The system discussed here was modified from an initial design described by Reece et al. (1995), applied to a stand of A. fasciculatum shrubs. The reader is referred to that paper for details of design, construction, materials and control systems. In this paper we describe our modifications from the original implementation, which were developed during a period of approximately 2 years of field use under a broad range of environmental conditions at the San Diego State University Sky Oaks Biological Research Station. The modifications described here have focused on reducing the amount of structure required to deliver the CO2-enriched air to the treatment area, increasing system reliability and effecting improvements in the software allowing the system to better maintain control of CO2 levels under the range of operating conditions.
The control software was modified from the original version as implemented by Reece et al. (1995). The software uses two algorithms to determine the required rate of CO2 injection. The first algorithm (mixing-model) depends on current wind speed. The second algorithm (PID) is a proportional-integral-derivative algorithm, a measurement and control methodology commonly used in industrial process control applications. Such CO2 control based on wind conditions in combination with a PID approach is conceptually similar to the control approach described by Hendrey et al. (1993) and by Miglietta et al. (1997). Our implementation of the algorithms will now be briefly described.
The first algorithm (mixing-model) uses current wind speed to predict the CO2 injection rate required to meet the downwind target CO2 concentration in the centre of the treatment ring. The algorithm does not use information about current CO2 levels in the treatment area. As implemented by Reece et al. (1995), this algorithm is based on current measured wind speed, ring dimensions and some assumptions about fumigation efficiency, and wind-flow characteristics derived from theoretical analyses described by Reece et al. (1995). Essentially, this algorithm considers a mixing process in which the incoming prevailing wind, at an assumed ambient CO2 concentration, is mixed with the high-CO2 air exiting the upwind vertical pipes. Using the measured wind speed and the effective cross-sectional area of the FACE array presented normal to the current wind direction, an effective incoming ambient air volume flow can be calculated. The product of this volume flow rate and the ambient CO2 concentration can then be used to estimate the CO2 addition required to elevate the incoming air volume to the target CO2 set point. This CO2 injection rate is the output of the mixing-model algorithm.
The second algorithm uses a ‘PID’ control approach. This approach comes from a well-understood body of engineering control and feedback theory and is commonly applied in industrial ‘process control’ applications (Shinskey 1988). In our application this algorithm makes use of feedback from a CO2 analyser that measures CO2 concentration in the centre of the treatment ring. This feedback provides an error term (measured p.p.m. minus target set point) which the PID algorithm tries to reduce to zero. Briefly, in the classical PID approach the control software calculates a proportional term P which depends on the current error, an integral term I which depends on the accumulated sum of the error terms over some previous time period, and a derivative term D which depends on the rate at which the process variable (CO2 p.p.m.) is changing. Thus, the PID algorithm continually adjusts it’s output based on current CO2 p.p.m. error or deviation from the set point (proportional term), on the past history of the error (integral term) and on the rate at which the error is changing (derivative term). The control output, in our case the flow rate of CO2 to be injected into the plenum, is calculated by multiplying each term (P, I and D) by a coefficient and then summing the terms to yield the final PID control output value.
Tuning a PID control system involves determining a set of coefficients for the P, I and D terms which gives the best overall control response. Although there are some engineering guidelines to selecting optimal coefficients (Shinskey 1988), in practice the best coefficients are often determined empirically by observing the system response and adjusting the coefficients. Our initial efforts in using the system focused on determining a set of tuning coefficients which gave the best performance in terms of response speed, minimizing under- and overshoot of the CO2 set point, and minimizing the longer-term deviations of the measured CO2 p.p.m. from the set point. Although the software allows for variable PID coefficients based on wind speed, in practice we use a constant set of coefficients.
The software implementation of Reece et al. (1995) provided for the use of both algorithms combined as a weighted sum of the result from each algorithm. That is, each algorithm was calculated for each time step in the measurement/control cycle and the result from each algorithm was multiplied by a weighting coefficient. The weighting coefficient may be thought of as a gain factor applied to the basic algorithm result. The weighted algorithm results were then added together to arrive at the final control output. However, when applied to a FACE type control problem such as ours, we found that no single set of constant algorithm weights could adequately accommodate the dynamic range of behaviour of our ‘process’. That is, a combination of weights which gave satisfactory control under one limited range of wind speeds often did not work well when wind conditions changed. We felt that the algorithm weights should be made to change adaptively under software control as wind conditions varied.
We then made program modifications which allowed us to enter relationships (not constants) driven by wind speed, for each algorithm weight. This allowed the program to modify it’s control behaviour adaptively according to changing wind conditions. In this approach, we specified a linear relationship for each mass consisting of a minimum and maximum wind speed as the independent variable, and a corresponding minimum and maximum value for the weighting coefficient as the dependent variable. Over the range of the specified relationship, the value of the weighting coefficient depends linearly on the wind speed. Above and below the maximum and minimum specified wind speeds the coefficient assumes the maximum or minimum specified coefficient value, respectively. Specifying the same value for the minimum and maximum coefficient values yields a constant-coefficient system as in the original software.
In actual operation of the control software, an algorithm can be used alone or algorithms can be added together in different proportions to arrive at an overall control result. We primarily control the system using the mixing-model algorithm as a first approximation to the required CO2 injection rate. This algorithm provides fast-response control, because it does not use any feedback from the CO2 concentration measured inside the treatment area. We use the PID algorithm (mainly the contribution of the I term) as a longer-term modifying correction to the result from the mixing-model algorithm. The resulting weighted combination of algorithms allows the system to respond quickly to short-term changes in wind conditions (mixing-model) and also provides the ability to use the feedback data from the measured CO2 (PID) to minimize longer-term accumulated errors. The weights for the mixing-model vary linearly from 1 to 13 over a wind-speed range from 0 to 10 m s–1. The PID algorithm weights vary from 1 to 6·5 over the same wind speeds. In actual use over typical wind speeds between 0·4 and 3·0 m s–1, the mixing-model contributes 55–62% and the PID algorithm 45–38% of the overall control system response, respectively. Similar to the approach of Hendrey et al. (1993) at wind speeds below the stall speed of the anemometer (0·4 m s–1, which can occur sporadically over 10–30 min periods during early morning) alternate vertical pipes are opened around the ring and control is by the PID algorithm only.
A main feature of the original design of Reece et al. (1995) was the so-called ‘dual-array’ configuration where, in addition to the usual configuration of plenum and vertical-pipe array releasing air at elevated CO2 concentration, there was a second, concentric outer plenum and array of vertical pipes releasing ambient air against the prevailing wind direction. The design intent of the outer array was to reduce the incoming momentum and thus reduce the diluting effect of the prevailing ambient air (S. Krupa, personal communication). The volume flow rate of the CO2-enriched air injected into the ring was intended to compensate, at least partially, for the momentum reduction at the outer array and thus help to maintain ambient wind speeds inside the FACE ring.
However, in practice this ‘dual-array’ approach (Reece et al. 1995) demonstrated several disadvantages. Among these was the presence of over twice the amount of structure (compared with a single-plenum design) required to implement the larger outer plenum and vertical pipe array. This additional structure tends to negate one of the main advantages of a FACE approach of minimal environmental alteration. In addition, the predicted (Reece et al. 1995) reduction of CO2 use was not realized in practice.
Prior to beginning regular fumigations, we evaluated the effect of the outer momentum-reducing plenum and pipe array in regard to possible reduction of CO2 use. During the period 12 February to 7 April 1995 we experimented with CO2 set points of 550 p.p.m. and 750 p.p.m. At each set point, we ran the system with, or alternately without, the outer array functioning. These experiments were carried out over a range in wind speeds of c. 0·5–4·5 m s–1, typical of wind speeds at the Sky Oaks site. Data including 1 min mean CO2 concentrations at the ring centre, CO2 injection rates, wind speed and wind direction were continuously recorded at 1 min intervals while the system was operating. At the conclusion of this evaluation period, the accumulated data record was analysed. During periods when the ± 10% performance was being met greater than 80% of the operating time, we categorized the associated CO2 injection rates with respect to whether the outer ring was in operation, or not. We then summarized those periods of CO2 use when the outer ring was in operation vs periods when it was not. Figure 1 is representative of this analysis. As expected, CO2 use increased with higher set point and at higher wind speeds; for example, at a wind speed of 3 m s–1, CO2 use was roughly 3 kg min–1 at the 750 p.p.m. set point compared with about 1·8 kg min–1 at the 550 p.p.m. set point.
Figure 1. . Comparison of the effect of the outer array in operation, or not in operation, on CO2 use over a range of wind speeds at a set point of 550 p.p.m. CO2 (upper graph) and at a set point of 750 p.p.m. CO2 (lower graph). The lines are linear least-square fits to the data sets, as discussed in the text.
Download figure to PowerPoint
However, contrary to initial expectation of Reece et al. (1995), we found that with the outer array in operation, CO2 use increased. Such a result might be inferred from numerical studies of wind flow and distribution of CO2 in a FACE system (Xu, Walklate & McLeod 1997). With the outer array in operation, the outward-directed flow could entrain air from inside the FACE ring and also increase turbulence and vertical transport. Both these effects would act to increase the CO2 use. The overall increase in CO2 use was c. 9% in the 550 p.p.m. treatment and 12% in the 750 p.p.m. treatment, averaged over all wind speeds during our trials. To better quantify these differences, we further analysed the data sets from Fig. 1 and summarize them in Table 1. This summary shows that the increased CO2 use when operating the outer array occurred in all wind speed classes up to 3 m s–1 as well as in the overall mean usage, for both the 550 p.p.m. and 750 p.p.m. treatments.
Table 1. . Summary of CO2 use in our 16-m inner-diameter FACE ring with and without the outer momentum-reduction array in operation, for the data set presented in Fig. 1. CO2 uses are averages over the wind speeds in each wind-speed class during all times of operation. Details are given in the text
Failure to achieve the reduction in CO2 use predicted by Reece et al. (1995) is presumably related to characteristics of the air flow over and through the FACE treatment ring caused by the presence of the additional outer ring structure and to effects of injecting an airflow from the outer vertical pipes into the incoming wind. The system operations log indicates that use of the outer array did not appear to affect the system’s ability to achieve the target set points. Because the main expected contribution of the outer array was a reduction in CO2 use, and because this reduction could not be achieved, we proceeded to remove the outer plenum and pipe array entirely, thus reducing the amount of structure of the system and converting it to a single 16-m diameter toroidal plenum and vertical pipe array. All subsequent results reported here are with the system configured as this single plenum and pipe array.
Following conversion to a single ring configuration, we ran the system continuously under a 13 h day–1 evaluation period from 1 January to 11 March 1997. CO2-use rates were summarized from system data files which included 1 min average CO2 injection rates (kg min–1) during the period. The injection rates were determined using the CO2-flow calibration equation for our BFD flow controller, as described later. Over this 70-day period, our CO2-use averaged 0·352 kg m–2 h–1 when normalized to a unit m2 of ring area. This CO2 use is higher than that reported for FACE experiments using an earlier design of Brookhaven National Laboratory (Hendrey 1993) at the University of Arizona Maricopa Agricultural Center. At Maricopa the CO2 use averaged over a growing season was 1400 kg day–1 (Nagy et al. 1993), operating over 12-h days in a 23-m diameter ring at 200 p.p.m. above ambient (comparable to our 550 p.p.m. conditions). Normalizing to a square meter of ring area, the Maricopa use is 0·281 kg m–2 h–1.
These CO2 usages cannot be strictly compared because site differences involving factors such as prevailing wind conditions, canopy structure (surface roughness and other canopy aerodynamic properties), atmospheric stability and lapse rate, and treatment area will affect differing CO2 requirements. For example, the higher CO2 injection rates at the Sky Oaks site would result in part from higher average wind speeds (1·98 m s–1) during the evaluation period, compared with the 1·01 m s–1 average wind speeds reported by Nagy et al. (1993) at the Maricopa site. We now describe further modifications and additions to the original system of Reece et al. (1995) which were required in order to improve system performance and better meet our experimental objectives.
CO2 FLOW CONTROL
In the original system described by Reece et al. (1995) and implemented at Sky Oaks, the flow rate of the injected CO2 was controlled by two mass flow controllers (MFC) (Model FC-2925 V, Tylan General, Rancho Cucamonga, CA, USA) operating in parallel, each controlled by a separate 12-bit digital-to-analog converter (DAC). A chief feature of the mass flow controllers is that, unlike volume flow controllers, MFC flow control is based on mass (mass time–1) rather than volume (volume time–1). MFCs are normally calibrated to read out in mass-equivalent volume flow rates at standard conditions of 273°K and 1 atmospheric pressure. Two MFCs were required because the maximum CO2 flow rate available in a single unit was 88 kg h–1. Given the size of our treatment area (16 m diameter), the set points we wanted to use (550 p.p.m. and 750 p.p.m.) and the expected range of wind speeds, we calculated that a total of 235 kg h–1 would be needed to maintain set points under wind conditions of up to 3 m s–1. Because the two MFCs connected in parallel still could not provide the required maximum flow, we initially modified the CO2 injection system by adding a ‘boost’ solenoid valve which dumped an additional flow of about 118 kg CO2 h–1, giving a total CO2 injection capacity of about 350 kg CO2 h–1.
CO2 was supplied from a tank of 29 500 kg capacity, passing through an electric vaporizer (30 KW, operating temperature 35 °C) and regulated to c. 3 bars at the mass flow controllers. This set-up performed satisfactorily but we experienced some problems with the MFCs in this application. On occasion, small CO2 crystals, liquid, or other contaminants apparently would form in the CO2 supply line upstream of the MFCs and would unavoidably cause internal damage to the MFCs. Additional filtering on the CO2 supply line reduced, but did not eliminate, these damaging events. Repairs were costly, and resulted in system downtime.
We considered alternative flow controllers in an effort to improve system reliability. Our design criteria included a flow capacity of up to 250 kg h–1 CO2, relatively fast response (better than 10 s), reasonable cost and relative immunity to damage from possible hazards in the CO2 supply line. A motorized metering valve as used by Miglietta et al. (1997) did not have sufficient response speed for our application. The motor-actuated rotary ramp valve used by Brookhaven National Laboratory (Nagy et al. 1993) was too costly. After reviewing various approaches, we decided to design a flow controller which would meet our criteria. We developed a simple and robust device consisting of eight solenoids connected in parallel. Each solenoid controls CO2 flow through an orifice. The orifice cross-section areas were chosen in a binary series, so that each succeeding orifice provides twice the flow of the preceding orifice. The parallel outputs of the orifices are plumbed into a common manifold, and distributed to the injection site in the plenum. The orifice sizes were chosen to provide a maximum flow rate of about 300 kg CO2 h–1 at a supply pressure of 3·4 bars. We can re-scale the flow rates by adjusting the pressure head with a pressure regulator.
While this ‘binary flow divider’ (BFD) can vary its flow only in discrete steps, compared with the MFC which can vary the flow continuously, the 256 flow increments over the full flow range provide more than sufficient resolution in this application. Moreover, the higher dynamic range (the ratio of the highest controlled flow to the lowest controlled flow) of the BFD (255:1) can possibly be an advantage over the MFC used here (10:1), under low wind conditions. Other advantages include relatively low initial cost (approximately US$700), low repair cost (solenoid replacement at US$65), lack of susceptibility to damage by hazards in the CO2 supply line, very fast response (c. 200 ms for a solenoid to open or close) compared with other control valves or MFCs, and simpler control (the flow rate is controlled through a single 8-bit digital output port, compared with the MFC system that required a continuously variable analogue voltage output provided by multiple DACs). The BFD has proved to be very reliable over more than a year of operation, during which time maintenance has involved replacement of four solenoids. Details of construction and materials for the BFD are available from the authors on request.
Because the BFD controller does not provide a flow rate output signal, we calibrated the BFD flow rate as a function of flow command value (0–255), using a rotameter calibrated for air placed at the outlet of the CO2 injection hose and exhausting to the atmosphere. Ten flow commands, from 0 to 180 at increments of 20, were programmed to the computer control port. At each flow command setting, the rotameter flow indication was recorded. After completing the measurements, the rotameter readings were corrected for density differences between CO2 and air, converted to volume flow rates at conditions of 273 K and 1 atmosphere, and then expressed as mass flow rates. The calibration under a 3 bar pressure head (Fig. 2), as used in the system currently, shows a linear flow response. Test calibrations at other higher and lower pressures also showed linear responses. Because we are not accounting for a total carbon balance on the FACE treatment ring, exact flow rates are not required. However, the well-behaved calibration provides good estimates of flow rate, useful for evaluating CO2 use under diverse operating conditions. There is also good agreement between periodic readings of CO2 amount in the CO2 source tank, and the accumulated CO2 use determined from the BFD flow controller (data not shown).
Figure 2. . Calibration graph of the BFD flow controller using a pressure head of 3 bars, typical of operating conditions. Other calibrations at higher and lower pressures showed similar linear responses, with slopes proportional to the pressure head. The Flow Command can be any integer in the range 0–255, sent by the control computer to the BFD control port. The commanded flow can thus be stepped across the full flow range in 255 increments of c. 1·2 kg h-1 each.
Download figure to PowerPoint
We characterized the performance of the FACE system both spatially and temporally. The spatial variation in CO2 concentration was determined in the following manner. Three 14-m line transects were established across the FACE treatment area. Each transect consisted of eight sample points spaced equidistantly at 2-m increments along the transect, plus a centre point, giving a total of 25 sample points. The transects all passed through the centre of the FACE ring, and were displaced at 60 ° intervals. All sample points were at the same 0·75 m height for all measurements, corresponding to the height of the shrub canopy.
At each sample point a vertically orientated sample tube was positioned. A T-fitting was connected to the end of the tube, such that air drawn in to the tubing would tend to be sampled in the horizontal. Each sample tube (equal lengths for all tubes) was connected to the inlet of a two-way normally off solenoid valve. The discharge ports of the solenoid valves were plumbed to a common manifold. The manifold was in turn connected to a pump operating at about 20 litre min-1. By injecting a high CO2 pulse at the sample tube inlets, it was empirically determined that the transit time from sample inlet to the CO2 analyser (model 6261, LI-COR, Inc., Lincoln, NE, USA) was c. 3 s. Under computer control, each of the sample solenoids could be energized in turn. After energizing a sample solenoid, the sample air was drawn through the pump, then filtered and directed to the CO2 analyser. To avoid pressure effects on the gas analyser, a ‘T’ was installed and a subsample of the airstream was passed through a rotameter to the analyser at c. 400 ml min-1. The excess flow was exhausted to the atmosphere.
Using this multi-sample system, the spatial variation in 1 min mean CO2 concentrations was characterized. To begin, the centre sample solenoid was energized by the computer. This sample was drawn for an initial 5 s period to purge the sample path. During the 5 s purge the CO2 analyser signal was not measured. Then, with the sample flow continually drawn through the pump, the CO2 signal from the analyser was read once each second during the following 60 s. These readings were summed and statistics were calculated to determine the mean and standard deviation of the 60 readings at the end of the 60 s sample period. These data (mean and standard deviation along with the date and time) were then written to the data file, the next sample solenoid was then energized, and the above process repeated. The sampling system ran continuously through the daily 12 h operating periods. A complete measurement cycle for all 25 sample points required c. 28 min, yielding 25 1 min means for each sample point over the daily operating period.
The grand means and grand mean standard deviations over an 18-day period from 12 to 29 December 1997 (n = 450 for each sample point) are shown as a spatial distribution (CO2 isolines) across the FACE ring (Fig. 3 upper graph). During these measurements the FACE ring control was based on a continuous air sample drawn at the same 0·75-m height as the sampling tube inlets and the operating set point was 550 p.p.m. CO2. Similar to results obtained by Miglietta et al. (1997) the topography at Sky Oaks is complex, resulting in complex wind-flow patterns. This effect tends to randomize the direction of gradients in CO2 concentration across the FACE ring, such that over time the treatment area with 1 min means remaining within the ± 10% criterion is nearly concentric with the FACE ring itself. The CO2 p.p.m. isolines in Fig. 3 show some effect of preferential wind direction from the north and south, but the ± 10% area (495–605 p.p.m.) appears as a roughly circular area about 11 m in diameter within the 16-m diameter FACE ring. Variability in the 1 min means, characterized here as standard deviations, showed a linear relationship to the mean CO2 concentration (Fig. 3 lower graph), which in turn depended on the position of the sample points. Those points showing higher CO2 concentrations, and associated higher variability, correspond to points located closer to the vertical vent pipes at the periphery of the ring.
Figure 3. . Upper graph: spatial analysis of CO2 p.p.m. concentrations measured along three transects (25 sample points total) across the FACE treatment ring. Details of the measurements are given in the text. The graph shows CO2 p.p.m. isolines of the grand means of c. 450 1 min averages at each sample point, obtained during 18 days of continuous measurements. The CO2 set point was 550 p.p.m. during these measurements, and the average wind speed was 2·2 m s-1. Lower graph: relationship of the 1 min grand mean standard deviation (y-axis) to the grand mean CO2 p.p.m. at a sample point (x-axis).
Download figure to PowerPoint
A representative daily course of 1 min mean values of CO2 concentration, wind direction and wind speed are shown in Fig. 4. The pattern shows typical light early morning winds with variability in direction, changing to higher wind speeds during noon and afternoon, with reduced variability in wind direction. The high concentration excursion that occurred in early morning resulted from reduced ability of the system to control CO2 under very low winds. Typical under these conditions is that a CO2 injection is not quickly mixed and diffused in the treatment area, with the result that any overshoot tends to linger. CO2 control under low wind is the most difficult control condition.
Figure 4. . Daily course of 1 min mean CO2 (p.p.m.), wind speed (m s-1) and wind direction (in degrees). The horizontal dashed and solid lines show the ± 10% and ± 20% limits, respectively, for the 550 p.p.m. set point. Typical reduced control capability during periods of low wind speed combined with changes in wind direction is evident from 07.15 to about 07.30 h.
Download figure to PowerPoint
In addition to the short-term spatial and temporal variation in CO2 conditions reported above, we summarized longer periods of CO2 control performance. From the data files produced each day we calculated 1-min and 10-min mean CO2 concentrations between 1 January and 11 March 1997. We then summarized these data by determining what fraction of the total fumigation time the measured CO2 concentration was within 10% or 20% of the set point. The 10% category would thus correspond to measured CO2 concentration lying in the range 495–605 p.p.m. and the 20% category would correspond to the range of 440–660 p.p.m. This analysis (Fig. 5) revealed that for the 1 min averages the system controlled CO2 to within 10% of the 550 p.p.m. set point 78% of the time, and to within 20% of the set point 95% of the time. The 10 min averages were controlled to within 10% of the 550 p.p.m. set point 87% of the time, and to within 20% of the set point 96% of the time. During this period, such as near Julian day 12, there were times of poor control owing to experimental suboptimal settings of algorithm coefficients, or possible equipment problems. Performance summaries for the Maricopa FACE site (Nagy et al. 1993), averaged over four FACE arrays, showed that they maintained 1 min averages within ± 10% and within ± 20% of the set point during 88% and 98% of the operating time, respectively.
Figure 5. . Seasonal course of CO2 control performance. The graphs show the percentage of operating time during which the 1 min or 10 min mean CO2 p.p.m. measured at canopy height (0·75 m) in the centre of the FACE ring was within the set point of 550 p.p.m. ± 10% or ± 20%. The upper two graphs show the performance of the 1 min means and the lower two graphs are for the 10 min means. As indicated in the lower right-hand corner of each panel, the 1 min means fell within the set point ± 10% and ± 20% during 78% and 95% of the operating time, respectively. The 10 min means fell within the set point ± 10% and ± 20% during 87% and 96% of the operating time, respectively. Reduced performance during days 9 through 12 is associated with a period of experimentation with algorithm settings.
Download figure to PowerPoint
The seasonal course of daily mean CO2 concentration measured in the centre of the FACE ring (Fig. 6) indicates that the concentrations nearly always remained within the 550 p.p.m. set point ± 10%. Over the time period covered in Fig. 6 the daily means departed above the set point by an average of 22 p.p.m. and departed below the set point by – 8 p.p.m., compared with 8 p.p.m. and – 9 p.p.m., respectively, at the Maricopa site (Nagy et al. 1993).
Figure 6. . Seasonal course of the daily average CO2 p.p.m. measured at 0·75-m height in the centre of the FACE ring. The daily averages are within the set point (550 p.p.m.) ± 10% nearly 100% of the operating time.
Download figure to PowerPoint
To evaluate possible physical effects of the FACE structure and operation on air temperatures, we set up 32 type-T (copper-constantan 30 AWG) thermocouples to be measured continuously during an 8-day period beginning on 4 February 1995. Fifteen thermocouples were placed inside the FACE ring (FACE) and 15 were placed outside the ring (ambient). At each location (FACE or ambient) the 15 thermocouples were distributed randomly over the vertical extent of the shrub canopy using a random number generator, covering a region 0·20–0·75 m above the soil surface. During these measurements, the outer array was still in place. The system was not actively fumigating with CO2 but the blowers for the inner and the outer array were set at their normal operating speeds. These measurements thus should reflect differences in radiative and advective exchange between the structure and blowers and the air temperature thermocouples inside the treatment area, compared with those air temperature thermocouples outside the area. Ten-minute means were recorded continuously for 8 days on CR-21X data loggers (Campbell Scientific, Logan, UT, USA) starting at 06.50 h on Julian day 39 of 1995 and continued until 08.40 h of Julian day 47. To correct for any offset between the FACE and the ambient data loggers, one additional thermocouple from each thermocouple array was brought to a common central location and thermally bonded to an insulated aluminum block. This common temperature was then used to correct the data for any measurement offset between the two data loggers.
Analysis of the overall comparison of temperature measurements (Fig. 7) showed that the FACE 10 min means were c. 0·29 °C warmer than the ambient means at a nominal ambient temperature of 0 °C. At 15 °C the FACE means were c. 0·56 °C cooler than the ambient means. We then partitioned the data into daytime (06.00–18.00 h) and nighttime (18.00–06.00 h) periods. Within the daytime or night-time periods we further partitioned the data into 5 °C temperature classes. This comparison, shown in Table 2, indicates that the temperature differences remained less than about 0·3 °C.
Figure 7. . Overall summary comparison of air temperatures measured inside (FACE) and outside (ambient) of the FACE treatment ring, consisting of 10 min means recorded continuously from arrays of 15 thermocouples arrayed over the vertical extent of the canopy at each location, during an 8-day period in February 1995.
Download figure to PowerPoint
Table 2. . Summary of air temperatures measured with thermocouples arrayed through the vertical extent of the shrub canopy (0·20–0·75 m above the soil surface) inside the FACE ring (FACE) and external to the ring (ambient). Details are given in the text
We feel these differences in air temperature probably should not greatly affect gas exchange and growth process differences in our plants. Temperatures of microphyllous leaves, such as those of A. fasciculatum, are closely coupled to prevailing air temperatures (Miller et al. 1981). Slightly warmer leaf temperatures would mean slightly higher transpiration rates (for similar stomatal conductances) during times when FACE temperatures were above ambient temperatures (0–5 °C) and slightly less when daytime ambient temperatures exceed 5 °C. Night-time growth and respiration could be slightly elevated in the FACE plants compared with ambient plants based on the small temperature increase. However, given the relatively small differences we measured, we feel it is more likely that other site differences, such as soil quality and soil moisture, would be more important than the temperature differences we measured here.
At the Maricopa FACE experiment, Kimball, Printer & Mauney (1993) measured canopy temperatures in cotton plants under elevated (550 p.p.m.) CO2 and in control cotton plants at ambient CO2, using infra-red thermometry. Over the bulk of the growing season, mid-morning canopy foliage temperature differences averaged about 0·8 °C higher in the 550 p.p.m. FACE cotton plants, compared with control plants at ambient CO2 levels. The higher FACE temperatures were attributed to partial stomatal closure under the elevated CO2 treatment, resulting in reduced transpirational cooling of leaves. The canopy temperature data of Kimball et al. (1993) are not discussed here as a comparison with our air temperature data; such discussion would be inappropriate given the differing measurement approaches. Rather, we simply point out that some leaf temperature effects, largely owing to CO2–stomatal interactions, are likely in plants subjected to elevated CO2 conditions.
Initially, to gauge some physiological responses of A. fasciculatum held at elevated CO2 levels, daily courses of photosynthesis and water potentials were measured just prior to and following a 6-week CO2 fumigation during the period 21 November 1995–11 January 1996. Prior to the fumigation period, three individual A. fasciculatum shrubs representative of the growth form were selected and three shoots within the shrubs were marked for measurement in the FACE treatment area. Control A. fasciculatum shrubs and shoots were similarly selected and marked in an adjacent area not influenced by the FACE ring. The same shoots on the same shrubs were used for all measurements. Photosynthesis measurements were made at ambient CO2 conditions (nominal 360 p.p.m.) using a LI-6200 Portable Photosynthesis System (LI-COR, Inc., Lincoln, NE, USA). Water potentials were estimated using a pressure chamber (PMS Instrument Co., Covallis, OR, USA).
Prior to beginning the fumigations, daily course photosynthesis measurements showed FACE A. fasciculatum shrubs with net photosynthetic rates consistently higher throughout the day compared with the control shrubs, with maximum rates of about 2·5 μmol m-2 s-1 (Fig. 8, upper left-hand panel). Predawn water potentials were lower (– 3·5 MPa) in the control shrubs compared with the FACE plants (– 2·5 MPa), suggesting drier soils for the control plants. These water potential differences persisted until mid-morning. The balance of the day showed similar water potentials for both control and FACE plants. Following the period of CO2 fumigation at 550 p.p.m., the photosynthetic and water potential measurements were repeated on the same shrubs and shoots, at ambient (360 p.p.m.) CO2 conditions on the FACE and on the control shrubs (Fig. 8, right-hand panels). Here, in contrast to the prefumigation measurements, the FACE shrubs exhibited consistently lower photosynthetic rates (0·8–1·6 μmol m-2 s-1) compared with the control shrubs (2–3·5 μmol m-2 s-1). Consistent differences in water potentials were also revealed, with the control shrubs showing the lower (more stressful) values through the daily course. This pattern would be consistent with a scenario in which reduced stomatal opening induced by the higher CO2 levels in the FACE ring may have reduced transpiration by FACE plants compared with control plants, resulting in a more favourable leaf water status (higher potentials) in the FACE plants. While these data from extant, naturally occurring individuals are preliminary, they suggest, in agreement with earlier work (Oechel et al. 1995), that A. fasciculatum, a widespread and often dominant component of the southern California chaparral vegetation (Hanes 1977), responds readily to elevated CO2 environments.
Figure 8. . Pre-fumigation (left-hand panels) and post-fumigation (right-hand panels) daily courses of net photosynthetic rate and water potential measured in the same shrubs of Adenostoma fasciculatum growing inside the FACE treatment ring or in adjacent control A. fasciculatum shrubs growing outside the FACE ring. Error bars are the standard errors (n = 3) in the means from the measured shrubs. Error bars smaller than the symbol size are not shown.
Download figure to PowerPoint