We measured leaf fluxes of COS, CO2, and water from 31 May to 6 July 2013 (day of year 151–187) at the San Joaquin Freshwater Marsh (SJFM, 33∘39′44.4′′ N, 117∘51′6.1′′ W). The SJFM is located near the campus of the University of California, Irvine, at 3 m above sea level and 8 km northeast of the Pacific Ocean . The SJFM is part of the University of California's Natural Reserve System. The site's history and management practices have been described in . Briefly, the SJFM is a mature freshwater marsh, the remnant of a once 2100 ha wetland along the San Diego Creek. Since the 1960s, the SJFM has been managed by flooding the area annually to a depth of approximately 1 m from December–January to March. The standing water recedes by evapotranspiration and subsurface drainage and eventually disappears by midsummer . A flux tower (5 m tall) is located on a floating wooden platform near the northeastern edge of the SJFM. The platform is surrounded by dense vegetation dominated by Typha latifolia (broadleaf cattail). In contrast to upland species in a mediterranean climate that grow in the rainy winter or early spring, the growing season of the marsh plants is summer due to the standing water.

Leaf fluxes of COS, CO2, and H2O were measured with a flow-through (dynamic) chamber (Fig. a). The cylindrical chamber (18 cm diameter, 38 cm height, 10.3 L volume) consisted of PFA Teflon film stretched between two aluminum rings connected by rods. The PFA film was laid inside the structure such that only the film was in contact with the sampled air. The chamber enclosed the upper sections of six tall T. latifolia leaves of the same plant with an average width of 1.5 cm. The leaves extended above and below the chamber. The total leaf area in the chamber was estimated as 409.5 cm2 by approximating the area of each leaf with a one-sided rectangle (length intersected by the chamber × width). Skirts of Teflon film were wrapped around the leaves to provide a seal at both ends of the chamber. Due to limitations on the sampling time of the COS analyzer, we did not install a replicate leaf chamber, but instead chose a high sampling frequency for the single leaf chamber.

Two fans were installed in the chamber for ventilation and mixing, respectively. On the inlet end, a high-speed axial fan (D344T, Micronel; 40×40 mm) provided ventilation to keep the chamber at ambient conditions (i.e., within 1 ppmv of ambient CO2, tested at the start of the campaign). A second, smaller flat fan (F62, Micronel; 16×16 mm), attached to a stainless steel rod, was placed near the center of the chamber for air mixing. During the measurement period, the ventilation fan was turned off and its opening served as the inlet to allow airflow through the chamber. The mixing fan, in contrast, was kept running at all times.

The chamber was connected via a 0.25 in. PFA Teflon tubing to a Quantum Cascade Laser (QCL) analyzer (CW-QC-TILDAS, Aerodyne Research Inc., Billerica, MA, USA), with a 1 µm Teflon filter attached at the inlet of the analyzer. The analyzer was placed in an instrument enclosure on the platform. Flow through the analyzer was provided by a Varian TriScroll 600 pump (Agilent Technologies Inc., Santa Clara, CA, USA). Flow rate in the sampling tube was 6.4 standard liter per min (sLm), which corresponded to a chamber air turnover time of around 1.5 min. The pump was placed next to the nearest main power line near the entrance to the marsh site, and connected to the analyzer by a 150 m long 2-in. vacuum line. A solenoid valve at the inlet to the QCL was used to switch from the sampling line to a stream of dry N2 (ultrahigh purity) for a 1 min background correction every hour. Data from the QCL analyzer were recorded at 10 Hz and stored on the QCL hard drive. The root-mean-square deviation of COS measurements at 10 Hz was 11–18 parts per trillion in volume (pptv).

Correction for water vapor effects on the dry mixing ratios of COS and CO2 was done in the TDLWintel data acquisition software on the analyzer . We did not use the same correction factors reported in for the same make of QCL analyzer; however, a mock run of data processing with CO2 concentration recalculated using their correction factor value resulted in a potential bias of only 0.12 % (r2=0.999). Thus, the flux uncertainty associated with the correction factor of water vapor effects was negligible (see the Supplement for details).

The leaf chamber was measured once per hour. Chamber operations were programmed on a CR1000 datalogger (Campbell Scientific, Inc., Logan, UT, USA). We monitored chamber air concentrations for a 5 min measurement period (i.e., while the ventilation fan was off), as well as the ambient air for 1 min before and after measurement periods (i.e., while the ventilation fan was running). Leaf fluxes were calculated from the transient changes with respect to the interpolated inlet (ambient) concentrations (Fig. b). The apparent fluxes from the chamber material (PFA), characterized post hoc, were negligible – the blank effects translated to apparent fluxes of 0.05 ± 0.29 pmol m-2 s-1 for COS and 0.02 ± 0.15 µmol m-2 s-1 for CO2 when normalized against the leaf area (see the Supplement).

Various sensors were installed to record environmental data, including photosynthetically active radiation (PAR) (SQ-215, Apogee Instruments), ambient air temperature and humidity (HMP45AC, Vaisala), and chamber air and leaf temperature (type T thermocouples, PFA coated). These data were recorded at 10 s intervals on the CR1000 datalogger. Because of a wider gap in the canopy to the west of the chamber than to other directions, the chamber received slightly more light in the afternoon than in the morning. To account for the heterogeneity of the light microenvironment of the chamber, the PAR sensor was collocated with the chamber. All sensor data are released alongside the flux data (see “Data availability”).

A mass balance equation is formulated for the gas species being measured (COS, CO2, or H2O), 1VdCdt=qCa-C+FA, where C (mol m-3) is the chamber headspace concentration of the gas, Ca (mol m-3) is the inlet (ambient) concentration, q (m3 s-1) is the inlet flow rate, V (m3) and A (m2) are the chamber volume and leaf area, respectively, and F (mol m-2 s-1) is the flux rate to be calculated. Solving the mass balance equation with the initial condition C(t=0)=Ca, we obtain 2Ct=-FAqexp⁡-qt/V+Ca+FAq. The flux rate F is 3F=qA⋅C-Ca1-exp⁡-qt/V. Let y^=C-Ca and x^=exp⁡-qt/V be the variables for the regression, hence, 4y^=FAq1-x^. The flux rate F is then solved from the slope of the regression y^∼1-x^. The standard error of the estimated F is also obtained from the regression. The flux calculation method described above does not require a steady state to be reached in the chamber. A typical example of the chamber measurement period with the fitted curve of COS concentration changes is shown in Fig. b.

All leaf flux and meteorological data have been quality checked and filtered. Conspicuously unrealistic data points in the meteorological data were removed. For the flux data, we used several independent criteria to filter measurements. First, measurement periods with a serious misfit in the shape of concentration changes during chamber closure or with strong drift in the ambient concentrations were discarded. Second, flux estimates associated with large root-mean-square errors between fitted and observed concentrations were also discarded. Next, outliers in flux data were detected using the Tukey's interquartile range method . In addition, strongly positive CO2 fluxes during the day and strongly negative CO2 fluxes at night were also removed. Only the data points that passed all these filtering criteria were kept in the final data for analysis. After the filtering, 73.9 % of COS flux observations and 54.3 % of CO2 flux observations were retained.

We used the LOWESS (locally weighted scatterplot smoothing) regression method to obtain smooth light response curves for COS flux, CO2 flux, and LRU (see Fig. ). The LOWESS regression method is a nonparametric method that does not require any a priori known relationship between the predictor (here, PAR) and the response variables (COS flux, CO2 flux, and LRU). At each point in the range of the predictor, a low-degree polynomial is fitted to all the neighboring points to estimate the least squares response, weighted by the distances between the neighboring points and the current point . The calculation was performed with the Python statsmodels package, version 0.8.0 .

During the campaign period in summer 2013 covering the peak growing season of Typha latifolia, meteorological conditions changed little except for a few cloudy days (8, 9, and 30 June 2013 in Fig. d), and the diurnal patterns of leaf COS, CO2, and H2O fluxes therefore also remained similar (Fig. a–c). The diurnal patterns of leaf fluxes and related variables are visualized with hourly binned medians and quartiles (Fig. ).

In the daytime, leaf uptake of COS and CO2 showed similar patterns (Fig. a, b), with uptake peaks in the morning and afternoon separated by a prolonged midday depression around local noon (13:00). The midday depression was up to 36 % for COS (5.5 pmol m-2 s-1 at 14 h vs. 8.5 pmol m-2 s-1 at 11 h) and 40 % for CO2 (3.7 µmol m-2 s-1 at 13 h vs. 6.1 µmol m-2 s-1 at 17 h), respectively. The morning peaks coincided for the two fluxes at around 11:00, whereas the afternoon peak occurred a bit later for COS (18:00) than for CO2 (17:00). The afternoon peak of CO2 flux was slightly stronger than its morning peak (Fig. b), probably because the chamber received slightly more light in the afternoon than in the morning (Fig. e). Leaf transpiration showed a decline at 11:00 (Fig. c), but with an earlier afternoon peak (16:00) that coincided with the maximum vapor deficit (Fig. f). Contrary to COS and CO2 fluxes, the diurnal pattern of water flux was strongly asymmetric due to the high vapor deficit in the afternoon (Fig. f).

In contrast to daytime fluxes, nighttime fluxes of COS and CO2 showed diverging patterns. At night, CO2 was emitted from leaf respiration (Fig. b), whereas COS uptake continued (Fig. a). Both fluxes had significantly smaller magnitudes than during the day, with CO2 emissions of around 1 µmol m-2 s-1, and COS uptake of around 2–3 pmol m-2 s-1. Note that although COS emissions were occasionally observed at night (Fig. a), they were likely caused by random error due to high flow rates (∼ 6 sLm), and the hourly medians indeed showed a robust pattern of nighttime COS uptake (Fig. a). When averaged over the whole campaign, nighttime COS uptake was 23 % of the total daily COS uptake by leaves. Nighttime transpiration was minimal (Fig. c) as the vapor deficit was close to zero at night (Fig. f).

COS flux was linearly correlated with CO2 flux (Fig. a), with r2=0.49 (p=7.6×10-64). The relationship between COS and water fluxes was nonlinear (Fig. b) and showed a wide spread in the daytime due to the asymmetric diurnal pattern of water fluxes (Fig. c). As a result, the correlation between them was lower (Fig. b), showing an r2 of 0.32 (p=4.7×10-57). The unbiased distance correlation (dCor; ) was also calculated as a more robust measure for the nonlinear correlation between COS and water fluxes, and dCor2=0.37. At night, COS fluxes showed stronger variability than water fluxes because vapor deficit that drives transpiration was small (Fig. f).

The midday depression was also evident in the light responses of fluxes. Both COS and CO2 uptake rates increased with PAR until they became light saturated, and then decreased at high light and high vapor deficit (Fig. a, b). According to the smoothed light response curves, at a typical midday light level (1800 µmol m-2 s-1), COS uptake drops by 37 % from the peak value of 7.5 pmol m-2 s-1 (at PAR = 493 µmol m-2 s-1) to 4.7 pmol m-2 s-1, while CO2 uptake drops by 31 % from the peak value of 5.3 µmol m-2 s-1 (at PAR = 740 µmol m-2 s-1) to 3.7 pmol m-2 s-1.

Stomatal conductance (gs,H2O) derived from water measurements showed a distinct period of midday depression in its diurnal pattern (Fig. a). gs,H2O was the highest in the early morning after daybreak, but started to drop quickly as the vapor deficit picked up, reaching its minimum at local noon (13:00). In the late afternoon, stomatal conductance slowly rebounded and remained relatively stable, but was still lower than the early morning level. Nighttime stomatal conductance was unable to be estimated from water measurements due to large uncertainty introduced by low vapor deficit and water flux.

The total conductance of COS (gtot,COS) exhibited broadly similar diurnal pattern to that of gs,H2O, but lagged by 1 h (Fig. a). A midday depression period was also visible in the diurnal trend of gtot,COS. At night, gtot,COS remained at a stable, low level.

The ratios of stomatal resistance to total resistance of COS (rCOS*) and of CO2 (rCO2*) indicated that stomatal limitation was the dominant component in the diffusional pathways of both gases during most of the daytime (Fig. b). Despite large uncertainties associated with these ratios, rCOS* was higher than rCO2* by 20–40 % around midday (10:00–13:00) at a significance level of p<0.05 (paired two-sample t-tests), indicating stronger stomatal limitation on COS uptake. However, in the late afternoon (15:00–17:00) the difference between stomatal limitations on COS uptake and on CO2 uptake was small and statistically insignificant (Fig. b).

The instantaneous leaf relative uptake (LRU) showed an asymmetric U-shape diurnal pattern (Fig. d). LRU had highest values of 2–3 (medians binned by the hour) near dawn or dusk, with a gradual decrease throughout the morning and early afternoon, and then had minima around 0.9 at 15:00.

The diurnal pattern of LRU (Fig. d) was consistent with the LRU response to PAR (Fig. c). With increasing PAR, LRU decreased to around 1.0 at PAR above 500–600 µmol m-2 s-1 (Fig. c). Surprisingly, the lowest LRU values during the day did not occur at the time of the highest PAR (Fig. d), but rather at the time of the highest vapor deficit (Fig. f) and moderately strong PAR (1000–1400 µmol m-2 s-1) due to the stronger stomatal limitation on fluxes as a response to the high vapor deficit. The timing of the lowest LRU (Fig. d), around 15:00, coincided with the timing of the highest vapor deficit.

The all-day mean LRU at this site showed large day-to-day variations (1.4–3.6) and also had large uncertainty due to the random error in nighttime CO2 fluxes (Fig. a). In contrast, the daytime mean LRU, averaged over the daylight period of 14 h, did not show strong variability (1.0–1.8) and had an average value of 1.2 across the campaign. The daytime mean LRU was consistently lower than the all-day mean LRU, since the latter included nighttime COS uptake and CO2 emissions (Fig. a). Following , a power law relationship was fitted between daytime mean LRU and daytime mean PAR: LRU=a⋅PARb (or rather, a linear model between ln⁡LRU and ln⁡PAR), which yielded a=24.0689 and b=-0.4620, with r2=0.28 and p=0.012 (Fig. b). On overcast days, the daytime mean LRU values were higher than on clear days (Fig. a), as is expected from the light response of LRU.

We have reaffirmed in field conditions that LRU decreases with increasing PAR (Fig. c), consistent with laboratory studies . The large sample size from high frequency measurements supported a robust analysis of LRU variability despite experimental limitations. Thanks to a strong diurnal variation of vapor deficit in this ecosystem, we were able to identify a further reduction in LRU caused by high vapor deficit – a secondary effect superimposed on the light dependence of LRU. But how are stomata responsible for the observed LRU responses?

Using the ratio of stomatal resistance to total resistance as a metric of the relative importance of stomatal limitation (Fig. b), we can recognize how the dynamics of stomatal vs. internal limitations regulate LRU. At the leaf scale, LRU manifests the ratio between the stomatal limitation on COS uptake (rCOS*) and that on CO2 uptake (rCO2*) (compare Eqs.  and to Eq. ): 14LRU≡gtot,COSgtot,CO2=0.83⋅rCOS*rCO2*, where 0.83 is the COS-to-CO2 ratio of diffusivity in air . The equation shows that LRU becomes smaller when rCOS* and rCO2* get closer, providing a simple mechanistic interpretation of LRU variability.

The light response of LRU arises from the fact that with respect to the same increase of PAR, the relative increase of COS uptake is less than that of CO2 uptake (Fig. a, b), i.e., 15∂LRU∂PAR<0⟺1FCOS∂FCOS∂PAR<1FCO2∂FCO2∂PARFCOS<0 and FCO2<0. Increasing PAR drives an increase in CO2 assimilation rates, which in turn leads to an increase in stomatal conductance to facilitate optimal CO2 uptake. This increase in stomatal conductance also enables higher COS uptake rates, but as COS hydrolysis is light independent , there is a proportionally less increase in COS than CO2 uptake. In other words, with the increase of PAR, both stomatal and biochemical limitations for CO2 assimilation are relaxed, whereas for COS only the stomatal limitation is relaxed. This explanation is supported by indirect evidence in rCOS* and rCO2*: from 06:00 to 13:00 there was a higher relative increase of rCO2* than that of rCOS* (Fig. b), which means the reduction of non-stomatal limitation – attributed mainly to the increases in biochemical reaction rates – is higher for CO2 than for COS.

Stomatal response to vapor deficit, such as the midday depression (Fig. a), is a well-known behavior that serves to optimize water cost against carbon gain e.g.,. However, the fact that vapor deficit has differential effects on COS and CO2 uptake appears puzzling, since it does not affect COS and CO2 biochemical reactions, and nor is it known to affect mesophyll conductance. A closer scrutiny of the stomatal limitations of COS and CO2 (Fig. b) shows that the difference between rCOS* and rCO2* became smaller during the period of peak vapor deficit (14:00–17:00). Although vapor deficit has the same effect on gs,COS and gs,CO2, it can change the proportion of stomatal vs. internal components in the total resistance to the uptake, because COS uptake is always more stomatal-conductance-limited than CO2 uptake (rCOS* always higher than rCO2* in Fig. b) – a direct consequence of the higher catalytic efficiency of CA than RuBisCO. Thus, vapor deficit controls LRU variability but is less influential than PAR.

Since the mesophyll conductance is also a component in the internal conductance, it is worthy of note that the increase of mesophyll conductance with leaf temperature may have contributed to the dynamics of stomatal vs. internal limitations over the course of the daytime, as is shown in , although we lack relevant data to separate biochemical limitation from mesophyll limitation.

During this campaign, nighttime uptake contributed to 23 % of the total daily leaf COS uptake. This fraction is comparable to those reported from a wheat field (29±5 %, ), an alpine temperate forest (25–30 %, ), a boreal pine forest (17 %, ), and a New England mixed forest (< 20 % after subtracting soil uptake, ). Collectively, these studies indicate that nighttime uptake is typically 17–30 % of the total canopy COS budget, a fraction too large to be ignored in ecosystem or regional COS budget. Understanding nighttime COS uptake is necessary for the success of COS-based photosynthesis estimates on daily and longer timescales.

The T. latifolia leaves showed a mean value of 5.0 mmol m-2 s-1 for the total conductance of COS (gtot,COS) at night (Fig. a). Assuming that the internal conductance of COS at night is the same as its daytime average, we obtain an estimate of nighttime gs,COS, 6.4 mmol m-2 s-1 (see the Supplement for detailed calculations). This estimate of the nighttime gs,COS is at the lower end of values reported from other ecosystems: 1.6 mmol m-2 s-1 for a New England mixed forest , 5–30 mmol m-2 s-1 for a Scots pine forest , 11.5 mmol m-2 s-1 for a wheat field , and 13–20 and 22–66 mmol m-2 s-1 for pine and poplar trees, respectively, in an alpine temperate forest . The nighttime stomatal conductance shows a large variability among different species.

In land biosphere models, nighttime stomatal conductance is often a fixed value regardless of plant type and water status, e.g., gs,H2O=10 mmol m-2 s-1 in the Community Land Model v4.5 . The fixed-value parameterization may introduce biases to the nighttime COS fluxes and long-term COS budget in regional simulations, which may in turn propagate into the COS-based photosynthesis estimates. Constraining nighttime COS uptake requires an understanding of the variability of nighttime stomatal conductance among plant species and ecosystem types. Water and COS flux measurements need to be used in conjunction to derive robust estimates of nighttime stomatal conductance. We expect COS measurements to be particularly useful for stomatal conductance estimates in tropical rainforests and other environments that experience high humidity conditions, provided that the variability of the internal conductance of COS is well understood.

LRU is an important empirical parameter used to derive ecosystem photosynthesis (also known as gross primary productivity, GPP) from COS measurements on spatial scales ranging from the ecosystem to the continent . Choosing a representative LRU for COS-based GPP estimation is crucial and challenging.

In addition to its environmental controls, LRU also varies among plant species . For the T. latifolia, the asymptotic LRU value at high light (PAR > 600 µmol m-2 s-1) is around 1.0 (Fig. c). This value is much lower than the mean LRU of 1.61±0.26 from laboratory measurements across a range of species , which has been used as a representative LRU in ecosystem-scale e.g., and regional-scale GPP inversion studies e.g.,. However, the low asymptotic LRU of T. latifolia is not surprising according to the mechanistic LRU model in , which describes that LRU is positively related to the ratio of intercellular CO2 to the ambient CO2 (Ci/Ca). As T. latifolia often has a high photosynthetic capacity e.g.,, its Ci/Ca ratio may be lower than other species, thus contributing to the low LRU. Additionally, it has been noted that the aerenchyma of T. latifolia serves as a conduit to transport reduced gases from the rhizosphere to the atmosphere , which may act as a hidden COS source. Although the presence of this mechanism cannot be ruled out with our method, as it is an intrinsic process of the marsh plant and part of the plant–atmosphere COS exchange, and therefore the LRU measured here remains relevant for larger scale applications in this, and similar, ecosystems. Relatively low LRU values have also been reported from other ecosystems, for example, 1.3 in a wheat field and 1.2 in a mixed temperate forest at high PAR . This suggests that for the success of COS-based GPP estimation, LRU needs to be locally constrained on the dominant species in an ecosystem, rather than assumed to be a constant.

For regional scale applications, the time-integrated LRU can be more relevant than the instantaneous LRU. Large scale patterns of COS and CO2 drawdown imprinted in an air parcel are spatiotemporally integrated features, because the transport of surface uptake signals to the planetary boundary layer takes time and may be affected by the entrainment with other parcels along the way. Our results of time-integrated LRU show that although daytime mean LRU and PAR are correlated, nighttime leaf respiration and COS uptake create large variability in the all-day mean LRU, which decouples it from PAR (Fig. b). This suggests that a bottom-up scaling is unlikely to offer reliable daily LRU values for regional scale applications. Instead, LRU that is diagnostically calculated from biosphere models such as the Simple Biosphere model would be more appropriate for COS–GPP inversion studies, provided that model parameterizations are validated against observations.