 cm depth) using four replicate soil water reflectometers (CS-615, Campbell Scientific Ltd., Edmonton, Alberta). The measurements made by CS-615 probes had been previously calibrated relative to manual soil volumetric measurements. To do this soil samples (0 15 cm depth) were collected on a weekly basis using a soil corer. The known volume of soil was weighed, dried at 105 8C for at least 48 h, and then re-weighed. The gravimetric moisture content was converted into volumetric measurement using the bulk density of the soil. Data are presented as available soil moisture (A w ), which was defined as the ratio of actual extractable water (difference between a given volumetric measurement and the minimum volumetric soil water content) to maximum extractable water (difference between maximum and minimum volumetric soil water content). The maximum (0.452 m 3 m 3, on June 19, 2002) and minimum (0.100 m 3 m 3, on October 22, 2001) volumetric soil water contents were those recorded during entire study period January 2001 December Total precipitation was recorded in 15-min intervals by a tipping bucket rain gauge (TE525, Texas Electronics Inc., Dallas, Texas). With the exception of the rain gauge, all data were recorded as half-hourly averages. All data were recorded on dataloggers (CR10 and CR10X, Campbell Scientific Ltd., Edmonton, Alberta). Precipitation data from AAFC was used in instances where meteorological data from the site was missing. Quality of the eddy covariance data was evaluated by comparing the sum of sensible and latent heat fluxes with the available energy (net radiation minus soil heat flux). These calculations were based on the 30-min average values for the parameters measured during May through September. There was a strong correlation between the sum of latent and sensible heat flux and available energy for both study years

5 L.B. Flanagan, B.G. Johnson / Agricultural and Forest Meteorology 130 (2005) (r 2 values of and for 2001 and 2002, respectively). The slopes of the relationships were and for 2001 and 2002, respectively. These data indicated values of energy balance closure typical of other long-term studies using eddy covariance (Baldocchi, 2003) Biomass measurements Replicate samples (n = 6) for total aboveground biomass were collected by clipping vegetation within a 20 cm 50 cm quadrat. The quadrats were placed in randomly selected 1 m 1.5 m sub-plots located within two larger 20 m 20 m plots, one northeast and the other southeast of the instrument hut. The samples were dried in an oven at 60 8C for at least 24 h and then weighed (Mettler PJ400, Greifensee, Switzerland) for above ground biomass. A biomass index (BIO) was defined as the ratio of the observed live biomass on any given date to the maximum live biomass measured during the entire study period January 2001 December 2002 (253.9 g biomass m 2, on July 16, 2002) Gap filling for eddy covariance data The eddy covariance data were screened for anomalous values outside the range normally encountered. Possible causes for such values can be sensor malfunction due to interference from dew, hoarfrost and birds. Breaks in data collection were caused by IRGA maintenance and calibration, power outages, pump failure, poor weather and the removal of anomalous values. Several strategies were used to compensate for missing data. Interpolated values were used to fill gaps that were 2 h or less. For larger gaps, the relationship between PPFD and net CO 2 flux was used to estimate missing data using a fitted second order polynomial equation. In cases where empirical relationships could not be developed due to missing meteorological data, mean diurnal variations were used to fill the missing data (Falge et al., 2001). The size of the data set used to develop these relationships was dependent on the size of the gap. Missing or rejected data occurred for a total of 7.6% of all possible 30-min time periods during the study period, January 1, 2001 to December 31, 2002 (5.9% in 2001 and 9.3% in 2002) Calculating ecosystem respiration and gross photosynthesis from eddy covariance data Using NEE values measured in darkness (PPFD < 1 mmol m 2 s 1 ) during high turbulence (friction velocity (u * ) 0.25 m s 1 ), relationships between nighttime CO 2 fluxes, soil temperature (4 cm) and soil moisture were investigated during 15-day periods. Justification for the choice of the threshold value of friction velocity was presented by Flanagan et al. (2002). In the absence of half-hourly values of soil moisture, daily values of soil moisture interpolated from manual measurements were used. Linear and multiple linear regressions between halfhourly values of dark respiration (the dependent variable) and soil temperature at 4 cm and moisture (the independent variables) were calculated. Exponential functions provided no better fit to the data than linear functions. The relationship with the highest level of significance was used to calculate ecosystem respiration during that 15-day period. These estimates of ecosystem respiration were used to replace CO 2 fluxes observed at night when turbulence was low and the air was poorly mixed (u * < 0.25 m s 1 ) and also to calculate daytime respiration (PPFD > 1 mmol m 2 s 1 ). Using the daytime respiration values, gross photosynthesis (GPP) could then be calculated by subtracting the modeled respiration rate from the corresponding daytime NEE flux. Screening data for high turbulence at night (measurements when friction velocity (u * ) 0.25 m s 1 ) resulted in removal of 61% and 60% of nighttime data during 2001 and 2002, respectively Comparison between chamber and eddy covariance measurements of respiration Calculations of seasonal variation in total ecosystem respiration were made using Eq. (1) and measurements of soil temperature. These calculations were done on a 30-min time-step and then integrated for a given day. Input values of R 10 and Q 10, for use in Eq. (1), were determined each day using the regression equations shown in Figs. 5 and 6, and daily average measurements of soil moisture and daily-interpolated measurements of the biomass index (Fig. 1). The daily-integrated respiration rates (g C m 2 day 1 ) calculated based on chamber measurements (hereafter

6 242 L.B. Flanagan, B.G. Johnson / Agricultural and Forest Meteorology 130 (2005) referred to as the chamber method) were then compared to calculations of the daily-integrated respiration rates from the gap-filled eddy covariance measurements (hereafter referred to as the EC method). We made the comparison between the two respiration methods only during the growing season because chamber measurements were restricted to this time period Estimating uncertainty in the flux measurements An analysis was conducted to evaluate the uncertainty in the integrated 6-month carbon budgets calculated using eddy covariance and chamber respiration measurements. For the eddy covariance measurements, this uncertainty analysis considered: (i) composite random error associated with the measurement equipment, flux footprint heterogeneity, and variation in turbulent transport; (ii) systematic errors caused by incomplete frequency response; and (iii) systematic errors associated with lack of nocturnal mixing and associated data screening procedures. This analysis does not include all the possible sources of error in eddy covariance flux measurements, but it does consider the major factors expected to introduce most uncertainty into carbon budget calculations (Aurela et al., 2002). The estimate of the composite random error made use of the Hollinger and Richardson (2005) repeated sampling method or daily differencing approach. In this approach, the random measurement uncertainty for the eddy covariance measurements (E EC ) was calculated based on the standard deviation (s) of differences between measurements made at the same flux site on different days (x 1 and x 2 ) when equivalent environmental conditions occurred: Fig. 1. Seasonal and interannual variation in: (a) available soil water content, (b) aboveground biomass, and (c) soil temperature. Biomass values are expressed as the ratio between a value measured on a given date and the maximum value (253.9 g m 2 ) recorded during Available soil moisture (A w ), was defined as the ratio of actual extractable water (difference between a given volumetric measurement and the minimum volumetric soil water content) to maximum extractable water (difference between maximum and minimum volumetric soil water content). The maximum (0.452 m 3 m 3, on June 19, 2002) and minimum (0.100 m 3 m 3, on October 22, 2001) volumetric soil water contents were those recorded during entire study period January 2001 December E EC ¼ p 1 ffiffi sðx 1 x 2 Þ (2) 2 Equivalent environmental conditions were defined as measurements made at the same time of day, with 30- min average PPFD, air temperature and wind speed within strict limits of similarity (differences in PPFD less than 75 mmol m 2 s 1, air temperature less than 3 8C and wind speed less than 1 m s 1 ) (Richardson et al., submitted for publication). This is a conserva-

7 L.B. Flanagan, B.G. Johnson / Agricultural and Forest Meteorology 130 (2005) tive approach to estimating random uncertainty as CO 2 flux measurement uncertainty was approximately 20 25% higher using the daily differencing approach than when uncertainty calculations were done using two towers sampling at the same site at the same time (Hollinger and Richardson, 2005). Richardson et al. (submitted for publication) calculated absolute values for the random uncertainties of 0.6 and 1.2 mmol m 2 s 1 during the growing season in dry years ( ) and wet years ( ) at the Lethbridge flux site, respectively. We used these uncertainties and the average (absolute value) of the NEE fluxes measured during the growing season (1.65 and 4.73 mmol m 2 s 1 ) to calculate relative random uncertainties of 36.4% and 25.4% during 2001 and 2002, respectively. An important systematic error that influences eddy covariance measurements is the loss of high frequencies contributing to turbulent fluxes associated with the transit time for air flow down long tubing connecting the closed path IRGA and the tubing inlet near the sonic anemometer head (Moncrieff et al., 1996). We made corrections for these errors as suggested by Moncrieff et al. (1997), and the magnitude of the corrections was typically 30%. Following Aurela et al. (2002) we assumed an uncertainty of 30% for the correction procedure, which resulted in an uncertainty estimate of 9% on the corrected flux caused by this systematic procedure. A second major systematic error is associated with the underestimation of nighttime fluxes and the procedures used to correct for the underestimation. As described above we applied a friction velocity threshold of 0.25 m s 1 in our correction procedure, and removed fluxes measured under low turbulent conditions (below the friction velocity threshold) and then replaced the missing values with modelcalculated result. We determined the effect of modifying this threshold friction velocity value on the carbon budget calculations and observed that the integrated NEE during April October varied from 53 to 41 g C m 2 over the range of friction velocity from 0.1 to 0.4 m s 1 in 2001; and the integrated NEE during April October varied from 350 to 327 g C m 2 over the range of friction velocity from 0.1 to 0.4 m s 1 in From these results we estimate an uncertainty of 12.8% (5.5 on the 6-month balance of 43 g C m 2 at a friction velocity threshold of 0.25 m s 1 ) during 2001 and 3.5% (11.5 on the 6-month balance of 327 g Cm 2 at friction velocity threshold of 0.25 m s 1 ) during 2002 for the procedures associated with correction of nighttime flux estimation. We combined the three types of errors affecting the eddy flux measurements using the error accumulation principle (combing them in quadrature) and obtained relative estimates of uncertainty of 39.6% in 2001 and 27.2% in These uncertainty estimates were used to assign error bars to all the NEE and GPP shown in Table 2. In addition, this assessment of uncertainty was used to assign error bars to the TER values in Table 2 calculated using the EC method. The estimation of uncertainty in the integrated 6- month TER values calculated using chamber respiration measurements was also affected by random and systematic errors. Lacking the information to evaluate the systematic errors associated with the LiCor chamber method, we applied an ad hoc estimate of 20% for the systematic instrument error. The random error uncertainty should be influenced by several factors including spatial variation in the chamber measurements, spatial variation in the aboveground biomass measurements (an input variable for calculation of R 10,seeFig.6),errors associated with the measurement of soil temperature and soil moisture, and errors associated with the regression equations used to calculate Q 10 and R 10 values (see Figs. 5 and 6). We quantified uncertainty associated with spatial variation in ecosystem respiration by using the coefficient of variation (standard deviation divided by the mean of replicate chamber measurements). This parameter was averaged for measurements made on different days throughout a growing season (25.3% for 2001 and 12.9% for 2002). Uncertainty in biomass measurements was determined in the same manner and the values we obtained were 32% in 2001 and 22% in We applied ad hoc errors of 10% for soil temperature measurements and 20% for soil moisture measurements in both 2001 and The uncertainty associated with application of the regression equations (E R ) shown in Figs. 5 and 6 was estimated by evaluating the difference between the individual data points (observed values (Obs)) and the predicted

8 244 L.B. Flanagan, B.G. Johnson / Agricultural and Forest Meteorology 130 (2005) values (Pred) calculated from the regression equations (Aurela et al., 2002): sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi X n ðobs PredÞ 2 E R ¼ (3) ðn 1Þn i¼1 This calculation resulted in an estimate of relative error of 8.2% and 8.6% for Q 10 during 2001 and 2002, respectively, and an estimate of 7.1% and 16.3% for R 10 in 2001 and 2002, respectively. We combined all these errors affecting the scaledup chamber respiration measurements using the error accumulation principle (combing them in quadrature) and obtained relative estimates of uncertainty of 51.8% in 2001 and 43.5% in These uncertainty estimates were used to assign error bars to the TER values in Table 2 calculated using the chamber method. 3. Results 3.1. Comparison of environmental conditions and biomass production There was a large difference in the amount of precipitation received during the two study years. In 2001 the growing season (April September) precipitation was 102 mm, which was less than one-half of the 30-year ( ) mean ( mm, (S.D.)), while 2002 received 537 mm, almost double the normal growing season precipitation. As a consequence soil moisture remained higher in 2002 than 2001 during most of the growing season (Fig. 1a). Daily-mean soil temperature was on average 4.6 8C warmer in 2001 than in 2002 during the growing season, although soil temperature differences increased to a maximum of C during an approximately 15-day interval (day of year ) in late July and August (Fig. 1c). Peak aboveground biomass production was 60% higher in 2002 and there was a much longer period of time in 2002 when vegetation was at or near peak biomass (Fig. 1b). In 2001 drought caused the end of the growing season, and this is the normal pattern at this site (Flanagan et al., 2002). By contrast, the end of the longer growing season in 2002 was brought on by frost rather than lack of soil moisture. Fig. 2. Diurnal variation in: (a) net ecosystem CO 2 exchange (NEE), (b) chamber measurements of ecosystem respiration rate, and (c) soil temperature near the time of peak aboveground biomass production in 2001 (June 7) and 2002 (July 11). NEE measurements are the mean S.E. for a 14-day period centered on the day chamber respiration measurements were made. Chamber respiration rate values represent the mean (S.E.), n = 6.

9 L.B. Flanagan, B.G. Johnson / Agricultural and Forest Meteorology 130 (2005) Fig. 3. Seasonal variation in the temperature response of ecosystem respiration rate during (a) 2001 and (b) Individual points represent the mean of measurements made at six different collars. The parameters (R 10, Q 10, r 2 ) of the regression lines displayed in the figure are shown in Table Chamber respiration measurements The rates of ecosystem respiration measured with the chambers showed strong diurnal patterns during time periods when live biomass was near peak levels in both years (Fig. 2b). The diurnal pattern of change in ecosystem respiration was strongly correlated with associated changes in soil temperature (Fig. 2c). Maximum respiration rates were higher in 2002 than in 2001 (approximately 9 versus 5 mmol m 2 s 1 ). Nighttime eddy covariance measurements were consistent with the chamber measurements of respiration, as ecosystem respiration measured by eddy covariance was also approximately double that recorded in 2001 (Fig. 2a). Net ecosystem CO 2 exchange measurement Table 1 Comparison of the calculated parameters (95% confidence limits) from non-linear regression fits to Eq. (1) based on diurnal chamber measurements of ecosystem respiration Date R 10 Q 10 r 2 April 26, May 2, May 10, May 23, June 7, June 13, June 20, June 27, July 4, July 11, July 24, August 9, August 27, September 10, September 24, May 28, June 5, June 13, June 20, June 25, July 11, July 25, August 9, September 16, October 8, R 10 is the standardized rate of ecosystem respiration at 10 8C (soil temperature at 4 cm depth). Q 10 is the temperature sensitivity coefficient of ecosystem respiration for a 10 8C change in soil temperature. The r 2 value is the proportion of variation in the measured data explained by the regression equation. Table 2 Comparison of the seasonal-integrated (April September) calculations of carbon exchange (g C m 2 period 1 ) for total ecosystem respiration (TER), gross photosynthesis (GPP), and net ecosystem exchange (NEE) using the chamber method and nighttime eddy covariance (EC) method for calculating ecosystem respiration Year and parameter Chamber method EC method Difference 2001 TER GPP NEE TER GPP NEE Uptake of carbon from the atmosphere is indicated by negative values and release of carbon from the ecosystem to the atmosphere is indicated by positive values. The error bars represent estimates of uncertainties for both the chamber and EC methods as described in Section 2.8.

10 246 L.B. Flanagan, B.G. Johnson / Agricultural and Forest Meteorology 130 (2005) showed that peak net uptake of CO 2 in 2002 was greater than twice as high as that during 2001 (Fig. 2a). Spatial variation in the chamber respiration measurements, as quantified by the coefficient of variation, was on average 25.3% and 12.9% during the growing seasons for 2001 and 2002, respectively. Slightly higher spatial variation was observed at the start and/or the end of the growing season. During the time of peak respiration fluxes, spatial variation was only approximately 18% and 10% during 2001 and 2002, respectively. This suggests that our limited sample of 6 replicate chambers was adequate to characterize seasonal variation in ecosystem respiration and its environmental control. The temperature response curves for ecosystem respiration (based on chamber measurements) changed dramatically during the growing season in 2001 (Fig. 3a; Table 1). In June 2001 the temperature response curve was quite similar to that recorded during June However, the slope of relationship between respiration and temperature declined as soil moisture was reduced through the growing season in 2001 (Fig. 3a). By contrast, there was little change in the slope of the respiration-temperature response curve during 2002, although the maximum rates of respiration increased from June through to August (Fig. 3b). In general Eq. (1) explained a high fraction of the variation in the measured respiration data (Table 1). The r 2 values for the fitted regression were lower later in the season (and sometimes earlier in the season) because of much lower diurnal variation in soil temperature and consequently lower variation in respiration rates (Table 1). The average (S.E., n = 10) temperature coefficient (Q 10 ) calculated for 2002 was and Q 10 values showed a slight decline throughout the growing season (Fig. 4a; Table 1). During 2001 the average (S.E., n = 15) Q 10 value was slightly lower , and Q 10 values tended to show a stronger decline as the growing season progressed than was observed in 2002 (Fig. 4a; Table 1). There was a positive linear relationship between Q 10 values and available soil moisture calculated from data in both 2001 and 2002 (Fig. 5). For data from both years combined, the average (S.E., n = 25) Q 10 value was The standardized rate of ecosystem respiration at 10 8C (R 10 values) showed different patterns of Fig. 4. Seasonal and interannual variation in: (a) the calculated temperature sensitivity coefficient (Q 10 ) of ecosystem respiration, and (b) the calculated standardized rate of ecosystem respiration (R 10 ). Fig. 5. The effect of changes in available soil moisture (A w ) on the temperature sensitivity coefficient (Q 10 ) of ecosystem respiration. The line represents a fitted regression for data from both years combined (Q 10 = A w , r 2 = 0.439).

11 L.B. Flanagan, B.G. Johnson / Agricultural and Forest Meteorology 130 (2005) Fig. 6. The relationship between changes in the standardized rate of ecosystem respiration (R 10 ) and seasonal changes in the product of aboveground biomass index (BIO) and available soil moisture (A w ) during both 2001 and The soil line represents a fitted regression for data from both years combined (R 10 = (BIO A w ) 0.535, r 2 =0.936). seasonal variation in 2001 and 2002 (Fig. 4b). During 2001 the seasonal variation in R 10 values closely followed the growing season variation in aboveground biomass with a peak near day 160. In 2002, R 10 values initially increased in a correlated pattern with growth of aboveground biomass and then reached a plateau and remained quite constant through the rest of the growing season (Fig. 4b). A large fraction (94%) of the variation in R 10 values calculated from data in both 2001 and 2002 could be explained by seasonal variation in the product of the aboveground biomass index and the available soil moisture index (Fig. 6) Seasonal variation and comparison between chamber and eddy covariance measurements of respiration Respiration rates varied strongly within a growing season and different seasonal patterns were observed between study years (Fig. 7). In 2001, respiration increased at a faster rate and reached peak values earlier in the growing season than was observed in This was consistent with the warmer soil temperatures observed in 2001 and the differences in timing of peak aboveground biomass production between the two study years (Fig. 1). In 2002, respiration varied substantially (Fig. 7b) while biomass measurements were relatively constant at Fig. 7. Comparison of seasonal and interannual variation in total ecosystem respiration rate calculated using Eq. (1), based on chamber measurements and the relationships shown in Figs. 5 and 6 (chamber method), and that calculated using nighttime eddy covariance measurements (EC method). The points represent dailyintegrated values based on calculations done on a 30-min time step. or near peak values (Fig. 1b). This variation in respiration was associated with significant changes in precipitation and soil moisture during the peak of the growing season in Peak daily-integrated values of respiration in 2002 were approximately twice as high as the peak values observed in 2001 (Fig. 7). In general there was very good correspondence between the seasonal patterns of respiration calculated using the chamber and EC methods. There was a positive correlation between the chamber and EC methods for calculating dailyintegrated TER rates (Fig. 8a and b). However, in both years the chamber method tended to calculate higher respiration values than the EC method under conditions when respiration rates were high, and the chamber method tended to calculate lower respiration

12 248 L.B. Flanagan, B.G. Johnson / Agricultural and Forest Meteorology 130 (2005) Fig. 8. (a and b) Comparison between total ecosystem respiration rate (TER) calculated using Eq. (1), based on chamber measurements and the relationships shown in Figs. 5 and 6 (chamber method), and that calculated using nighttime eddy covariance measurements (EC method). The points represent daily-integrated values based on calculations done on a 30-min time step. Also shown are comparisons between calculations of daily-integrated values of gross photosynthesis (c and d; GPP) and net ecosystem exchange (e and f; NEE) done using the two different methods for determining respiration rates. The daily-integrated values of NEE are influenced by the method of respiration rate calculation because of the rejection and replacement of nighttime eddy covariance measurements made during periods of low turbulence (friction velocity less than 0.25 m s 1 ).

13 L.B. Flanagan, B.G. Johnson / Agricultural and Forest Meteorology 130 (2005) values than the EC method when respiration rates were low (Fig. 8a and b). This effect was most prominent in When TER was integrated over the entire growing season period (April September), the chamber method produced higher values than the EC method by approximately 4.5% and 13.6% during 2001 and 2002, respectively (Table 2). The differences between the values calculated with the two techniques were much smaller than the estimated uncertainties associated with both techniques (Table 2). Integrated calculations of GPP and NEE can also be influenced by the procedures used for calculating respiration rates. We determined the consequences of the chamber method relative to the (night-time) EC respiration method for determining integrated values of NEE and GPP. In 2001 the chamber method produced a more negative seasonal-integrated value of NEE (a larger net sink for carbon), despite the higher seasonal-integrated respiration rate calculated using the chamber method (Table 2). This result was possible because the seasonally-integrated TER included regression-based respiration calculations during both daytime and nighttime periods, while the integrated NEE values only made use of the regression-based respiration calculations during calm periods at night. During a number of low turbulence nighttime periods in 2001, replacement of eddy covariance data with respiration values from the chamber method regression equations resulted in lower respiration rates than were predicted by the EC method regression equations. This can be seen in the small cloud of points that fall below the 1-to-1 line at values between 0 and 1 mmol m 2 s 1 in the 2001 NEE comparison graph (Fig. 8e). In 2002, a different pattern was observed where the higher TER calculated using the chamber method resulted in a less negative seasonal-integrated NEE (a smaller net sink for carbon) than the EC method (Table 2). While the direction of change in NEE differed between the 2 years, the absolute magnitude of the difference between the seasonal-integrated NEE calculated using the two respiration procedures was quite similar in 2001 and 2002 (10 and 12 g C m 2 season 1, respectively). However, the percentage change was much larger in 2001 (23.2%) than in 2002 (3.7%) because of the smaller net sink observed in that year (Table 2). There was a strong correlation between dailyintegrated values of GPP calculated using the two methods (Fig. 8c and d). However, the chamber method produced more negative values than the EC method for the seasonal-integrated GPP by approximately 7.3% and 6.7% in 2001 and 2002, respectively (Table 2). The differences in GPP observed between the two techniques were much smaller than the uncertainties estimated for each technique (Table 2). 4. Discussion Ecosystem respiration is dependent on autotrophic (plant) and heterotrophic (microbe) activity, and both of these are controlled by environmental conditions (primarily temperature and water availability), and supply of carbohydrate and other substrates (Raich and Schlesinger, 1992; Davidson et al., 1998; Janssens et al., 2001; Reichstein et al., 2002a,b). By making diurnal measurements of ecosystem respiration at intervals when water availability and plant activity were held relatively constant, we were able to quantify the temperature sensitivity (Q 10 ) of ecosystem respiration. Our average values of Q 10 ( ) were in the range expected for biological processes and at the mid-point of the range of values ( ) reported in review studies of soil and plant respiration (Raich and Schlesinger, 1992; Tjoelker et al., 2001). In addition we observed a reduction in Q 10 in response to declines in soil moisture. Similar responses have also been reported by Xu and Qi (2001) for soil respiration in a Sierra Nevadan forest and by Reichstein et al. (2002a,b) and Xu and Baldocchi (2004) for ecosystem respiration in Mediterranean evergreen forests and a Mediterranean annual grassland, respectively. The temperature sensitivity of respiration can also change with changes in temperature (Lloyd and Taylor, 1994; Kirschbaum, 1995). Tjoelker et al. (2001) demonstrated that Q 10 declines with increases in temperature for plant respiration, and Xu and Qi (2001) showed a negative correlation between soil temperature and Q 10 for soil respiration. It is difficult, however, to separate the interaction between increases in soil temperature and reductions in soil moisture for their effects on Q 10 and soil respiration in many field studies in temperate ecosystems (Davidson et al., 1998; Xu and Qi, 2001). The lower average Q 10 we observed in 2001 was likely due to the reduced soil moisture in 2001 compared to 2002,

14 250 L.B. Flanagan, B.G. Johnson / Agricultural and Forest Meteorology 130 (2005) but the associated slightly higher temperatures in 2001 may have also been a contributing factor. Calculation of R 10, a standardized rate of respiration at 10 8C, provides insight into seasonal and interannual changes in ecosystem respiration (mostly) independent of changes in temperature (Lloyd and Taylor, 1994). The R 10 value can be thought of as a measure of the capacity of an ecosystem for respiration that will depend on factors influencing both autotrophic and heterotrophic respiration. The magnitude of autotrophic respiration is influenced by the amount and activity of plants and so reflects changes in environmental conditions (including temperature) that control plant growth and development, photosynthesis and carbon allocation patterns. Heterotrophic respiration is dependent on the supply of respiratory substrates (primarily from plant litter and plant root exudates) as well as environmental conditions that control microbial growth and development, and the supply and quality of respiratory substrate provided by plants, particularly plant roots (Raich and Schlesinger, 1992; Raich and Tufekcioglu, 2000). Our data illustrated that approximately 94% of the seasonal and interannual variation in R 10 data was explained by the product of changes in soil moisture and aboveground biomass (Fig. 6). This observation suggests that the amount of plant aboveground biomass is not only a good proxy for estimating total autotrophic activity, but it also acts as a good proxy for heterotrophic activity as well. In addition, soil moisture appears to be the dominant environmental factor in this semi-arid grassland that influences changes in both autotrophic and heterotrophic respiration during the growing season, when temperature effects are held constant. Consistent with our study, Janssens et al. (2001) showed that soil and ecosystem respiration were strongly correlated to photosynthesis in several European forests, and that forest productivity had a larger effect than temperature in explaining variation in respiration rates among the different forests. In a Mediterranean annual grassland, Xu and Baldocchi (2004) also showed that the interaction between soil moisture and plant activity was the dominant control on the magnitude of ecosystem respiration. The highest rates of respiration observed by Xu and Baldocchi (2004) generally occurred under cool temperatures (approximately 10 8C) in the winter when soil moisture and plant activity were highest. Only very low respiration rates (normally less than 1 mmol m 2 s 1 ) were measured in summer when soil temperature was high (approximately 30 8C) but soil moisture was very low and the grass community was dormant. The magnitude of our average respiration rates compare well with other studies in grassland ecosystems. When soil moisture and plant biomass were both high, we measured ecosystem respiration rates of 9 mmol m 2 s 1, similar to the maximum values measured in a California annual grassland (Xu and Baldocchi, 2004), a tall grass prairie (Suyker and Verma, 2001) and in sites from the northern portion of the Great Plains prairie (Gilmanov et al., 2005). In contrast some grassland and forest systems have much lower respiration rates. The maximum rate of ecosystem respiration in an approximately 70-year old mixed Pinus sylvestris and Quercus robur forest in Belgium was only 5 mmol m 2 s 1 (Carrara et al., 2004). Under well-watered conditions in Oklahoma rangeland of the southern Great Plains, Meyers (2001) also measured peak ecosystem respiration rates of approximately 5 mmol m 2 s 1. When comparing among ecosystems, temperature and soil moisture status, and other soil factors that affect the production and consumption of organic matter are the dominant controls on the magnitude of soil and ecosystem respiration (Raich and Tufekcioglu, 2000; Xu and Baldocchi, 2004; Xu et al., 2004). However, Raich and Tufekcioglu (2000) also showed that in paired comparisons, soil respiration rates were approximately 20% higher in grasslands than in forests growing under similar conditions. They attributed this result to higher allocation of carbon to roots in grassland ecosystems. Our daily-integrated estimates of respiration based on chamber measurements were generally higher than that determined by the EC method (Fig. 8a and b), although the differences between the two techniques were smaller than the estimated uncertainties for both techniques. Wohlfahrt et al. (2005) observed agreement to within 35% (the level of uncertainty) among comparisons of several approaches to quantifying ecosystem respiration in a mountain grassland including eddy covariance, ecosystem chamber and scaled up leaf and soil chamber measurements. There was much better agreement (less than 5% difference) for three methods, scaled up chambers, respiration