Low-cost chamber design for simultaneous CO2 and O2 flux measurements between tree stems and the atmosphere

Abstract Tree stem CO2 efflux is an important component of ecosystem carbon fluxes and has been the focus of many studies. While CO2 efflux can easily be measured, a growing number of studies have shown that it is not identical with actual in situ respiration. Complementing measurements of CO2 flux with simultaneous measurements of O2 flux provides an additional proxy for respiration, and the combination of both fluxes can potentially help getting closer to actual measures of respiratory fluxes. To date, however, the technical challenge to measure relatively small changes in O2 concentration against its high atmospheric background has prevented routine O2 measurements in field applications. Here, we present a new and low-cost field-tested device for autonomous real-time and quasi-continuous long-term measurements of stem respiration by combining CO2 (NDIR-based) and O2 (quenching-based) sensors in a tree stem chamber. Our device operates as a cyclic-closed system and measures changes in both CO2 and O2 concentration within the chamber over time. The device is battery powered with a >1-week power independence, and data acquisition is conveniently achieved by an internal logger. Results from both field and laboratory tests document that our sensors provide reproducible measurements of CO2 and O2 exchange fluxes under varying environmental conditions.


Introduction
Stem CO 2 efflux is an important part of the carbon balance of forest ecosystems, as it accounts for 5-42% of the total ecosystem respiratory fluxes in forests (Lavigne et al. 1997, Damesin et al. 2002, Chambers et al. 2004, Ryan et al. 2009, Trumbore et al. 2013, Yang et al. 2016. It is typically measured by using chambers of various designs and measurement principles (e.g., Xu et al. 2000, Pumpanen et al. 2004, Maier and Clinton 2006, Etzold et al. 2013, Hilman and Angert 2016, Katayama et al. 2016, Brändle and Kunert 2019 and then often assumed equal, or at least proportional, to the rate of actual respiration in the underlying tissues. This assumption neglects the fact that local CO 2 emission is the combination of respiratory CO 2 production and a number of post-respiratory processes (Teskey et al. 2008, Trumbore et al. 2013. Key processes are the transport of dissolved CO 2 in the xylem both away from or towards the site of measurement (McGuire and Teskey 2004, Teskey and McGuire 2007, Teskey et al. 2008, Bloemen et al. 2013, photosynthetic re-assimilation in chloroplasts of sub-corticular cells (Pfanz et al. 2002, Teskey et al. 2008, Ávila et al. 2014, Cernusak and Cheesman 2015, De Roo et al. 2020, non-photosynthetic refixation by parenchyma cells within the xylem, cambium and phloem via the enzyme phosphoenolpyruvate carboxylase (PEPC) (Gessler et al. 2009, Hilman et al. 2019 or axial diffusion of CO 2 in the gas phase (De Roo et al. 2019). All these processes can be highly variable over time and may differ between plant organs. Thus, while chambers can provide accurate flux measurements, these fluxes can temporarily differ significantly from local stem respiration rates.
Aerobic respiration not only produces CO 2 but also results in an anti-correlated uptake of O 2 , as O 2 is consumed as the electron acceptor at the end of the mitochondrial electron transport chain to form H 2 O. To date, stem O 2 uptake rates have rarely been measured because the high background of O 2 in ambient air (20.95 vol.% or 209,500 p.p.m.) makes the detection of O 2 concentration changes in stem chambers (typically a few hundred p.p.m. over tens of minutes in many chambers) technically challenging. Differential fuel-cell analyzers (e.g., Stephens et al. 2007, Battle et al. 2019) are able to detect very small changes in atmospheric O 2 (down to several p.p.m.), but require costly infrastructure and high maintenance for application in the field, and usually have a very limited application radius around the position of the analyzer. This can be overcome by laboratory measurements of discrete flask samples from the field, which allow for decentralized measurements over wider areas. Still requiring typically costly analyzers, this approach is usually limited by the number of flasks and the required processing time of the samples and thus typically results in low temporal resolution (Seibt et al. 2004, Hilman et al. 2019. Hilman and Angert (2016) presented an intermediate approach ('direct discrete method'): they used low-cost chambers that were installed independently of each other on several trees and O 2 measurements were carried out with a portable optical fiber system. While this approach allows measurements over a wider area and immediate results, this method cannot easily be automated and requires manual measurements, thereby again limiting the temporal and spatial resolution. Cavity-enhanced Raman multi-gas spectrometry (CERS) has been used to measure quasi-continuous fluxes of O 2 in pine (Pinus sylvestris L.) branches (Keiner et al. 2013, Fischer et al. 2015; however, due to the high sensitivity of the CERS to changes in temperature and air pressure, the methodology is not easily applicable under field conditions. Like with CO 2 , O 2 fluxes are affected by processes other than respiration. For example, O 2 can also be transported to or from the site of respiration by xylem water. However, since O 2 is ∼ 30 times less soluble in water than CO 2 (Dejours 1981), this effect is considerably smaller. Stem photosynthesis is usually considered to play a minor role in stems of older trees (Wittmann and Pfanz 2008, Rosell et al. 2015, Tarvainen et al. 2018 but would result in a release of O 2 and an anticorrelated CO 2 consumption. The only O 2 -exclusive metabolic processes we are aware of is lignification (Amthor 2003) resulting in O 2 consumption. While the actual amount of O 2 consumption by lignification is unknown, it seems unlikely to result in large changes of O 2 concentration in mature trees. Differences between actual respiration and measured fluxes therefore have to be expected for both gases.
The ratio of CO 2 release to O 2 uptake (respiratory quotient, RQ) depends on the stoichiometry of the respiratory substrate. For example, the stoichiometric RQ for complete oxidation is ∼1 for carbohydrates, ∼0.8 for proteins and ∼0.7 for lipids. Thus, the measured RQ has been used to identify respiratory substrates (Stiles andLeach 1933, Lambers et al. 2008). Plant respiration is commonly assumed to be dominated by carbohydrate catabolism, but shifts to lower RQ and δ 13 C of respired CO 2 have been used to infer a switch to lipid respiration for plants under stress and carbon starvation (Tcherkez et al. 2003, Fischer et al. 2015. The simultaneous measurements of CO 2 and O 2 fluxes therefore provide a more robust estimate of actual respiration rates as well as information on the stoichiometry of the respired substrate. The much smaller solubility of O 2 provides the potential to assess the influence of post-respiratory processes on CO 2 in the stem (Angert and Sherer 2011, Angert et al. 2012, Trumbore et al. 2013, Hilman and Angert 2016, Hilman et al. 2019.
Our aim was to develop and test a portable, weatherproof, low-cost and fully autonomous stem chamber design that allows simultaneous in situ measurements of CO 2 and O 2 fluxes from tree stems. The data presented here demonstrate the reliability and robustness of the individual sensors as well as the complete chamber design and are based on various laboratory tests and field measurements. Our new tool can improve our understanding of respiratory fluxes in tree stems. Given its low cost, it allows large-scale assessments of ecosystem carbon fluxes with sufficient replication, and the use of O 2 sensors in addition to CO 2 sensors represents a substantial improvement for assessing the importance of tree physiological factors in ecosystem carbon fluxes.

Gas sensors
In the final version of our chambers, we measured CO 2 concentration with the COZIR non-dispersive infrared (NDIR) absorption sensor (Gas Sensing Solution GSS, Cumbernauld, UK), O 2 concentration with the LuminOx Optical fluorescence quenching sensor (sealed, LOX-02-S; SST Sensing Ltd, Coatbridge, UK) and H 2 O concentration with a high precision humidity sensor (Digital Humidity Sensor SHT-85 (RH/T), Sensirion, ZH, Switzerland). Measurements of H 2 O concentration are necessary for the correction of O 2 measurements (see 'Correction of measurement data (O 2 ) for the dilution effect of changing H 2 O and CO 2 concentrations'). Initially, we instead used the relative humidity sensor integrated in the COZIR. However, detailed laboratory tests revealed significant mismatches between known and measured humidity and a very slow reaction time for the COZIR built-in sensor. Therefore, we switched to the more accurate and rapidly responding SHT-85 sensor, using the manufacturer's calibration (detailed humidity test results from Tree Physiology Volume 41, 2021 the laboratory and field can be found in S1 and S2 available as Supplementary data at Tree Physiology Online). The COZIR has a CO 2 measurement frequency of 2 Hz and includes a temperature ( • C) and relative humidity (%) sensor. Using the internal filter feature, we set the sensor to report running means of the last 50 measurements or 25 s. We do this to reduce high frequency noise and smooth the CO 2 readings, though this leads to an overall slower response time to concentration changes (details see GSS Sensor User's Manual 2015). The COZIR sensor allows for one-point calibration (see 'Multiple sensor calibration and testing unit').
The LuminOx O 2 sensor measures the partial pressure of O 2 (ppO 2 ; (mbar)), the total pressure (pO 2 ; (mbar)) and temperature (TO 2 ; (C o )). The O 2 sensors do not allow for changing the manufacturer-supplied calibration parameters unique to each sensor, which are determined by exposing the sensor to different oxygen concentrations, temperatures and barometric pressures in an environmental chamber. Device specifications are provided in the datasheet of the supplier (see also Table 1).

Chamber design and measurement principle
We designed a modular measurement system (Figure 1; see S3 available as Supplementary data at Tree Physiology Online) consisting of (i) the chamber module for creating a gas tight measurement headspace on the stem surface; (ii) a waterproof housing for the CO 2 and O 2 sensors mounted on top of the chamber; (iii) a separate housing for a pump; and (iv) a waterproof transport-case containing the power supply and an Arduino ® logging and control unit. Our device operates as a closed-cycle system, alternating between incubation periods for measuring CO 2 /O 2 /H 2 O concentration changes over time, and periods for flushing the chamber headspace with ambient air.
The chamber module is made of a 5-mm-thick polyethylene high-density sheeting and is 10 cm wide and 20 cm long, mounted on a 4-cm-thick closed-porous cell foam (EPDM, ethylene propylene diene monomer rubber). The foam is placed between the plate and the tree stem prior to fixing the chamber module with three ratchet straps. If required for an air-tight fit, the bark underneath the foam can be smoothed with sandpaper or an angle grinder, but this requires extreme caution to avoid any damage to the underlying cambium and phloem. Putty butyl sealant (Teroson RB IX, Henkel, Düsseldorf, Germany) is then applied around the edges of the foam to cover potential small leaks associated with remaining bark irregularities. Installed chambers are tested for leaks by blowing high CO 2 air around the edges while monitoring the headspace CO 2 concentration. Leaks are closed with putty and by refastening the straps until repeated tests show no further leaks. Because chamber headspace volume varies with stem geometry and ratchet strap tension from ca 75 to 112 cm 3 , it is measured for each installation by filling the headspace (in situ) with water from a calibrated syringe. To do so, we use two syringe needles (inserted from the top of the chamber, between bark and foam, to reach the headspace volume), one for injecting the water, the other to vent air from the chamber until water droplets appear. Sensors should be removed during headspace measurement.
The CO 2 , O 2 and H 2 O sensors are placed inside a waterproof (IP66 standard) housing (acrylonitrile-butadiene-styrene copolymers; dimensions: 10 cm L × 10 cm W × 7.5 cm H) with a removable lid. For easier handling, the sensor box is permanently welded to the base plate (polyethylene highdensity) of the chamber module. The sensors are placed over drill holes on 1-cm-thick EPDM rubber rings, allowing gas diffusion from the chamber headspace into the sensors. The O 2 sensor has a sealed sensor base (LOX-02-S) to guarantee that gas from the chamber headspace cannot leak through the sensor into the ambient air. For the CO 2 sensor, such a design is currently not available, so we seal the sensor base by applying hot glue around the electronic pins. The polyethylene plate and the sensor box are fully covered with adhesive aluminium Tree Physiology Online at http://www.treephys.oxfordjournals.org Figure 1. Custom-made modular stem gas exchange system featuring a CO 2 and an O 2 sensor for repeated cyclic measurements of changes in gas concentration over time within a closed dynamic stem chamber installed on the tree stem. A pump automatically flushes the chamber headspace in between measurement cycles. The device is battery powered with up to 10 days operating time. An Arduino ® controls switching between measurement and flushing mode and logs the data on an SD card. © Annett Börner foil to prevent heating from sun exposure. We further reduce potential temperature variations by installing the chambers on the side of the tree receiving the least direct sunlight during the day (i.e., north in the northern hemisphere). The air inlet and outlet of the chamber are connected to the pump via metric tubing (PVC tubing, RS Components GmbH, Frankfurt am Main, Germany) with a diameter of 6 mm (outer) x 3 mm (inner). The sensors (CO 2 , O 2 and H 2 O) are connected to a 5 V power bank (52,800 mAh, Li-ion type, MP-50000, XTPower, Seattle, WA, USA) in a separate waterproof transport-case (72,601 Outdoor Dry Box, Dyntronic GmbH, Glashütten, Germany), allowing the system to remain operational unattended for up to 10 consecutive days. The chamber operation and data logging are controlled by a custom-made Arduino ® (Arduino Mega 2,560 Rev3, Arduino S.r.L.) device (see S4 available as Supplementary data at Tree Physiology Online). The Arduino ® processor-based device enables us to program the measurement time interval, incubation time, duration of chamber headspace flushing and to define a CO 2 concentration threshold in which chamber flushing is desired. Our measurement approach relies on the ability to accurately measure concentration change rather than absolute concentrations; hence, we focus here on the rates of change of both gases over time. For a component list of the stem chamber (main parts), see S5 available as Supplementary data at Tree Physiology Online.

Multiple sensor calibration and testing unit
We built a multiple sensor calibration and testing unit, consisting of a gas-tight chamber with a volume of 1945 ml featuring two valves for flushing and a series of slots for simultaneous operation of up to 10 LuminOx and 10 COZIR sensors (see S3 available as Supplementary data at Tree Physiology Online). All 20 sensors are connected to a computer via multiple USB hubs.
For calibration of the COZIR sensors, we used one reference gas of known CO 2 concentration (approx. half of the maximum sensor range, here referred to as 'span' gas). First, we flushed the calibration unit for 10 min at 2 l min −1 with span gas, then closed it and let the reading stabilize (SD ≤ 30 p.p.m. for CO 2 for at least 10 measurements) before the respective calibration parameter was adjusted according to the span gas concentration (program: Microsoft Visual Basic). The calibration program allows reading and logging of data, provides access to the filter setting of the COZIRs and for defining stabilization criteria, and (when readings stabilize) sensors can be calibrated to known gas standard. The COZIR sensors allow to automatically store the new calibration parameters internally.
Direct calibration of the LuminOx sensors was not possible since there is no option to adjust their internal calibration parameters. We therefore conducted indirect calibration by comparing the sensors' measurement to known O 2 concentrations Tree Physiology Volume 41, 2021 Implementing O 2 sensor in stem respiration studies 1771 of gas mixtures being measured. The focus was on accurately measuring changes of the O 2 concentration rather than the absolute concentration since concentration changes over time are the parameter the chamber flux measurements are based upon. We placed the LuminOx sensors in the calibration unit and exposed them to synthetic air with an O 2 concentration of approx. 20.95% O 2 . We diluted the headspace concentration by injection of 30 ml pure N 2 into one port of the unit while extracting 30 ml of the unit's air on the opposite side to keep pressure in the chamber constant. We recorded the measurement after equilibration (i.e., SD ≤0.01% for at least 10 measurements) and repeated the dilution several times.

Testing of the sensors under laboratory and field conditions
We performed the following tests to validate accuracy and linearity of the COZIR and LuminOx sensors: measurements of a range of known concentrations (CO 2 , O 2 ) under (i) standard laboratory conditions; (ii) under varying temperatures; (iii) at increasing time intervals since last calibration (sensor drift); (iv) measuring the headspace over germinating wheat seeds as a biological model system of carbohydrate catabolism (with an expected ratio of CO 2 production: O 2 consumption of 1); (v) direct comparison of the COZIR sensor with another widely used commercially available CO 2 sensor (GMP252, Vaisala GmbH, Helsinki, Finland) under field conditions; and finally (vi) a field application test of the complete chamber setup.
Measurements of known gas concentrations under standard laboratory conditions As a reference point, sensor readings were tested against a range of known concentrations. Sensors were mounted in the calibration and testing unit, which was subsequently flushed with known gas concentrations. For the COZIR sensor, we used calibrated reference gas bottles (Westfalen AG, Münster, Germany) with CO 2 concentrations of 420, 2944 or 6020 p.p.m.. For O 2 , we followed the same procedure as described in the 'Multiple sensor calibration and testing unit'. We performed three dilution steps (20.95, 20.68, 20.41 and 20.16% representing the mean O 2 concentration from 20 LuminOx sensors at 25 • C).
Measurements of known gas concentrations: effect of temperature changes and sensor drift over time For extended field application, it is important to test whether temperature changes or sensor drift over time affect the gas measurements. All tests were performed using 10 CO 2 and/or 10 O 2 sensors. For the temperature test, the equipment (including the reference gas bottles) was set up in a phytochamber where we could control ambient temperature. Tests for the COZIR and LuminOx sensors were done separately. Tests started by adjusting the phytochamber temperature to either 5, 10, 20 or 25 • C (for COZIR sensors) or 5, 15 or 25 • C (for LuminOx sensors), followed by a period for equilibration of all materials and gas cylinders. We then followed the same procedure as explained in the 'Measurements of known gas concentrations under standard laboratory condition' (repeated for the different temperature levels).
Sensor drift was determined by repeatedly testing field installed sensors over a period of 3 weeks without re-calibration, each time measuring known gas concentrations (CO 2 : 420, 1430, 2944 and 6020 p.p.m.; O 2 : 20.95, 20.68 and 20.41%). Sensor readings were tested before initial installation see "Measurements of known gas concentrations under standard laboratory conditions". The sensors used in this drift test were installed and recorded data in the field between successive tests. They were tested in the lab 14, 18 and 22 days after initial installation and then re-installed in the field without re-calibration.
Incubation of germinating wheat seeds We tested sensor performance by measuring the CO 2 emission and O 2 uptake of germinating wheat seeds. While absolute fluxes in this setup are unknown, the ratio of CO 2 production to O 2 consumption (respiratory quotient, RQ) is expected to be 1 since wheat seeds exclusively use carbohydrates as respiratory substrates (Stiles andLeach 1933, Lambers et al. 2008). Wheat seeds were soaked in water overnight and placed in the calibration and testing unit (laboratory conditions, 25 • C). Headspace concentrations were measured for 80 min (n = 10 for COZIR and LuminOx sensors, respectively). Corrections for the dilution effect on O 2 (by CO 2 and H 2 O) were implemented (see 'Correction of measurement data (O 2 ) for the dilution effect of changing H 2 O and CO 2 concentrations').
Sensor comparison under field conditions: COZIR vs Vaisala GMP252 COZIR measurements were compared to the Vaisala GMP252 NDIR CO 2 sensor frequently used for xylem CO 2 measurements inside the stem (e.g., Saveyn et al. 2008, Cerasoli et al. 2009, Bloemen et al. 2014, Salomón, Valbuena-Carabaña, Teskey, et al. 2016, Fan et al. 2017). The Vaisala GMP252 has a measurement range of 0-10,000 p.p.m. CO 2 with accurate p.p.m.-level CO 2 measurements (device specifications are provided in the datasheet of the supplier: Vaisala GMP252 Carbon Dioxide Probe 2020). We designed a special version of our chamber system for simultaneous measurements of both the COZIR and the Vaisala GMP252 sensor in the same chamber under field conditions. Three such chambers were installed at three positions on the same tree (Prunus avium L., mean stem diameter: 105 cm) and measured CO 2 efflux for 1 week in September 2019 (14th-20th) in Jena, Thuringia, Germany.
Field application test Parallel to our sensor test in the laboratory, we tested several chambers under field conditions. These field tests (see 'CO 2 and O 2 flux measurements in the field') were conducted in the Thuringian Forest; Germany Tree Physiology Online at http://www.treephys.oxfordjournals.org

Correction of measurement data (O 2 ) for the dilution effect of changing H 2 O and CO 2 concentrations
Oxygen as a non-trace gas is sensitive to concentration changes of any other gas (Keeling et al. 1998). Measured apparent O 2 concentrations thus have to be corrected for changes in CO 2 , H 2 O and even for changes in O 2 itself (self-dilution) (Keeling et al. 1998, Bugbee andBlonquist 2006). We would like to clarify that these corrections have nothing to do with a crossinterference of other gases on the sensor signal, and we are not aware of any cross-interference issues in the sensors we used.
Previous publications usually linked the correction with simultaneously expressing the O 2 concentration changes relative to an arbitrarily defined reference concentration in per meg (= per million), thus following the original method as described by Keeling et al. (1998). This conversion undeniably has clear advantages when aiming at reporting and comparing measurements of absolute atmospheric O 2 concentrations and their changes over time. However, considering the focus and target group of our new application, we decided to go for a more intuitive approach that does not require the reader to get acquainted with the per meg scale. Since we are interested in simultaneous concentration changes of CO 2 , O 2 and H 2 O during separate incubation cycles, we refrained from expressing all concentrations relative to a fixed standard, and rather decided to express all concentrations as relative to the starting concentration ([CO 2 ] t = 0 , [O 2 ] t = 0 , [H 2 O] t = 0 ) during a measurement cycle. The measured apparent change in the O 2 concentration (δO 2,app ) at any given time during a measurement cycle is correcting for the effect of any observed changes in the concentration of CO 2 and H 2 O relative to the starting concentration (δCO 2 and δH 2 O). These corrections are proportional to the mole fraction of O 2 in the gas mixture, i.e., X O2,t = 0 , which equals the apparent O 2 concentration in p.p.m. divided by 10 6 .
Tree Physiology Volume 41, 2021 Since relative humidity is 100% at saturation vapour pressure, the ambient partial pressure of H 2 O (ea, hPa) can be calculated as a function of the current relative humidity ea = es × RH. (3) The current concentration of H 2 O thus equals the ratio of ea to the total atmospheric pressure of all air inside the chamber (P), which we report in p.p.m. here H 2 O (p.p.m.) = ea P × 10 6 . (4)

Flux calculation and apparent respiratory quotient
The CO 2 and O 2 fluxes (F, μmol m −2 s −1 ) were calculated according to the following equation: where C/ t (hereafter referred to as slope) is the change in concentration of gas C (in p.p.m.) over time t (s −1 ) for CO 2 and O 2 , respectively. V is the volume of the chamber (m 3 ), P the barometric pressure (kPa), R the molar gas constant (0.008314 m 3 kPa K −1 mol −1 ), T the temperature (K) and A the stem surface area (0.0028 m 2 ). P and T are recorded from the sensors. We assumed linearity in the first 20 min of measurement (time is required to measure changes in the O 2 and CO 2 concentration of at least 1000 p.p.m., 0.1%) and therefore used the slope of the linear regression to calculate CO 2 (increasing concentration) and O 2 (decreasing concentration) change over time. The negative slope of O 2 is always given as absolute value.
Only measurements with correlation coefficient (R 2 ) > 0.96 for both O 2 and CO 2 were used. Lower correlation coefficients were discarded. For field data, the first 5 min of each measurement cycle (user-defined) were discarded after we noticed these data are noisy, probably due to pressure fluctuations after the pumping period. The sensors' readings were extracted every 10 s. The ratio of CO 2 efflux and O 2 influx results in the Apparent Respiratory Quotient (Angert and Sherer 2011; ARQ; Eq. (6)) as we do not measure the actual RQ of respiring cells (see 'Incubation of germinating wheat seeds') but the stem equivalent of RQ ARQ = CO 2 production O 2 consumption .

Statistical analysis
Data from temperature and drift tests were analyzed by regression analysis. As shown in Eq. (5), flux rates for the chambers were calculated from concentration changes over time. Thus, sensors have to reliably measure concentration changes. We estimated how much the slope of a linear regression of measured concentration vs known concentration was affected by temperature, or time (drift). We examined the slope of the regression, where slope of 1 represents a perfect agreement, slope <1 means the sensors underestimating the true concentration change and vice versa. Slopes measured by individual sensors (n = 10) were tested for temperature and time effects. For the wheat seed respiration test, linear regressions and comparisons among slopes of CO 2 increase and O 2 decrease over time were performed, too. For normally distributed data (after checking the assumption of heteroscedasticity with a Levene test) one-way ANOVA was used for comparison among slopes. Correlation between COZIR and Vaisala GMP252 was evaluated by Pearson's correlation coefficient. Data from CO 2 and O 2 flux measurements in the field (see 'CO 2 and O 2 flux measurements in the field') were combined to 4 h mean for further analysis. All statistical analyses were conducted in R 3.4.4 (RStudio Team 2016).

Measurements of known gas concentrations
Measurements of known CO 2 concentrations by the COZIR sensors showed good accuracy and precision (Figure 2A). Using 10 sensors, the linear regression for measured vs known concentration had a slope of almost unity (0.99) and variability between the individual sensors was very small as reflected in the R 2 of 1.0. For the LuminOx sensors ( Figure 2B), measurements of known O 2 concentrations also resulted in a slope for the linear regression close to 1 (0.98), indicating that the sensors can reliably detect relative concentration changes over the tested range. However, the variability between individual sensors in terms of absolute concentration measurements was far greater than for the COZIR sensors, with individual sensors being off by almost ±0.5%. This results in a low R 2 of the regression of only 0.68.

Measurements of known gas concentrations as affected by temperature changes, and sensor drift over time
Changes in temperature did affect O 2 but not CO 2 measurements. For CO 2 measurements, linear regression parameters of measured vs known concentrations show no statistically significant effect of temperature with mean slopes (±SD) of 0.96 ± 0.05, 0.97 ± 0.03, 0.99 ± 0.01 and 0.97 ± 0.02 at 5, 10, 20 and 25 • C, respectively (Table 2). For O 2 , differences in temperature were highly significant with mean slopes (±SD) of 0.81 ± 0.06, 0.86 ± 0.03 and 0.97 ± 0.02 at 5, 15 and 25 • C, respectively ( Table 2). The sensors themselves do not differ significantly (ANOVA, COZIR: F = 0.12, P = 0.73; LuminOx: F = 0.43, P = 0.52). Based on these findings, we formulate the following temperature correction for the O 2 sensors: Tree Physiology Online at http://www.treephys.oxfordjournals.org Table 2. Rate of change (slope) for different temperature levels and sensor drift over time (mean ± SD). Analysis of variances (ANOVA) for temperature and drift effect on COZIR and LuminOx reading (n = 10) is shown. Significant results are shown in bold.

COZIR-AH1 LuminOx
Effect (temperature/duration) mean slope ± SD ANOVA (F, P) mean slope ± SD ANOVA (F, P) where SC is the slope correction factor that should be multiplied with the measured slope in stem chamber incubation in temperature T. As fluxes were estimated based on slope changes over time (see 'Flux calculation and apparent respiratory quotient'), only a slope correction was considered. Time elapsed since last calibration had an effect on CO 2 measurements when exceeding 22 days, i.e., sensor drift affected the CO 2 sensors after ca 3 weeks (Table 2). Shorter time intervals (14 and 18 days) showed no significant effect on CO 2 measurements. Mean slopes (±SD) from linear regression parameters of measured vs known CO 2 concentrations were 0.97 ± 0.05, 0.96 ± 0.05 and 0.95 ± 0.04 after 14, 18 and 22 days of field operation, respectively. We found no effects of sensor drift over time for the O 2 sensors, with mean slopes (±SD) of 1.02 ± 0.10, 1.03 ± 0.10 and 1.03 ± 0.08 after 14, 18 and 22 days of field operation, respectively. The sensors themselves do not differ significantly (ANOVA, COZIR: F = 1.06, P = 0.44; LuminOx: F = 1.36, P = 0.28).

Comparison of CO 2 measurements between COZIR and Vaisala sensors
We found a significant correlation between simultaneous measurements of COZIR and Vaisala GMP252 (Pearson correlation coefficient, R = 0.95; P < 0.001; Figure 4). However, efflux rates measured with the COZIR sensor were consistently lower than rates measured with the Vaisala (up to 11%).

CO 2 and O 2 flux measurements in the field
During a typical measurement cycle ( Figure 5A and B), CO 2 rapidly increases from atmospheric levels (∼400 p.p.m.) to ∼6000 p.p.m. (depending on season and time of day) within a 45-min period while the net O 2 decrease is ∼0.7%. Following a typical cycle, it takes ∼15-min flushing period (starting at 20:40h in Figure 5) to reach initial concentrations again. For analysis, we focus on the initial 20 min of measurement (beginning 5 min after pumping stopped), for which we assume linearity. In our example, over the 20-min period, we observed changes in the O 2 and CO 2 concentration of 4000 p.p.m. and 0.4%, respectively ( Figure 5C and D). Relative humidity can vary up to ∼8% over the 20-min time interval (change of 1560 p.p.m. at standard condition of 20 • C, 945 hPa).
Calculated fluxes (4 h mean) from the field application test on three poplar trees are presented in Figure 6. Two correction steps for O 2 fluxes were implemented: correction (i) dilution effect on O 2 (CO 2 and H 2 O) and correction (ii) temperature effect on LuminOx readings. Dilution correction (including selfdilution) results in an increase of the mean daily fluxes by 5.6 ± 2.2% ( Figure 6A), and the subsequent temperature correction from LuminOx readings as determined by our laboratory test results in an increase of the fluxes (daily mean increase of 12.7 ± 4.5%; Figure 6A). Over the 2-week measurement period, CO 2 efflux is lower than O 2 influx ( Figure 6B); this would result in daily mean ARQ of 0.63 ± 0.06.

Discussion
In our study, we were able to show that the combination of three low-cost sensors (CO 2 , O 2 , H 2 O) allows reliable and quasicontinuous measurements of CO 2 and O 2 stem gas exchange under field conditions. Being affordable, highly mobile and independent of additional infrastructure like local power supply or external logging devices makes our setup highly attractive for application in remote ecosystems or for measuring many individuals and/or widely dispersed trees simultaneously.

Technical aspects and sensor performance
The sensors installed in our chamber design produced robust measurements. Obviously, such low-cost devices have caveats that one has to be aware of. Our chambers followed a non-steady state incubation design, aiming to measure concentration changes of several thousand p.p.m.; therefore, we only tested the sensors' performance of measuring relatively big changes of concentration. Especially for the O 2 sensors, measurements of known concentrations revealed high variability between sensors and-for individual sensors-a significant offset between measured and known concentrations, making accurate measurements of absolute concentrations questionable. Accuracy for O 2 measurements in terms of absolute values is 20 times lower than for CO 2 (see sensor specifications in Table 1).
For the COZIR sensors, one important limitation was the effect of sensor drift over time (Table 2). According to our findings, the sensors can be operated without loss of precision for a maximum of 18 days before a new calibration is required. Since calibration has to be done in the laboratory, it is recommended to keep additional calibrated sensors in stock for rapid exchange in the field. Especially when planning to operate a high number of chambers simultaneously, one should consider manufacturing a multi-sensor calibration unit similar to ours for efficient re-calibration.
Direct comparison of the COZIR sensor to the more expensive Vaisala GMP252 showed an offset between the two sensor types, with the COZIR measuring on average 11% smaller fluxes than the Vaisala (Figure 4). Since this comparison was performed under field conditions with unknown concentrations, it is impossible for us to conclude which of the sensor types has the better accuracy. We did not test or re-calibrate the Vaisala in the lab, instead we relied on the manufacturer calibration. The COZIR sensors, on the other hand, were calibrated and thoroughly tested under various laboratory conditions, but calibration was done in dry air, so it is possible that humidity Tree Physiology Online at http://www.treephys.oxfordjournals.org  Tree Physiology Volume 41, 2021 affected the measurements under field conditions. However, the respiration measurements of germinating wheat seeds matched expectations (see below), which indicates that the COZIR works reliably under a realistic humidity range of typical field measurements (50-85%). Assuming that the differences between the Vaisala and the COZIR sensor were due to measurement errors of the COZIR sensor, we would have to consider correction of the field test by up to 11%, thus reducing the observed imbalance between measured O 2 and CO 2 fluxes. We would like to point out, however, that this offset is smaller than the observed differences in the field application test.
Field tests in a tropical rainforest (Tanguro ranch, Matogroso, Brazil) indicated complete sensor failure at relative humidity levels between 95 and 100%. One possible solution for tropical applications of stem chambers can be to use humidity traps as shown by Brecheisen et al. 2019 for a field-portable soil gas analyzer. Brändle and Kunert (2019) presented a stem chamber design with an implemented low-cost CO 2 sensor type MH-Z14A (Winsen Electronics Technology Co., Ltd, Zhengzhou, China). They found good agreement of their device with a portable infrared gas analyzer (Li-8,100; Li-COR Inc., Lincoln, NE, USA) under tropical rainforest conditions at high temperature and high humidity. However, their chamber design was still limited to CO 2 measurements.
We found two problems that were specific to the LuminOx sensors. First, changes in temperature did affect LuminOx sensor reading. At lower temperatures (5-15 • C), O 2 fluxes based on sensor readings significantly underestimated actual O 2 fluxes and required a temperature correction (Eq. (7)). Our data indicated that the effect of temperature on sensor readings can differ between individual sensors. In our approach, we used the average correction determined by measuring 10 individual sensors, but correction parameters for individual sensors deviated from the mean by as much as ±8% (see S7 available as Supplementary data at Tree Physiology Online). In order to avoid this deviation, correction functions for each sensor individually can be applied. We only tested sensor performance between 5 and 25 • C, so it remains uncertain how the sensors behave at temperatures outside of this temperature range. Second, the fact that the sensor does not support adjustment of the manufacturer calibration parameters makes working with this sensor less convenient. However, since we observed no critical sensor drift as with the COZIR sensor, this issue was less of a problem, but it should still be kept in mind when considering using this sensor. In addition to the abovementioned temperature correction, the O 2 sensors also require a dilution correction to compensate for apparent changes in O 2 concentration resulting from concentration changes of other gases (mainly CO 2 and H 2 O). Please note that this is not a correction resulting from technical issues of the sensors, but is a general requirement when measuring concentrations of non-trace gases like oxygen. In the dilution correction, O 2 self-dilution outweighed the dilution by other gases (CO 2 and H 2 O), resulting in an increase of ∼ 6% (two-week average) for calculated O 2 fluxes after correction. In our application, we used the relative humidity sensor SHT-85 successfully as this sensor responded quickly (within seconds) to changes in relative humidity, whereas the integrated COZIR RH sensor often underestimated actual humidity levels in the chambers as their response time is very slow to the increasing humidity (see S1 and S2 available as Supplementary data at Tree Physiology Online).
To evaluate actual sensor performance under realistic conditions, we measured respiration of germinating wheat seeds. This test allowed testing the sensors over a wider range of concentration changes and within a realistic humidity range. Wheat seeds are a suitable biological model system for this purpose as their carbohydrate-based respiration during the initial germination implies equal CO 2 and O 2 fluxes. Results from these wheat seed measurements confirmed that the sensors can reproduce expected values and work under field-humidity levels ( Figure 3). The test also underlined the shortcomings of the low-cost sensors with respect to absolute concentration measurements: while the sensors showed similar concentration changes (slopes) during the incubation, absolute concentration measurements at any given time were subject to major offset biases, especially for the O 2 sensors (see also Figure 2).
To ensure gas tightness, we decided to seal our chambers by means of closed-porous cell foam. The area of the tree stem covered by foam was relatively large compared to the chamber headspace area. Gas exchange for any live tissue underneath the area covered by the foam has to occur via an alternative surface, and some of it will occur via the chamber headspace surface. As a rule of thumb, one may assume that roughly half of the area covered by the foam should be considered as effectively being part of the chamber area. In any case, assuming that the resulting effect is identical for CO 2 and O 2 , we postulate that the area covered by foam has no impact on the gas exchange ratio or the ARQ.

Beyond CO 2 : potential application of simultaneous CO 2 and O 2 flux measurements in ecosystem and ecophysiological research
The combination of simultaneous CO 2 and O 2 measurements in a one-chamber design allows new additional research questions to be addressed. It could be used for the detection of respiratory substrate shifts during stress by calculating the ratio of CO 2 efflux to O 2 influx like demonstrated by Fischer et al. (2015) in a greenhouse experiment using Raman spectroscopy. Embedded in the correct experimental design, it could also help to quantify actual rates of local in situ respiration by disentangling respiratory CO 2 production and O 2 consumption from the effect of other post-respiratory processes (see Hilman et al. 2019 for a Tree Physiology Online at http://www.treephys.oxfordjournals.org detailed discussion). Using simultaneous measurements of CO 2 and O 2 fluxes in multiple tree species, they observed a significant mismatch in the amount of CO 2 emitted vs the amount of O 2 consumed, which they interpreted as the effect of a variety of whole-tree processes on locally measured CO 2 concentrations, like non-photosynthetic refixation or stem xylem transport of CO 2 away from (e.g., to canopy) and to (e.g., from roots) the site of measurement. Data from our initial field test also indicate mismatches between CO 2 and O 2 fluxes ( Figure 6B), which could be further explored in future experiments. Experimental approaches may include, for example, simultaneous CO 2 and O 2 measurements at different stem heights and in the canopy to quantify the effect of vertical gas transport. Combining flux measurements with 13 C isotope labeling of stem tissue in the dark could help to quantify the postulated non-photosynthetic CO 2 uptake in tree stems due to PEPC activity. With slight modifications, our chamber design may also be useful for measuring other ecosystem components like soil, root, branch or leaf fluxes. Furthermore, data provided by our device can serve as important input and calibration variables for mechanistic models of tree and stem functioning. For instance, Salomón et al. (2020) developed TreSpire, a process-based model, which couples carbon and water fluxes at the organ (stem) level. Implementation of combined CO 2 and O 2 data can provide crucial information to constrain the model parameter space. In this way, key parameters used to estimate overall tree respiration at large spatial scales-growth respiration coefficient, respiration sensitivity to temperature (Q 10 ) and basal maintenance respiration (Atkin et al. 2017)-could be accurately estimated. Such insights are much needed to improve model predictions and advance our understanding on stem respiration, for which currently measurements of CO 2 efflux at breast height are commonly used as estimates for whole-tree respiration, even though stem CO 2 efflux does not reflect respiration rates of underlying tissues (e.g., Teskey et al. 2008, Trumbore et al. 2013, Darenova et al. 2018).

Conclusion
We present a versatile low-cost chamber setup for measuring CO 2 and O 2 fluxes between tree stems and the atmosphere. Adaptation of the general setup to other applications (e.g., soil or branch measurements) should technically be relatively easy. We showed that low-cost sensors are prone to drift over time and/or require temperature correction. Our study also points out that O 2 sensors require dilution correction to get accurate O 2 data that are not biased from concentration changes of other gases (CO 2 and H 2 O). Using both CO 2 and O 2 measurements in the correct experimental design provides additional information on tree physical and physiological processes like xylem CO 2 transport, post-respiratory enzymatic fixation of CO 2 and subcortical photosynthetic uptake of respired CO 2 .

Supplementary data
Supplementary data for this article are available at Tree Physiology Online.