Characteristics of free air carbon dioxide enrichment of a northern temperate mature forest

Abstract In 2017, the Birmingham Institute of Forest Research (BIFoR) began to conduct Free Air Carbon Dioxide Enrichment (FACE) within a mature broadleaf deciduous forest situated in the United Kingdom. BIFoR FACE employs large‐scale infrastructure, in the form of lattice towers, forming ‘arrays’ which encircle a forest plot of ~30 m diameter. BIFoR FACE consists of three treatment arrays to elevate local CO2 concentrations (e[CO2]) by +150 µmol/mol. In practice, acceptable operational enrichment (ambient [CO2] + e[CO2]) is ±20% of the set point 1‐min average target. There are a further three arrays that replicate the infrastructure and deliver ambient air as paired controls for the treatment arrays. For the first growing season with e[CO2] (April to November 2017), [CO2] measurements in treatment and control arrays show that the target concentration was successfully delivered, that is: +147 ± 21 µmol/mol (mean ± SD) or 98 ± 14% of set point enrichment target. e[CO2] treatment was accomplished for 97.7% of the scheduled operation time, with the remaining time lost due to engineering faults (0.6% of the time), CO2 supply issues (0.6%) or adverse weather conditions (1.1%). CO2 demand in the facility was driven predominantly by wind speed and the formation of the deciduous canopy. Deviations greater than 10% from the ambient baseline CO2 occurred <1% of the time in control arrays. Incidences of cross‐contamination >80 µmol/mol (i.e. >53% of the treatment increment) into control arrays accounted for <0.1% of the enrichment period. The median [CO2] values in reconstructed three‐dimensional [CO2] fields show enrichment somewhat lower than the target but still well above ambient. The data presented here provide confidence in the facility setup and can be used to guide future next‐generation forest FACE facilities built into tall and complex forest stands.


| INTRODUC TI ON
The 'greening' of the terrestrial surface across planet Earth has been driven by changes to the dynamics of vegetation and their interactions, to a large extent, with increasing levels of carbon dioxide (CO 2 ) in the atmosphere (Forzieri, Alkama, Miralles, & Cescatti, 2017;Zhu et al., 2016). The land carbon sink currently absorbs 20%-30% of CO 2 released by human activities (Le Quéré et al., 2018) and a large proportion of this absorption is by woody vegetation (Gaubert et al., 2019). This sink activity is largely ascribed to the fertilization effect of increasing atmospheric CO 2 concentrations (Schimel, Stephens, & Fisher, 2015), especially through the stimulation of growth and carbon sequestration in established, mature forest ecosystems (Luyssaert et al., 2008). However, the future magnitude of the land carbon sink, as atmospheric CO 2 continues to increase (at least until mid-21st century), is uncertain. Modelling of future C-uptake rates ranges from 0% to 30% of human CO 2 emissions, across the suite of Earth systems models used by the Intergovernmental Panel on Climate Change, Working Group 3 (Friedlingstein et al., 2014).
The uncertainty in the sensitivity of the land C sink to increasing atmospheric CO 2 is due, in large part, to a lack of experimental data on mature forest ecosystems under future elevated [CO 2 ] (e[CO 2 ], Ellsworth et al., 2017;Norby et al., 2016). In the northern hemisphere, where about 40% of the net uptake occurs, highly seasonal mature forests dominate the land carbon sink (Luyssaert et al., 2008).
A somewhat different ('WebFACE') free-air methodology targeted canopy exposure of mature trees, but did not quantify the CO 2 field around the treated trees (Klein et al., 2016) and was not suitable for biogeochemical budget studies.
Since the closure of important 'first-generation' forest FACE experiments-'Duke FACE' in an evergreen loblolly pine stand (Hendrey et al., 1999), the Oak Ridge National Laboratory FACE in a young deciduous sweetgum plantation (Norby et al., 2006), and the AspenFACE that followed aspen and poplar seedlings over a decade (Dickson et al., 2000;Kubiske et al., 2015)-the scientific community has advocated for large-scale, ecosystem-plot-sized FACE experiments in important forest ecosystems (Calfapietra et al., 2010;Norby et al., 2016).
The 'EucFACE' experiment in an open, Mediterranean-type sclerophyll forest in Australia (Drake et al., 2016) is the first such 'second-generation' forest FACE, which has been operating since September 2012; the Birmingham Institute of Forest Research (BIFoR FACE), which is the focus of this study, is the second (Norby et al., 2016). The forest stand in BIFoR FACE has the most complex canopy structure of all forest FACE experiments to date, dominated by up to 25 m tall mature pedunculate oak (Quercus robur), with distinct mid-and understoreys formed mainly by sycamore maple (Acer pseudoplatanus) and common hazel (Corylus avellana), as well as dense ground cover vegetation (Norby et al., 2016). As a deciduous forest ecosystem, it has very variable leaf area index (LAI) during the active vegetation and CO 2 fumigation period, from very low at the spring flush to very high LAI (>5 m 2 /m 2 ; Norby et al., 2016) during the main summer assimilation period (MacKenzie et al., 2019).
The structural and temporal characteristics of the BIFoR FACE forest pose specific problems for the CO 2 exposure system, which are not directly comparable to those in 'EucFACE', with its evergreen and much sparser canopy . Oak Ridge National Laboratory FACE was perhaps most comparable, being a deciduous forest plantation with a high LAI of about 5.5, but the trees were younger and smaller, and the e[CO 2 ] canopy volume was smaller and more uniform in each experimental patch (Norby et al., 2006). FACE facilities have a simple scientific aim-that is, to subject ecosystem patches to consistent e[CO 2 ]-but are complicated to engineer. In order to meet the science aim without altering other environmental parameters, the CO 2 fumigation must be accomplished using infrastructure that minimally influences canopy structure, environmental aerodynamics and microclimate. BIFoR FACE consists of nine experimental patches of forest, three infrastructure arrays dosing air +CO 2 , three infrastructure arrays dosing with ambient air only and three noninfrastructure patches (see Section 2.2). Fumigation of 30 m diameter patches (see Table 1) is accomplished using approximately circular arrays of 16 free-standing lattice towers, supporting perforated pipes from which premixed air/CO 2 is released from the upwind quadrant (see Section 2.4).  1995;Hendrey et al., 1999). Similar analyses for smaller systems used, for example, on crop canopies, are not directly applicable, as they use pure CO 2 injection, and variability in those latter systems is mostly analysed on a two-dimensional basis (Mollah, Partington, & Fitzgerald, 2011). Hendrey et al. (1999) found that in the one array analysed in detail, FACE control was satisfactory within 90% of the entire volume, at least in the longer term (i.e. averaged over ~232 days). Hendrey et al. (1999) also found that CO 2 consumption was positively related to wind speed and photosynthetically active radiation (PAR).
We analysed the full performance data of the facility for the first full season of fumigation in BIFoR FACE to address the following questions: subcommunity Holcus lanatus (W10d) classification (Rodwell, 1991 October 27, see Section 2.2) +14.5 ± 4.0°C (mean ± SD; using 1 min averages). The mean annual wind speed at the mature oak canopy (24-26 m height) was 2.2 ± 1.0 m/s (mean ± SD, temporal variation from the location of array 4 and calculated from 1 min averages). For annual air temperature, PAR, wind speeds and rainfall, see Figure S1.
The 3(c) (see Figure 1). The greenness index (gcc, Toomey et al., 2015) from the PhenoCam is shown in Figure 2 for 2017, which is typical of the four seasons to date for which we have greenness phenology data. A maximum gcc was observed on 15 May and minimum on 19 November, which corresponded to visual observations of canopy closure and leaf fall respectively.

| Array infrastructure
The design-and-build approach for the facility (

| FACE arrays and experimental control
There are six FACE infrastructure arrays, each array comprising 16 peripheral towers with a central tower and a 15 m radius space for research. For separation distances between arrays, please see Table 2. Tower heights are designed to be ~1 m above the canopy height in each array and so vary between 24.7 and 27.3 m, as specified in Table 1. Towers are parallel in elevation, square in plan, and use an equilateral truss design (known as 'Warren Truss'; Griggs, 2015). An example array plan and front elevation is presented in Note: Rows = fumigated arrays, Columns = control arrays. Direction is defined from the control array so that, for example, from the centre of array 2(c), the centre of array 1(f) is 91 m in direction 135° (south east).

TA B L E 2 Distance and direction matrix between FACE arrays
F I G U R E 3 Example of a BIFoR FACE array. (a) Plan view of the structures encircling a 30 m diameter experimental patch. Dotted arrow shows radius from the centre to outer plenum edge (R ≈ 20 m) and the dashed arrow shows the radius from the centre to the internal edge of the research array (R ≈ 15 m). Note the nonuniform distribution of towers. This was due to the existing tree infrastructure (roots, trunk and canopy) that had to be avoided when installing the towers. (b) Front view showing screw pile system penetrating into the bedrock to provide secure anchoring and avoiding the need for guy cables to support the towers. The north tower is evident at the top of the plan view and in the centre of the front view, with a vertical pole containing a two-dimensional ultrasonic anemometer (fumigated arrays only) e[CO 2 ] is achieved by a computer controlled system that introduces varying volumes of pure CO 2 gas into a fixed volume of ambient air (contained and circulating the periphery of the array inside a torus whilst being continuously replenished using an air intake fan), to a concentration of ~30 mmol/mol (3% by volume). The highly enriched air-CO 2 mixture is then released from the VVPs in the upwind quadrant of the array. Process control uses a proportional-integrative-differential algorithm (Hendrey et al., 1999), to achieve a more uniform and efficient mixing effect within the treatment array (Lewin et al., 1994).
Ambient CO 2 is measured using infrared gas analysers (IRGA, LiCor 840A, LiCor Lincoln) with inlets situated in the control arrays at ~24 m. The process-control algorithm selects the lowest 1 min low-pass filter average from the three control arrays and assigns this as the set point ([CO 2 ] set ) from which the treatment is calculated (i.e. ambient +150 µmol/mol). Analysers were calibrated every 2 weeks between 0 and 1,000 µmol/mol using ultrapure N 2 (Air Liquide) and certified 1,000 µmol/mol CO 2 in compressed air (Air Liquide) for the first 4 months, and monthly thereafter.
Daily fumigation times, on in the morning and off in the evening, were determined by solar elevation. This was calculated continuously by an FCP built-in procedure based on Doggett (1987). The solar elevation used at BIFoR is −6.5° (roughly civil twilight). Predawn start up (i.e. at <0 degrees solar elevation), allows the arrays to attain the fumigation target before photosynthesis is significant. Array pairings start in sequence 1(f) + 3(c), 2(c) + 4(f) and 5(c) + 6(f) with a 5 min time lag. The planned operation times are, therefore, not exactly equal across the three pairings, varying by 51 hr over the season (Table 3).
The FACE Control Program (FCP) is designed to halt fumigation when canopy-top, 1 min average, air temperature is <4°C, and resumes fumigation when the air temperature is ≥5°C. Carbon uptake under these conditions is considered negligible (Hughes, 1966  This 32-point system is installed in all six infrastructure arrays. For more details on the multiport sampling procedure, see Hendrey et al. (1999) and Lewin et al. (1994).

| Statistical analyses and graphical applications
Data analysis and statistical calculations were conducted using a combination of Microsoft Excel 2013 and R (R Team, 2017; see also Bivand, Pebesma, & Gomez-Rubio, 2013;Pebesma & Bivand, 2005).

| Engineering performance
The Note: Minimum operating hours vary by ~2 hr due to array-specific engineering failures that were confined to single days of restricted running times. See Figure 1 for array locations.
TA B L E 3 Planned operation hours, per array pairings of fumigation (f) and control (c), over the 2017 operating season 206 operating days. Start date was determined using current and previous years' phenological observations taken at the site ( Figure 2).
The facility-average target was for 2,994 hr of operation (out of a total of 4,944 hr including night-time). Based on previous work (Dodd et al., 2005;Hart, 1988;Hughes, 1966;Johnson et al., 1998), there was no CO 2 fumigation for the 1,950 night-time hours (i.e. solar elevation ≤−6.5°). Planned, actual and average daily operation times and total downtime per array are documented in Table 3.
The main FACE fumigation system was functionally operational for 2,928 hr (97.7% uptime) with ~66 hr of downtime due to 'engineering faults' (a term we use here to cover mechanical, CO 2 supply, electrical and software issues) and wider environmental conditions (e.g. high winds and low temperature, see above). Over the operating season, a total of 17 hr were lost due to engineering faults or necessary infrastructure upgrades, accounting for 0.6% of downtime.
These events were sometimes isolated to one FACE array at a time, allowing the rest of the facility to operate (

| Experimental performance
The enrichment set point (target) for the facility (+150 µmol/mol above ambient) is determined using a moving 5 min average, relative to the ambient concentration in the control arrays measured in real time and automatically fed into the FCP algorithm. To look at CO 2 distributions in more stringent detail, statistics reported below as determined using the control array ambient reference sampling ports, was 400 ± 17.0 µmol/mol (mean ± SD). The annual average enrichment value achieved was +147 ± 21 µmol/mol (mean ± SD, Figure 4). Table 6 reports the summary distribution statistics for the 1 min average [CO 2 ] and the calculated enrichment value of the two treatment levels for all arrays.
In line with previous studies, we set the a priori goal for acceptable performance of the 1 min average e[CO 2 ] to remain within ±20% of the set point, for at least 80% of the operation time (Hendrey et al., 1999;Miglietta et al., 2001). BIFoR FACE achieved its enrichment set point at 97% of the operation time across the three fumigated arrays, that is, well above target. A more stringent goal of being within ±10% of the enrichment target was achieved during 82% of the time (see Figure 5).
The averaged array performances shown in Figure 5 demonstrate the relative stability of the facility over the growing sea- The average % deviation from target, using hourly averages of the 1 min enrichment data during fumigation periods, was +5 ± 14.5%, +7 ± 10.4% and +7 ± 12.1%, that is, not statistically different from zero, for arrays 1(f), 4(f) and 6(f) respectively. Across all the fumigation arrays, the deviation was 6 ± 12.4%, that is, also not statistically different from zero. Figure S2 shows the monthly distributions of variance between the target (red line) and actual enrichment hourly averages for all hours when fumigation was scheduled to operate. Negative outliers represent events such as engineering failures (e.g. loss of CO 2 feedstock in August); positive outliers represent calm weather events (wind speeds <0.4 m/s) resulting in short-term over-dosing.
There are three groups of outlier events (April, June and August) across the three fumigated arrays that indicate under performance was an issue at some point in those months (rather than short-term deviations from the target. These were due to engineering shutdowns and all events have been catalogued (Hart, Miles, Harper, & MacKenzie, 2017). The cluster of negative outliers in August, for all three treatment arrays, was caused by a UK-wide shortage of CO 2 causing the facility to shut down operations due to low liquid storage levels. These shutdowns were managed to prevent fumigation shut down over an entire solar day and maximize fumigation exposure over the solar maxima. A critical shortage occurred on 6 August 2017, when the system was shutdown at 1200 hr and did not restart until 7 August at 0800 hr.
'Grab' e[CO 2 ] samples are recorded as 4 s averages of 1 Hz measurements at the FACE array control ports. The 1 and 5 min averages are automatically calculated from these data to manage the facility and help assess its day-to-day performance ( Figure 6) shown for all times when the FACE system was scheduled to operate and, hence, includes all engineering failures and automatic shutdowns due to inclement weather ambient. The grab, 1 min, and 5 min enrichment averages are very similar, the distribution of 5 min averages having a higher peak and slightly narrower width. The density distribution is centred below zero, indicating that the system was generally slightly below target.

| Cross contamination of control arrays
It is important to quantify the level of cross contamination between fu- FI G U R E 6 Kernel density plot of the facility average instantaneous grab, 1 min, and 5 min average [CO 2 ] measurements at the control port (~24 m height) for the fumigated arrays over the seasonal operating period. The filled areas correspond to points between the 25th and 75th percentiles, see Table S2 for   of the time and minima ranged between 0 and −11 ± 3.7 µmol/mol (mean ± SD) averaged across the three control arrays.
These data would suggest that CO 2 released from the fumigated arrays may, very occasionally, travel several tens of metres in or over the canopy, depending on wind speed and direction.

| Three-dimensional [CO 2 ] fields inside the arrays
The multiport samplers allow for a more detailed analysis of the spatial distribution of the [CO 2 ] within each array (see Section 2.5).
For consistency, only ports at the same heights were included in F I G U R E 7 Wind speed and direction for 5 min averages for instances when Δ[CO 2 ] ≥ 10 µmol/mol occurred in control arrays, relative to the facility reference control set point, [CO 2 ] set . Data points represent enrichment values with respect to [CO 2 ] set and are sorted by taking the mean value for each 1° of cardinal direction and for each wind speed increment of 0.25 m/s. Distance and direction to the fumigated arrays are denoted by red arrows. Length of the arrow indicates distance between arrays (see Table 2 for measured distance and azimuth direction matrix) F I G U R E 8 Histogram of cross-contamination incidents in all control arrays (2(c), 3(c) and 5(c)). Δ[CO 2 ] is defined in the main text this analysis. It takes 32 min to poll all 32 inlets (providing ~2 measurements per hour per port), during which time the flow of CO 2 into the array will be adjusted many times by the FCP, so measurements are not instantaneous snapshots of the three-dimensional field. Figure 9 shows season-average height 'slices' through the array, using only daytime data collected during fumigation, spatially interpolated using kriging and linear fitting of the variogram.
CO 2 mixing ratios are higher on the 2 and 24 m horizontal planes than on the 10 m plane. Red to blue areas in Figure 9 denote fixed sampling positions where e[CO 2 ] were consistently high or low (compared to the target concentration) across the season, presumably as a result of imperfect mixing in the lee of a tree stem. e[CO 2 ] was higher in the south-west quadrant for all three arrays, corresponding to the prevailing winds.
Detailed performance statistics for each of the analysed cross sections are provided in Table S3. To summarize: at 24 m, the e[CO 2 ] field was 100 ± 19%, 88 ± 8% and 101 ± 25% of the target for Arrays 1(f), 4(f) and 6(f) respectively. For array 1(f), eight of the nine inlets were within ±20% of the operating target over the seasonal average.
All nine inlets were within ±50% of the target. The high enrichment deviations that are clustered between the west and north perimeters are likely caused by the position of the array on the south-east edge of the forest. For array 4(f), five of the nine inlets were within ±20% of the operating target over the seasonal average. All nine inlets were within ±50% of the target. For array 6(f), eight of the nine inlets were within ±20% of the operating target over the seasonal average. All nine inlets were within ±50% of the target.
At 10 m, the e[CO 2 ] field was 64 ± 10%, 56 ± 7% and 70 ± 16% of the target for Arrays 1(f), 4(f) and 6(f) respectively. For array 1(f), none of the nine inlets were within ±20% of the operating target over the season. All nine inlets were within ±50% of the target. For array 4(f), one of the nine inlets was within ±20% of the operating target. Five of the nine inlets were within ±50% of the target. For array 6(f), two of the nine inlets were within ±20% of the operating target. Eight of the nine inlets were within ±50% of the target.
At 2 m, the e[CO 2 ] field was 107 ± 16%, 70 ± 7% and 100 ± 16% of the target [CO 2 ] for arrays 1(f), 4(f) and 6(f) respectively. For array 1(f), eight of the nine inlets were within ±20% of the operating target over the season. All nine inlets were within ±50% of the target. For array 4(f), one of the nine inlets was within ±20% of the operating target. All nine inlets were within ±50% of the target. For array 6(f), all nine of the inlets were within ±20% of the operating target. 3 µmol/mol CO 2 below target) across the season (Table 6, above).
A vertical profile of the Δ[CO 2 ] in the fumigation and control arrays are measured using eight fixed points at the centre of each array ( Figure 10). The seasonal Δ[CO 2 ] vertical profiles for the treatment arrays (measured from just above ground level to above canopy) show a nonuniform distribution (Figure 10a). The red dashed line in Figure 10a shows the target set point that should be achieved , 4(f) and 6(f) respectively ( Figure 10). These statistics agree with the single-point FCP data discussed in Section 3.2 and demonstrate that the entire upper storey canopy is being adequately enriched.
For 0.25 < height < 10 m, Δ[CO 2 ] was 78%, 62% and 85% of target, for arrays 1(f), 4(f) and 6(f) respectively. These results indicate that the coppice canopy and understorey were exposed to a signif- Contamination of the ambient control arrays by e[CO 2 ] from the treatment arrays is rare and short-lived, being mostly governed by short periods when winds shift away from the predominant south-westerlies.
Control arrays are within 10% of the ambient set point-defined as the lowest 1 min low-pass filter average amongst the control array measurements-98.8% of the time. When contamination does occur, only 13% of such events produce 1 min average [CO 2 ] perturbations in excess of 15 µmol/mol. Array 2(c) experienced more frequent cross contamination events, but of a much lower intensity, than control arrays 3(c) and 5(c). Array 5(c) suffered more extreme enrichment events, with treatment array 6(f) providing the majority of the source.
4. How does the enrichment achieved vary throughout the canopy volume?
We have captured what we believe to be the most comprehensive data on the three-dimensional distribution of [CO 2 ] in a forest FACE facility.
Each array contains 32 gas-sampling inlets, placed at the array centre and at 6 and 12 m distance in each of the cardinal compass directions, at approximately 2, 10 and 24 m above ground. For operational reasons, [CO 2 ] tend to be lower at 10 m height than above or below. The median [CO 2 ] values in the reconstructed [CO 2 ] fields show enrichment lower than the target but still well above ambient.
We continue to monitor performance of the facility overall, and to make measurements of [CO 2 ] with high temporal and spatial frequency in the arrays allowing for a more detailed assessment of the three-dimensional FACE statistics. This will also be a valuable 'tracer' data set to derive forest canopy turbulence and mixing statistics, particularly when combined with sonic anemometer data from instruments within and around the BIFoR FACE forest patch.
Based upon the facility design and the continuous monitoring of the engineering control systems, in line with the local environmental conditions, BIFoR FACE operated within its design parameters for the majority of 2017. BIFoR FACE has demonstrated over its first operation season that it will provide an extensive, consistent and reliable data set for the analysis of e[CO 2 ] in a seminatural, temperate, deciduous, mature forest. These data, and ongoing sample collections, will provide an essential resource for modelling the potential impacts and effects of e[CO 2 ] on other similar landscapes.

ACK N OWLED G EM ENTS
The BIFoR FACE facility is a research infrastructure project supported by the JABBS Foundation and the University of Birmingham.
Applications to work at the facility should be addressed to A. Rob MacKenzie. The facility has received support for scientific studies from the JABBS Foundation and The John Horseman Trust. P.B., G.C. and S.K. acknowledge support from UK Natural Environment

CO N FLI C T O F I NTE R E S T
All listed authors declare no conflict of interest.