Introduction
Wetlands and freshwaters are among the main sources for methane (CH4),
which is one of the major greenhouse gases (Dengel et al., 2013; Bastviken et
al., 2011; IPCC, 2013). In wetland ecosystems, CH4 is released via three
main pathways: (i) diffusion (including “storage flux”, in terms of rapid
diffusive release from methane stored in the water column), (ii) ebullition
and (iii) plant-mediated transport (e.g. Goodrich et al., 2011; Bastviken et
al., 2004; Van der Nat and Middelburg, 2000; Whiting and Chanton, 1996). The
magnitude of CH4 released via the different pathways is subject to
variable environmental drivers and conditions such as water level,
atmospheric pressure, temperature gradients, wind velocity and the presence
of macrophytes (Lai et al., 2012; Tokida et al., 2007; Chanton and Whiting,
1995). As particularly ebullition varies in time and space (Maeck et al.,
2014; Walter et al., 2006), total CH4 emissions feature an extremely
high spatial and temporal variability (Koch et al., 2014; Repo et al., 2007;
Bastviken et al., 2004). Hence, attempts to model CH4 emissions based on
individual environmental drivers are highly complex. To identify relevant
environmental drivers of CH4 emissions, the separation of measured
CH4 emissions into the individual pathway-associated components is
crucial (Bastviken et al., 2011, 2004). Moreover, the understanding of the
complex processes determining the temporal and spatial patterns of CH4
emissions is a prerequisite for upscaling field-measured CH4 emissions
to the landscape or regional scale, and thus for adequately quantifying the
contribution of wetland CH4 emissions to global greenhouse gas (GHG)
budgets (Walter et al., 2015; Koebsch et al., 2015; Lai et al., 2012; Limpens
et al., 2008).
However, field studies measuring CH4 release above shallow aquatic
environments or flooded peatlands generally measure total CH4 emissions
as a mixed signal of individual CH4 emission components, released via
all possible pathways (i.e. diffusion, ebullition and plant-mediated
transport). Studies separately measuring temporal and spatial patterns of
CH4 emissions resulting only from either ebullition or diffusion are
rare. Measurements of CH4 ebullition can be performed using manual or
automatic gas traps, as well as optical and hydro-acoustic methods (Wik et
al., 2013, 2011; Maeck et al., 2014; Walter et al., 2008; Ostrovsky et al.,
2008; Huttunen et al., 2001; Chanton and Whiting, 1995), often requiring
considerable instrumentation within the studied system. Diffusive CH4
fluxes are commonly either derived indirectly as the difference between total
CH4 emissions and measured ebullition, or directly obtained based on the
use of bubble shields or gradient measurements of CH4 concentration
differences (DelSontro et al., 2011; Bastviken et al., 2010, 2004). A
graphical method to separate diffusion, steady ebullition and episodic
ebullition fluxes from the total CH4 flux was presented by Yu et
al. (2014), using a flow-through chamber system. However, performed at the
laboratory scale for a peat monolith, measurement results as well as the
applied method were lacking direct field applicability. A first simple
mathematical approach for field measurements to separate ebullition from the
sum of diffusion and plant-mediated transport was introduced by Miller and
Oremland (1988), who used low-resolution static chamber measurements.
Goodrich et al. (2011) specified the approach using piecewise linear fits for
single ebullition events. However, the static threshold to determine
ebullition events, as well as low-resolution measurements, limited the
approach on estimating ebullition events which were characterized by a sudden
concentration increase of ≥ 8 nmol mol-1 s-1. This prevents a
complete and clear flux separation. Moreover, CH4 flux separation
approaches based on manual chamber measurements with rather low temporal
resolution fail to capture the rapidly changing absolute and relative
contributions of the pathway-associated flux components both in time and
space (Maeck et al., 2014; Walter et al., 2006).
Hence, there is a need for a non-intrusive method for separating
pathway-associated CH4 flux components. Improvements in measurement
techniques, particularly by using high-resolution gas analysers (e.g. eddy covariance (EC) measurements), allow for high-temporal-resolution records of CH4 emissions
(Schrier-Uijl et al., 2011; Wille et al., 2008). Recently, a growing number of experimental GHG studies have
employed
automatic chambers (ACs) (Koskinen et al., 2014; Lai et al., 2014; Ramos et al., 2006), which can provide flux data with an enhanced temporal resolution and
capture short-term temporal (e.g. diurnal) dynamics. In addition, AC
measurements can also represent small-scale spatial variability, and thus
identify potential hot spots of CH4 emissions (Koskinen et al., 2014;
Lai et al., 2014). AC systems therefore combine the advantages of chamber
measurements and micrometeorological methods with respect to the
quantification of spatial as well as temporal dynamics of CH4 emissions
(Savage et al., 2014; Lai et al., 2012).
Combined with a high-resolution gas analyser (e.g. cavity ring-down
spectroscopy), AC measurements provide opportunities for (i) detecting even
minor ebullition events, and (ii) developing a statistically based flux
separation approach. This study presents a new calculation algorithm for
separating open-water CH4 fluxes into its ebullition- and
diffusion-derived components based on ebullition-related sudden
concentration changes during chamber closure. A variable ebullition filter
is applied using the lower and upper quartile and the interquartile range
(IQR) of measured concentration changes. Data processing is based on the
R script developed by Hoffmann et al. (2015). The script was modified for
the purpose of CH4 flux calculation and separation, thus including the
advantages of automated and standardized flux estimation. We hypothesize
that the presented flux calculation and separation algorithm together with
the presented AC system can reveal concealed spatial and temporal dynamics
in ebullition- and diffusion-associated CH4 fluxes. This will
facilitate the identification of relevant environmental drivers.
Material and methods
Automatic chamber system
In April 2013, an exemplary measurement site was equipped with an AC system
and a nearby climate station (Fig. 1). The AC system consists of four
rectangular transparent chambers, installed along a transect from the
shoreline into the lake. Chambers are made of Lexan polycarbonate with a
thickness of 2 mm and reinforced with an aluminium frame. Each chamber
(volume of 1.5 m3; base area 1 m2) is
mounted in a steel profile, secured by wires, and lifted/lowered by an
electronically controlled cable winch located at the top of the steel
profile. All chambers are equipped with a water sensor (capacitive limit
switch KB 5004, efector150) at the bottom, which allows steady immersion (5 cm) of the chambers into the water surface.
Hence, airtight sealing and constant chamber volume are ensured during the study period, despite
possible changes of the water level. All chambers are connected by two tubes
and a multiplexer to a single Los Gatos fast greenhouse gas
analyser
(911-0010, Los Gatos; gas flow rate: 5 L m-1), which measures the air
concentration of carbon dioxide (CO2), methane (CH4) and water
vapour (H2O). To ensure consistent air pressure and mixture during
measurements, chambers are ventilated by a fan and sampled air is
transferred back into the chamber headspace. However, due to the large
chamber volume, mixture of the chamber headspace took up to 30 s. As a
result of this, most peaks due to ebullition events were directly followed
by a smaller decrease in measured CH4 concentration. This indicates a
short-term overestimation of the ebullition event (peak), which was
compensated after the chamber headspace was mixed properly (decrease). This
signal in the observed data is hereafter referred to as overcompensation.
Concentration measurements are performed in sequence, sampling each chamber
for 10 min with a 15 s frequency once per hour. When switching from
one chamber to another, the tubes were vented for 2 min using the air
of the open chamber to be measured next. Between two measurements at the
same chamber position, each chamber was vented using the internal fan
throughout the entire 50 min. A wooden boardwalk north of the
measurement site allows for maintenance access, while avoiding disturbances
of the water body and peat surface.
Transect of automatic chambers (ACs) established at the measurement
site. The arrow indicates the position of the climate station near
chamber II.
Flow chart showing the principles of the calculation of
CH4diffusion and CH4total (dashed boxes) as
well as subsequent CH4 flux separation (dotted box).
Time series plot of recorded concentrations (ppm) within the chamber
headspace for (a) a simulated ebullition event and (b) an
exemplary field study CH4 measurement. Time spans dominated by diffusive
CH4 release are marked by (c, d) black dots, enclosed by the 25
and 75 % quantiles ±0.25 IQR of obtained concentration changes, shown
as black dashed lines. Unfilled dots outside the dashed lines display
ebullition events (see also Goodrich et al., 2011; Miller and Oremland,
1988). Grey shaded areas indicate the applied death band at the beginning of
each measurement (25 %). Negative ΔCH4 values indicate a
overcompensation due to (temporally) insufficient headspace mixing.
Flux calculation and separation algorithm
CH4 flux calculation and separation was performed based on an
adaptation of a standardized R script (Hoffmann et al., 2015). Figure 2 shows a
flow chart of the flux calculation algorithm and the principle of the
performed CH4 flux separation. To estimate the relative contribution of
diffusion and ebullition to total CH4 emissions, flux calculation was
performed twice (Fig. 3), once for the total CH4 flux
(CH4total) and once for the diffusive component of
CH4total (CH4diffusion), by adjusting selected
user-defined parameter setups of the used R script. First of all, a death
band of 25 % (user defined) was applied to the beginning of each flux
measurement, thus excluding measurement artefacts triggered by the process
of closing the chamber. On the remaining flux measurement data sets a
variable moving window (MW) with a minimum size of 5
(CH4diffusion; user defined) and 30 consecutive data points
(CH4total; user defined) was applied. This generated
several data subsets per flux measurement for CH4diffusion and one
data subset for CH4total. Subsequently, CH4 fluxes were
calculated for all data subsets per flux measurement using Eq. (1), where M
is the molar mass of CH4, A and V denote the basal area and chamber
volume, respectively, and T and P represent the inside air temperature and
air pressure. R is a constant (8.3143 m3 Pa K-1 mol-1).
rCH4µg C m-2s-1=M×P×V×δνR×T×t×A
In the case of CH4total, δν is calculated as the
difference between the start and end CH4 concentration of the enlarged
MW (30 consecutive data points; 7.5 min). To avoid measurement artefacts
(e.g. overcompensation), being taken into account as start or end
concentration, measurement points representing an inherent concentration
change smaller or larger than the upper and lower quartile ±0.25 times
IQR (user defined) were discarded prior to calculation of
CH4total. In the case of diffusion, δν is the slope
of a linear regression fitted to each data subset. The resulting numerous
CH4diffusion fluxes calculated per measurement (based on the
moving window data subsets) were further evaluated according to different
exclusion criteria: (i) range of within-chamber air temperature not larger
than ±1.5 K; (ii) significant regression slope (p≤0.1); and
(iii) non-significant tests (p > 0.1) for normality
(Lilliefors' adaption of the Kolmogorov–Smirnov test), homoscedasticity
(Breusch–Pagan test) and linearity. In addition (iv) abrupt concentration
changes within each MW data subset were identified by a rigid outlier test,
which discarded fluxes with an inherent concentration change outside of the
range between the upper and lower quartile ±0.25 times (user defined) the
interquartile range (IQR). Calculated CH4diffusion fluxes
which did not meet all exclusion criteria were discarded. In the case of more
than one flux per measurement meeting all exclusion criteria, the
CH4diffusion flux with a starting CH4 concentration
being closest to the atmospheric CH4 concentration was chosen. Finally,
the proportion of the total CH4 emission released via ebullition was
estimated by subtracting identified CH4diffusion from the
calculated CH4total following Eq. (2).
CH4ebullitionn=∑i=1nCH4total-CH4diffusion
Since no emergent macrophytes were present below the automatic chambers,
plant-mediated transport of CH4 was assumed to be zero. The same
accounts for negative estimates of CH4 released through ebullition.
The used R script, a manual and test data set are available at
https://zenodo.org/record/53168.
Verification of applied flux separation algorithm
A laboratory experiment was performed under controlled conditions to verify
the used flux separation algorithm. In order to artificially simulate
ebullition events, distinct amounts (5, 10, 20, 30 and 50 mL) of a gaseous
mixture (25 000 ppm CH4 in artificial air; Linde, Germany) were
inserted by a syringe through a pipe into a water-filled tub (12 L) covered
with a closed chamber (headspace V=0.114 m3; A=0.145 m2). The
water within the tub was not replaced during the laboratory experiment, thus
ensuring CH4 saturation after the first simulations of ebullition
events. Airtight sealing was achieved by a water-filled frame, connecting
tub and chamber. The chamber was ventilated by a fan and connected via pipes
to a Los Gatos greenhouse gas analyser (911-0010, Los Gatos), measuring
CH4 concentrations inside the chamber with a 1 Hz frequency (Fig. 4).
To ensure comparability between in vitro and in situ measurements, data processing was
performed based on 0.066 Hz records. The expected concentration changes
within the chamber headspace as the result of injected CH4 were
calculated as the mixing ratio between the amount of inserted gaseous
mixture (25 000 ppm) and the air-filled chamber volume (2 ppm).
Exemplary field study
Ecosystem CH4 exchange was measured from beginning of July to end of
September 2013 at a flooded former fen grassland site, located within the
Peene river valley in Mecklenburg-West Pomerania, northeast Germany
(53∘52′ N, 12∘52′ E). The
long-term annual precipitation is 570 mm. The mean annual air temperature is
8.7 ∘C (DWD, Anklam). The study site was particularly influenced
by a complex melioration and drainage programme between 1960 and 1990,
characterized by intensive agriculture. As a consequence, the peat layer was
degraded and the soil surface was lowered by subsidence. Being included in
the Mecklenburg-West Pomerania mire restoration programme, the study site
was rewetted in the beginning of 2005. As a result, the water level rose
above the soil surface, thus transforming the site into a shallow lake.
Exceptionally high CH4 emissions at the measurement site were reported
by Franz et al. (2016), who measured CO2 and CH4 emissions using
an eddy covariance system, and Hahn-Schöffl et al. (2011), who
investigated sediments formed during inundation. Prior to rewetting, the
vegetation was dominated by reed canary grass (Phalaris arundinacea), which disappeared after
rewetting due to permanent inundation. At present, the water surface is
partially covered with duckweed (Lemnoideae), while broadleaf cattail
(Typha latifolia) and reed mannagrass (Glyceria maxima) are present next to the shoreline (Franz et al., 2016; Hahn-Schöffl et al., 2011). However, below the chambers, no
emergent macrophytes were present throughout the study period.
Scheme of experimental setup used for the simulation and
determination of ebullition events with a Los Gatos fast greenhouse gas (FGG)
analyser (911-0010, Los Gatos). The crimped area represents water-filled
tub.
Temperatures were recorded in the water (5 cm above sediment surface) and
different sediment depths (2, 5 and 10 cm below the sediment–water
interface), using thermocouples (T107, Campbell Scientific). Additionally,
air temperature at 20 and 200 cm height, wind speed, wind
direction, precipitation, relative humidity and air pressure were measured
by a nearby climate station (WXT52C, Vaisala). Water table depth was
measured by a pressure probe (PDCR1830, Campbell Scientific). All parameters
were continuously recorded at 30 min intervals and stored by a data
logger (CR 1000, Campbell Scientific) connected to a GPRS radio modem.
Results and discussion
Verification of the flux separation algorithm
A good overall agreement was found during the laboratory experiment between
CH4ebullition fluxes calculated for the simulated ebullition events
and the amount of injected CH4. This supports the assumption of using
sudden changes in chamber-based CH4 concentration measurements to
separate diffusion and ebullition flux components and shows the accuracy of
the presented algorithm (Fig. 5). However, when applied under field
conditions, flux separation might be biased due to a steady flux originating
from other processes than diffusion through peat and water layers, such as
the steady ebullition of microbubbles (Prairie and del Giorgio, 2013;
Goodrich et al., 2011). To minimize the potential impact of the steady
ebullition of microbubbles on calculated CH4diffusion, the
concentration measurement frequency during chamber closure should be
enhanced. This allows identifying and filtering small-scale differences
within measured concentration changes using the variable IQR criterion,
which thereby reduces the detection limit of ebullition events.
Scatter plot of the amount of injected CH4 and the corresponding
calculated CH4 ebullition event. The solid black line indicates the
1 : 1 agreement. The linear fit between the displayed values is represented
by the black dashed line, surrounded by the 95 % confidence interval
(grey shaded area).
Time series of (a) total CH4 emissions with
proportions of ebullition (grey bar) and diffusion flux components (black
bar) during the study period from July until September 2013.
Figure 6b and c show the separated flux components
(b ebullition and c diffusion), together with the
development of important environmental parameters, which are assumed to
explain their specific dynamics (a water level, b RH and
wind speed and c sediment (solid line) and water temperature
(dashed line)). Pie charts represent the biweekly pooled diurnal cycle of
measured CH4 fluxes. Slices are applied clockwise, creating a 24 h
clock, with black and light grey slices indicating hours with CH4 flux
above and below the daily mean, respectively.
Application to an exemplary field study
Time series of measured CH4total fluxes, integrated over the
four chambers of the transect, as well as the respective contributions of
ebullition and diffusion, are shown in Fig. 6. Apart from short-term
measurement gaps, a considerable loss of data occurred between 27 July
and 7 August 2013 due to malfunction of the measurement equipment.
CH4total fluxes observed by the AC system and calculated
with the presented algorithm were comparable to CH4 emissions measured
during the study period by a nearby eddy covariance system (Franz et al.,
2016). This indicates the general accuracy of
the used measurement system and calculation algorithm.
Observed CH4total fluxes showed distinct seasonal patterns
following the temperature regime at 10 cm sediment depth. This is in
accordance with Christensen et al. (2005) and Bastviken et al. (2004), who
showed that biochemical processes driving CH4 production are closely
related to temperature regimes, determining the CH4 production within
the sediment. In addition to seasonality, CH4total also featured
diurnal dynamics, with lower fluxes during daytime and higher fluxes during
nighttime, which were most pronounced during July and early September (Fig. 6). During August, the diurnal variability was superimposed by short-term
emission events and high amplitudes in recorded CH4total. Similar
to CH4total, diffusive fluxes also showed a distinct
temperature-driven seasonality as well as clear diurnal patterns throughout
the entire study period (Fig. 7). However, compared to the diurnal
variability of CH4total fluxes, a pronounced shift of maximum
emissions from early morning to nighttime hours was revealed for
CH4diffusion during August 2013 (Figs. 6, 7).While maximum
CH4diffusion fluxes during July were recorded during early morning
hours (approx. 03:00 to 06:00 CET), a shift to the nighttime was observed for
August (max. from 21:00 to 00:00 CET). During September maximum fluxes shifted
back to the early morning, with maximum fluxes between 00:00 and 09:00 CET (Fig. 6). This could be explained by differences in turbulent mixing due to
changing water temperature gradients. During daytime, the surface water is
warmed, thus preventing an exchange with the CH4-enriched water near
the sediment, which results in lower fluxes for CH4diffusion.
During nighttime, when the upper water layer cools down and mixing is
undisturbed, enhanced CH4diffusion fluxes can be detected. These
dynamics are more pronounced during warm days, explaining the seasonal
shift, and concealed during periods with a high wind velocity. The obtained
diurnal trend is in accordance with findings of Sahlée et al. (2014) and
Lai et al. (2012), who reported higher nighttime and lower daytime CH4
emissions for a lake site in Sweden and an ombrotrophic bog in Canada,
respectively. However, an opposing tendency was found by Deshmukh et al. (2014), who reported higher daytime and lower nighttime CH4 emissions
from a newly flooded subtropical freshwater hydroelectric reservoir within
the Nam Theun river valley, Laos. In contrast to diurnal trends obtained for
CH4total and CH4diffusion, estimated ebullition events
occurred erratically and showed neither clear seasonal nor diurnal dynamics.
Nonetheless, periods characterized by more pronounced ebullition seemed to
roughly follow the sediment temperature-driven CH4 production within
the sediment as, for example, reported by Bastviken et al. (2004) (Fig. 5). This is
confirmed by a distinct correlation between daily mean sediment temperatures
and corresponding sums of measured ebullition fluxes (r2: 2 cm = 0.63; 5 cm = 0.63; 10 cm = 0.62). Moreover, fewer and smaller ebullition
events were detected in times of reduced wind velocity and high relative
humidity (RH) (e.g. 10–11 September and 18–19 September 2013). However, at the level of single flux
measurements, no significant dependency was found between the recorded
environmental drivers and CH4 release via ebullition. The relative
contributions of diffusion and ebullition were 55 % (min. 33 to max.
70 %) and 46 % (min. 30 to max. 67 %), respectively. This is
in accordance with values reported by Bastviken et al. (2011), who compiled
CH4 emission estimates from 474 freshwater ecosystems with clearly
defined emission pathways. A similar ratio was also found by Tokida et al. (2007), who investigated the role of decreasing atmospheric pressure as a
trigger for CH4 ebullition events in peatlands.
Monthly averaged diurnal cycle of diffusive CH4 fluxes
indicating differences in magnitude and amplitude as well as a shift in
minimum and maximum daily CH4 fluxes over the course of the study period.
Comparison of flux data among the four chambers reveals considerable spatial
heterogeneity within the measured transect (data not shown). Monthly averages of
diffusive, ebullition and total CH4 emissions for all four chambers of
the established transect as well as statistics showing the explanatory power
of different environmental variables are summarized in Table 1. With respect
to total CH4 emissions, neighbouring chambers generally featured high
differences in CH4 fluxes, with no obvious trend along the transect.
The same holds true for derived ebullition and diffusive CH4 flux
components. After separation into diffusion and ebullition, flux-component-specific
dependencies on different environmental drivers were revealed (Table 1).
Monthly averages ±1 standard deviation of hourly CH4
emissions (mg m-2 h-1) for the chamber transect (from chamber
I–IV, starting near the shoreline). Average standardized (beta) coefficients
and Nash–Sutcliffe efficiency (NSE) based on linear regressions and multiple
linear regressions between different environmental drivers and daily subsets
of calculated CH4 emissions are shown below. Monthly averages as well as
statistics are separated according to diffusion, ebullition and total
CH4 flux. Superscript letters indicate significant differences between chambers and the p values of applied
linear and multiple linear regressions (MLRs).
Month
Chamber
CH4diffusion
CH4ebullition
CH4total
mg m-2 h-1
July
I
4.6bd ± 3.1
5.5 ± 7.0
10.1bd ± 7.8
II
1.8acd ± 1.5
3.7 ± 6.9
5.5acd ± 7.1
III
6.1bd ± 4.0
4.7 ± 6.9
10.7bd ± 8.2
IV
8.7abc ± 5.9
4.7 ± 5.3
13.3abc ± 7.6
August
I
5.1 ± 5.9
5.0bd ± 6.8
10.1 ± 10.0
II
3.7 ± 5.0
2.9ad ± 6.0
6.5 ± 8.6
III
5.7 ± 4.9
5.8bd ± 7.4
11.5 ± 9.5
IV
6.1 ± 6.8
3.0ac ± 5.0
9.1 ± 9.4
September
I
2.3bd ± 2.0
1.8bd ± 3.9
4.1bd ± 4.8
II
2.6a ± 2.7
1.1ac ± 3.0
3.7ac ± 4.4
III
3.9d ± 3.9
5.4bd ± 6.9
9.3bd ± 8.8
IV
1.3ac ± 1.6
0.7ac ± 3.4
2.1ac ± 4.0
Mean
5.1 ± 5.7
4.2 ± 6.5
9.2 ± 9.6
Driver
CH4diffusion
CH4ebullition
CH4total
Average standardized (beta) coefficient of
daily data subsets
wind velocity
-0.4e
-0.1
-0.3e
relative humidity (RH)
0.5f
0.1
0.4e
air pressure
0.0
-0.1
0.0
water level
-0.5f
-0.1
-0.4e
air temperature (2 m)
-0.6f
-0.1
-0.4e
water temperature (5 cm)
0.1e
0.1
0.1e
sediment temperature (2 cm)
0.3e
0.0
0.2e
Δ water-air temperature
0.6f
0.1
0.4e
average NSE of MLR
0.72
0.30
0.51
Significant difference (Tukey HSD test; α≤0.1) between
a chamber I, b II, c III and d IV. Significant dependency with average
e p value < 0.2 and
f p value < 0.1.
Overall performance
Compared to direct measurements of diffusion or ebullition (e.g. Bastviken
et al., 2010, 2004) the presented calculation algorithm features two major
advantages. On the one hand it allows deriving ebullition and diffusion flux
components based on the same measurement and spatial entity, which prevents
an interfering influence of spatial heterogeneity on observed flux
components. This is not the case for flux separation based on a combination
of different measurement devices, such as automatic chambers and bubble
traps, which need a sufficient number of repetitions and degree in data
aggregation to reduce the bias, emerging from the spatiotemporal
heterogeneity of erratically occurring ebullition events. On the other hand,
the solely data-processing-based flux separations approach allows for an
application when the use of direct measurement systems for either
ebullition (gas traps, funnels) or diffusion (bubble shields) might be
limited. This is in particular the case when measuring at wetland ecosystem
with a varying water level, such as at the exemplary study site (22 to 35 cm).
During the summer months of 2009 and 2016 the water level dropped
substantially, being either next to or even below the surface (data not
shown). This limited a potential application of bubble traps and shields to
periods with a sufficient water level, despite ebullition from the water-saturated sediment during periods with low water level. In addition to that,
the AC system and presented flux separation algorithm allows for parallel
measurements of different trace gases (e.g. CO2 and CH4) at the
same chamber position.
However, flux separation using the presented algorithm might be biased by
steady ebullition of microbubbles and frequently occurring strong ebullition
events. Steady ebullition of microbubbles results in an overestimation of
CH4diffusion and underestimation of
CH4ebullition, an effect that might be reduced by
enhancing the measurement frequency and thus the sensitivity of the variable
IQR filter. Compared to that, frequently occurring strong ebullition events
might disable the calculation of CH4diffusion, which hampers
flux separation for the corresponding measurement. Out of 14 828 valid
automatic chamber measurements during the exemplary field study, the
algorithm failed to calculate CH4diffusionduring
170 measurements. This equals 1.15 % of all measurements. Taking into
account that the presented measurement site is characterized by rather large
CH4 emissions (Franz et al., 2016) and
frequently occurring ebullition events (Fig. 3), this limitation seems to be
negligible.
Compared to other data-processing-based approaches for CH4 flux
separation (e.g. Goodrich et al., 2011; Miller and Oremland, 1988), the
presented algorithm calculates an integrated ebullition flux component. This
ensures a reliable flux separation, despite potential measurement
artefacts such as overcompensation or incomplete ebullition records.
Accounting for the few prerequisites (high-resolution closed chamber
measurements) as well as mentioned advantages, an application of the
presented approach to open-water areas of a broad range of wetland
ecosystems and automatic closed chamber systems is stated.
Conclusions
The results of the laboratory experiment as well as the estimated relative
contributions of ebullition and diffusion during the field study indicate
that the presented algorithm for CH4 flux calculation and separation
into diffusion and ebullition delivers reasonable and robust results.
Temporal dynamics, spatial patterns and relations to environmental
parameters well established in the scientific literature, such as sediment
temperature, water temperature gradients and wind velocity, became more
pronounced when analysed separately for diffusive CH4 emissions and
ebullition. The presented algorithm will be applicable as long as the
underlying closed chamber measurements deliver continuous high-resolution
records of CH4 concentrations and air temperature. However, steady
ebullition of microbubbles might yield an overestimation of the
diffusive flux component, whereas continuously strong ebullition events
might totally prevent flux separation. Hence, the application and adaptation
of the presented algorithm for different wetland ecosystems and automatic
chamber designs is needed. Obtained results should be further validated
against direct flux measurements using, for example, bubble traps or barriers. This
will allow evaluating the generalizability and applicability to other
freshwater and wetland ecosystems as well as chamber designs.
Despite the mentioned shortcomings, the presented calculation algorithm for
separating CH4 emissions increases the amount of information about the
periodicity of CH4 release and may help to reveal the influence of
potential drivers as well as to explain temporal and spatial variability
within both separated flux components. In future, the implementation of
CH4 released through plant-mediated transport into the flux separation
algorithm should be addressed. This could be realized by complete chamber
measurements with CH4 concentrations measured in different water and/or
sediment depth, which will allow the direct derivation of CH4diffusion.
In a next step, the remaining two flux components could be separated using
the presented algorithm.