Notice: Wiley Online Library will be unavailable on Saturday 27th February from 09:00-14:00 GMT / 04:00-09:00 EST / 17:00-22:00 SGT for essential maintenance. Apologies for the inconvenience.
A high-frequency response relaxed eddy accumulation flux measurement system for sampling short-lived biogenic volatile organic compounds
Robert R. Arnts,
National Exposure Research Laboratory, Human Exposure and Atmospheric Sciences Division, US EPA, Research Triangle Park, North Carolina, USA
Corresponding author: R. R. Arnts, National Exposure Research Laboratory, Human Exposure and Atmospheric Sciences Division, US EPA, 109 T W Alexander Drive, Mail Drop E205-03, Research Triangle Park, NC 27711, USA. (firstname.lastname@example.org)
 A second-generation relaxed eddy accumulation system was built and tested with the capability to measure vertical biogenic volatile organic compound (VOC) fluxes at levels as low as 10 µg C m−2 hr−1. The system features a continuous, integrated gas-phase ozone removal procedure to allow for the measurement of highly reactive species such as β-caryophyllene and polar terpenoids such as linalool. A two-component internal standard continuously added to the accumulators was used to correct for switching-induced volumetric errors and as a check on VOC losses exceeding accumulator tube adsorption limits. In addition, the internal standards were used to demonstrate that accumulators quickly return to target flow rates at segregation valve switching frequencies up to at least 0.8 Hz. The system was able to measure daytime hourly fluxes of individual biogenic VOC including oxygenated terpenoids, monoterpenes, and sesquiterpenes.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 Relaxed eddy accumulation (REA) or conditional sampling theory has provided an important avenue to the measurement of trace gas area source atmosphere-biosphere exchange. While eddy correlation methods are the preferred technique when fast (~10 Hz) chemical sensors are available (usually for a single component), REA can utilize slower sensors to measure single or multicomponent fluxes [Businger, 1991]. With REA, updrafts and downdrafts are sorted into collection reservoirs over an integration period (typically 30–60 min) sufficient to sample a representative range of eddy sizes. If the target analyte(s) are not sufficiently stable or easily contaminated, on-site analysis may be performed. When the analytes are stable in the reservoirs, analysis may be performed in the laboratory where stable environmental conditions permit application of sophisticated separation and detection methods.
 REA theory [Businger and Oncley, 1990], derived from eddy correlation theory, states that the vertical flux F can be determined by the product of a constant, β, times the standard deviation of the vertical wind velocity σw (where the mean vertical wind velocity is zero) and the difference between the mean updraft concentration and the mean downdraft concentration :
Furthermore, the sorting of up and downdraft samples shall be performed at constant flow rate. While the vertical wind speed component can be easily measured using commercial sonic anemometers, the determination of the concentration differential poses three significant challenges. First, the sorting of updraft and downdraft samples into their respective accumulators should be performed without cross-contamination from the other direction. Since the decision to direct the sample to one direction or the other requires a finite time period to measure the wind velocity and direction and another time period to compute and execute a sorting command, sorting errors can arise due to the time lag. The second major challenge is to quickly establish a constant flow rate in each accumulator when switching between accumulators. Without constant sampling, eddies will not be equally sampled, and the resulting concentration differential may be biased. The third major challenge in measuring biogenic volatile organic fluxes is the loss of highly reactive compounds in the accumulators due to reaction with co-sampled ambient ozone.
2 Previous Design Approaches
2.1 Directional Sorting
 General guidance for measuring fluxes in the boundary layer specifies that eddies should be sampled at 10 Hz to ensure accurate characterization of the scalar gradient. A review of previous REA systems reveals widely differing approaches to achieving this goal through hardware and software designs. These stem from consideration of issues of analyte stability/storability, the selection of an appropriate analytical technique and the resulting sample volume required to accurately measure the analyte concentration differential. With systems herein referred to as Type I designs, the accumulator samples air from the anemometer measurement zone concurrently with measurement of the vertical wind velocity. Consequently, any decisions to start or stop directional routing will be late in implementation and improper binning of the samples will always occur for which no correction can be made. Whether or not this causes a significant reduction in measured flux is largely dependent on the relative distribution of small and large eddies measured during the flux measurement period. If the surface is rough as observed over forest canopies, the most significant eddies tend to be large, resulting in long accumulation intervals (10s of seconds). Under these conditions, improper sorting of samples for one to several 100 ms increments relative to much longer accumulation periods may not significantly impact the measurement. However, if the surface is smoother such as over agricultural crops, mean accumulation intervals will tend to be much shorter and the percentage of improperly sorted samples relative to the properly sorted samples will increase. Numerous examples of Type I design can be found [Valentine et al., 1997; Ciccioli et al., 2003; Darmais et al., 2000; Pryor et al., 2002; Nemitz et al., 2001; Majewski et al., 1993; Baker, et al., 1999; Beverland et al., 1996; Schery et al., 1998; Amman, 1998; Graus et al., 2006; Meyers, et al., 2006; Gaman et al., 2004; Haapanala et al., 2006; Skov et al., 2006]. These are often used where it is critical to avoid analyte losses to inlet tubing and valves—e g., nitric acid, ammonia, reactive gas mercury, and particles [Pryor et al., 2002; Meyers et al., 2006; Skov et al., 2006; Schery et al., 1998; Nemitz et al., 2001; Gaman et al., 2004; Baum and Ham, 2009]. The impact of improper binning on flux accuracy for any set of measurement conditions has been estimated for some of these designs by numerically simulating the effect of a given lag on co-collected sensible heat flux data [Amman, 1998; Beverland et al.1996; Schery et al., 1998; Graus et al., 2006; Majewski et al., 1993; Meyers et al., 2006].
 In Type II REA system designs, a constant flow inlet tube(s) prior to the accumulator routing (segregation) valves can be used to create sufficient delay to synchronize directional transitions with accumulator routing. There are several examples of such systems using plastic bags to collect sample [Oncley et al., 1993; Pattey et al., 1993; B. Baker et al., 1999; Haapanala et al., 2006; Delon et al., 2000; Brut et al., 2004], extractive-type (adsorbent tubes) systems [Nie et al., 1995; Olofsson et al., 2005] and on-line analysis [J. Baker et al., 1992; Cobos et al., 2002; Valverde-Canossa et al., 2006; Bash and Miller, 2008; Schade and Goldstein, 2001].
2.2 Achieving the Constant Flow Rate Sampling Requirement
 Since all REA system designs require valves to start and stop sampling to implement sorting of updraft and downdraft samples, it is essential that the pressure differential across the valve inlet(s) and exit be equivalent regardless of their position. If the two valve state (sampling and not sampling or neutral gas) pressure differentials are not equal, then each valve transition will cause flow surges or back surges to stabilize pressures thus violating the constant flow requirement. This issue can be most severe in systems where no neutral (below threshold) channel is used and the flow is completely stopped and started through the segregation valves. Systems using whole air accumulators, such as Summa cans or polymer bags where it is not desirable to dilute the sample with zero air, are more susceptible to this issue. REA systems using extractive-type accumulators which pass the bulk air components, such as adsorbent tubes, coated denuders, or filter packs, allow system designs of continuous flow of sample and analyte-free zero gas. Thus, constant flow can be more easily achieved through judicious placement of three-way valves, flow controllers, and pumps with extractive-type accumulators.
2.3 Valve Switching Transients
 Common to all REA design types are accumulator volumetric errors introduced by flow interruption caused by valve switching. Depending on the valve design, flow may be completely or partially restricted as the actuating mechanism moves from one state to the other [Nie et al., 1995]. Typically, upon activation, the actuator held in place by spring tension is moved by application of a magnetic field supplied by a coil. The turn-on response time, as defined by the valve industry, is the time between the application of the coil energizing voltage until the downstream pressure reaches 90% of the applied pressure. This time may be further decomposed to (1) an initial lag between the start of the applied coil voltage and the start of the valve plunger movement (start of flow disturbance) and (2) the completion of the movement of the plunger to its energized state. If the duration of the initial lag is known, this can be accounted for by synchronizing the arrival of eddy samples at the segregation valves. The lag and duration of the second part may be reduced through application of programmed voltage-switching signals (high triggering voltages followed by lower holding voltages). The valve response to switching off the coil voltage is usually not symmetrical with the on response. The return of the plunger is governed by the collapse of the coil magnetic field and the spring tension on the plunger. In a two-segregation or three-segregation valve system, this can lead to multiple valve ports being open during part of the switching cycle. While designers of other REA systems sometimes note a valve response time and perhaps a synchronization delay to offset the electronic-flow delay, the disturbance to flow constancy and volumetric accuracy impact of the unavoidable flow transient are not explicitly discussed.
2.4 Analyte Losses
 The absorptivity and reactivity of the target analytes help determine the choice of materials that will contact the sample flow path. Fluorinated polymers such as Teflon are often selected due to their low absorptivity and freedom from out-gassing of volatile organic compounds (VOCs). As a consequence, ambient ozone which is often present at levels of 20–100 ppbv is also efficiently conducted through the REA plumbing. Since ozone is quite reactive toward many olefinic hydrocarbons, especially many monoterpenoids and sesquiterpenoids, losses can result in the plumbing and accumulators. These losses can be severe, particularly when the accumulators are of the extractive type, such as a porous polymer adsorbent, e.g., Tenax-TA. In this case, the surface buildup of a reactive compound on an adsorbent, continuously bathed in co-sampled ozone, creates more opportunity for loss than in the gas phase. One remedy is to insert an ozone-removing filter upstream of the accumulator such as several plies of manganese dioxide-coated copper screens [Ciccioli et al., 1999; Christensen et al., 2000; Darmais et al., 2000; Graus et al., 2006; Haapanala et al., 2006], KI [Park et al., 2010] or sodium bisulfite-coated filter paper [Helmig, 1997; Pollmann et al., 2005]. However, with less volatile or polar compounds, these filters may also remove the target analytes [Hoffman, 1995].
 Our second-generation type II REA system described herein incorporates (1) synchronous routing of 100 ms eddy capture samples to extractive-type accumulators using continuous flow from either sample or neutral (zero) gas, (2) internal standard gases added to the sample stream to allow for volumetric corrections made necessary by valve switching transients, (3) an integrated gas-phase ozone removal system that reduces VOC ozone loss to negligible levels, and (4) continuous data logging of REA operations and micrometeorological sensor array signals.
3 Design Criteria
3.1 Matching Flux Detection With Sample Volume and Analytical Method
 To design an appropriate flux measurement system, we must first answer the following questions: What compounds do we want to measure? At what rates do we expect these compounds to be emitted to the atmosphere? Global estimates of biogenic nonmethane VOC emissions have identified isoprene as the single largest (~35–50%) contributor to the budget [Guenther et al. 1995; 2006]. Uncertainty of those estimates have been significantly reduced as a result much research over the past 20 years. However, the balance of the VOC budget is comprised of contributions from many compounds, especially those broadly classified as monoterpenoids and sesquiterpenoids and which are recognized as significant contributors to formation of secondary organic aerosol. These estimates, largely derived from branch enclosure measurements, are assigned large errors due in part to foliage disturbance issues surrounding these measurements. Hence, our objectives here are to focus on those compounds which would benefit from nondisturbing micrometeorological-based measurements. A survey of reported flux estimates for major compounds within these classes finds fluxes in the range of 10–10,000 µg C m−2 hr−1 over forests and croplands [Guenther et al., 1995; Rinne et al., 2000; Helmig et al., 1998; Gallagher et al., 2000; Valentini et al., 1997; Plaza et al., 2005; Ciccioli et al., 1999].
 To measure fluxes as low as 10 µg C m−2 hr−1, the sampling and analytical technique employed must accurately measure the mean concentration differential of these compounds between the up and down accumulators. Businger and Delany  found that the sensor resolution, R, necessary for a flux measurement accurate to ±10%, can be determined from the equation:
where Fc is the flux and AP is the atmospheric parameter. AP is defined as (βσw)−1 which ranges from 2 to 6 s m−1 when β = 0.58. To facilitate easier measurements, the mean concentration differential may be increased by rejecting vertical velocities below 0.6σw (small eddies) and using a reduced β of 0.39 (computed according to Baker ). Thus, a new AP range from ~3 to 9 s m−1 is computed. Figure 1 illustrates the estimated chemical resolution needed to measure fluxes during optimum daytime mixing conditions. Gray areas to the right of the solid squares at a resolution of ~1 ppt C (0.5 ng C m−3) indicate fluxes ranging from about 2 to 6 µg C m−2 hr−1 should be measurable depending on mixing conditions.
 To resolve 1 ppt C, sufficient sample volume must be collected to deliver target analytes to the analytical detector within the range of accurate quantitation. Typically, this is defined as the limit of quantitation (LOQ) where the signal must be greater than 10 times the noise level of the detector. For our gas chromatographic system (model 5890 Hewlett-Packard gas chromatograph with flame ionization detection), approximately 15 L of sample are required to align the chemical resolution requirement with the LOQ of the chromatographic detector.
3.2 Accumulator Sizing
 Candidate accumulator designs must be able to deliver the analytes to the chromatographic system while preserving analyte integrity. Extractive-type adsorbent-based accumulators can efficiently retain target analytes from large volumes of air and can be integrated into system designs which recover quickly from switching transients and thus meet the constant flow sampling requirement. In choosing the proper adsorbent or adsorbent combination, it is important to consider the factors which govern retention of analytes on adsorbents. Adsorbents typically possess uniform, high specific surface areas (area/mass) and are chosen by their attractive properties to the analytes and their lack of retention of the bulk air components, e.g., O2, N2, CO2, and H2O. In addition, they must be capable of releasing these compounds without alteration when the adsorbent bed is heated in the laboratory.
 Passage of an air stream containing a mixture of analytes of differing volatility and polarity through an adsorbent bed creates a competition for the available surface area of the adsorbent. Compounds most attracted to the adsorbent surface are deposited first while less attracted compounds move further into the adsorbent bed. As sampling continues, more surface attractive compounds displace less attractive compounds. If the sampling temperatures rise above laboratory conditions, the rate of migration for any adsorbed compound will increase at approximately twice the rate for every 10°C rise in temperature. In addition, all adsorbed compounds will slowly migrate in the direction of the gas flow in proportion to the flow rate. Thus, sufficient adsorbent must be used to account for the total sample volume (sample + zero gas) and any temperature excursions, both field and laboratory, that would cause deeper penetration of the analytes into the adsorbent bed. Our system was set up to operate for a 55-min flux integration period at 1.00 L min−1 or 55 L total (sample + zero gas). With a 0.6σw velocity filter, each accumulator will collect sample 30% of the time (55 L/0.3 = 16.5 L). Thus, the adsorbent tube was sized to retain 55 L plus additional adsorbent to ensure against a 10°C elevation (35°C)—for a total of 110 L. As a quality control measure to ensure against exceeding these design limits, we used a set of internal standards consisting of methylcyclohexane and 2-methylheptane added continuously to the inlet sample line. These are less well retained by the adsorbent beds than the least well-retained prominent monoterpene (methylcyclohexane < 2-methylheptane ≪ α-pinene). If the 2-methylheptane/methylcyclohexane ratio increases above the introduced ratio (~1.2), this may indicate that breakthrough of methylcyclohexane is occurring.
 Extensive evaluations of Tenax and graphitized carbon adsorbents were carried out to evaluate suitability for capture and recovery of biogenic compounds [Arnts2010]. The popular graphitized carbons were found to be unsuitable because of a tendency to catalyze transformations of olefins and alcohols. Tenax-TA, however, was found to retain and thermally release these compounds without the artifacts of the graphitic materials. To determine sufficient VOC breakthrough characteristics of Tenax-TA, multicomponent mixtures of terpenes, sesquiterpenes, aromatics, and alkanes were sampled by small adsorbent tubes at various sample volumes. The tubes were then thermally desorbed and analyzed to determine VOC recoveries. The least well-retained biogenic compound was α-pinene. The maximum volume without breakthrough of the α-pinene was used to scale up from the small (5.33 mm ID) to the larger REA tubes (12.7 mm ID) using the same linear sampling rate (13.2 cm s−1). The Tenax-TA bed depth was increased by a factor of 1.5 (5 g) and thus, up to 119 L (at 25°C) could be sampled with the REA tubes. The Tenax-TA bed was held in place by 2 micron pore size stainless steel wire mesh disks that have been passivated using the Sulfinert process. A second adsorbent bed of friable Carbotrap (see following discussion) was retained by a compressed wafer of stainless steel wool. Glass and silica wool bed retainers, which are usually used in adsorbent tubes, were found to take up unacceptable quantities of water.
3.3 Controlling VOC Losses From Ozone
 Canopy-level flux measurements are representative of the actual biogenic emission to the atmosphere only if there are no sources or sinks occurring during transport to the point of measurement. With transport time scales on the order of 30–90 s, most of the biogenic terpenoids will be diluted but not significantly reacted. However, some compounds such as γ-terpinene and β-caryophyllene can be drastically oxidized in the presence of ambient ozone. In addition, the unreacted portion of those compounds, surviving the transport to the flux measurement point of collection, can be further lost if ambient ozone is allowed to contact the VOC adsorbed on the surface of the accumulator substrate, i.e., Tenax-TA. Since the probability of ozone-VOC contact increases as the VOC is accumulated on the adsorbent surface, the rate of loss increases [Arnts, 2008] offsetting the advantage of larger sample volumes. Transport losses due to ozone can be estimated through modeling if ambient ozone concentrations and accompanying turbulence measurements are available. Residual VOC sampled by the REA inlet can be preserved by immediately blending in another VOC to scavenge the co-sampled ozone. To do this, trans-2-butene (10% in nitrogen at ~100 cm3 min−1) was continuously added to reduce competition of terpenoid for ozone to a negligible level. To allow sufficient time for the scavenging process to occur, the sample passes through heated (40°C) inert Sulfinert coils (5 s residence time). Trans-2-butene is not efficiently retained by the accumulator adsorbents (Tenax-TA/Carbotrap) and thus passes through the accumulator tubes.
3.4 Volumetric Corrections for Switching Transients
 Despite a design which facilitates rapid switching of sample-zero gas and minimal flow disruptions, there remains an unavoidable switching transient associated with the time required to change valve states. We tested a Teflon-body three-way solenoid valve (model 225T031, 12 VDC, Neptune Research, Caldwell, NJ, USA) using a fast hot wire anemometer (Series 50, Thermo Systems Inc., St. Paul, MN, USA) and an oscilloscope. These tests indicated that actuator movement started about 6 ms after start of the triggering voltage and completed about 3 ms later. Upon cessation of the applied voltage, the valve returns to its ground state driven solely by the tensioning spring on the actuator. This return is not as crisp as the coil driven opening. The impact of these transients on volumetric accuracy and hence on flux accuracy is not constant since the duration of the sample periods will vary. At high switching frequencies (≤10 Hz), the relative contribution of the disturbance is larger for short periods than for longer periods of sampling where stasis is undisturbed until the next switch.
 In the actual field system, the switching transients are not detected by the mass flow controllers due to the dampening effect of the reaction coil, tubing, and accumulator volumes coupled with the inability of the flow controllers to measure and respond to short-term fluctuations (manufacturer specification: <2 s to within 2% of set point for MKS model 1100 mass flow controllers). Thus, the inability of the mass flow controllers to accurately record the instantaneous flow, coupled with valve opening-closing asymmetry discussed above, may lead to inaccuracy in volume determination based on mass flow controller logs alone. To remedy this, we used a standard analytical chemistry approach of adding a known quantity of internal standard to the sample and assaying the resulting concentration to calculate the volume of the sample. To implement this, we continuously introduced a stable, relatively inert internal standard gas to the main sampling stream (methylcyclohexane/2-methylheptane mentioned earlier). These compounds are not present in the zero gas (nitrogen) or in ambient air above 1 ppb C, are not reactive with ozone, and are easily separated by gas chromatography. Note that the Tenax-TA in the accumulator tubes will retain most of methylcyclohexane and 2-methylheptane, but some breakthrough will occur. To guard against this, we added 1.5 g of Carbotrap adsorbent to the rear of the REA tubes to capture any residual internal standard. Both compounds are easily captured and recovered by thermal desorption. To obtain the corrected REA sample volume, we compared the resulting internal standard quantity in switched sample (flux measurement) with that of the result of a nonswitched measurement (continuous, constant flow rate sampling) of known volume. Since we also know the concentration gases in the internal standard cylinder and the flow rates of the internal standard into the inlet and inlet flow rate, we have the second method to calculate the nonswitched internal standard concentration and thus a second way to calculate the corrected volume of the switched (flux) samples. Park et al.  also used an internal standard in a similar manner to adjust for channel bias.
4 System Design and Operation
4.1 Gas Handling System
 Figure 2 illustrates the key components of the gas management system. Air is sampled at 13.5 L min−1 from the plane of the vertical wind velocity measurement zone (about 20 cm behind and about 6 cm above the sonic sensor measurement zone) through an open face, 47 mm, Teflon membrane (20–30 micron) particle filter (PF). Flow rates and tubing diameter were selected to maintain turbulent conditions (Reynolds number ~5000) and promote minimal attenuation of concentration fluctuations [Lenschow and Raupach, 1991]. The sample flow is immediately mixed with internal standard and an ozone scavenger to produce approximately 25 ppb C each of methylcyclohexane and 2-methylheptane and about 600 ppm of trans-2-butene in the sample stream. The segregator (module) consists of two Teflon union crosses and three sample valves where 1.0 L min−1 is drawn from the main flow by one of the three valves (SVL, SVN, SVR). During the 400 ms required to reach the segregator valves from the inlet, the data system determines which path is appropriate for each arriving 100 ms segment of inlet air. The segregated air samples then pass through a heated (40°C) Sulfinert-treated (passivated) stainless steel coil (RCL or RCR) to allow sufficient time (5 s) for ozone scavenging by trans-2-butene. The ozone-depleted sample then passes through a six-port Teflon manifold valve, MV (model 18-132-900 General Valve, Fairfield, NJ, USA or model 225T091 Neptune Research, West Caldwell, NJ, USA) which routes the sample to the current adsorbent tube accumulator. The sample path is heated (40°C) from the PF inlet through the upstream manifold valves (PF-HP-SVs-RCs-MVs). In addition, the internal standard gas supply lines were heated to 40°C from the gas cylinder to the sample air inlet. The accumulator tubes are clamped into an aluminum manifold which is cooled by Peltier-type thermoelectric coolers.
 The three segregator valves are also supplied with (VOC free) nitrogen (ZG) at 2 L min−1 so that the two valves not being sampled will maintain a continuous gas flow of 1 L min−1. Pressure sensors (PSM and other unlabeled black squares) upstream and downstream of the valves allow the operator to minimize pressure differentials which would cause over or under sampling following a switching event.
 The mass flow controllers are housed in a separate, insulated temperature-controlled stainless steel box to minimize drift. To reduce mass flow dependency on humidity, the sample passes through silica gel desiccant trap (not shown) before entering the flow controllers [Lee, 2000]. Fluorocarbon polymers used in mass flow controllers were found to be swelled by contact with trans-2-butene, so capillaries and metal frits were used to govern flow. Ten percent trans-2-butene in nitrogen was prepared from 99+% grade (Sigma-Aldrich) by dilution in a high pressure aluminum cylinder.
4.2 Electronic Systems
 The electronics for the REA measurements and the ancillary hardware necessary for energy balance and atmospheric stability measurements are outlined in Figure 3. While much of the system was under computer control, several systems operated autonomously. These include the four individual temperature regulators mounted in the REA box on the tower: (1) the inlet line, (2) the segregator module and manifold valves, and (3) and (4) the reaction coils which were all set to 40°C using PID-type controllers (Watlow, Richmond, VA, USA). The shed, near the base of the tower, holds the Supervisor computers, the Controller computer with Data Packer 4 (DP4), the mass flow controller's power supply/readout units (model 247C, MKS Instruments) and the atmospheric pressure sensor.
 In order to affect uninterrupted segregation valve decision making, i.e., avoiding computer operating system interrupts, a specialized printed circuit board was developed by Hampton Technologies to generate sonic triggering commands using a precision on-board clock. The sonic anemometer (model SAT-3V, Applied Technologies Inc., Longmont, CO, USA) was operated at 10 Hz and set to external triggering. After triggering, the sonic performed 100 ms averaging in the first time frame, computed the three velocity components and temperature in the next 100 ms, and reported them to Data Packer 1 (DP1) in the third 100 ms time frame. Simultaneously, DP1 received analog data from the H2O/CO2 analyzer and performed parallel 100 ms averaging. DP1 reported the resulting data back to the Controller computer. Simultaneously, DP2 performed 10 s averaging of voltages from a pyranometer, net radiometer, a PAR sensor, a vertical array of precision thermocouples, segregator valve zone pressure sensors, and accumulator tube manifold temperature. DP3, at ground level, also performed 10 s averaging of voltages from soil heat flux sensors and soil thermocouples while DP4 performed 10 s averaging of seven mass flow controller outputs and of an atmospheric pressure sensor. The data packers avoided analog voltage drops associated with long field cables via use of digital communication by either serial binary or ASCII code. DP data packets were received by the Controller computer which correlated the various averages with the proper time stamp before writing the data to a file and forwarding the data to the supervisor computer.
4.3 Data Acquisition and Control System
 The integrated data acquisition and control system was built by Applied Technologies Inc. with the primary sonic monitoring and valve control architecture evolved from a system initially developed by one of us (Hampton) at the NCAR and further revised by Hampton Technologies Inc. (Longmont, CO). The system used two personal computers: (1) a primary controller (MS-DOS based) which interacted directly with the sonic anemometer, segregation valves, and data packers and (2) a supervisory computer which recorded the data stream from the controller. The controller coordinated both the sonic data stream and all of the Data Packer communications, sorted them to their proper files, and coordinated assignment of proper time stamps. Separate fast (10 Hz) and slow (0.1 Hz) files were created and passed to the supervisor PC where calibration factors were applied to the raw data and then converted to appropriate units for recording to files and for real-time graphical displays.
 The supervisory PC REA interface program allowed the operator to perform a variety of functions including entry of sensor calibration factors, real-time graphical display of all current sensor outputs, internal pressure balancing across the segregator valves, entry of site conditions, accumulator tube leak checking, entry of sonic-valve synchronization delay time, entry of time base for rolling vertical velocity statistics, entry of accumulator tube tracking numbers, flux measurement duration, and percent of σw to exclude from sampling. In order to keep the accumulator tubes within analyte adsorption limits on hot summer days, the operator could enter tube manifold temperature set points to maintain temperatures (typically 25°C) of the tube arrays (no lower than 5°C above the dew point to avoid condensation). The real-time displays of radiation, wind direction-wind speed, and vertical temperature allowed the operator to judge whether conditions were conducive to flux measurements and to observe the environmental conditions during data collection.
 To address the previously stated issue of the need to achieve equivalent pressures across the two upstream ports of the segregator valves, the zero gas mass flow controller was calibrated so that the zero gas was supplied at twice the sample flow rate. Thus, during flux measurements, all segregator valves had flow rates of 1 L min−1 with one segregator valve drawing a sample while the other two drawing zero gases. To check this and make any fine tuning adjustments to the zero gas flow, the supervisory PC allowed the operator to perform a dynamic switching test to observe the equivalency of pressures across the zero gas and sample side ports of the segregator valves. The operator could select any switching interval from 100 ms to 10s of seconds. By observing the pressures on both sides of the valves during a switching cycle, the operator could fine tune the mass flow controllers to achieve pressure equivalence.
 After completion of calibrations and system QA checks, up to six sets of fresh accumulator tubes could be loaded into the manifold. A leak checking routine was available to verify all tubes were securely installed. Finally, the operator entered tracking numbers for each tube and their associated manifold port position (left or right side and port number), the duration of the measurement (usually 55 min for a flux measurement), the type of measurement (flux, system balance, or contamination), and the delay (if any) between measurements. For flux runs, the percentage of σw to exclude from sampling was also entered. When the data acquisition was started, the system created data files with names that include the start time. Simultaneously, it began triggering the sonic anemometer and receiving data until the specified buffer (usually 5 min) had filled and the starting mean vertical velocity and standard deviation had been established. The first set of tubes was automatically selected and randomly assigned as up or down for the length of the measurement period (a check on channel bias). Sorting the inlet gas into up and down draft samples then began based on using the percentage exclusion of the rolling σw .The valve actions were lagged according to entered synchronization delays. During the measurement period, the operator could observe all mass flow rates, radiation profiles, and wind direction/velocity distributions as well as view the rolling vertical velocity with imposed velocity exclusion thresholds. At the conclusion of each measurement, the associated files were closed and the next run was started.
4.4 Field Measurement Conditions
 The REA system was set up on a 26 m walk-up tower located in a Loblolly pine (Pinus taeda) plantation at the Duke University Blackwood Division Free Atmosphere Carbon Exchange (FACE) site near Chapel Hill, NC. The gas inlets and the sonic anemometer were mounted at the end of a boom supported by a 3 m mast positioning the inlet 3 m away from the tower at the 24 m level (canopy height ~20 m). The understory canopy consisted of sweetgum (Liquidambar styraciflua), tulip poplar (Liriodendron tulipifera), red maple (Acer rubrum), and dogwood (Cornus florida). A temperature-controlled instrument shed near the base of the tower was used to house the data acquisition and control system and mass flow controller electronics. The REA system was operated on days selected for optimum predicted meteorological conditions throughout 2007 over temperatures which ranged from 0 to 37°C. A full discussion of those results and their value in estimating emission model accuracy will be the subject of a companion paper. Measurements were set up to integrate fluxes over a 55 min period with up to six consecutive runs made before reloading the system with fresh sample tubes. The inlet sample line (PFA Teflon, 5/32″ ID × 7 m) and total flow rate were selected to match the 406 ms system delay (400 ms DAQ time delay plus another 6 ms to account for the valve response time delay) in processing the data stream from the sonic anemometer. The accumulator flow rates were set to 1.00 L min−1. Sampling decisions for activation of the eddy capture valves were based on exceedance of a rolling minimum vertical velocity: (1) the 300 s rolling mean and standard deviation of the vertical velocity (σw) were computed, (2) the current velocity is corrected for the rolling mean velocity, and then (3) compared with 0.6σw. The velocity threshold was used to increase the concentration differentials of the VOC by excluding small eddies which usually have a minor contribution to the total flux. The use of the rolling threshold was intended to help remove out of plane horizontal wind contributions to the vertical velocity component. The 5 min duration was deemed long enough to not exclude important eddies but sufficient to remove horizontal wind contamination of the vertical velocity resulting from sloping fetches.
 In addition to the flux measurements, 1 h (60 L) continuous samples were periodically drawn for gas chromatographic-mass spectroscopic analysis to aid in identification of emissions. The larger volume was necessary to compensate for the lower full scan mass spectrometer sensitivity compared to the flame ionization detector. Also, a parallel emission study of the loblolly pine foliage using Teflon bag enclosures without the presence of ozone was conducted and used to help link ambient measurements to their source [Geron and Arnts, 2010].
4.5 Analysis of Accumulator Tubes
 After sampling, tubes are capped and returned to the laboratory for analysis. A custom-built thermal desorption apparatus was constructed to process the nonstandard large-size tubes [Arnts, 2010, supplementary materials therein]. In brief, the system allowed up to 16 tubes to be thermally desorbed, cryogenically focused, and injected into one of the two gas chromatographs for quantification (flame ionization detection) and identification (mass spectral matching with the NIST/EPA data base). The sample path through the system was passivated with Restek's Sulfinert process and heated to prevent sorptive losses. The processing of the tubes includes steps to remove residual water vapor and trans-2-butene before high temperature thermal desorption. Also, an n-alkane internal standard was added to aid in retention time adjustment and peak identification. The flame ionization detector was calibrated with a Scott prepared mixture of isooctane and 2,5-diemethylhexane in nitrogen which was calibrated against a NIST-certified propane in nitrogen standard. Based on mass spectral analysis of field samples, pure samples of the tentatively identified compounds were obtained and used to prepare dilute mixtures in the laboratory. These mixtures were then sampled and analyzed to compare with the field samples. Retention time matches provided additional confidence in peak identification of field samples. For compounds for which no commercial sources were available, retention time indices were compared with published values in the NIST data base to help with identification (http://webbook.nist.gov/chemistry/cas-ser.html).
4.6 Data Analysis
 Text data files (10 and 0.1 Hz) from the REA were read into a custom MATLAB data reduction program and REA time stamps were checked for missing or excess data. The 10 Hz sonic anemometer (u, v, w, T) data were cleaned of electronic spikes and replaced with the previous good data points when data points exceeded six standard deviations from their respective averages (methodology used by real-time controller PC). The mean standard deviation of the vertical velocity flux measurement period (55 min), σw, was computed from the cleaned data. Because the measurements were made according to Pattey et al.  recommendation that small eddies be excluded using a vertical velocity threshold of 60% of σw, the β coefficient must be adjusted to compensate for the resulting increased concentration differential. We used Baker's method  to compute reduced β. The nominal concentration data were extracted from the analytical files. To correct these data for switching-induced volumetric errors, the internal standards introduced into the sample were compared with the internal standard present in an unswitched (continuous) run of known volume. Concentration differentials were then computed by subtracting the down concentration from the up concentration for each measured compound. The biogenic fluxes were calculated using equation 1 and the β coefficient computed using Baker's method. Fourier analysis of the sonic velocity data was performed to evaluate the size of eddies contributing to the total flux. Turbulent statistics were also calculated to check that the assumptions for measuring eddy covariance were satisfied [Moncrieff et al., 2004].
 The use of a rolling vertical velocity-based threshold (in our case a 5 min rolling average) has the potential to exclude some low-frequency contributions to the flux. To test this, we computed sensible heat fluxes by three methods from forty nine 10 Hz sonic data sets collected in March, April, July, and November using the vertical velocity directly, using a 5 min rolling mean vertical velocity as performed by the REA and using a planar-fit vertical velocity [Lee et al., 2004]. Comparing the nonrolling w heat fluxes with the rolling w heat fluxes shows the effect of using rolling means on the fluxes, namely if there was an observable effect on the fluxes due to the possible loses of low-frequency eddies. The planar-fit method adjusts for long-term site, tower, and anemometer alignment effects for a sonic anemometer at a fixed location. Comparison of the planar fit with the rolling mean will show any effect of low-frequency eddies and any uncompensated anemometer alignment effects. The calculated planar-fit vector includes only daytime conditions.
 Lastly, to examine the sensitivity of the flux measurements to errors in synchronizing sampling decisions with eddy segments arriving at our point of sample sorting, we again used the 49 sonic w and T data sample periods to simulate flux loss. To do this, we recalculated the heat flux by shifting the phase of T data by increments of 100 ms (the timing of sonic data packets) up to 1000 ms. The resulting shifted fluxes were compared with the in-phase (synchronized) data sets.
 Our first step in reducing the data to fluxes was to examine the ratio of the two internal standard compounds introduced during sampling. The carbon ratio of the 2-methylheptane (2MH) to methylcyclohexane (MCX) in the original gas was 1.2:1. If collection and recovery from the accumulator tubes is complete, the ratio should not change. If the adsorption capacity of the adsorbent is exceeded (i.e., due to excessive hydrocarbon loading) or the temperature of the tube during sampling is excessive, then the least well-retained component will preferentially be lost. In our case, the MCX will be lost first, resulting in an increase in the 2MH/MCX ratio. In Figure 4, we plotted the ratio of 2MH/MCX versus the temperature of the tube manifold during collection. The data indicated that when temperatures exceeded ~27°C, in some runs, the MCX decreased relative to the 2MH demonstrating some loss of MCX. Therefore, in instances where the ratio exceeded 1.4, we used the 2-methylheptane alone to compute volumetric corrections for the associated flux measurements. Typically, the tube manifold temperature control should have prevented the tubes from exceeding 25°C. However, there were a few periods when the dew point was so high that the coolers used a higher set point to avoid condensing water in the tubes.
 Next, we examined whether there was any evidence that the sample volumes collected were greatly modified by the pressure transients during switching and thus a need for a correction factor. Such a correction would be most important at high switching frequencies where the flow disruption would be a significant portion of the sampling period. There are two independent ways to calculate the sample volumes: by summing the volume of samples flowing through the sample tubes and by measuring the volumes using the two internal standard gases. Hence, we plotted the ratio of volumes determined by these two methods. The volume using the flow method was calculated by summing of the product of the number of up and down 100 ms time intervals and the average flow rate. The volume using the internal standard method was calculated from the volume of each internal standard using the GC-FID. The ratio of the two methods versus the number of 100 ms valve transitions during sample period is plotted in Figure 5. The mean ratio was 1.04 ± 0.14 indicating that on average the volume determined by use of the internal standards was about 4% lower than the strict time-flow rate product. No frequency-dependent response is evident from this data—even at the highest switching frequency around 2800 (corresponding to mean sampling durations of about 1.2 s or 0.8 Hz). Clearly, the system design is quite adequate for sampling above forest canopies where the eddies tend to be large with very little contribution from the high frequencies. Data above and below the optimum 1/1 ratio may be due to some drift from the set point of the accumulator mass flow controller. Schade and Goldstein  used methacrolein as an external standard in a similar manner to adjust for channel bias. They believed that the methacrolein did not exhibit mixing gradients in the constant flux layer and should therefore be present at equivalent concentrations in the up and down draft samples. Systematic deviations from this equivalence during the course of their study were interpreted to be indications of valve failure. These ratios were used to correct the affected samples for volume. Park et al.  used an internal standard in a similar manner to check for channel bias. Our outliers in Figure 5 could not be associated with any systematic valve failure. These points appear to be random when viewed against the diurnal and season long time series. Possible causes of these deviations could be due to incomplete valve operation due to particles lodging/dislodging in the valve diaphragm or undetected tube desorber-gas chromatographic problems (problems with constant liquid nitrogen delivery during column refocusing) causing incomplete tube-to-GC transfer of the internal standards. Regression analysis of the separate up and down ratios indicated no significant correlation with switching frequency. Furthermore, Fourier analysis of the sonic velocity data indicates that there is very little contribution from eddies shorter than 1 s over the forest canopy
 Flux measurements over smoother surfaces such as agricultural crops may require better sampling accuracy than over forest canopies. To evaluate if our system would still be accurate at a switching frequency of 10 Hz, a supplementary test was devised to test whether sample durations of 100 ms would still be accurately sampled. For this test, a square wave-switching function was supplied to the segregation valves for a 21 min period and molecular sieve-packed accumulator tubes were used to collect water vapor for the left sample, right sample, and neutral (below threshold) channels. A direct sample from the main line was also sampled to obtain a continuous measurement of the water vapor without a switching transient disturbance. With alternating switching intervals of 100 ms (neutral to right, right to neutral, neutral to left, left to neutral, etc.), the three channels were within 1% of their expected water accumulation and the total accumulation of the summation of the switched channels was within 5% of the continuously collected channel. This demonstrated that sample/zero switching transitions were followed by immediate resumption of stable flow after the valve switching transient had passed. Thus, pressure surges or back streaming were minimized through system pressure balancing.
 To test whether the use of the 5 min rolling means, as used by the REA system, removed any low-frequency contribution to the fluxes, we calculated the heat fluxes with and without rolling temperature means using sensible heat fluxes as a surrogate for biogenic VOC fluxes. The data we used included periods with mostly unstable to neutral conditions with sensible heat fluxes ranging from negative to ~500 w m−2 and with mean wind velocities ~0.3–2.7 m s−1. The standard and rolling heat fluxes (ρcpw′T′), where ρ is the density of dry air and cp is the heat capacity of dry air, were computed from the raw sonic data using the eddy correlation method with the T synchronized with the w and with the T lagged by 300 s with the w, respectively. For a second comparison, we computed the heat flux using the standard eddy correlation method with the w sonic data rotated using the planar-fit method to remove terrain effects. These data are plotted in Figure 6. The plot of the standard eddy correlation (nonrolling w heat fluxes) versus rolling heat fluxes indicates that there is no detectable difference between the two heat flux methods which, in turn, indicates that there is no detectable loss due to the removal of low-frequency eddies. The comparison of the heat flux using rolling w data versus planar fit w indicates that, in our case, the heat fluxes were not significantly affected by local long-term factors (terrain slope). A more sophisticated REA system might incorporate site-specific long-term factors by using planar-fit vertical velocities in determining the thresholds for switching the valves. This would, of course, be limited to long-term measurement sites where direction-specific streamlines could be established. This should be considered carefully since Finnigan et al.  has shown that using such averaging and coordinate rotation techniques may, in some conditions, filter out long wave contributions to the eddy fluxes.
 The delay times used to synchronize the switching of the sampled gas into the proper updraft and downdraft accumulators were calculated from the sum of the inlet filter, tubing, and fittings swept volume times the total flow rate. Since we were operating under turbulent flow conditions, gas velocity profile across the tubing will be uniform except at the walls. This minimizes the attenuation of concentrations by smearing as the gas passes through the tubing. We estimated an uncertainty of our swept volume of about 5.7% (dimensional uncertainty of tubing, Tee unions, and inlet filter) and perhaps as high as 7% flow rate uncertainty due to influence of varying humidity on the main line mass flow controller [Lee, 2000]. This yields a maximum uncertainty of ±13.7% in the timing (100 ± 13.7 ms). To test the sensitivity to uncertainty in timing as result of imprecision in volume and flow rate, we simulated the effect of 100 ms incremental lags on the eddy covariance heat fluxes. Figure 7 shows a plot of the results of flux underestimates that would result from lags as high as 1 s. With the exception of a single period late in the day when the heat flux was negative (loss of turbulent mixing), fluxes gradually decrease with increasing lag. Clearly, with our uncertainty estimate under 100 ms, our measurements show little to no effect resulting from this uncertainty.
 To illustrate capabilities of our system and their potential application to the improvement of biogenic emission inventories, we present some data from our 2007 measurement campaign. Figure 8 presents measured α-pinene fluxes as a function of air temperature. We have also included emission rates predicted by using the latest MEGAN2.1 monoterpene emission algorithm coefficient and our local basal emission rates from enclosure-based measurements for Loblolly pine reported by Geron and Arnts . These rates are scaled up to canopy emissions by assuming biomass densities of 500 g m−2 in winter, 750 g m−2 in spring and fall and 1000 g m−2 in summer [Kinerson et al., 1974]. Enclosure-based measurements (typical of the methodology used in generating emission inventories) can suffer from many difficulties including disturbing the branch or sapling emission process, rapid temperature elevation in the enclosure, and inadequate sampling to represent a mature canopy. Use of micrometeorological techniques, such as REA, avoids these disturbance issues by operating in the free atmosphere and integrating fluxes from whole canopies. While the emission algorithm [Guenther et al., 1993] predicts an exponential increase which is observed in the REA data, the enclosure-based predictions are significantly higher—probably due to the issues mentioned above. The enclosure-based measurements also have difficulty making precise measurements during spring bud break as is reflected in the very high and variable basal emission rates [Geron and Arnts, 2010]. Increased emissions during bud break do, however, appear to be reflected in the spring REA measurements as evidenced by the spike in fluxes between 25 and 30°C.
 To further illustrate the capability of the system, we present two consecutive days of flux measurements in winter and summer from our 2007 data base in Figure 9. To simplify plotting, the sum of the 24 biogenic VOCs measured were combined into common groups: monoterpenes (α-pinene, β–pinene, limonene, β-phellandrene, myrcene, camphene, p-cymene, sabinene, α-phellandrene), oxygenated monoterpenes (linalool, camphor, α-terpineol, terpin-4-ol, nopinone, verbenone), and sesquiterpenes (β-caryophyllene, cis-α-bergamotene, δ-cadinene, copaene, β-cedrene, α-muurolene, α-humulene, trans-α-bergamotene). Most of the winter emissions were monoterpenes with little contribution from the oxygenated terpenoids and sesquiterpenes. 2-Methyl-3-buten-2-ol was absent in the winter emissions while constituting the single largest emission in the summer. Note that it is likely that some accumulator tube breakthrough of 2-methyl-3-buten-2-ol occurred in July and the fluxes should therefore be viewed as probably low. The sesquiterpenes make significant contributions to the flux in the summer during the warmest part of the day.
 REA flux measurements can only reflect the actual canopy emissions if there are no sources or sinks between the emitting surface and the point of measurement—especially potential chemical loss during transport from reaction with ambient oxidants (O3, OH, and NO3). Several studies have estimated canopy loss of biogenic VOCs from these sinks [Rinne et al., 2012; Stroud et al., 2005; Forkel et al., 2006; Fuentes et al., 2007; Wolfe and Thornton, 2011]. The reactive losses of these compounds are largely governed by their transport time and their individual reactivity with the three oxidants. Using the model of Rinne et al. , we calculated the Damköhler number (ratio of the time scale of turbulent mixing to the chemical lifetime of the compound) for 13 of the most prominent compounds we observed on 4 July 2007. The chemical portion of this term was calculated using available rate constants [Atkinson, 1997; Atkinson and Carter, 1984; Baker et al., 2004; Hoffmann et al., 1997; Klawatsch-Carrasco et al., 2004; Wells, 2005; Rudich et al. 1995; Fantechi et al., 1998; Jones and Ham, 2008; Reissell et al., 2001; Calogirou et al., 1999; Bouvier-Brown et al., 2009], ambient ozone measured by local monitor along with OH, and NO3 concentrations estimated by Stroud et al.  from the same location in July 2003. The mixing time scale was estimated from z/u* where z is the measurement height (24 m) and u* is the friction velocity. The mixing scale was in the range of 35–40 s on the example day. Using the lookup table in Rinne et al. , we estimated the flux/emission ratio (predicted flux at the inlet versus emitted flux in the canopy) with table entries of 7 for the leaf area index (assumes canopy density of Scots pine is similar to Loblolly pine), a constant vertical oxidation rate, and a z/h ratio of about 1.25 (measurement height/canopy height). We found that measured fluxes were predicted to be ≥93% of emissions for camphene, p-cymene, 2-methyl-3-buten-2-ol, camphor, β-phellandrene, linalool, α-pinene, and β-pinene and within 80–87% for sabinene, nopinone, α-copaene, terpin-4-ol, α-terpineol, and limonene. Very significant losses were estimated for α-bergamotene (24%) and myrcene (31%) and especially for β-caryophyllene, α-phellandrene, and α-humulene (60%). Note that oxidant rate constants were not available for verbenone, δ-cadinene, β-cedrene, and α-muurolene.
 The REA system described herein demonstrated capabilities to measure important reactive biogenic compounds, especially precursors of secondary organic aerosol, volatile organic fluxes including monoterpenes and sesquiterpenes, as well as their polar oxygenated species, at levels as low as 10 µg C m−2 hr−1 (15 L sample size and detection by GC with flame ionization detection). The system demonstrated these capabilities over ambient temperatures ranging from ~0 to 35°C. Key design features which facilitated this performance included (1) integration of gas-phase ozone removal to permit VOC collection without on-adsorbent losses, (2) a short inert inlet operated under turbulent flow (Reynolds number ~5000) to effect efficient transmission of sampled air to the segregator valves with minimal axial mixing (smearing), (3) mass flow control for all critical gas streams, (4) a heated sample path through the tubing, segregator valves, and reaction coils, (5) cooled accumulator tubes, (6) temperature control of mass flow controller module to stabilize flow control, (7) a two-component internal standard which permitted correction of accumulator volume for flow switching transients and monitoring for valve failure and VOC adsorbent breakthrough, and (8) a 12-tube magazine of high-capacity adsorbent tubes with two sets of six-port manifold valves to permit up to six consecutive flux measurements without operator attention. While there have been some previously reported VOC REA systems (excluding the gas-phase removal approach) that have incorporated one or more of these elements, there has been only one report of REA measurements which included observations of sesquiterpene fluxes [Ciccioli et al., 1999]. That system used manganese dioxide-coated screens to reduce ozone, a methodology that is recognized to also remove oxygenated terpenoids and sesquiterpenes [Hoffman, 1995; Olofosson et al., 2003; Graus et al., 2006; Arnts2008]. Ciccioli et al.  placed the adsorbent tubes near the sonic anemometer to minimize any losses due to long inlet tubing, but it created less precise sorting of up and down drafts due to computational lags. However, the use of REA systems over forest canopies likely rendered these errors unimportant due to the controlling large (long) eddy sweeps. Also, only single flux measurements were made with their system before operator attention was required to change out the adsorbent tubes. Similar direct inlet, single measurement systems have been used to measure monoterpene fluxes [Valentini et al., 1997; Ciccioli et al., 1999; 2003; Darmais et al., 2000].
 Inherent in all REA implementations are certain analyte selectivity and minimum detectable flux considerations. As we have noted, only Ciccioli et al.  had similar measurement objectives to ours. We note that the only VOC REA systems where minimum detectable fluxes were estimated were that of Haapanala et al. . So, it is perhaps more appropriate to comment on design differences that could be important in achieving a given set of measurement objectives. For the nonpolar and polar C10 and C15 monoterpenes and sesquiterpenes, it is important to account for their reactivity, especially with ozone, and their affinity for sorption to surfaces. Thus, systems which minimize exposure to inlet tubing, valves, flow controllers, and pumps [Valentini et al., 1997; Ciccioli et al., 2000; Darmais et al., 2000] before sample collection will likely be better suited than those which do not [Graus et al., 2006; Haapanala et al., 2006; Park et al., 2010]. Passage of the sample stream through flow controllers and pumps with their associated glass and metal surfaces, along with their larger internal volumes, introduces more opportunity for sorption of analytes and internal mixing of eddy segments. Sorption of analytes to long inlet tubing and components in single inlet systems could retard passage of some analytes and cause asynchronous eddy capture valve switching with resulting flux underestimation. Systems with dedicated up and down inlet lines are less likely to suffer from these issues since the eddy sample has already been sorted into its proper bin and each accumulator will integrate most of the slow-moving analytes as long as the sampling period is much greater than their inlet transit period.
 Virtually all of the previous VOC REA systems with sample volumes above 1 L use some type of in-system or external adsorbent trap to provide sufficient sample volume for gas chromatographic separation and detection at sub-ppb levels. Typically, these are in the 2–5 L range and are limited by the quantity of adsorbent (~300 mg maximum of Tenax or a graphitized carbon black or a combination of the two). These adsorbent tubes are usually prepacked commercially available or custom made to mate with available thermal desorption and concentrator apparatus, i.e., Markes International, PerkinElmer, or Chrompack. Our system used larger, custom-packed tubes with greater than 10 times more adsorbents than previous systems and allowed much larger sample volumes. This provided us with sufficient on-column analyte delivery to operate well above the limit of detection of the flame ionization detector and sufficient sample to identify small chromatographic peaks by mass spectrometry. These large sample volumes were a key advantage for making measurements of highly reactive compounds during cold weather when VOC gradients are small.
 We acknowledge Gary Zimmerman, Steve Osborne, and Herb Zimmerman of Applied Technologies, Inc. for their technical assistance with this project. The Duke Forest site was supported by the Office of Science (BER), U.S. Department of Energy, DE-FG02-95ER62083. The United States Environmental Protection Agency produced the research describe here. It has been subjected to Agency's administrative review and approved for publication.