Soil fluxes of COS were determined using a dynamic, flow-through chamber approach. A commercially available Aerodyne quantum cascade laser (QCL, Aerodyne Research, Inc., Billerica, MA, US) was used to quantify COS and CO2 mixing ratios in the effluent of a laboratory-based apparatus (Fig. 3). The precision of the instrument was 8 ppt COS at 1 Hz and 2 ppt COS for observations averaged over 50 s, with an absolute calibration accuracy of 5 % COS (Commane et al., 2013). Fluxes determined here required quantifying only relative mole ratios. Fluxes were calculated using an equation adapted from de Mello and Hines (1994): F=V(Cf-Ci)msoil-1. F is the COS or CO2 exchange rate in pmolgasmin-1gdrysoil-1. Ci is the mixing ratio of the compound entering the chamber, determined by analyzing the gas stream bypassing the chamber headspace. Cf is the mixing ratio of the compound exiting the 1 L PFA chamber headspace. V represents the sweep rate of the total air through the chamber, measured by the mass flow meter upstream of the QCL, typically 0.29 to 0.31 Lmin-1, and converted to pmolmin-1. The value msoil is the amount of dry soil enclosed inside the chamber in grams. The flow of the system was driven by a vacuum pump downstream of the QCL. The instrument also measured H2O and applied an adjustment for dilution by water vapor, generally a less than 1 % correction. According to the NIST spectral database, strong water vapor lines do not overlap the COS and CO2 lines at 2052.256 and 2052.096 cm-1. A more detailed description of the instrument can be found in Commane et al. (2013), with a further exploration of the instrument limitations by Kooijmans et al. (2016). Ambient laboratory air was used as the sweep gas for the incubations performed here. Observation of a dry nitrogen stream between incubations was used to correct for instrument baseline drift. While ambient COS mixing ratios had small variation (510 ppt with 80 ppt standard deviation), some of the CO2 fluxes were uninterpretable because of variations in ambient CO2 mixing ratios, Ci. CO2 fluxes that could not be distinguished from 0 are graphically presented at 0.

Each F quantification is generated from 80 min of 1 Hz air analysis: 10 min of nitrogen gas, 10 min of bypass air, and 40 min of air directed through the soil chamber, followed by repeating the analysis of nitrogen gas and bypass air for 10 min each. To check for baseline stability, the ambient air and nitrogen gas were both analyzed for 10 min before and after each chamber measurement. Air flow was directed through the chamber and the effluent analyzed for 40 min to promote soil equilibration within a dynamic headspace. COS mole ratios were generated by monitoring the appropriate absorption lines. The average COS reported over the last several minutes of chamber flow-through and bypass were corrected for zero drift using the nitrogen (COS-free) signal, then used as Cf and Ci, respectively, in Eq. (2). COS fluxes are reported in pmol COS per gram of dry weight soil per minute (pmol COS g-1min-1); negative values indicate uptake of COS, when Cf<Ci.

The temperature of the chamber was manipulated from 10 to 40 ∘C with a constant-temperature water bath. For higher-temperature observations of soil fluxes from the soy field soil, the incubation chamber was placed in a container of water on a hotplate. The actual soil temperature was recorded by a small, self-contained temperature data logger with a stainless steel outer casing (iButtons, Maxim Integrated, San Jose, CA, US). In order to prevent the soil from drying out during the analysis, a length of Nafion tubing was placed upstream of the chamber inside a container of distilled water in the same water bath. Even with this precaution, soil samples still dried slightly during the experiment. Samples were weighed daily, and soil moisture content was altered or maintained by adding distilled water. When water content was changed, soil samples were held at 20 ∘C and COS flux observations continued for at least 12 h.

To explore the sensitivity of COS uptake to chamber COS mole fractions, we performed a series of incubations with a freshly collected soil sample from near the original soy field site. The soil sample was air-dried to approximately 2 % VWC then incubated with ambient sweep air and COS-free zero air, which contained 300 ppm CO2 and no detectable COS. The difference between the two treatments helped characterize the effect of COS concentration on observed COS fluxes.

Performing soil incubation experiments allowed for precise manipulation of environmental variables to reveal underlying patterns in soil COS exchange. Soil in situ has an important dimension not represented by these laboratory experiments: depth (Ogee et al., 2015; Sun et al., 2015). Data from this study could represent COS exchange from only the top layer of soil. Nonetheless, it would be enlightening to compare controlled experiments to data collected in the field.

A further experiment was performed to estimate the relationship between laboratory, per-gram measurements and field, per-area measurements. Soy field soil was gradually added to a 20 ∘C incubation chamber, starting with 50 g and increasing to 300 g. While the total COS emissions increased with every soil addition, the flux per gram soil increased linearly between 50 and 100 g, then demonstrated saturation behavior with samples greater than 100 g. Thus, all fluxes were scaled up to 100 g and assumed to represent a soil footprint equal to the area of the incubation chamber base, 0.00779 m2. In short, fluxes were multiplied by a factor of (100 g) (0.00779 m-2) (60 s min-1)-1 or 214 gminm-2s-1.

The total net COS flux observed from the soils is thought to be the combination of abiotic and biotic fluxes. FCOS,soil=FCOS,biotic+FCOS,abiotic FCOS,soil is the net flux of COS, whereas FCOS,biotic and FCOS,abiotic represent the contribution of biotic and abiotic processes, respectively. The flux units used here were transformed as described in Sect. 2.2 from pmolCOSmin-1gdrysoil-1 to pmolCOSm-2s-1. Two models were fitted to soy field COS soil flux observations to explain FCOS,biotic and FCOS,abiotic separately. First, air-dried soil COS measurements were described using an exponential equation, as in Maseyk et al. (2014). FCOS,abiotic=αexp⁡(βTsoil), where Tsoil was the temperature of the soil in ∘C, and α and β were parameters determined using the least-squares fitting approach. These driest measurements were assumed to represent the observable fluxes with the least influence from microbial uptake of COS while keeping the soil in tact. This calculation was used to examine potential effects of soil COS production during changes in soil moisture. To disentangle biotic fluxes, the soy field FCOS,abiotic was then subtracted from FCOS,soil observations over a range of temperatures to yield FCOS,biotic, as in Eq. (3).

To explain FCOS,biotic, we used a model that was originally developed for soil NO production in Behrendt et al. (2014). Previous work (Van Diest and Kesselmeier, 2008) had used a similar NO soil flux model. The overall form of the equation is the product of a power function and an exponential function, Eqs. (5) and (6). a=ln⁡FoptFθgln⁡θoptθg+θgθopt-1-1FCOS,biotic=Foptθiθoptaexp⁡-aθiθopt-1 Here a is the curve shape constant, Fopt and Fθg are the COS fluxes (pmol COS m-2s-1) at soil moistures θopt and θg (percent volumetric water content, % VWC), Fopt is the maximum biotic COS uptake, and θopt>θg. FCOS,biotic is the COS uptake for a given soil moisture θi after subtracting FCOS,abiotic within the specified temperature range. The two models for FCOS,biotic and FCOS,abiotic could then be used to predict soil COS fluxes for a given temperature and soil moisture condition.

Ecosystem COS flux, FCOS,ecosystem, is the sum of leaf COS uptake, FCOS,leaf and soil COS exchange FCOS,soil, including litter. Two approaches were used to explore the error introduced by calculating GPP from ecosystem COS exchange without correcting for FCOS,soil.

The first method sought to calculate temporal variability in the relative importance of FCOS,soil. We used GPP estimates for the soy field FLUXNET site (US-Bo1) based on half-hourly CO2 eddy flux covariance measurements and a respiration model (Reichstein et al., 2005), restricted to values greater than 25 µmolCO2m-2s-1 to include only midday fluxes when photosynthesis was high. FCOS,leaf was anticipated from these reported GPP values, using Eq. (1) with relative uptake of 1.8 (Stimler et al., 2011) and an ambient mixing ratio of CO2 at 380 ppm and of COS at 500 ppt. The model described in Sect. 2.3 was used to generate FCOS,soil estimates from field soil moisture and temperature data collected at the site. Estimates of FCOS,leaf and FCOS,soil were then added together and used to calculate new GPP estimates with Eq. (1). The difference between the reported GPP estimates and estimates using FCOS,ecosystem instead of FCOS,leaf in Eq. (1) was then evaluated.

Secondly, we examined the spatial importance of reported FCOS,soil from the few values reported in the literature, relying on a similar concept as the global calculation above. Using the biome GPP estimates from Beer et al. (2010), we back calculated anticipated estimates of FCOS,leaf using Eq. (1). For this purposefully simple calculation, we assume a 100-day growing season with 12 h of light per day to convert between annual estimates of GPP and field measurements calculated in s-1 units, though this obviously does not represent the diversity of biome carbon assimilation patterns. For each biome where data existed, a range of FCOS,ecosystem was calculated as the estimated FCOS,leaf added to the range of reported FCOS,soil from previous studies. A GPP estimate was then made using Eq. (1) with FCOS,ecosystem in place of FCOS,leaf. The percentage difference between the GPP estimate in Beer et al. (2010) and this new GPP estimate was then evaluated.

Altering the mixing ratio of COS in the chamber sweep air had a strong effect on COS exchange with a soy field soil sample. The slopes of the linear regression lines in Fig. 4b represent the change in COS flux of the soil sample divided by the change in ambient COS. The slopes are all negative and become strictly steeper as temperature increases. Under ambient and zero air treatments, the soil sample showed exponentially higher net COS emissions with temperature. Apparent uptake increased with more available COS in the headspace. The linear regression intercepts in Fig. 4b and graphed separately in Fig. 5b represent the theoretical flux we would expect if there were no COS in the chamber at all. This soil sample exhibited net emissions of COS at all temperatures; therefore, the headspace always contained some small amount.

The COS mixing ratios observed in laboratory air during the entire course of experiments was 510±80 ppt. The mixing ratio observed at the outlet was 470±95 ppt COS. To calculate the maximum anticipated effect of this range, we used the maximum slope observed for the linear relationship described in Kesselmeier et al. (1999) for soils at 17 ∘C at a specific volumetric water content: FCOS,uptake = 0.006 × [COS] - 0.32, where soil COS uptake, FCOS,uptake, is reported in pmolCOSg-1h-1 and [COS] is the mixing ratio of COS in ppt. The variability of COS mixing ratios in the soil chamber calculated by this method would cause a variability of ±0.019 pmol gram dry soil-1 min-1. By our simplified scaling presented in Sect. 2.2, this translates to 4.1 pmolCOSm-2s-1.

Overall, desert and rainforest samples had the smallest magnitude net COS exchange rates. The temperate forest samples showed the largest net uptake during the first trials, when the soil sample was at field soil moisture, 41 % VWC. Of the small fluxes presented in Fig. 6, temperate forest soils also had the largest net production when the soil sample was in its hottest and driest state (Fig. 6b, 38 ∘C and 5 % VWC). Samples from the oak savannah displayed variable fluxes (Fig. 6c). Observations with the soy field soil generated mostly net production of COS, often 10 times greater than fluxes from other soil samples (Fig. 7).

COS fluxes tended towards more positive fluxes with hotter temperatures (Figs. 6 and 7). Soils incubated at 40 ∘C exhibited net COS production while incubations at 10 ∘C yielded net COS consumption in a majority of cases. Except for the desert site, the areas where these soils were collected rarely experienced such high maximum soil temperatures, if at all (Fig. 1).

The temperate forest showed the highest CO2 fluxes, with increasing fluxes for increasing temperatures and soil moisture (Fig. 6e), contrasted by the small fluxes from the rainforest and desert soils (Fig. 6d). The savannah soils exhibited an optimum temperature for CO2 fluxes near approximately 30 ∘C (Fig. 6f).

The soybean agricultural soil incubations yielded net COS emissions for the majority of trials, with a larger range than the other soils investigated: -0.04 to 0.09 pmolCOSg-1min-1 when incubated between 10 and 40 ∘C. When samples of the agricultural soil were heated further, COS net production persisted. To determine the contribution of soil organic matter in the sand-sized fraction (SSF), coarse litter >53 µm was removed from one subsample and incubated as before. COS net emissions were higher compared to non-sieved samples at similar temperature and water content (Fig. 7).

Soil COS fluxes had a more complicated relationship with soil moisture. When soil samples were waterlogged, net COS exchange shifted towards zero compared to drier trials. For the most part, drier soils have net emissions of COS, except in the case of the varied fluxes from the oak savannah soil (Figs. 6 and 7). In oak savannah soil, increases in soil moisture led to increases in COS uptake. When soil moisture was increased further to near 40 % VWC, COS exchange returned to near zero. The savannah site was expected to experience this range of soil moisture (Fig. 1). In contrast, where dry rainforest soil experienced an increase in net COS production, rainforest soil rarely experiences near-zero soil moisture (Fig. 1). Increasing water content to field levels, the rainforest soil COS exchange returned to near zero. This does not take into account the fluctuations in soil moisture and redox potential experienced in a rainforest in situ. Temperate forest soils appear to experience net COS uptake except under very dry or unusually hot conditions (Fig. 6b).

To observe changes in COS fluxes during changes in soil moisture (i.e., as would happen in situ via precipitation), COS exchange was recorded for at least 12 h after soil moisture was changed during the course of the experiment (Fig. 8). The rainforest and savannah fluxes showed no discernible pattern in fluxes after water additions. For one series of observations with rainforest soil, the Nafion tubing was removed and the soil dried slowly over time, continuing to show little variability. In contrast, the temperate forest and soy field soils (Fig. 8a) responded with a large variability in COS fluxes after soil moisture manipulation, taking several hours to reach a consistent flux value. There was an overall negative relationship between soil moisture and net COS production for the soy field soil samples, but the link between soil moisture and COS fluxes for soils collected at other sites is not as clear.

To test whether the COS mixing ratio controls the variation seen in Fig. 8, we examined the mixing ratio of COS exiting the chamber versus COS exchange (Fig. 9), and we find that no clear relationship emerges. Additionally, high COS production does not appear to obscure the relationship between COS ambient mixing ratios and COS uptake. As a thought exercise to demonstrate this, we estimated the COS production component of soil fluxes and subtracted it from the net fluxes depicted in Fig. 9, shown in Fig. 10. When soils were air-dried then incubated, a net COS emission was observed with a positive relationship to temperature ranging from 10 to 40 ∘C for all the samples except the desert soil. Using Eq. (4) and least-squares regression, a curve was generated for all soil types investigated other than the desert soils, shown in Fig. 11 and Table 2. We did not generate enough data to characterize the relationship in savannah soils. Correcting for COS production in this way does not change the overall relationship between incubation COS mole fraction and observed COS fluxes. The production of COS is assumed to be insensible to the concentration of COS the soil experiences, depending here only on temperature. From examination of Fig. 11, it can be seen that the correction for abiotic production at 20 ∘C is a small portion of the overall magnitude of the fluxes. Using this purposefully simple model (Eq. 3) to subtract out the effects of COS production vertically shifts the data and does not change the pattern of the relationship, as shown in Fig. 10.

COS fluxes change over time after a change in soil water content was not consistent for given changes in soil moisture. However, when water was added to dry soil (<10 % VWC), many soil subsamples exhibited the pattern in Fig. 12b: CO2 fluxes remained consistent, while COS fluxes increased immediately after water addition, then slowly decreased over many hours. This is contrasted by Fig. 12a, where both COS and CO2 fluxes demonstrate some variability after changes in water content.

Net COS fluxes were a balance of abiotic and biotic processes. If we assume that incubations of air-dried agricultural soils were representative of an abiotic COS production or desorption (less some physical limitations), we can calculate the relationship between abiotic COS production and temperature for agricultural soil (plotted in Fig. 13a). We fitted Eq. (4) to the data scaled as described in Sect. 2.2 and using a least-squares approach, much like in Maseyk et al. (2014). The resulting Eq. (7) had an r2 value of 0.9. FCOS,abiotic=0.437exp⁡(0.0984Tsoil) There were more cold (<15 ∘C) incubations performed than hot (>35 ∘C) incubations, and some of the coldest incubations were excluded from the fit to give appropriate weight to the hottest incubations.

Subtracting the dry soy field soil signal component from all other COS incubation results, we found the biotic and physically limited flux component (Fig. 13b). The COS incubation observations had been converted to pmolm-2s-1 units, binned by incubation temperatures as <20, 20–30, and >30 ∘C, fitted to Eq. (6) and plotted in Fig. 13b. The resulting parameters are shown in Table 3. For the purposes of generalizing the equation to any temperature and moisture content pairing, θg was held constant at 35 % VWC; thereafter the data were binned by different temperature increments to discern how Fopt, Fθg, and θopt in Eqs. (5) and (6) change with temperature. More data need to be collected to create a robust model; however, we think this is a worthwhile attempt at capturing variability. Fopt=-0.00986Tsoil2+0.197Tsoil+-9.32θopt=0.287Tsoil+14.5Fθg=-0.0119Tsoil2+0.110Tsoil+-1.18 The total flux FCOS,soil can be calculated as the sum of fluxes generated by biotic and abiotic processes.

Using this framework of equations, we estimate the influence of large soil COS fluxes on GPP estimates. We used data reported for the Bondville FLUXNET site, US-Bo1, the soy field site in this study. The model shown in Fig. 13 and described in Eqs. (3)–(10) was based on flux observations from soil collected at this site. There are well-known uncertainties associated with reported GPP from flux towers (Desai et al., 2008). However, since we have no in situ measurements of COS from the site, these data are used as a starting point for calculating theoretical error potentials.

Two GPP estimates are presented in Fig. 14a: the first represents GPP estimates with COS leaf uptake fluxes alone, the second was based on theoretical net COS fluxes, including both leaf and soil COS exchange calculated with Eq. (3). The difference between the 1-day moving averages (Fig. 14b) signifies how GPP could have been over- or underestimated if net ecosystem COS fluxes were used as leaf uptake fluxes, ranging from -5 to +25 %.

To explore the possible spatial variation in soil COS exchange influence on the GPP proxy, we perform a similar calculation (described in Sect. 2.4) using in situ soil fluxes from previous studies (Table 4). The potential error in GPP estimates based on these sparse measurements ranges from -220 to +119 %. More observations and modeling of soil COS exchange for different ecosystems could ameliorate this large error.

Previous studies have shown that the interaction between net fluxes and ambient concentration of COS is linear (e.g., Conrad, 1994; Kesselmeier et al., 1999). COS soil fluxes have a demonstrated “compensation point”, the atmospheric concentration of COS where the net flux of a specific system is 0. At concentrations below the compensation point, net emission to the atmosphere is observed; net consumption is observed when ambient concentrations are higher than the compensation point. We believe that the variability in fluxes due to changes in soil moisture in Fig. 8 masks the effect of changes in COS chamber mixing ratios. In Kesselmeier et al. (1999), the authors used the mole fraction of COS exiting the incubation chamber as a measure of the well-mixed ambient environment actually experienced by the soil. The relationship between the observed COS fluxes after soil moisture change and the COS mixing ratio exiting the chamber is depicted in Fig. 9; all incubations depicted took place at 20 ∘C. If the controlling variable of the net fluxes in Fig. 9 was ambient COS, one would expect a strong inverse linear relationship where higher concentrations of COS result in higher uptake of COS at a particular soil moisture state. Instead, we see a positive relationship between COS mole fraction and COS flux. This is not surprising, because higher soil COS production leads to more COS leaving the chamber. Perhaps there is a dampening effect on COS fluxes, where net COS production by soils increases soil COS consumption, but the overall effect is overwhelmed by high COS production. In other words, the net production reported here may be in reality higher at the lower ambient COS mixing ratios that would be encountered by unenclosed soils in the field. However, when COS production was estimated and taken into account (Fig. 10), the overall pattern was unchanged.

The soil samples in this study were incubated under flowing air. The soil and headspace air were assumed to be in equilibrium after 30 min. If that were true, adsorption and desorption should no longer contribute to the soil flux: equal amounts of COS should adsorb and desorb. The uptake difference between the zero air and ambient air treatments in Fig. 4 indicates that some uptake process was affecting net soil fluxes, even in a very dry soil.

Multiple mechanisms determined the net COS exchange from soil, which were affected by soil water content and temperature. There are three proposed abiotic processes: COS production from abiotic degradation of soil organic matter (Whelan and Rhew, 2015), the physical limitations of water restricting air exchange between soil pore spaces and the chamber headspace (Van Diest and Kesselmeier, 2008), and adsorption/desorption of COS onto soil grains. The biotic uptake of COS by soils is theorized to be via enzymes present in the microbial community that are similarly responsible for COS uptake in plants (Kesselmeier et al., 1999; Protoschill-Krebs et al., 1996). There is no known biotic COS production mechanism in soils.

Taking these routes of COS exchange into account, we can explain qualitatively the fluxes observed here. For example, hot, dry soil appeared to produce the highest net COS emissions. Dry soil has a smaller active microbial community (Manzoni et al., 2011), and biotic uptake would be small. Higher temperatures should yield more thermal degradation of organic matter, resulting in higher COS production. In this study, when soy field soils were heated from 40 to 68 ∘C, COS net emissions continued, suggesting that the trace gas production here had no optimum temperature and was most likely abiotic (Conrad, 1996). Simultaneously, COS within the soil would exchange with the chamber air without the added tortuosity of water-filled pore space. The overall result is more COS produced abiotically, less COS consumed biotically, and the resulting COS excess diffusing quickly out of the soil. After wet-up, the temperature response curve shifts towards a COS sink, though it often retains a similar shape. When soil moisture is increased further, soil pore spaces are effectively cut off from the chamber headspace. When waterlogged, the soil exhibits COS fluxes nearer to 0 regardless of temperature. This reasoning evidently holds across the temperate forest, savannah, and agricultural soil investigated here.

The desert soil samples, however, demonstrated near-zero COS exchange at field moisture and COS uptake when wetted. Since these soils are frequently hot and dry, it could be that there is not sufficient remaining organic material to abiotically degrade into COS, or there are not enough clay or silt surfaces for COS to adsorb/desorb. The behavior of the desert soil resembles the soil COS exchange observed in Van Diest and Kesselmeier (2008) and Kesselmeier et al. (1999), which both investigated exclusively sandy soils.

For the agricultural soils studied here, it appears that some soil interaction produced much more COS than other soils investigated. Large COS emissions were also observed from a wheat field soil in China (Liu et al., 2010) and the previously mentioned wheat field in Oklahoma (Billesbach et al., 2014; Maseyk et al., 2014; Whelan and Rhew, 2015), but not from the sandy arable soil in Germany, Finland, and China (Van Diest and Kesselmeier, 2008), where only net COS uptake was observed. While Melillo and Steudler (1989) found increases in forest soil COS production coincident with nitrogen fertilizer application, the composition of fertilizer used at the sites discussed above is unknown to us. It is unclear what is particular about the agricultural soils in the study by Van Diest and Kesselmeier (2008) that should result in only soil COS net consumption.

Two hypotheses emerge from the theoretical framework detailed above. The first is that all soils experience large COS production from thermal degradation of soil organic matter or desorption from soil surfaces, but most or all COS generated is usually consumed by in situ microbial communities. The agricultural soils collected in Oklahoma and Illinois undergo pesticide/herbicide applications and irrigation during the course of their management that may limit the diversity and size of the microbial community (Griffiths and Philippot, 2013) and the magnitude of the microbial COS sink. This idea is partially supported by Whelan and Rhew (2015), where autoclaved agricultural soils only experienced net COS production, though autoclaved soils are known to emit COS (Kato et al., 2008).

The second hypothesis suggests that the accessibility of the agricultural soil organic matter allowed more abiotic COS production than in forest or savannah soils. This could also be due to agricultural land management practices, which tend to break down soil aggregates and destabilize soil organic matter (Sollins et al., 1996). Accessibility, rather than litter quality, could explain why we see a similar COS production from agricultural fields with different crop cover, i.e., wheat (Billesbach et al., 2014; Liu et al., 2010) and soy/corn (this study). However, this still does not explain the biotically driven net COS uptake patterns found in arable soils by Van Diest and Kesselmeier (2008) and Kesselmeier et al. (1999), who report COS fluxes that resemble more the desert soil fluxes investigated here.

These two hypothesis may both influence COS exchange simultaneously. When the course litter and sand (>53 µm) fraction was removed from a soy field soil sample, COS production increased per gram of incubated sample (Fig. 7). This implies that the origin of the COS emissions resides in the silt and clay-associated fraction of organic matter, which has been shown to consist of plant matter that has undergone some microbial processing (Six et al., 2001, 2002). The combination of microbial activities and increased accessibility of organic matter to degradation may lead to large COS emissions from soils. While these mechanisms may explain differences between managed and non-managed soil COS exchange, we still lack a hypothesis for the difference between the small sinks in European arable soils and the temperature-driven sources in US and Chinese arable soils.

The drawdown of COS over North America has been observed from aircraft vertical profiles, appearing to scale with GPP-based uptake of COS by plants (Campbell et al., 2008). Data presented here indicate soil COS emission was maximum during high-temperature incubations, coincident with some surface temperatures observed during the North American growing season. We generated a model in Sects. 2.3 and 3.2 to calculate COS fluxes for US agricultural soils, taking these large emissions into account. Relating laboratory measurements to in situ observations has inherent problems, so we present this as a theoretical exercise investigating the possible magnitudes of soil COS exchange on broader scales.

We plotted our equation with one developed by Maseyk et al. (2014) from fluxes (Fig. 15a) and environmental parameters (Fig. 15b) recorded in situ at a wheat field in Oklahoma over the course of that study in 2012. The COS flux model developed by Kesselmeier et al. (1999) is displayed using the same input variables, assuming a constant ambient COS mixing ratios of 500 ppt and a standard flux of 75.3 pmolm-2s-1 (Fig. 15b). This last equation can only predict COS soil uptake and has been used to model soil COS exchange globally (Kettle et al., 2002).

Key patterns emerged from examining differences between the observations and predictions over the course of the campaign in Maseyk et al. (2014) (Fig. 15), noting first that the model presented by Kesselmeier et al. (1999) and the model presented here were not parameterized using soil from this site. The fact that there are any similarities at all between the model outputs and observations is encouraging for future modeling efforts. None of the three models captured the large emissions observed before day of year (DOY) 130 when wheat was present in the field and higher soil moisture occurred. None of the models captured the large swings from COS source to sink found during large temperature fluctuations between DOY 110 and 115. After DOY 130, the wheat senesced and was harvested, resulting in hot and dry soils. The simple model from Maseyk et al. (2014) reproduced the COS soil flux variability better under these conditions. The Kesselmeier et al. (1999) model generated some variability, but could not predict any soil COS emissions. This study's model overlapped both the uptake model's variability during wheat senescence and the high emissions predicted by Maseyk et al. (2014) after wheat harvest.

There are several explanations for the discrepancies between models and flux observations. Both this study and the Kesselmeier et al. (1999) model were based on idealized laboratory conditions, not taking into account interactions with soil COS exchange at different depths. No doubt COS is produced or consumed in all layers of soil, not just at the surface, but soil incubations were purposefully designed to avoid these issues. Additionally, there is variability in both soil moisture and temperature even over the area of the soil plot: a heterogeneous soil may experience variations in these parameters on a small scale (Entin et al., 2000). Also, soil temperature was measured at 5 cm, generally cooler than the observed surface temperature for the site (Maseyk et al., 2014). While there was not enough variability in soil moisture and temperature to perform a similar treatment as shown in Fig. 13 for field observations, we believe the hybrid model presented here will lead to new investigations that close the gap between lab-based COS observations and COS exchange at larger scales.

The main motivation of this work was to make progress towards better estimates of GPP. The drawdown of COS over the continents appears to be associated with the uptake of carbon dioxide (Campbell et al., 2008). For some of the biomes explored here, like deserts, soil COS exchange under field conditions may actually be negligible compared to plant uptake. On the other hand, recent work has suggested that soil COS fluxes in agricultural areas might be large and need to be taken into account (Billesbach et al., 2014; Maseyk et al., 2014). The model presented in this study anticipates these agricultural soil COS fluxes using commonly measured variables. With such a correction, applying the COS-GPP tracer will be more feasible to constrain GPP estimates on regional scales.

Taking COS soil fluxes into account when estimating GPP can avoid over- and underestimations of carbon fluxes presented in Table 4 and Fig. 14. Observations are still scarce: despite a plea for data from desert soils by Kettle et al. (2002), we were not able to find such a study in the literature over 10 years later. Boreal forest soil COS exchange estimates are represented by a single study performed at a single site in Sweden over the course of 2 months in 1993 (Simmons, 1999). Modeling efforts suggest large COS fluxes in the tropics (Berry et al., 2013; Suntharalingam et al., 2008) and tropical forests and savannahs are associated with 60 % of global terrestrial GPP (Beer et al., 2010). However, there remains a dearth of observations in tropical latitudes.

This magnitude of avoidable error suggests that soil fluxes are not negligible; however, the uncertainty of GPP at regional to global scales is much larger. The error introduced by large soil emissions from cropland soils to COS-GPP estimates can be avoided by characterization and correction of COS fluxes. This study's approach deconvolves the production rates seen to dominate the net COS flux in Maseyk et al. (2014) and the small uptake rates observed in sandy soils by Van Diest and Kesselmeier (2008).