Physiological responses of a black spruce forest to weather

. We used eddy covariance to measure the net exchange of CO2 between the atmosphere and a black spruce (Picea mariana) forest in Manitoba for 16,500 hours from March 16, 1994 to October 31, 1996. We then partitioned net exchange into gross photosynthesis and respiration by estimating daytime respiration as a function of temperature, and used these data to define the physiological responses of the forest to weather. The annual rates of gross production and respiration by the forest were both around t C ha-(cid:127) yr -(cid:127). Both photosynthetic and respiratory response were reduced in winter, recovered with warming in spring, and varied little in summer. Respiration in mid summer increased with air temperature (Tair) at a Q (cid:127)0 of around 2 to a rate of 2-8/.tmol Tai r < 0øC, T(cid:127) r from 0 ø to 14øC, and was relatively insensitive to T(cid:127)r > 14øC. Gross CO2 uptake at T(cid:127)r > 14øC


Introduction
Boreal forests cover -8% of the Earth's land surface and contain -13% of the carbon in the terrestrial biosphere [Whittaker, 1975;Schlesinger, 1977]. Increasing concentrations of CO2 and other trace gases in the atmosphere are expected to impact the climate of the boreal zone over the next century [Mitchell et al., 1990] and, as a result, to alter the rates of photosynthesis and respiration by boreal forests [Bonan and Shugart, 1989;Oechel and Billings, 1992;Oechel et al., 1993]. To the extent that these changes in activity translate into changes in carbon balance, a positive or negative feedback to the concentration of atmospheric CO2 may result. However, an improved understanding of the physiological sensitivity of boreal forest CO 2 exchange to weather is required before the effects of climate change on carbon balance can be predicted reliably. Most investigations of forest CO2 exchange have relied on models to extrapolate short-term gas-exchange measurements made with small chambers. This approach has contributed greatly to understanding, but uncertainties are inevitable when small-scale observations are aggregated to whole ecosystems, and short-term data to annual balances [Whittaker and Marks, and continuous feather moss. Low-lying areas were dominated by sparse, 1-to-6-m-tall, chlorotic spruce and continuous sphagnum moss. Approximately 45% of the area within 500 m of the tower was feather moss, 45% sphagnum moss, and 10% fen [Harden et al.,

this issue].
The flux measurements were made from a 31-m-tall, 30-cmcross-section tower (Rohn 25G, Peoria, Illinois). The data acquisition equipment and some of the instruments were housed in a climate-controlled hut 20 m northeast of the tower base.
Power was provided by a diesel generator which operated continuously 300 m east. A network of boardwalks around the site minimized disturbance, and access to areas south, west, and north of the tower was strictly limited. The data acquisition and control systems were fully automated, allowing extended periods of unattended operation. The raw digitized signals were stored on a hard drive and transferred by a local employee to a removable cartridge every 3-7 days for immediate shipment to our laboratory. The site was visited by science personnel every 5 weeks for maintenance and as needed for repairs.

Measurements
The turbulent fluxes of sensible heat, latent heat, CO2, and momentum at 29 m were determined following Wofsy et al. [1993] and Goulden et al. [1996b]. The signals directly associated with the flux calculations were digitized and recorded at 4 Hz. Wind and temperature were measured with a three-axis sonic anemometer (Applied Technologies, Boulder, Colorado) pointed due west (270ø). The anemometer incorporated online corrections for flow distortion [Kaimal et al., 1990], and the effect of crosswind on the calculation of temperature [Kaimal and Gaynor, 1991].
The mixing ratios of COg and HgO were monitored by sampling 16-18 standard liter min-• (slpm) from 0.5 m east of the vertical axis of the anemometer through four Teflon filters (3-•m pore size, 50 mm diameter) configured in parallel. The inlet filters were changed at least once every 5 weeks. Sample air was drawn into the instrument hut through a 50-m-long, 0.64-cm inner diameter (id) high-density polyethylene tube. A subsample was drawn through a thermostated block to stabilize temperature [Webb et al., 1980] and then through a CO2/H20 infrared gas analyzer (IRGA; Model 6262, LI-COR, Lincoln, Nebraska) at 4 slpm. The main sample tube was replaced with Teflon PFA in May 1996, resulting in an improved response for fluctuations in water vapor. The pressures in the main sample line and immediately after the IRGA were monitored and controlled, maintaining the IRGA cell at 53 kPa (MKS Instruments, Andover, Massachusetts). The gain of the IRGA was determined every 3 hours by addition of 4% COg at 40 and 80 standard mL min -1 to the main sample stream. The IRGA zero was determined every 3 hours by passing sample air through a series of soda line and Mg(C104)g traps. The raw IRGA signals were recorded (COg 1S, H20 1S), and the effects of water vapor on COg measurements removed and the signals linearized during subsequent processing [LI-COR, 1991]. The COg flux was calculated as the 30-min covariance of the vertical wind velocity w' and the COg concentration after subtracting the linear least squares regression c'. The time lag (typically, 6 s) was determined by maximizing the correlation between w' and c'. The flux was rotated to the plane with zero mean vertical wind [McMillen, 1988]. We used a second HgO/CO 2 IRGA to sequentially measure the mixing ratios at six levels through the canopy (0.3, 1.5, 4.6, 8.4, 12.9, and 28.8 m) every 12 mins. The profile IRGA was calibrated every 3 hours using a pair of COg mixtures traceable to the 1993 Scripps-World Meteorological Organization (WMO) standard. The hourly change in COg beneath 29 m (storage [Wofsy et al., 1993]) was calculated by interpolation. The eddy flux and storage observations were summed to calculate the hourly net ecosystem exchange (NEE [Wofsy et al., 1993]), except for a few intervals when storage data were unavailable and just eddy flux was used (less than 2% of the growing season).
The data required to characterize the physical environment were recorded at 0.5 Hz. The photosynthetically active photon flux density (PPFD) above the forest was measured with a silicon quantum sensor (LI-COR, Lincoln, Nebraska). Mosssurface PPFD was measured at eight locations along a topographic gradient with quantum sensors. The net radiation at 29 m was measured with a thermopile net radiometer (REBS Q*6, Seattle, Washington). Air temperatures at 1, 8, and 28 m altitude were measured with ventilated thermistors. Soil temperatures at 5, 10, 20, 50, and 100 cm beneath the moss surface were measured at five sites along a topographic gradient with precision thermistors.

Analysis
Our analysis proceeded by (1) examining the raw flux data for errors associated with wind from behind the tower, calm conditions, and the damping of high-frequency fluctuations and for intervals with malfunctioning instruments, and then excluding these periods or correcting for these errors in subsequent analyses; (2) separating the observed net COg exchange into whole-ecosystem respiration and gross uptake by estimating daytime respiration as a function of temperature; and (3) using the large sample of observations to isolate the environmental factors controlling photosynthesis and respiration. We describe the first and fourth aspects of step 1 and all of step 2 here. The second and third aspects of step 1 are discussed in the Results, and step 3 is implicit in the Discussion.
We rejected flux data when the sonic anemometer or IRGA malfunctioned or when background COg changed rapidly. Problems with the IRGA were identified by monitoring the flow and pressure at the instrument and by comparing the signals with simultaneous measurements made with the profile instrument. Additional problems with the IRGA were identified with an automated calibration every 3 hours. The sonic anemometer developed problems with spiking occasionally due either to precipitation or malfunctioning transducers. (Spikes are step changes in wind or temperature signal that occur when the instrument misidentifies the return pulse [Kaimal and Finnigan, 1994].) We therefore adapted the processing code to determine the number of spikes in each interval and to recalculate the turbulent fluxes after filtering out spikes. We subsequently discarded the sensible heat and momentum fluxes when the spike rate was greater than 5% and water vapor and COg fluxes when the spike rate was greater than 20%. We also rejected periods with unreasonable means or variances in the temperature measured by the sonic based on comparison with an adjacent thermistor, and unreasonable horizontal wind based on comparison with an adjacent cup anemometer.
An analysis of turbulence statistics as a function of wind direction revealed anomalies such as a systematic decrease in crw/u* during neutral and unstable periods from a mean of 1.25 when the wind came from in front of the tower (135ø-45 ø) to 1.1 when the wind came from behind the tower (45ø-135ø). Similarly, an analysis of energy exchange revealed anomalously poor closure of the energy budget during periods with wind from behind the tower, a sector that included the generator, instrument hut, and access trail. We therefore excluded from subsequent analyses observations with wind from 45 ø to 135 ø .
Net exchange can be separated into respiratory and photosynthetic fluxes by several approaches: an independent set of eddy covariance instruments may be deployed beneath the canopy to measure forest floor respiration [Baldocchi and Meyers, 1991], a series of automated chambers may be deployed to continuously measure soil and stem respiration, as was done at the site by Goulden and Crill [1997] and Lavigne et al. [this issue], or day respiration R may be estimated from night measurements of NEE and the diel course of temperature [Goulden et al., 1996b]. We used the third method in conjunction with information on the diel variation in respiration from the second approach, since this approach provides the greatest likelihood that similar footprints are considered and minimizes biases between methods.
Forest respiration R was calculated assuming an exponential dependence on air temperature at 28 m altitude with a Q•0 of 2.0. We determined the fit between the NEE during windy nights and air temperature at 28 m altitude for a series of  [Goulden and Crill, 1997] respiration measurements made at the site over diel courses. The key assumptions in the approach are (1) that the eddy flux measurements after removal of periods with u* < 0.2 m s-• are not biased from day to night as discussed in the Results, and (2) that the temperature measured at 28 m is closely related to the activity-weighted temperature of the forest. The second assumption is most likely to fail if appreciable biomass is colder than the temperature at 28 m during the day due to storage, leading to an overestimation of daytime respiration, or colder than the temperature at 28 m during the night due to atmospheric inversion, leading to an underestimation of daytime respiration.
We checked for the sensitivity of calculated R to our selection of 28 m temperature by recalculating the regressions and daytime rates of R using the temperatures at 1 m elevation and also at 5 cm depth in the moss layer. The runs calculated with 1 m temperatures indicate the errors if respiration from boles and leaves dominates whole-forest respiration, providing an upper bound on the estimates of daytime R. The runs calculated with 5-cm-deep temperatures indicate the errors if respiration from moss and soil dominates whole-forest respiration, providing a lower bound. We found that calculated daytime R was not biased at or below 20øC by our choice of temperature location. At 30øC the rates of daytime R calculated using the 28 m temperature were 2 /•mol m -2 s -• less Gross ecosystem exchange (GEE) was calculated as the difference between NEE and estimated respiration: Gross ecosystem exchange should equal the net rate of carboxylation and oxygenation by ribulose-l,5-bisphosphate carboxylase. Gross exchange does not include dark respiration and is not equivalent to measurements of net assimilation made with leaf or canopy chambers. Net or gross "exchange" of CO• into the forest was considered negative, and "exchange" out of the forest positive. However, we discuss "photosynthesis," "uptake," and "ecosystem production" as processes with positive signs.

Accuracy of Flux Measurements' Underestimation of High-Frequency Fluctuations
The closed-path IRGA and long sampling tube resulted in an underestimation of water vapor and CO2 flux due to the damping of high-frequency fluctuations [Moore, 1986]. The time constant for the damping of CO2 fluctuations, as determined by analyses of cospectra, was 0.6 s, and the time constant for the damping of water vapor fluctuations was 1.25 s. We simulated the loss of flux (Loss ..... ) associated with these responses for each 30 min run by recalculating the sensible heat flux after numerically slowing the response of the temperature detector with a low-pass analog of a R-C filter, and adjusting for the lag introduced between w' and T'

Accuracy of Flux Measurements: Nocturnal Measurements During Calm Periods
A selective measurement bias between day and night can also result if CO2 "leaks" from the site during calm nocturnal periods [GouMen et al., 1996b]. In principle, NEE (eddy flux plus directly measured storage) should be insensitive to reduced turbulence, since storage increases to offset any reduction in turbulent flux (as reported by Grace et al. [1996]). However, an analysis of nocturnal CO2 exchange at NSA-OBS revealed that decreased eddy CO2 flux at u* < 0.2 m s-I was offset only partially by increased storage, resulting in a reduction in NEE (Figure 3a). Simultaneous measurements with chambers indicated that the CO2 efflux at the moss surface was not similarly affected by turbulence [Goulden and Crill, 1997] and hence that the decline in NEE was a measurement artifact.
The efflux of CO2 immediately following resumption of mixing was not unusually high (Figure 3b) above the forest (Figure 4). The coherence was especially impressive considering that we did not account for heat storage and also that scatter is inevitable due to sampling uncertainty. Unfortunately, the absolute agreement between measurements was less gratifying with an 18% disparity between H + ,•E and R nc t. The inequality varied through the year, reaching a maximum in spring when energy was used during melt, a minimum in summer, and increasing again in fall despite energy release by freezing. A comparable 5-15% discrepancy between measured turbulent exchange and net radiation has been observed above other [Lee and Black, 1993 The disparity continued year-round, and was reduced but not eliminated in late afternoon, implying that it was not caused entirely by changes in storage. Similarly, the discrepancy was independent of Bowen ratio, implying that it was not caused by a bias peculiar to one of the turbulent exchange measurements. Energy closure did improve with increasing u*, a trend that could indicate stationary circulation patterns during calm periods which transport flux that is not included in the detrended covariance [Ban' et al., 1994; Lee and Black, 1993]. We also observed a systematic difference in closure as a function of wind direction, a pattern that could indicate spatial heterogeneity in albedo and net radiation. However, we are unable to account fully for the bias between measurements and must conclude that our turbulent exchange measurements may underestimate both day and night surface activity by 10-20%. Nonetheless, the strength of the correlation between H + and Rnet, establishes the integrity of the measurements and their usefulness for characterizing the proportional responses of exchange to the physical environment.

Annual Integrated Activity
We integrated the observations of NEE, GEE, and R to calculate the annual net ecosystem production (NEP), gross The NEP represents the small residual of -4000 kg C ha-• yr-• gained during daylight and -4000 kg C ha-• yr-• lost at night. A seemingly minor systematic bias of 20% in the measurements from day to night will cause an error in the calculated NEP of 800 kg C ha -• yr-• [Goulden et al., 1996b]. We assume that our correction for the damping of high-frequency fluctuations and our exclusion of calm periods remove any day-to-night bias from the measurements. Each of these corrections decreased the calculated NEP by -500 kg C ha -• yr-•, underscoring the need to pursue these biases aggressively when integrating eddy flux. We emphasize that the reliability of eddy covariance at night requires further investigation and that day-to-night biases represent the greatest uncertainty in the use of eddy covariance to calculate NEP. In contrast, the bias of 10-20% in turbulent flux indicated by the energy budget (Figure 4), assuming it was consistent from day to night, would have a small effect on the assessment of NEP. early September, and ceased with freezing in late October (Figure lc). Respiration was low, but not negligible, in winter, increased a few weeks after warming in early June, decreased moderately in early September, and decreased considerably with freezing in late October (Figure lb). The annual cycles of respiration and photosynthesis reflected the direct effects of temperature on metabolism and also changes in the physiological responses of the forest to weather.
Moderate rates of uptake in 1996 were first observed a few weeks after above-freezing maximum temperatures became common, but before snow melt was complete ( Figure 5). Photosynthetic uptake increased through May, with the largest rises following nights with above-freezing air temperatures (e.g., May 11 and 24). The spring increase in respiration was delayed at least 1 week relative to the increase in photosynthesis during each of the 3 years (Figures 5 and la) 6, lc, and 5). These trends were not correlated with any contemporaneous aspects of the physical environment that we could discern and appear to reflect changes in the physiological state of the forest. The relationship between nocturnal respiration and air temperature also changed through the season. Respiration was low and insensitive to air temperature in spring (Figure 7a) and high and uniformly sensitive to air temperature in summer (Figures 7b and 7c). Late summer respiration (Figure 7c The 1 /xmol m 2 s i rise from early to late summer that was insensitive to diel and synoptic changes in temperature suggests increased emissions from a region of stable temperature such as deep in the soil . Nocturnal eddy flux measurements are noisy, and we were unable to find consistent relationships between respiration and aspects of the physical environment other than temperature. The lack of uptake during periods with favorable conditions in early spring (Figures 5 and 6) may be a direct effect of soil and stem frost or, alternatively, of changes in leaf biochemistry associated with dormancy [Larcher, 1995]. Photosynthetic capacity appeared to increase with the onset of above-freezing nights, allowing uptake in mid-May that was approximately 60% of the mid summer rate. The photosynthetic response then increased gradually, and the respiratory response increased rapidly, to reach maxima in early June (Figures 5, 6,  and 7). The comparatively low rates of respiration in late May (Figure 5), despite moderate rates of photosynthesis, may reflect the need for carbohydrate replenishment prior to growth resumption. Both respiratory and photosynthetic response decreased in the first week of September (Figures 6 and 7), possibly due to changes in allocation or shifts in biochemistry in preparation for winter. The timing of this reduction was remarkably consistent from year to year. The increases in response around June i and decreases in the first week of Sep-  Hourly photosynthesis under bright light was negligible at air temperatures below 0øC, increased linearly with air temperatures from 0øC to 14øC, and was relatively insensitive to temperatures above 14øC (Figure 10). Gross photosynthesis was not obviously reduced at high temperature, though wc cannot eliminate the possibility of a modest (1-2 •mol m s -') depression at air temperatures of 25-30øC, since our estimation of respiration during these periods is uncertain (sec Methods). Our measurements do not excludc the possibility of decreased leaf net photosynthesis with warming, since GEE does not include dark leaf respiration (see Methods).
Photosynthesis was not reduced during periods of high evaporative demand (Figure 11), though, as with the analyses at high temperature (Figure 10), we cannot exclude a modest depression. Nonetheless, vapor pressure deficit (VPD) exceeded 4 kPa during only 4% of the daylight summer periods, and a reduction in gross exchange of 2 •mol m ?-s • during these periods would have reduced annual gross production by less than 1%. Similarly, we did not observe obvious effects of soil drought during dry periods in August and September 1994 and June 1995 (Figure l c), a pattern consistent with previous investigations of black spruce [Oechel and Lawrence, 1985].

Empirical Model of CO 2 Exchange
The overall response of CO: exchange to the physical environment was remarkably simple [Sinclair et al., 1976 overall response of CO2 exchange may be so simple that it creates a problem in the use of eddy flux data to validate models, since flawed algorithms may nonetheless match observations over certain ranges. Regression analysis provides a benchmark against which sophisticated models of ecosystem CO2 exchange can be evaluated [/lber et al., 1996]. Complex models that have been tuned to NSA-OBS should explain more than 72% of the variation in hourly exchange. Models that have not been tuned, and therefore are presumably more general, should outperform the regression model from NSA-OBS when applied to other black spruce sites. The physiological responses of the forest determine the direct sensitivity of gross production to climate. Photosynthesis was not limited markedly by high evaporative demand ( Figure  11) or soil water depletion, and modest changes in either aspect of climate should not impact gross production appreciably. Similarly, increased cloudiness should have little effect, since the concomitant increase in diffuse light would largely offset the reduction in light (Figure 8) The simplicity and consistency of CO2 exchange at Harvard Forest and NSA-OBS arguably represent our most significant finding from eddy covariance to date. Nonlinear and spatially variable physiological responses coupled with spatial and temporal heterogeneity in the physical environment could have resulted in complex or seemingly random responses at the whole-forest level. Furthermore, heterogeneity in vegetation associated with local drainage or history could have caused marked variation in exchange as the region of forest sampled shifted with wind and atmospheric stability. However, we found that the physiological responses of leaves and soil microbes scale remarkably well to the stand at both sites and hence that models of CO2 exchange that treat complex ecosystems as simple integrated units are justifiable. In fact, the