Nitrous oxide measurements were carried out using a newly developed semiautomatic chamber system (Fig. 1). It consisted of aluminum guiding racks (length 2121, width 936, height 3033 mm) with aluminum soil collars (length 750, width 750, height 160 mm; inserted 0.10 m into the soil), and opaque PVC chambers (color: white; interior dimensions: length 777, width 777, height 565 mm) (Ps-plastic, Eching, Germany). Subtracting inside items such as an axial fan, screws, supporting racks, and tubes, the chambers have a headspace volume of 0.33 m3 and cover a surface area of 0.56 m2. Depending on vegetation height, extension modules (interior dimensions: length 730, width 730, height 360 mm) can be connected to the chambers (total headspace volume is then 0.54 m3) if needed over taller vegetation, but they were not used in this study. EPDM gaskets (20 mm × 15 mm) were attached to the bottom of each chamber in an aluminum u-channel to ensure gas-tight closure when chambers were operating. Up to three chambers can be combined into one system (Fig. 1b) with a joint control unit and autosampler or analyzer. Two custom-made temperature probes (Pt100) were installed inside and outside of each chamber to measure ambient air temperatures. Chambers were ventilated during measurements using an axial fan, which was mounted to produce a horizontally oriented airflow alongside the chamber walls to minimize interference with the natural steady-state soil efflux but to maximize proper mixing of the chamber headspace as was described in Drösler (2005). The air was sampled from the top center of the lids. Chamber operation was controlled by a logic module (Millenium 3, Crouzet, Hilden, Germany). An autosampler consisting of a membrane pump (operated at 0.8 L min-1; NMP 830 KNDC, KNF Neuberger, Freiburg, Germany), an absorber to avoid water condensation within tubes (3.2 mm ID, 6.4 mm OD) (BEV-A-Line, ProLiquid GmbH, Überlingen, Germany), and valves, and an exchangeable rack for 162 headspace vials (20 mL, WICOM WIC 43200, Maienfeld, Germany) was connected to the chamber system. Chambers were lifted and moved down by a 24 V (DC) motor winch and were directed to the soil collar by the aluminum rack. After measurement events, the chamber was lifted to 1.18 m above ground and dragged backwards at a 45∘ angle to keep the soil and vegetation inside the soil collar under as natural conditions as possible (e.g., prevention of shading and undisturbed throughfall). To avoid pressure changes when setting the chamber on the collar, the chamber had a 1.5 m pressure compensation tube leading from the inside through the side wall of the chamber to the outside. Information about our chamber system including the construction plan is available to the scientific community and can be requested from the authors.

Two field campaigns were conducted in fall 2012 in Braunschweig, Germany, and in spring 2013, at Risø Campus, Technical University of Denmark, using both GC and QCL chamber setups (see Table 1 for additional information). The chamber architecture was identical during the two campaigns. Sites and time periods were selected with the aim to compare chamber system performance under high- and low-flux conditions. Due to low temperatures and the lack of fertilizer application, we expected a low-exchange regime during the Braunschweig campaign, whereas higher fluxes were expected at Risø (higher temperatures and a substantial amount of fertilizer applied).

During parallel operation of GC and QCL, chambers were closed for 60 min at both sites to measure the concentration increase. When only QCL measurements were conducted, i.e., at Risø at DOY < 105.5 and > 108.5, chambers were closed for only 10 min. For the GC setup, four air samples were taken at 0, 20, 40, and 60 min after chamber closure to calculate one flux rate. Air samples (20 mL) were pumped through the tubing system using a membrane pump (3.2 L min-1; NMP 830 KNDC, KNF Neuberger, Freiburg, Germany) and were injected into septum-capped vials. Two cannulas were automatically inserted through the septum, one cannula acting as sample air inlet until overpressure was established and the other cannula acting as outlet for cycling the air back to the chamber. Air samples were stored in the exchangeable rack of the autosampler unit and were analyzed in the GC lab of the Thünen Institute using a GC-2014 (Shimadzu, Duisburg, Germany; modified according to Loftfield et al., 1997) with an electron capture detector for N2O analysis. Performance of the GC system was checked weekly by conducting 10 consecutive measurements of a standard gas with ambient N2O concentration (320 ppb). Samples were only analyzed if the coefficient of variation of peak areas during this test was below 3 %.

Parallel to the autosampler setup for GC analysis, we operated our chamber system directly connected to a QCL (continuous-wave quantum cascade laser absorption spectrometer, model mini-QCLAS, Aerodyne Research Inc., Billerica, Massachusetts, USA; see Nelson et al. (2004) for principle of operation) in a thermo-controlled housing. Briefly, the laser is thermoelectrically cooled (Thermocube) to 25 ∘C, uses a 76 m path length, 0.5 L volume, and multiple-pass absorption cell for sampling, and operates at 40 Torr. It provides a measurement precision of 0.04 ppb (1σ) within an averaging interval of 1 s. Calibration is performed by continuously aligning the N2O absorption peak of the sampled air to the standard of the HITRAN database (Rothman et al., 2009). A dry vacuum scroll pump (BOC Edwards XDS10, Sussex, UK) maintained a steady flow rate of 1.0 L min-1. After passing the QCL cell, the sample air was cycled back to the respective chamber to avoid underpressure conditions and unintentional sucking of soil air into chambers. Data were stored on the QCL's internal hard drive at a frequency of 10 Hz.

The detection limit (LoD) of our QCL and GC setups could be estimated using our campaign data assuming stationary conditions during the low-flux campaign in Braunschweig. Taking the whole campaign into account, the calculated standard deviations were 2.5 and 7.5 µg m-2 h-1 for QCL and GC measurements, respectively. Thus, the resulting 2σ uncertainty range for QCL was 5.0 and for GC 15.0 µg m-2 h-1. If only the first quarter of the Braunschweig campaign data are taken, i.e., a period where environmental conditions were less variable than over the whole campaign, the calculated standard deviations were 1.3 and 6.5 µg m-2 h-1 for QCL and GC measurements, respectively. Thus, the resulting 2σ uncertainty range for QCL was 2.6 and for GC 13.0 µg m-2 h-1. These estimates can be regarded as an upper flux detection limit. A theoretical lower flux detection limit solely depends on the sensitivity of the analyzers. Precision of the QCL is 0.03 and 0.01 ppb when averaging over 1 and 60 s, respectively. Table 2 summarizes features of the chamber-analyzer system used in this study.

GC-based N2O fluxes using linear, robust linear (Huber, 1981), and modified Hutchinson–Mosier regression (HMR; cf. Pedersen et al., 2010) were calculated as described in Leiber-Sauheitl et al. (2014) after converting molar concentrations into mass concentrations using temperature but no pressure correction. Briefly, nonlinear flux estimation with the HMR method (R Core Team, 2012; HMR package version 0.3.1) was performed when four data points were available and all of the following criteria were met: (1) the HMR function could be fitted, (2) Akaike information criterion (AIC; Burnham and Anderson, 2004), which is a measure of (relative) model quality (i.e., gives fit quality penalized by the model's degrees of freedom) and can be used to compare the quality of different model fits to the same data set, was lower for HMR fit than for linear fit, (3) p value of flux calculated using HMR was lower than that from robust linear fit, and (4) the HMR flux was less than 4 times larger than the robust linear flux. Otherwise, robust linear regression or ordinary linear regression were used when four or three data points were available, respectively.

QCL-based fluxes were estimated using two different methods. We applied the nonlinear HMR model with a slightly modified parameterization (Eq. 1 this study; cf. Moffat, 2012) to the 60 min data set of a full chamber cycle (10 min cycle in Risø at DOY < 105.5 and > 108.5) and compared these fluxes with those resulting from an application of linear regression when only the first 3 min of data after chamber closure were used (cf. Sect. 3.2).

To investigate the frequently observed nonlinearity in chamber field data, we computed a quantitative parameter κ describing the curvature in N2O concentration increase (or decrease) over time (60 and 10 min QCL data only). Based on the assumption of exponential gas concentration changes in the chamber (cf. Nakano et al., 2004) using ct=cmax1-exp⁡-kcmaxt+c0, with c(t) being the N2O concentration in the chamber at a certain point in time, cmax the maximum possible concentration range, c0 the measured concentration at t=0, and k the initial flux F0 divided by the effective chamber height h, we estimated the N2O soil–atmosphere flux as the first derivative of Eq. (1) evaluated at t=0, i.e., c′(t)t=0=k, and the curvature parameter κ as the second derivative of Eq. (1) evaluated at t=0, i.e., c′′(t)t=0=-k2cmax=κ. Units for concentrations c(t), cmax, and c0 are grams per cubic meter, units for k are grams per cubic meter per second, and units for κ are grams per cubic meter per square second. Negative values of κ correspond to concave curvature indicating a plateauing, i.e., saturating concentration increase over time. Standard errors in this study were calculated as the parameter errors from the respective regression model with the algorithm being based on the Levenberg–Marquardt method (nlsLM function in R package “minpack.lm”, R Core Team, 2012). Standard errors are solely associated with the flux calculation method and not with any kind of observational errors or issues related to measurement performance such as changes in flow rate, temperature sensitivity of the QCL, pump performance, or changes in chamber volume due to rough soil surfaces or plants in the chamber.

Significantly different patterns in chamber N2O concentration changes during the Braunschweig and Risø campaigns were observed (Fig. 2). While increases in the order of 10 to 20 ppb per hour (one chamber cycle) were found for the grassland site in Braunschweig, steep concentration increases measured on the harvested willow field at Risø were almost exclusively higher than 100 ppb per hour and reached maximum rates of over 650 ppb per hour in the period from DOY 105.5 to 108.5. For the low-exchange regime in Braunschweig, GC-based data points were highly scattered and rarely showed a clear increasing (or decreasing) tendency making flux calculations difficult. For the high-exchange regime at Risø, GC-based concentration data mostly showed well-defined increases and were similar to those obtained by the QCL system (cf. Sect. 3.3). The latter showed a precise and robust performance with clear base line levels and obvious chamber cycles during both campaigns. None of the QCL-based measurements revealed concentration decreases, i.e., negative fluxes (N2O uptake), while chambers were closed.

Results of the investigation on quantifying the curvature in c(t), expressed as κ, are given in Fig. 3. Extremely low absolute κ values between -10-4 and -100 – indicating quasi linearity in ∂c/∂t – were almost exclusively found under low-flux conditions, whereas fluxes > 100 µg N m-2 h-1 were only observed when κ was < -101 (Fig. 3a).

With the QCL's high time resolution – in this study operated at the analyzer's maximum frequency of 10 Hz – we compared N2O flux estimates based on 60 min (DOY 105.5 to 108.5) and 10 min (DOY < 105.5 and > 108.5) closure periods calculated by the modified HMR approach with those flux estimates that are based on the first 3 min of concentration data only and were calculated by linear regression. The Risø data set was used for this comparison because both high and low fluxes were observed. Flux estimates of the two approaches matched reasonably well; significant differences were only observed at very high rates (Fig. 4a, b). In total, 85 % of the variance in N2O fluxes from 3 min closure could be explained by fluxes from 60 and 10 min closure (Fig. 4b). The relatively high slope of 1.80 was mainly caused by three exceptionally high fluxes where the 60 min method considerably overestimated values of the 3 min method. Standard errors of N2O fluxes from both 3 and 60 min closure were extremely low, i.e., in the order of 0.2 % of the fluxes (Fig. 4c) with median values of 0.17 and 0.06 µg N m-2 h-1 and arithmetic means of 0.21 and 0.20 µg N m-2 h-1 for the 3 and 60 min closure flux estimates, respectively.

For better comparison with other studies, we also compared HMR-based fluxes with robust linearly calculated fluxes from our GC measurements when the full 60 min cycle was taken into account. A linear regression analysis (data not shown) resulted in a slope of 0.97 and an R2 value of 0.86 under the high-flux regime in Risø with the data set of robust linearly calculated fluxes being the independent variable. The mean campaign flux value from HMR-based calculations was 22 % higher than the average campaign value of the robust linear method. The difference between the two methods was even higher under the low-flux regime in Braunschweig. Slope and R2 value of a linear regression analysis were 1.82 and 0.42, respectively. Despite the high slope value, the mean campaign value of the robust linear method only reached 51 % of the value obtained from the HMR method.

A further intriguing analysis shows that standard errors were found to be invariant on QCL sampling frequency (Fig. 5). We simulated different sampling times ranging from 1/10 of a second to 25.6 s, which corresponds to a frequency of 0.0390625 Hz, by excluding the respective intervals from the original 10 Hz data set. Results show that the median of the standard error of the fluxes remains stable over a wide range of measurement frequencies. At a frequency class of 0.15 and lower (three boxes on the right-hand side of Fig. 5), which corresponds to a sampling time of ∼ 5 s and higher, lower and upper quartile values begin to deviate and the median changes slightly.

Time series of N2O fluxes and their associated standard errors using both the GC and the 3 min QCL linear regression method during the Braunschweig and Risø campaigns are given in Fig. 6. QCL fluxes in Braunschweig were at a constantly low level ranging between 2 and 16 µg N m-2 h-1, whereas GC-based fluxes at the same site were scattered between -13 and 39 µg N m-2 h-1. A linear regression revealed no significant relationship between GC and QCL fluxes with a very low coefficient of determination of 0.036 (Fig. 7a). While standard errors of the QCL method were always below 0.6 µg N m-2 h-1, values of the GC method were distributed between 0.5 and 22.0 µg N m-2 h-1. Although higher variability and higher standard errors in GC-based fluxes were evident, mean N2O flux rates of the entire observation period were almost identical when comparing the two analyzer types. In total, 6.42 ± 5.98 and 7.77 ± 0.13 µg N m-2 h-1 were found for the GC and the QCL method, respectively.

Under the high-exchange regime at Risø, N2O fluxes of the two analyzer types matched considerably better (Fig. 6d). Although the willow field was already fertilized on DOY 99, N2O fluxes did not start to increase until DOY 105 when a sharp rise in air temperature was observed. GC-based fluxes were lower than QCL-based fluxes (slope = 0.50) as in most cases a nonlinear model could not be fitted with only four data points. A linear regression between GC and QCL fluxes revealed a coefficient of determination of 0.48 (Fig. 7b). Standard errors of the QCL method were again extremely low, i.e., < 1 % of the flux value, and were always below 1.0 µg N m-2 h-1, while those from the GC method were on average in the range of 5 to 10 % of the flux value. Parallel operation of both methods was conducted from DOY 105 to 108. During this period, the campaign means were 117.8 ± 0.2 and 77.4 ± 8.2 µg N m-2 h-1 for the QCL and GC method, respectively.

As standard errors of QCL-based N2O fluxes were on a constantly low level, no dependency on flux value was observed in any of the campaigns (Fig. 7). The same was evident for GC-based fluxes in Braunschweig. At Risø, however, a slight but nonsignificant tendency of higher standard errors at higher flux rates was found. Only 8 % of GC data from Braunschweig met the criteria for flux calculation using the HMR model. At Risø, 38 % of GC data allowed for HMR flux calculation indicating that higher-exchange regimes favor the usage of an exponential model when using the GC method.

Neither at Risø nor during the Braunschweig campaign was soil N2O uptake observed when using QCL-based measurements. Only very few cases (n=5) c′(t) were initially found to be negative; however, these data, which exhibited abnormally high standard errors, were discarded due to mechanical malfunctioning of the chamber system as a result of non-closure caused by distorted guiding racks through very high wind speeds at Risø (cf. Sect. 3.3).

Regarding GC-based data in our study, 2 out of 37 fluxes in Risø were negative. Note that GC-based fluxes in Risø were only determined between DOY 105.5 and 108.5 when fluxes were elevated due to fertilizer application. In Braunschweig, however, nearly 25 %, i.e., 50 out of 201 flux rates, from the GC setup showed N2O uptake with only 3 of the 50 negative flux rates being significant (p < 0.05; p values not corrected for multiple testing).

An investigation of the diurnal variability of N2O fluxes showed that during the Braunschweig campaign – although only small differences were observed – the highest fluxes were found during midday and early afternoon (∼ 8.7 µg N m-2 h-1), while the lowest N2O efflux was measured shortly before midnight and before sunrise (∼ 7.2 and 7.3 µg N m-2 h-1, respectively; Fig. 8), thereby following a commonly observed temperature-driven pattern (cf. Sect. 4.4). In Risø, however, we found the lowest fluxes of ∼ 18.2 µg N m-2 h-1 at midday and the highest fluxes when it was dark, peaking before midnight at ∼ 32.0 µg N m-2 h-1 (only data of DOY < 105.5 and > 108.5 were taken to exclude fertilizer effects). Error bars in Fig. 8 indicate the standard error of the mean from all flux values in each bin. Each bin contains fluxes from 3 h periods, i.e., from 00:00 to 03:00, 03:00 to 06:00, 06:00 to 09:00 CET, etc. The mean values in Fig. 8 are plotted in the center of each bin. Fluxes were binned due to irregular starting times of new chamber cycles. In general, a new chamber cycle could be started each full hour, but to get a more robust diurnal pattern, we decided to bin data in the abovementioned 3 h containers. While the diurnal variation of N2O fluxes from the Risø campaign is significant (p = 0.0059), the diurnal variation found during the Braunschweig campaign is not, as the difference between mean minimum and maximum values is lower than the upper flux detection limit of ∼ 2.6 µg N m-2 s-1.

The high time resolution of QCL data allowed for a closer look at the shape of the concentration increase. The general form of the curve is determined by the rate of transport of a diffusing trace gas into the chamber headspace, which declines throughout deployment because any increase in the headspace concentration results in a corresponding decline in the vertical concentration gradient driving that transport (Rolston, 1986; Hutchinson et al., 2000; Livingston et al., 2006). The change in the rate of transport is the initial curvature kappa, i.e., the second derivative of the concentration change at t=0.

The fact that extremely low negative κ values between -10-4 and -100 – indicating quasi linearity in ∂c/∂t – were almost exclusively found under low-flux conditions, whereas fluxes > 100 µg N m-2 h-1 were only observed when κ was < -101 (Fig. 3a), means that at higher fluxes the curvature in c(t) is concave, suggesting concentrations that tend to plateau over time with the saturation effect becoming larger at higher flux rates. Near-zero fluxes, however, corresponding to κ values around zero, indicate no considerable changes in N2O concentrations and thus hardly any alteration of the vertical concentration gradient over time. Furthermore, closure time was found to have an impact on the magnitude of κ (Fig. 3b). Longer chamber deployment led to higher curvature in c(t), which was expected as concentration gradients decline over time when a considerable flux is measured (cf. Hutchinson and Mosier, 1981; Livingston et al., 2006; Pedersen et al., 2010).

Our results imply that at low to moderately high flux rates < 200 µg N m-2 h-1 (cf. Fig. 4d) and/or short chamber closure, the slight nonlinearity in concentration change when calculating fluxes is of minor importance and the application of linear models is acceptable, particularly with regard to other commonly observed errors such as those originating from soil disturbance, chamber placement (Christiansen et al., 2011), temperature, and pressure and humidity perturbations (Parkin and Venterea, 2010). At higher fluxes, however, significant curvature in c(t) expressed by large negative κ values will most likely lead to a substantial underestimation of fluxes when using linear regression instead of applying an exponential model for flux calculation (cf. Matthias et al., 1978; Jury et al., 1982; Anthony et al., 1995; Kroon et al., 2008; Sect. 3.2). In principle, several other reasons making flux determination with linear or exponential models problematic may technically be found. These are exponentially increasing N2O concentrations after chamber closure due to possible dispersion effects leading to biased analyzer readings when the elevated gas concentration is initially not uniformly mixed with the air inside the tubing, placement of the sample tube inlet at the top of the chamber lid leading to an establishment of a temporary concentration gradient in a weakly mixed chamber atmosphere, or an insufficient dimension of the pressure compensation tube leading to a pushback of air into the uppermost soil layer at the moment when the chamber is set onto the lid. However, none of these were observed during our campaigns, thereby indicating a robust setup and chamber design for reliable N2O flux calculations.

We also investigated the possible effect of ambient wind speed and direction on concentration buildup characteristics (Fig. 3c and d, respectively) as differences between turbulence conditions outside the chamber may possibly vary from those conditions inside the chamber under changing wind speed. Theoretically, pores in the uppermost soil layer might be ventilated under high wind speed when no chamber is in place; thus, a close coupling of the flux to the atmosphere exists. Consequently, the establishment of a steady-state flux may be more postponed under these high wind speed conditions once the chamber is put onto the soil frame. Such time delay caused by a slow filling up of the previously ventilated pore space in parallel to the diffusion into the chamber might in principle explain exponentially increasing concentrations. However, κ values (Fig. 3c, d) and fluxes (not shown) were independent of both wind speed and direction, which is a further indicator that the chosen chamber design and setup can be used over a wide range of environmental conditions and neither seem to affect concentration buildup characteristics nor resulting flux magnitudes.

Reviewing past decades of field chamber measurements for studying the soil–atmosphere exchange of N2O, several challenges and shortcomings emerged such as a limited number of replicates or the disturbance of the soil microenvironment due to chamber coverage and soil collar insertion (e.g., Hutchinson and Mosier, 1981; Parkin and Venterea, 2010). One way of getting a higher temporal resolution and thereby a higher number of replicates, and keeping soil disturbance as low as possible is to reduce the chamber closure period, which is also expected to decrease deviation from linearity in concentration increase.

The overestimation of the 60 min method compared to the 3 min method as shown in Fig. 4b, which causes a relatively high slope of 1.80, was mainly caused by three exceptionally high fluxes. In addition to any form of unintended interferences with the “natural steady-state flux” (for example, disturbances through macrofauna, fluctuating pump performance, or analyzer malfunctions due to internal recalibration during chamber deployment), much higher 60 min-based HMR fluxes compared to 3 min-based linear fluxes may be observed when one of the two following concentration increase patterns are observed.

A slow initial increase in concentrations followed by a steeper rise after some minutes. The slope of the linear fit will then be much lower than the one from the HMR fit (linear fit at t0).

As shown in Fig. 4c, standard errors of N2O fluxes from both 3 min and 60 min closure were extremely low, i.e., in the order of 0.2 % of the fluxes with median values of 0.17 and 0.06 µg N m-2 h-1 and arithmetic means of 0.21 and 0.20 µg N m-2 h-1 for the 3 and 60 min closure flux estimates, respectively. In comparison, Cowan et al. (2014a) also find low flux uncertainty of less than 1 to 2 µg N m-2 h-1. This implies that limiting the chamber closure period to 3 min in beneficial in two ways. On the one hand, the soil column of interest is less disturbed by shorter coverage and/or the number of replicates can be significantly increased. As these measurements are automated, no further manual work is required. On the other hand, standard errors of fluxes remain extremely low. However, it is recommended to extend the chamber closure period to a minimum of 5 and a maximum of 10 min as slightly delayed concentration increases under low-flux regimes may occur (in ∼ 5 % of the cases in our study) and would lead to an underestimation of 3 min linear fluxes (see Fig. S1 in the Supplement). We therefore recommend skipping the first 2 min of data to guarantee undisturbed conditions that might have been caused at the moment when the chamber is set on the soil collar. The “dead time” of the system, i.e., the time that passes between the moment when an air sample leaves the chamber and the moment when it reaches the analyzer, was ∼ 10 s – given a tube length of 10 m, a flow rate of 1 L min-1, and an ID of the tube of 4.6 mm – and was already considered in the recommendation.

Standard errors of N2O fluxes were found to be invariant on QCL sampling frequency (Fig. 5). The conclusion we can draw from this finding is that chamber operators – in the case of an analyzer with a precision like the QCL presented in this study being available – can reduce their sampling time down to 5 s without risking an increase in the standard error of the flux, which would still be on a much lower level than those obtained from GC measurements (cf. results in Sect. 3.2).

Our comparison of GC vs. QCL fluxes revealed that despite much higher precision, robustness, and temporal resolution in QCL measurements, GC is still a useful method to determine the average campaign N2O soil efflux. Although single flux values particularly under low-exchange regimes did not match well, campaign means and medians were similar to those obtained by the QCL method. Under high-exchange regimes, however, flux patterns matched considerably better but resulted in larger absolute errors when comparing the campaign average, thereby leading to systematic errors (in our case an underestimation) when using the GC method at high N2O fluxes for the assessment of N balances. However, given the fact that the bulk of the annual efflux occurs after management events on a relatively short timescale (Flechard et al., 2007; Skiba et al., 2013), usage of a GC-based system will be prone to large uncertainties (cf. Fig. 7).

While only 8 % of GC data from the Braunschweig campaign met the criteria for flux calculation using the HMR model, 38 % of GC from the Risø data allowed for HMR flux calculation, indicating that higher-exchange regimes favor the usage of an exponential model when using the GC method. Similar findings (37 % allowance for nonlinear model application) were reported by Petersen et al. (2011). Forbrich et al. (2010) found percentages of 13.6, 19.2, and 9.8 % of GC measurements on hummocks, lawns, and flarks, respectively, which were best fitted with an exponential model. Their measurements, however, were made for methane fluxes and under an even larger ∂c/∂t range than was prevalent in our two campaigns. The fact that higher fluxes in our study were associated with lower standard errors and accepted HMR application corresponds well with κ findings in Sect. 3.1 indicating that higher curvature in c(t) coincided with higher fluxes (Fig. 3b).

In general, chamber architecture is essential for headspace concentration buildup patterns given certain enclosure times, activity levels and headspace mixing. Our new chamber system performed well during the two campaigns for both analyzer setups. Through its specific design with not only vertically but also horizontally moving chambers, it will keep the soil column under relatively natural conditions. The only problem emerged at Risø when the guiding racks were slightly distorted under high wind speed conditions, i.e., when half-hourly means of wind speed were higher than 6 m s-1. However, this problem could easily be fixed by tightening the guy wires that are attached to the aluminum rack. Commonly observed winter problems such as unnatural accumulation of snow inside the chamber and rime ice formation on the guiding racks and soil frame were not tested within this study but will likely affect the ease of operation during harsh winter conditions.

From an ecological point of view, QCL measurements offer a new opportunity for robust quantification of soil N2O consumption. As N2O uptake via denitrification exists in theory and could be shown under controlled lab conditions (e.g., Firestone and Davidson, 1989), it has been a major challenge to measure reliable fluxes in the field due to the fact that the magnitude of N2O uptake by soils is usually very low (Schlesinger, 2013) and thereby problematic to be determined by GC measurements that are subjected to low signal-to-noise ratios (e.g., Brümmer et al., 2008).

Our QCL-based measurements under the given soil, temperature, and moisture conditions at Risø and Braunschweig did not result in any soil N2O uptake fluxes. In the study by Cowan et al. (2014b), approx. 10 % of their fluxes on grazed grassland and barley sites were negative. However, only 4 out of 115 negative fluxes were above the LoD of the method, which was estimated to be 4 µg N m-2 h-1, thus being similar to ours (cf. Table 2).

GC-based data in our study showed 2 out of 37 and 50 out of 201 negative fluxes in Risø and Braunschweig, respectively. In Risø, only 3 of the 50 negative flux rates were found to be significant (p<0.05; p values not corrected for multiple testing), thus stressing the challenge of a robust determination of soil consumption of this important greenhouse gas when using the common vial–GC approach. Due to the fact that in this study no N2O soil uptake was found when using the much more reliable QCL setup, a further investigation of this topic on a variety of soil types under different land uses, plant communities, and climatic conditions is highly desired.

Besides investigating possible N2O soil uptake, the QCL methodology offers the opportunity to study diurnal variability of N2O fluxes. In a recent study by Shurpali et al. (2016), it has been pointed out that neglecting diurnal variations leads to uncertainties in terrestrial N2O emission estimates, and they should therefore be taken into account carefully when calculating nitrogen budgets. Similar to our study (Fig. 8), Shurpali et al. (2016) found reversed diurnal patterns under differing flux magnitudes. Intriguingly, when mean N2O fluxes were in a range between 12 and 35 µg N m-2 h-1, in both this study (Risø low-flux regime) and Shurpali et al. (2016), the highest fluxes were found during nighttime and the lowest fluxes around midday. A contrasting diurnal pattern was observed when fluxes were lower than during the Risø campaign, i.e., in Braunschweig (7.2 to 8.7 µg N m-2 h-1), or much higher due to fertilizer application as in Shurpali et al. (2016) (230 to 475 µg N m-2 h-1). In the latter campaigns, mean N2O fluxes were highest at midday and lowest during the nighttime, which corresponds to earlier findings (e.g., Christensen, 1983; Du et al., 2006; Parkin and Kaspar, 2006; Brümmer et al., 2008; Alves et al., 2012) where temperature was proved to be the main controlling factor for N2O soil–atmosphere exchange. Our study highlights that through their high time resolution, QCL-based measurements will not only help enhance process understanding of N2O exchange by disentangling the strength of different drivers of N2O production like temperature, soil moisture, nitrogen availability, and microbial activity, but they also have the potential to provide new insight into bidirectional exchange characteristics of other trace gases such as CH4, which can be sampled simultaneously with our chamber system depending on the analyzer type used.