Eddy covariance for quantifying trace gas fluxes from soils

Soils are highly complex physical and biological systems, and hence measuring soil gas exchange fluxes with high accuracy and adequate spatial representativity remains a challenge. A technique which has become increasingly popular is the eddy covariance (EC) method. This method takes advantage of the fact that surface fluxes are mixed into the near-surface atmosphere via turbulence. As a consequence, measurements with an EC system can be done at some distance above the surface, providing accurate and spatially integrated flux density estimates. In this paper we provide a basic overview targeting scientists who are not familiar with the EC method. This review gives examples of successful deployments from a wide variety of ecosystems. The primary focus is on the three major greenhouse gases: carbon dioxide (CO2), methane (CH4), and nitrous oxide (N2O). Several limitations to the application of EC systems exist, requiring a careful experimental design, which we discuss in detail. Thereby we group these experiments into two main classes: (1) manipulative experiments, and (2) survey-type experiments. Recommendations and examples of successful studies using various approaches are given, including the combination of EC flux measurements with online measurements of stable isotopes. We conclude that EC should not be considered a substitute to traditional (e.g., chamber based) flux measurements but instead an addition to them. The greatest strength of EC measurements in soil science are (1) their uninterrupted continuous measurement of gas concentrations and fluxes that can also capture short-term bursts of fluxes that easily could be missed by other methods and (2) the spatial integration covering the ecosystem scale (several square meters to hectares), thereby integrating over small-scale heterogeneity in the soil.

1 Introduction 20 Soils are highly complex physical and biological systems which have been challenging scientists who study soil processes in situ. Even simple soils that purely consist of quartz sand, exhibit a heterogeneity that leaves experimental scientists with an impressively large uncertainty when measuring gas fluxes from the soil to the atmosphere under controlled conditions (Gao et al., 1998a, b;Pumpanen et al., 2004;Pihlatie et al., 2013). 25 Thus, in addition to well established techniques to measure soil-atmosphere exchange processes, the technical advances to measure turbulent gas fluxes in the atmosphere above the soil surface with the eddy covariance (EC) method has been enthusiastically adopted even in soil process investigations. In this review we will provide the basic background why atmospheric flux measurements can provide sensitive, accurate, and timely quantitative information on soil processes that are otherwise 30 difficult to measure with traditional techniques. At the same time we will discuss the limitations of the eddy covariance technique in soil process research studies and provide suggestions, how to design field experiments when using the eddy covariance technique and when to rather deploy traditional enclosures or chambers for flux measurements. Finally, we highlight potential future applications and developments.

2 How it works
The basic concept of the eddy covariance method is the measurement of a turbulent flux in the nearsurface atmosphere above an ecosystem and thus also above the soil surface. The measurements follow a systematic sampling approach where air parcels moving past a wind velocity sensor combined with a gas concentration sensor or inlet of relatively small dimension are sampled with a sufficiently 40 high time resolution to resolve all -or almost all -turbulent transport motions that turbulently move the entity of interest up and down in the atmosphere (Baldocchi and Meyers, 1998). To be more precise, the movement is chaotic and three-dimensional (3-D), where the vertical movements of air is of highest interest for soil scientists because this is the direction in which a gas produced in the soil can escape the soil towards the atmosphere. Similarly, the corresponding flux can be in the opposite Because the turbulent flux is directly measured by EC, as compared to inferential methods, where the flux is deduced from e.g. the change of concentration over time inside an enclosure, it is important to understand which assumptions are actually made if such a (direct) point flux measurement is interpreted as the (inferential) flux density of a representative soil surface upwind of the EC sensors.

The basics of atmospheric turbulence
Without a minimum understanding of turbulent motion in the near-surface atmosphere it remains Equation (1) describes how the concentration (c) changes over time (t) inside an imaginary small volume of air (see example in Fig. 1d). Due to physical mass conservation considerations, this temporal change of concentration ∂c/∂t must have a reason, given on the right hand side of the equation.
As Eq. (1) shows, the changes of c are strongly driven by the 3-D movements, the so-called eddies (turbulent whirls) leading to short-term fluctuations in all three wind speed components. Due to the fact that eddy covariance measurements are chaotic (nonlinear) time series measurements, an 90 attempt to linearize the system helps both the understanding of the concept and the calculations.
Following the decomposition concept by Reynolds (1895), each turbulently varying variable is decomposed into a mean component (denoted with an overbar, as in c) and an instantaneous deviation from that mean (denoted by a prime, such as in c ). A single measurement c hence equals c + c . In micrometeorology and ecosystem sciences an averaging period of 30 min has become the standard, 95 but depending on the research questions addressed, the averaging is also done at shorter or longer time intervals (Lenschow et al., 1994).
In principle, Eq. (1) is quite simple and straight forward: the right hand side states that the gas of interest (expressed by c) is transported in all three directions of space (x, y, z), and if there are source and sink terms (S oc and S ic ) the concentration of c in the air volume can further change due 100 to these. Following Reynolds (1895), all transport terms in Eq. (1) -which are the product of a wind vector component multiplied with c -are split in two parts, one which groups the means (e.g. u c), and one which groups the turbulent deviations (e.g. w c ). Technically, the terms u c , v c and w c are statistical covariances, which have given this micrometeorological measurement technique the name "eddy covariance". Older literature often used the term "eddy correlation". Both are correct, 105 and the correlation r w,c is related to the covariance w c (see Section 2.4).
As long as all transport terms in Eq. (1) do not change in space (i.e., ∂/∂x, ∂/∂y and ∂/∂z are zero in all cases), then there is no change in c over time, unless a local source or sink is active. For the volume shown in Fig. 1d it is safe to assume that CO 2 (used here as an example) is neither taken up nor produced inside this volume, and the air stream used by the gas analyzer is so small that it assumptions -must be made to allow us to quantify surface fluxes with one single eddy covariance system as shown in Fig. 2. 2.2.1 Assumption of stationarity of the turbulence field and negligible horizontal flux divergence Although we may obtain a technically perfect measurement for the sample volume (Fig. 1d), relating 125 this measurement to a surface area is only possible if we can make a space-for-time substitution approach known also as ergodicity (Katul et al., 2004) or Taylor's frozen turbulence field hypothesis (see Stull, 1988, for an in-depth discussion). In simple words: under ideal conditions the turbulence field is stationary in such a way that we would get exactly the same readings no matter where we placed our EC system around the location of interest. Under such conditions, the horizontal advection 130 terms ( ∂ ∂x u c and ∂ ∂y v c) become zero, and also the horizontal flux divergence terms ( ∂ ∂x u c and ∂ ∂y v c ) become zero. If this is the case, then also vertical advection ( ∂ ∂z w c) can be neglected. With these assumptions, Eq. (1) expressed for the EC measurement height can be greatly simplified to 135 with the negative sign indicating the convention that turbulent fluxes directed from the atmosphere towards the soil surface are negative. In simple words, Eq.
(2) shows that if the term w c decreases with height, then c increases over time, and if w c does not change with height, then ∂c/∂t = 0.
The vertical flux divergence term ∂ ∂z w c remains in the equation due to the simple fact that the soil surface is a discontinuity in the system where turbulent motion dominates all gaseous transport in 140 the atmosphere, but in the soil pores only laminar (nonturbulent) flow and hence molecular diffusion can occur. If we simplify Eq. (1) for the soil surface where turbulence vanishes, using the same simplifying assumptions as for Eq. (2), then we are left with where Q c is the net flux of component c (in units of c per unit air volume and per unit time, e.g.

145
µmol m −3 s −1 ) at the soil surface (Fig. 2). This is not yet the flux density, F c , which is the flux per unit surface area. Thus, to yield F c from an EC system, Eq. (1) must be intergrated from the ground surface (z = 0) up to the EC height z ec .

Assumption about vertical concentration profile
If ∂c/∂t differs at the soil surface (not inside the soil volume, but just above the physical soil surface 150 which is difficult to define in field research) from the value obtained at the EC measurement height, then a profile measurement of concentrations is recommended. In practice, however, over soils and short-statured vegetation it is safe to assume that ∂c/∂t ≈ const for 0 < z ≤ z ec , and hence Figure 3 adopted from Stull (1988) illustrates this aspect very nicely: in the lowest centimeter 155 above the ground the molecular flux that dominates the gas flux inside the soil volume is incorporated into the turbulent flux in the atmosphere in such a way, that a single EC system at some distance above the ground is able to accurately quantify the gas exchange F c at the soil surface.

Importance of measurement height
In principle, the lower the measurement height above the ground, the easier it is to assume that be-160 tween the ground surface and EC measurement height there are no confounding processes distorting the flux measurements. Due to the size of the sensors (Fig. 1d) and their response times and sampling intervals, it is however necessary to keep a distance from the ground surface such that no relevant eddy sizes are missed by the EC system. An additional reason may also be that the lower the measurement height, the smaller the footprint area represented by such a measurement (Sect. 3). If we 165 however measure at greater height, our approximations made to yield Eq. (4) may become problematic: within the atmospheric boundary layer, which is the layer of air touching the soil surface and which is mixed by atmospheric turbulence, a given flux density F c at ground surface almost linearly decreases with height until the flux vanishes at around inversion height. In Fig. 3a this inversion height is drawn at 1 km height above the ground surface, a typical value for daytime conditions. In a 170 desert or semi-arid region, the height may be higher, and under cloudy conditions or at night it can be substantially lower. If we assume that EC flux measurements are accurate to ±20 %, unless special, restrictive data filtering is applied (e.g., Foken and Wichura, 1996), then it would be challenging to resolve the vertical gradient with normal EC instruments within the lowest 10-20 % of the actual atmospheric boundary layer. If its height is 1000 m, then a 10 m tall tower would still measure 99 % 175 of the expected flux. Contrastingly, if e.g. at night the boundary layer is only 20 m, then the 10 m tower most likely only sees around 50 % of the surface flux. Using a measurement height of 2 m, however, brings the measurements back to the range of experimental uncertainty.
Such considerations imply that the entrainment of the gas of interest from the free atmosphere to the atmospheric boundary layer is small. For CO 2 this is typically the case, but model runs by Of most concern is, that the assumptions made above to simplify Eq. (1) are challenged by nocturnal atmospheric conditions, both over flat ground, where a meandering low-level jet may lead to intermittent turbulence (Mahrt, 2014) or by stagnant air pooling in valleys and topographic depres-190 sions in nonlevel terrain. Aubinet et al. (2012) address all these general issues with eddy covariance measurements in specific systems.

Uncertainty in flux estimates
An important aspect of any measurement technique is the realistic quantification of its measurement uncertainty. Many studies on this topic exist for CO 2 fluxes, which have become standard in ecosys-

195
tem research, but less information is available for other gases and particles. CO 2 fluxes appear to be a comparably simple case since the two main processes -assimilation and respiration -occur continuously and should be similarly represented in EC data. CO 2 flux measurements tend to be quite predictable based on simple light response and respiration models (see Sect. 4.1.1). Futhermore, a frequency analysis of CO 2 fluxes can help to identify issues in the measurements for which quality Contrastingly, effluxes of CH 4 and N 2 O are not necessarily resulting from continuous processes, but may appear in bursts (Fig. 4), from temporally active hot spots, or in response to short-term events (e.g. first precipitation after a drought period). In such cases, a frequency analysis of the 205 time series may show some erratic behavior in both the spectra of the w and c components and the cospectrum of w c , and thus it is not easily determined, how the flux signal can be separated from the noise in the flux data. Wienhold et al. (1996) suggested the use of the cross-covariance function to identify the flux peak from the variance associated with noise at time lags between w and c that are outside the technically realistic range of delays between the wind speed and gas concentration 210 measurements. Statistically, one can also test whether the correlation coefficient between w and c is statistically significant : covariances w c are related to variances of the individual components (σ 2 w and σ 2 c ) via Pearson's correlation coefficient r w,c (e.g. Wilks, 2006), Student's t test can be used to test whether r w,c significantly differs from a zero (random) corre-215 lation, where n is the number of independent samples in the period of the time series that is averaged to obtain the EC flux. In extension to what  showed, it should be recalled that EC measurements obtained with a high temporal resolution tend to heavily oversample the component of interest, and thus the raw number of data records N in a time series does not correspond with the number of independent samples in the same time series n. Wilks (2006) named this the variance inflation factor due to serial dependence (autocorrelation) in the time series, which can be used to derive n from N as 225 where ρ 1 is the lag-1 autocorrelation coefficient. To determine the band of insignificant r w,c around zero, Eq. (6) can be rearranged  to solve for r w,cp for a specific p value, where t p is the t value for the significance level p. As an example: in a 10-Hz time series which is averaged to 30-minute intervals, N =18,000. If ρ 1 is assumed to be 0.7 (this value must be determined 230 individually for each averaging period), then the effective sample size n is approximately 3176. For r w,c = 0 at a significance level of p < 0.05 the critical t value is |t p | = 0.675 (two-sided t test).
Hence |r w,cp | = 0.012, and all covariances derived under conditions where −0.012 ≤ r w,c ≤ 0.012 should be considered insignificantly different from a zero flux (at p < 0.05) since the correlation coefficient r w,c is not significantly different from zero.

235
It should be noted that such statistical assessments only address the random error uncertainty of the flux; however, in most cases it is the systematic errors (e.g., during rainy periods measurements from open-path instruments need to be discarded) that introduce much higher uncertainty, and for the systematic errors no generally applicable quantification procedures exist. For CO 2 fluxes, however, some protocols have been established on how to deal with the typical systematic problems (see 240 Aubinet et al., 2012). Further developments will be needed to establish protocols for CH 4 , N 2 O and other fluxes where the general understanding of the systematic errors is still much more limited than in the simpler case of continuous, nonintermittent and nonepisodic CO 2 fluxes.

Flux footprint
Eddy covariance measurements have the great advantage that they can be performed in a noninvasive 245 and nondestructive way. Only a minor disturbance is needed for placing a tripod or tower at a field site, but since the sensors measure the turbulent flux in the atmosphere, their measurements are in the vast majority of cases unrelated to the position of the tripod or tower itself, but reflect the surface conditions of a so-called "footprint area" upwind of the sensors.
Under the simplifying assumptions made in Sect. 2.2 it is possible to relate the point measurement 250 obtained from eddy covariance to an upwind surface area that influences the measurements. To estimate the area of the flux footprint a wide variety of models is used. The most detailed information can be gained by inverse models. These models invert the wind field and the computer model collects gas molecules artificially emitted by an EC system when they hit the ground. From such information a weighting function can be derived to estimate the percentage of influence of a given square meter surface, typically leading to rejection of these data.
Good practice is to check the energy budget closure measured with an EC system, which is recommended for beginners to assertain that their EC system technically works as expected. However, this requires additional measurements of net radiation, ground heat flux, latent and sensible heat flux.
The last component is always measured by the ultrasonic anemometer-thermometer employed in an 290 eddy covariance system, but the other three components may require additional instruments, and ultimately the general recommendation is to not correct trace gas fluxes, even if the energy budget is unclosed (Foken et al., 2012, their Sect. 4.2.2). The checking of the energy budget closure still helps for a critical evaluation of the technical operation of an EC system.

Experimental approaches 295
The understanding of the footprint concept of atmospheric measurements in general, and of eddy covariance flux measurements in particular, is essential for the establishment of feasible field experiments (see Sect. 4.2). The low predictability of wind direction and turbulence, in combination with the relatively large dimensions of the footprint area and its temporal variability normally do not allow for small-scale measurements in conventional factorial designs of manipulative experiments.

300
However, for ecosystem-scale survey type experiments eddy covariance systems have proved their quality and usefulness. The decision for a specific experimental design is strongly related to the research question and processes that are addressed. Hence, we first provide a short overview over the key processes that have been investigated with the use of eddy covariance flux measurements, followed by a more general classification of approaches and experimental designs.

Measuring net fluxes of trace gases
Due to the fact that the eddy covariance technique is commonly applied to understand the ecosystemscale processes it is crucial to understand that this technique is only capable of measuring a net flux of a specific greenhouse gas. That is, the difference between two or more counteracting gross fluxes is measured. In this review we primarily focus on CO 2 , CH 4 and N 2 O. For these, a basic knowledge 310 of the soil processes involved in the exchange of a trace gas is needed to correctly interpret eddy covariance fluxes. Only few specific experimental set-ups allow measurements of a gross flux component such as soil respiration. Examples are in-canopy (or below-canopy) eddy covariance setups (Meyers and Baldocchi, 1993;Baldocchi and Vogel, 1996;Baldocchi et al., 1997;Blanken et al., 1998;Law et al., 1999;Pihlatie et al., 2005b;Misson et al., 2007;Mammarella et al., 2010; Emmel of the major processes involved in CO 2 , CH 4 and N 2 O production or consumption will be provided in the following paragraphs.

CO 2 fluxes 325
Two counteracting processes, respiration and photosynthesis drive the net ecosystem exchange of CO 2 (NEE CO2 ). Both processes can easily be distinguished by the fact that photosynthesis is only active during the day, whereas respiration dominates the net flux measurements in the dark. Nighttime EC measurements, however, are more difficult to perform and may need specific filtering and corrections (e.g. Acevedo et al., 2009), before they can be used for further analyses (Section 2.4).

330
Respiration is also active during the day, and some discussion has been focusing on the question, whether respiration fluxes are higher during the day due to higher (soil) temperatures during the day as compared to night, or whether daytime respiration rates should actually be lower than at night due to the so-called Kok effect of light inhibition of plant respiration (e.g., Atkin et al., 2000).
The most common approach to study soil CO 2 production processes during the day is to extrapo-335 late nighttime measurements using a regression approach based on major driving variables, such as soil temperature and soil moisture (Lloyd and Taylor, 1994;Reichstein et al., 2005;Coleman and Jenkinson, 2008). Doing so ignores the possible light inhibition effect of plant respiration during the day, but for soil scientists interested in CO 2 production from the soil this may not be of concern. Another approach is the partitioning the net flux of CO 2 via daytime measurements and a light-response 340 curve approach (Gilmanov et al., 2003;Yi et al., 2004;Gilmanov et al., 2007;Desai et al., 2008;Lasslop et al., 2010). Still, it is noteworthy that measured ecosystem respiration consists of several contributing sub-processes, and hence measurements can not directly be linked to the soil only (i.e., to below-ground processes). The major components of ecosystem respiration measured by eddy covariance are autotrophic respiration (CO 2 originating from plants as a by-product of photosynthe-345 sis) and soil respiration. The latter consists of several sub-terms such as heterotrophic respiration (namely basal and litter respiration) and the mycorrhizosphere respiration, which again consists of root respiration, rhizomicrobial respiration and mycorrhizal respiration. Details on each of these processes, driver variables, and their relative contribution to the net flux of CO 2 from the soil have been studied in detail across many ecosystems (e.g., Högberg and Högberg, 2002; Bhupinderpal-Singh

370
Besides specific case studies in forests (e.g. Pilegaard et al., 2003;Pihlatie et al., 2005b;, investigations were also carried out in grasslands (Scanlon and Kiely, 2003;Leahy et al., 2004;Neftel et al., 2007;Soussana et al., 2007;Zeeman et al., 2010), croplands including rice paddies Alberto et al., 2009;Smith et al., 2010;Du and Liu, 2013), savannas (Hanan et al., 1996;Williams et al., 2004;; Hutley et al.,  Another successful approach is to use natural gradients or "treatments" such as in the study by Rocha and Shaver (2011) who established three eddy covariance towers in an area where a huge tundra fire in northern Alaska has left its trace in the landscape: one tower was placed on the most 430 severily burnt surface, a second one was placed on a moderately burnt surface where the fire only destroyed the surface plant material but not the surface soil layer, and the third tower was placed next to the burnt area as a control ("unburnt"). Similarly, chronosequences after forest management The key issue for success with such experimental designs is the synchronous measurements of two or more eddy covariance towers. Due to the huge variability in atmospheric weather conditions, it is always much more difficult to find significant differences between sites that do not cover the 440 same time period, than if the time periods coincide and statistical testing for differences can be made using the paired-samples approach. Due to serial autocorrelation in time series data and spatial autocorrelation between sites, a correction is needed in all statistical testing (e.g., Legendre and Legendre, 1998;Wilks, 2006, see also Section 2.4).
Such a correction takes care of the fact that eddy covariance data heavily oversample the process 445 of interest in the same way as was described in Section 2.4, but now for the comparison of 30-min averaged flux time series. This is not specific to EC measurements, but to any regularly spaced time series of measurements. Namely, when comparing the performance of EC systems with chamber measurements it will be essential to consider serial autocorrelation in statistical testing of EC fluxes.

Survey-type experiments using eddy covariance 450
Manipulative experiments are typically limited to paired-site studies, and hence the most widespread experimental approach seen in the scientific activities employing EC measurements is rather conforming to a survey-type approach. Even if a specific research group is only focusing on one single site, sooner or later the wish for comparisons with other sites measuring the same gas fluxes arises.
Since such comparisons were not planned a priori when the experimental design was established, 455 such comparisons do not normally conform to the rather strict control mechanisms employed in manipulative experiments.
Namely, if measurements from different sites and different years should be compared, then the data analysis is most successful with a survey-based approach, using natural gradients and contrasts of environmental variables and ecosystem parameters that were measured or estimated in addition to 460 EC flux measurements. Often, the first step is to relate near-surface measurements to remote sensing products using regression analysis, be it simple linear or multivariate and nonlinear approaches. The goal is to find functional relationships that in the ideal case can be used as time-for-space substitu-tions to translate the information content of a time series to what a snapshot of the spatial distribution would reveal, had it been possible to take such a snapshot.

465
Aircraft flux measurements directly aim at providing a spatial survey of fluxes. By carefully considering the spatio-temporal variability of fluxes and using increasingly elaborate computer models to interpolate aircraft flux measurements in space and time, it is now possible to assess spatial variability of soil effluxes over larger regions (e.g. Desjardins et al., 1995;Zulueta et al., 2013;Hiller et al., 2014). This comes at the expense of a relatively coarse spatial resolution due to the measure-

475
Roving towers (also mobile towers, portable towers) are an approach to assess spatial variability of soil surface fluxes by placing a tower on different surfaces for a few days or weeks at each location (Eugster et al., 1997;Billesbach et al., 2004). While a few days are generally already helpful for assessing surface energy fluxes, the use of roving towers for CO 2 fluxes is more challenging due to the phenology and seasonal variations in these fluxes (Eugster et al., 1997). The first successful 480 deployment of roving towers as replicated measurements of CO 2 fluxes on four arable plots using only two EC towers was reported by Davis et al. (2010). Such innovative experimental concepts will most likely see further developments and applications in the future, but will always remain a compromise between a permanent tower and a full spatial flux survey as can be done with an aircraft.
In general, the greatest strength of EC measurements over other techniques is the continuous 485 measurements at a high temporal resolution. This is an essential aspect to quantify episodic effluxes that can only be captured via the systematic time-series measurements of EC systems. Substantial fluxes may result from periods with episodic flux bursts, which are not easily captured by a standard chamber measurement approach. However, special care must be taken in cases of varying footprints in highly heterogeneous landscapes: here apparent episodes and flux periods may simply relate to  Table 1 provides an overview over advantages and disadvantages of EC flux measurements in comparison to conventional chamber flux measurements.

Future directions and challenges
Even though important developments have been made in measuring concentrations of particularly CH 4 and N 2 O at suitable time resolution (e.g. field deployable laser absorption spectrometers) to perform EC measurements, large uncertainties remain in measuring the net exchange of both greenhouse gases. This is predominantly due to heterogeneous source distributions within the flux footprint. Additional reasons for difficulties in interpreting flux measurements of CH 4 and N 2 O occur due to the much smaller concentrations in the atmosphere compared to CO 2 (small changes in small numbers vs small changes in big numbers) as well as the behavior of both scalars under varying turbulent conditions (Hörtnagl and Wohlfahrt, 2014;Kroon et al., 2010;Mammarella et al., 2010;Pattey et al., 2006). The complexity of the underlying processes in combination with the fact that 505 multiple contributing processes occur often simultaneously poses further challenges. Fortunately, some of these challenges can be overcome by applying the EC technique in combination with traditional techniques such as flux chambers. Thereby, the combination of both techiques should not be taken as a validation method Pihlatie et al., 2005b;Smith et al., 1994) but rather taken as a supplement to understand the differences between plot-scale versus landscape- and Wang et al. (2013b). Furthermore, emissions that may contribute to the overall ecosystem flux but do not originate from the soil, e.g. plant mediated transport of trace gases (Pihlatie et al., 2005a) need to be quantified.
Another way forward towards understanding soil processes is the combination of EC measure-520 ments with state-of-the-art stable isotope measurements (Chun-Ta et al., 2003;Sturm et al., 2012).
Stable isotopes have been shown to provide a powerful tool to identify hotspots of consumption and production of trace gases in the soil, leading to a more complete understanding of interacting soil process at larger scales. Besides identifying source processes (Baggs, 2008) an additional advantage when combining EC with isotope measurements, is the potential of partitioning net fluxes into the 525 contributing gross components (Ogee et al., 2003;Knohl and Buchmann, 2005;Baggs, 2008). Still, studies combining online measurements of isotopologues and EC remain rare and have in most cases focused on CO 2 only (Saleska et al., 2006;Wehr et al., 2013). Studies focusing on N 2 O and CH 4 are likely to become available in the near future. Namely with the development of new laser absorption spectrometers capable of measuring the isotopic signatures of carbon, nitrogen and oxygen 530 in methane and nitrous oxide (Mohn et al., 2008;McManus et al., 2010;Mohn et al., 2013) new insights into soil processes producing these trace gases become possible (Wolf et al., 2014(Wolf et al., , 2015. It has to be noted that measurements of isotopes provide an additional challenge to EC flux measurements, and detailed in-depth knowledge about isotopic fractionation effects is essential (Friedman and O'Neil, 1977). Still there might be large potential to derive new in-depth knowledge by large 535 scale (e.g. ecosystem) labeling approaches.
Also the power of measurement networks providing ecosystem greenhouse gas flux data has only partly been elaborated to study soil processes (Sanderman et al., 2003). Data originating from such networks (e.g., ICOS: Integrated Carbon Observatory System in Europe, and NEON: National Ecological Observatory Network in the USA, respectively) will provide standardized flux data of trace 540 gases at different spatial scales (EC and chamber data) that will be publicly available and ready to use. The great advantage of such networks is threefold: (1) data are becoming directly available after having passed basic quality checks without any further need for in-situ data collection; (2) they cover a wide range of ecosystems; and (3) they provide ancillary data, which are essential for the interpretation of flux measurements.

545
Besides data driven approaches, combining the EC technique with modeling approaches to reliably estimate the contribution of heterogeneously distributed source and sinks of specific trace gases to the net flux can be achieved (Goeckede et al., 2006;Massman and Ibrom, 2008;Vesala et al., 2008) which will allow for deployments of EC towers in more complex environments than ever before (e.g., small-scale multicrop systems as often used in organic farming). Furthermore, deploy-550 ments of EC towers within or below the canopy as well as at ecosystem edges (Rogiers et al., 2005;Kirton et al., 2009) bear great potential in studying soil processes at the ecosystem scale. Still, each application in heterogeneous terrain involves complex micrometeorological conditions and thus requires a careful interpretation of the measured fluxes. Modeling activities may not only focus on the source distribution within a specific area but may be carried one step further by simulating net trace 555 gas emissions from the soil for areas where it is impossible to measure trace gas fluxes with experimental approaches. Such process-based biogeochemical models (e.g. DAYSCENT, DNDC, PaSim, to name only few of them) can be validated in similar ecosystems with in-situ eddy covariance flux measurements beforehand and then applied at the landscape scale.
Last but not least, extending the application of the eddy covariance technique to other -often 560 even less abundant -scalars besides CO 2 , CH 4 and N 2 O has become reality. For instance, EC measurements of carbonyl sulfide (COS), originally considered an independent tracer for photosynthesis (Asaf et al., 2013;Wohlfahrt et al., 2012), revealed additional COS release from the soil and litter and thereby lead to new research questions Maseyk et al., 2014). In the future, COS measurements may lead to a better understanding of both, ecosystem carbon cycling as well as 565 ecosystem sulfur cycling (Kesselmeier et al., 1999;Billesbach et al., 2014). In addition to COS studies that aimed at measuring other compounds such as ozone (O 3 ) and reactive components as part of nitrogen cycling (NO x , which is the sum of NO and NO 2 ; NO y , which is the total of NO x plus additional products of atmospheric oxidation of NO x , namely HNO 3 , HONO, NO 3 , N 2 O 5 , HNO 4 , PAN, RONO 2 , ROONO 2 ; and ammonia, NH 3 ) with micrometeorological approaches have become 570 available in the past decades (Munger et al., 1996;Eugster and Hesterberg, 1996;Hesterberg et al., 1996;Munger et al., 1998;Famulari et al., 2004;Horii et al., 2004;Jones et al., 2011;Ferrara et al., 2012;Ammann et al., 2012;Marr et al., 2013;Geddes and Murphy, 2014). In particular the ongoing development of easy-to-deploy quantum cascade laser spectrometers capable of measuring various scalars in the field opened the door for fast-response concentration measurements (Bruemmer et al.,575 2013; Famulari et al., 2004;Skiba et al., 2009;Sutton et al., 2007). Still, several limitations have to be overcome when measuring reactive compounds (Marx et al., 2012). These include fast transformation from one compound into another as well as fast cell contamination due to the stickiness of ammonia, to give only two examples (Bruemmer et al., 2013).
The focus on these less abundant gas species (e.g. COS) and reactive compounds urges the need 580 for in-depth knowledge of turbulence as well as fluid dynamics both highly relevant when applying the EC technique. This aspect and the multitude or research areas that are involved when aiming at understanding soil trace gas emissions with micrometeorological approaches ask for further and intensified collaborations between often separated research communities.

585
The eddy covariance method is a micrometeorological technique to quantify surface flux densities of many trace gases produced by soil organisms or taken up by soils and the vegetation. The EC method's main advantage compared to other methods is its spatial scale of integration, ranging from several meters squared to a hectare and more, depending on measurement height, and its temporal coverage that includes all turbulent time scales that are relevant for trace gas mixing in the near-590 surface atmosphere from fractions of seconds to a typical 30 min averaging interval and longer.
Still, the method is not yet at a level where the uninitiated scientist can simply buy a system off the shelf, install it, and be happy with the data streaming in (although most vendors promote exactly this approach). In contrast to conventional meteorological measurements (e.g., air temperature), for which robust, reliable sensors exist that are easy to use and maintain, present-day eddy covariance 595 flux systems still require a fair share of technical and micrometeorological knowledge to operate them reliably and satisfactorily under field conditions and over longer time periods.
In practice, this method substantially enlarges the soil scientist's toolbox and does not necessarily replace any of the existing traditional methods, namely chambers and enclosures. Thus, the EC method is preferrably used in combination with other techniques and methods that fill the gap where 600 eddy covariance flux measurements are not helpful. The key issues to keep in mind to succeed with eddy covariance flux measurements is an adequate design of field experiments that rather should conform to paired-site and survey-type investigations, whereas its application in fully manipulative factorial designs with small spatial treatment units remains a challenge to be solved with future developments. As eddy covariance systems may drop in costliness both in investment and main-which would be required for full factorial manipulative experiments with several treatments that also include replications. Meteorol., 149, 1737Meteorol., 149, -1750Meteorol., 149, , 2009.   shows the ultrasonic anemometer's sensor head (a) with two inlets of tubes leading air to (b) an enclosed-path CO2 and H2O analyser, and (c) to a closed-path CH4 analyzer. The ultrasonic anemometer has three pairs of sensors (pairs shown by arrows), in a configuration that allows to measure the 3-dimensional wind vector in the air volume indicated by (a). Due to sensor separation between wind vector and gas concentration measurements, the eddy covariance "point" measurement is representing a somewhat larger volume of air indicated by (d).
at instrument level at soil surface Figure 2. Under the assumption of negligible horizontal and vertical flux divergence, the covariance w c measured at the instrument height zEC is a direct measured of the net flux SO c − Si c at the soil surface. This simplified assumption can be made for chemically inert components c that do not have any sources and sinks above the soil surface, so that the term SO c −Si c is zero at the instrument height zEC . Correspondingly, there is no turbulent flux at the soil surface and inside the soil volume, hence w c is zero at the soil surface and below.   treatments (c) are placed close to the eddy covariance system (round circle) but outside the main footprint area to minimize disturbance.