Small lakes represent an important source of atmospheric CH
Due to its higher global warming potential, methane contributes about 20 %
of the overall greenhouse effect (Cicerone and Oremland, 1988; Wuebbles and
Hayhoe, 2002; IPCC, 2013). Lakes and wetland ponds have strong potential
impacts on the methane budget (Tranvik et al., 2009) due to their anoxic
sediment conditions and often high organic matter content (Zehnder, 1978).
Lake methane fluxes and their temporal patterns are still poorly
constrained and form a major gap in the northern C budget (Rasilo et al.,
2015). Over the past decade, new evidence has demonstrated that these
systems have been underestimated in their contribution to the northern
carbon balance (Kortelainen et al., 2006; Juutinen et al., 2009; Holgerson
and Raymond, 2016; Wik et al., 2016b). Lakes and wetland ponds can form
high CH
Small shallow lakes have high methane emission potential for several
reasons. First, methanogenesis is sensitive to temperature conditions
(Zeikus and Winfrey, 1976; Dunfield et al., 1993; Schulz et al., 1997;
Yvon-Durocher et al., 2014) and shallow lakes warm up quickly during
the summer season. Second, the small water volume coupled with high organic
carbon content promotes the formation of anoxic hypolimnion and is related
to increased concentrations and fluxes of CH
Methane fluxes from small lakes have great spatial variability (e.g.,
Casper et al., 2000; Dzyuban, 2002; Kankaala et al., 2004;
Bergström et al., 2007), which hampers flux upscaling and modeling of
the processes required for making regional and global estimations. This high
variability results from multiple environmental controls, including
biological (system productivity, organic matter loading and its
mineralization, methane production, and oxidation by different groups of
microorganisms), physical (temperature, mixing rate, stratification,
diffusion and bubble transport rate) and chemical factors (such as
concentrations of methane, oxygen, inhibitors and pH) (Rudd and Hamilton,
1978; Bastviken et al., 2004; Lofton et al., 2014; DelSontro et al., 2011,
2015, 2016). Due to these variations, the effect of different control
factors is complicated and still insufficiently known. In particular we must
develop robust relationships between lake CH
The excess water supply and flat topography with impeded drainage provides
favorable conditions for wetland and lake formation in West Siberia
(Terentieva et al., 2016). A few studies of methane emission from lakes of
this region (Gal'chenko et al., 2001; Repo et al., 2007; Glagolev
et al., 2011; Sabrekov et al., 2013) indicate that the methane flux from
middle taiga lakes is 10 times lower in magnitude than from southern taiga
lakes. These findings agree with data showing strong latitude gradient of
methane release from permafrost lakes in the Northern Hemisphere (Holgerson
and Raymond, 2016). Despite many studies addressing local spatial
variability in methane emission from lakes, there is still a gap of
knowledge about methane emission from lakes and, even more rare, their
environmental controls on a regional scale (West et al., 2015). An
understanding of the nature of this difference is important for modeling and
reliably estimating regional methane emission from lakes. In this study, we
have the following objectives:
to estimate methane fluxes from small lakes of the middle and southern
taiga to detect key environmental controls of methane emissions on both regional
(between zones) and local (within each zone) scale to improve the precision of methane emission modeling.
In order to complete these tasks, the following methodology was applied. The
best way to take into account the key processes is to construct a
process-based model founded on established physical, chemical and biological
dependences. Regression procedures, commonly used for identification of
environmental controls, cannot fully reflect complicated interactions
between different controls and thus mask a lot of important details.
Rather, with process-based modeling, it is possible to test which
dependences and which parameters are reliable (at least, for a certain
climate zone) and which are not. After taking into account well-known
dependences, a comparison of predicted and measured values can help to find
new potentially important controls.
Studied lakes in MT (
Since the focus of our work is on lake methane emissions, it is necessary to examine their huge temporal variability and episodic peaks attributed to bubbling, which is challenging to quantify (Walter et al., 2006; Wik et al., 2013). It can be suggested that these stochastic emissions can be modeled using self-organized criticality (SOC) theory (for details on this theory see, e.g., Bak et al., 1988; Bak, 1996; Turcotte, 1999), which can simplify scaling the flux measurements across larger areas. Bubbling is similar to systems showing SOC behavior, in which constant external force leads to rapid changes in nonlinear interactions after reaching a certain threshold (Bak et al., 1988). Systems with SOC behavior occur in many disciplines, including physics, biology and economics (Bak et al., 1988). It has been argued that earthquakes, landslides, forest fires and species extinctions are examples of SOC in nature (Bak, 1996). To the best of our knowledge, no applications of SOC to bubbling in lakes have been previously published. However, such a behavior of gas bubbles in foam (Kawasaki and Okuzono, 1996) and in different artificial systems (see, for example, Juodis et al., 2006; Petrashenko et al., 2005) is well known. Therefore, we test whether the measured flux values demonstrate SOC behavior and examine the consequences for upscaling.
Basic characteristics of study lakes (mean values of three measured values – bottom, middle, surface).
The studied lakes are located in two different boreal zones (both with
a subarctic climate according to Köppen climate classification) of the
Western Siberian Lowlands (Russian Federation) (Fig. 1). The northern study
area is in the middle taiga zone (referred to hereafter as MT) about
20–30 km southwest from Khanty-Mansiysk (61
The MT lakes are mostly acidic, have low ionic strength and are surrounded by wetlands (Fig. 1, Table 1). Four lakes were selected to cover the range of sediment properties in this zone. Lake Muhrino has peaty sediments with high mineral content (sandy bedrock), lakes Babochka and Lebedinoe both have mineral-free peaty sediments, and Lake Bondarevskoe has sapropel (flocculated humic material) sediments. The ST lakes are more diverse due to high groundwater mineralization. Lakes Bakchar-ryam, Plotnikovo and the three Bakchar-forest lakes (1–3 in Fig. 1) represent mesotrophic or eutrophic lakes surrounded by soils rich in clay and grasslands. Lakes Gavrilovka-1 and Gavrilovka-2 represent lakes with low nutrient concentrations influenced by groundwaters with a high pH. Lakes Bakchar-bog-1 and Bakchar-bog-2 represent acidic humic wetland lakes with low pH, low ionic strength and low nutrient concentrations. Finally, lake Ob' Floodplain represents floodplain lakes (oxbows) with extremely high nutrient concentrations.
Field investigations were carried out during summer 2014. We conducted 190
methane flux and 170 carbon dioxide flux measurements, with 70 and 60 in MT
and 120 and 110 in the ST, respectively. The total field measurement time
varied from 4 to 10 h per lake, while the average was 6 h. All measurements
were carried out between 10:00 and 20:00 LT; each lake was visited one time. All measurements were conducted using
a boat to prevent any influence on the lake vegetation or sediments. CH
During the chamber measurements, near-surface water (10 cm depth) was
sampled for dissolved CH
We tested for potential diurnal variability in stratification and vertical
mixing due to the difference between day and night temperatures that can
occur in shallow lakes by examining the lake temperature and
oxidation-reduction potential profiles. We found that in this study's lakes,
these terms do not show strong vertical gradients, and the lakes all belong
to the class of continuous cold polymictic lakes (Wetzel, 2001). We are
therefore confident that our single daytime measurements will not generate a
bias in the measured flux estimates (Ford et al., 2002). During flux
measurements there were no periods with strong thermal stratification, as
temperature gradients between surface and bottom water never exceeded
2
Submerged funnel gas collectors analogous to those in other measurement
campaigns (Huttunen et al., 2001; Repo et al., 2007) were used to monitor
CH
At each lake site, environmental characteristics were measured at three depth
levels – 20 cm below water surface, the lake profile midpoint and 10 cm
above sediment depth. At each level, we measured air and water temperatures
(
It is important to notice that the obtained flux and supporting data give
only a momentary snapshot of methane emission from a certain lake section.
While the spatiotemporal variability in CH
Schematic representation of the model structure. The
one-dimensional column is divided into lake water and sediments. The forcing
consists of lake and sediment depths, the water
All statistical analyses were performed using the Statistica 8 software
(StatSoft, USA). Ordinary least square regression (
To analyze the zonal difference between fluxes, a process-based model reproducing the effect of the main environmental controls that are well known from literature (such as temperature, pH, lake area and depth, DOC concentration, etc.) was developed. The model is designed to couple the processes of production, consumption, and transport of methane and consumption and transport of oxygen in the water column and sediments of shallow boreal lakes. The model structure is similar to other methane emission models for wetlands (Walter and Heimann, 2000; Tian et al., 2010; Meng et al., 2012) and lakes (Stepanenko et al., 2011, 2016; Tan et al., 2015). The model structure is represented in Fig. 2, and a full description is given in Appendix A. Input model parameters include the temperature profile of the lake, concentrations of DOC, total phosphorous and the pH value where these values are measured in the near bottom water. The parameters are assumed to be representative for the sediment layer, lake and sediment depth, latitude, and wind speed at the 10 m height. The model outputs are methane and oxygen concentration profiles, methane ebullition rate, and diffusive flux of methane to the atmosphere.
The model is constructed similarly to other modern models that have shown
good ability to predict methane emissions from lakes and ponds (Stepanenko et
al., 2011, 2016; Tan et al., 2015). There are several differences between our
approach and these models. First, in our model, the necessary parameters were
each obtained from published literature for the appropriate climate zone
(where possible) and averaged across different sources. There was no
calibration of model parameters because we try to test how current
scientific knowledge about the methane cycle in boreal lakes can simulate the
chamber-measured methane fluxes. In order to avoid using different calibrated
constants relating the dependence of methane production from substrates
(Stepanenko et al., 2011; Tan et al., 2015), DOC was selected as a single
proxy for substrate of methane emission (Tian et al., 2010). In order to
avoid calibrating the strongly variable temperature dependence of methane
production and to take into account its potential climatic differences, a
climate-sensitive approach was used (see Appendix B). Second, unlike previous
models (Stepanenko et al., 2011, 2016; Tan et al., 2015) we have added the
influence of pH through a nondimensional scaling factor (Appendix B). Third,
gaseous molecular methane diffusion in lake sediments is included in contrast
with the previous models (Stepanenko et al., 2011, 2016; Tan et al., 2015).
It was introduced because initial numerical experiments demonstrated that
taking into account only liquid CH
Surface-dissolved water CH
Summary of field flux observations (empty cells indicate no data).
This choice of the model framework is based on the data availability, which covers a mix of both spatial and seasonal conditions. Recent models for lake methane emissions (Stepanenko et al., 2011, 2016; Tan et al., 2015) are validated mostly against seasonal time series taken at singular locations. Thus, it is not clear whether the influence of spatial variability can be explained according to modern knowledge about environmental controls of methane emission or whether there are controls that are valuable on different spatial scales but not included in models. For example, controls that are relatively stable for a single lake and on a seasonal scale (climate, lake pH and trophic state, sediment porosity) may not be relevant on greater spatial scales. Since this paper's obtained flux data cover regional and local spatial variability, we use simple empirical relationships for controls that are known to be important on these scales: temperature (on a climate-sensitive basis), pH and DOC concentration (Le Mer and Roger, 2001; Nazaries et al., 2013; Serrano-Silva et al., 2014). The microbial communities of methanogens and methanotrophs and their dynamics were not simulated (as performed, for example, in Grant and Roulet, 2002, and Kettunen, 2003) despite their importance because it is currently not possible to obtain reliable estimates of microbiological parameters for lakes with different pH and trophic states. Therefore, we compromise between the model's complexity (which cannot be overly detailed due to the challenge of obtaining reliable data for validation) and data availability (i.e., that the model should describe the influence of important and measured controls on the scale of the data that are present).
The partial differential equations were solved with MATLAB v. 7.8.0 (Mathworks, USA). A bootstrap method (Efron and Tibshirani, 1986) was implemented to find the uncertainty bounds on the modeled fluxes, as follows. First, artificial errors were introduced for each model parameter using their given standard deviations and a normal distribution. Then, 1000 iterations of these noisy parameter values were used to generate noisy flux estimates, and the uncertainty on the predicted flux value was derived as the standard deviation of these outputs.
A summary of methane flux measurements is presented in Table 3. The median
methane fluxes were 0.3 and 4.1 mgCH
Simple regression showed that there were several correlations between
environmental variables and either average or median CH
The ST lake methane flux is more variable (both for individual lakes and for
data combined for all lakes) than the MT lake fluxes (see Table 3). For
example, the median coefficient of variation for average flux values from ST
lakes (0.87) is more than twice the value from MT lakes (0.36). CH
Flux data for
Since multiple linear regression did not reveal statistically significant dependences with two or more independent variables from the environmental factors listed in Sect. 2.2.2, the multiple effect of environmental controls is confounding. Further analysis was provided using a process-based model (see Sect. 2.2.3 and Appendix A), which reproduced the methane and oxygen production, consumption and transport in lake water and lake sediments.
Observed versus predicted values of methane flux. Whiskers denote
The modeling results are presented in Fig. 4 and in Table 4. The predicted
fluxes fit the observed values for ST lakes quite well (
Summary of modeling results.
Methane concentrations appear to be strongly underestimated (4–6-fold) for
those ST lakes in which it was measured (Table 2). Our numerical experiments
showed that the CH
Thus, the main differences between observed and predicted methane emissions
are that the model
overestimated fluxes for MT lakes by more than 1 order of magnitude underestimated concentration of dissolved methane in both MT and ST
lakes (4–6-fold).
Additionally, the data showed extremely high variability in fluxes from ST
lakes. Without additional flux monitoring and a greater focus on the driving
process mechanisms, it may be that this experimental dataset is not suitable
for a model comparison or validation effort. In the discussion section we try
to suggest where these discrepancies have come from and how they can be
explained.
Summary for temperate and boreal lakes with an area
The obtained data indicate that CH
Comparison with measurements from small (
The significant differences in measured CH
We therefore focus on MMPR, which may actually be lower in MT lakes compared
to ST lakes due to substrate availability. While the mean difference in DOC
between zones is not significant (15 mg L
This low-production, low-ebullition hypothesis is supported in this study by
measurements with bubble traps (see Table 3): the ebullition flux in MT lakes
is less or equal to the diffusion flux calculated as the difference between
the flux measured by static chambers and the flux measured by bubble traps.
Meanwhile, the ebullition flux in ST lakes is many times higher than the
diffusive flux from both model predictions and measurements. Certainly, our
field experiments covered a relatively short period and were insufficient for
exhaustively estimating methane emission pathways because we lacked the time to
measure more bubbles. However, data by Repo et al. (2007) obtained in similar
lakes with bubble traps during a month or more in summer are in good
agreement with our measurements: no bubbles caught in a lake with a peat
bottom (similar to lakes Lebedinoe and Babochka in the current study in which
bubbles were also not detected) and small fluxes in lakes with sandy bottoms
(0.04–0.4 mgCH
One can try to estimate the impact of these two possible causes:
climatic and trophic. There are no data about the MMPR in lake
sediments in West Siberia but we can estimate the climatic impact
driving MMPR differences using data for ST and MT wetlands. According to
Kotsyurbenko et al. (2004), the methane production under optimal temperature
conditions and without substrate limitation measured in ST wetlands is
110 mgCH
Therefore, the sum of the climatic and trophic impacts gives a
12-fold reduction for the MMPR value for MT lakes in comparison with ST
lakes. If we presume that the model's MMPR value is typical for ST lakes, the
MMPR for MT lakes should be 2.60 mgCH
Model calculations show that only on average 22 % (12–40 %) of total
produced methane is oxidized (see Table 3). The latter value is lower than
the experimentally measured oxidized CH
The first possible reason for these differences is that the model has
underestimated methane oxidation. Indeed, a comparison of half saturation
constants for methane oxidation from different studies showed that this
constant for highly productive CH
This pattern could also be explained by the underestimated gas-filled
porosity in lake sediments, which is an important control of dissolved
CH
Accumulation of free gas affects the tortuosity of the sediment and leads to
an underestimated diffusion coefficient for dissolved gas (Flury et al.,
2015). The gas-filled porosity influence on methane cycling in lakes can be
tested with a quick numerical experiment as follows. Consider doubling the
gas-filled porosity for ST lakes to 0.05. This value is still typical for
natural shallow lake sediments. For example, according to Valsaraj et
al. (1999) the maximal gas-filled porosity is 0.07, a value more than 2 times
higher than the 0.025 used by default in our model. In this higher-porosity
case, the oxidized fraction of produced methane will increase to average
49 % (over a wide range from 19 to 90 %). The concentration of
dissolved methane will increase to average 27.7 mgCH
It could be concluded that the natural variability in gas-filled porosity in the sediments can strongly influence the ratio between diffusive transport and ebullition and, hence, the fraction of oxidized methane and total emissions. This variation may result from the extremely nonlinear influence of relatively low values of gas-filled porosity on gas diffusivity (Sallam et al., 1984; Flury et al., 2015). This nonlinearity is related to interconnected water films causing disconnectivity in gas-filled pore space and, thus, reducing gas diffusivity (Moldrup et al., 2003). Unfortunately, data about gas-filled porosity in lake sediments are very sparse and it is difficult to provide a comprehensive analysis of this parameter's influence on methane emission from lakes.
The power-law dynamics of methane emission from ST lakes (Fig. 3) are similar to dynamic system behavior in the SOC theory (Bak et al., 1987, 1988; Jensen, 1998; Turcotte, 1999). SOC is based upon the idea that complex behavior can develop spontaneously in certain multicomponent systems whose dynamics vary abruptly. The paper by Bak et al. (1987) contained the hypothesis that systems that (i) are driven by some external force and (ii) consist of nonlinear interactions amongst their components may generate a characteristic self-organized behavior. The self-organized state into which systems organize themselves has properties similar to equilibrium systems at their critical point; thus, they are described as having SOC behavior (Bak et al., 1987). SOC dynamics are assumed to evolve through the contribution of processes on different timescales. The processes driven externally are typically much slower than the internal relaxation processes. A prototypical example is an earthquake, driven by stress that has slowly accumulated in the Earth's crust due to tectonic activity. This slowly built stress is subsequently released very quickly (in seconds or minutes) in an earthquake (Jensen, 1998). There is an analogous situation in lake sediments, as they become saturated by methane. Methane molecules and energy input continue much longer and more continuously than the release of bubbles and relaxation to the new steady state (Scandella et al., 2011).
The separation of relevant timescales is generated by the threshold
responses – which build up over time – and metastability, which awaits a
triggering event. In lake sediments, the situation is generated by
microorganisms that produce and emit methane molecules into the surrounding
lake water. The methane concentrations increase slowly until a solubility
limit is reached. In this moment a new phase in the form of a bubble is
produced. Then the methane concentration inside the bubble slowly continues
to increase until the moment when pressure in the bubble is high enough to work against forces preventing its release to the atmosphere (Scandella et
al., 2011). When a critical pressure is exceeded, bubbles very quickly leave
sediments via the previously formed channel. The applied force – the buildup
of the CH
The actual force that the generated bubble of CH
There are several practical consequences of SOC behavior of methane emission in lakes. The high values of SD in Fig. 4 show not the low accuracy of measurements but natural spatial and temporal variability in methane emissions from lakes. Short-term measurements can produce uncertainty if they are extrapolated to a long time period or season. Controls found to be important from short-term measurements may be unreliable on other spatial or temporal scales. Whole season, multiyear measurements in three lakes in northern Sweden, made by Wik et al. (2013, 2014), confirm this hypothesis. Each season of their measurements has a unique type of seasonal dynamic with a unique pattern of peaks and falls related to temperature and atmospheric pressure dynamics (Wik et al., 2013). However, the whole season methane budget clearly linearly correlates with seasonal energy input to lakes (Wik et al., 2014).
Another practical consequence is in the upscaling of flux measurements in
lakes for large regions. Once we determine that the probability distribution
law is relevant across all the ST lakes, we can use it for upscaling. The
mean value for a power-law distribution is (Newman, 2005):
Despite this stochastic behavior of emissions, our modeled flux values are in
good correspondence with measured fluxes. There are several reasons for this
agreement. According to the probability law distribution identified for ST,
10 or more flux measurements, as we have performed, allow detection of high-flux moments (for example, for our ST flux power-law function the probability
of detecting a flux with a magnitude from 10 to 20 mgCH
Comparison of observed and predicted fluxes can help to reveal other
important methane emission controls on a spatial scale. There are two strong
site discrepancies for our model: CH
The Ob' Floodplain lake has the highest trophic state (in terms of P
concentration) in our sample (see Table 1). Therefore, it is natural to suggest that
higher trophic states produce higher MMPRs (in this case approximately
50 % higher) and hence higher emissions for this lake. Phosphorous does not
directly influence methane production but strongly positively correlates with
chlorophyll concentration, indicating productivity of algae, and with
sediment respiration, indicating higher intensity of organic matter
decomposition and higher oxygen consumption by sediments, as reviewed by Pace
and Prairie (2005). Higher algae productivity (West et al., 2015) and peat
decomposition supply methanogenesis with fresh organic substances, while
lower oxygen concentration leads to decreasing methane oxidation.
Moreover, temperature dependency of CH
We decided not to compare residuals for the MT lakes because of the small
sample size and, as mentioned in Sect. 4.2, possible differences in
gas-filled porosity. The latter parameter needs special investigation since
now, without further datasets, it requires near arbitrary selection. It is
interesting to compare our CO
Another possible important control of CH
A study of small-size bodies of water in the non-permafrost region of West Siberia has demonstrated that lake and pond methane fluxes vary on both regional and local spatial scales. Based on the presented model's calculations, it can be suggested that it is possible to predict fluxes for individual lakes within the same climate zone with a fair agreement by taking into account such established controls as temperature, pH and substrate availability. Individual characteristics of lake origin and development, such as sediment gas-filled porosity, trophic state and organic matter quality, can also have crucial effects on methane emission.
To successfully predict CH
The constructed ab initio model is much more primitive than more complex
recent models (Tan et al., 2015; Stepanenko et al., 2016), but it does not
include calibrated parameters because all parameters can be adopted from the
literature as average values from several literature sources for the suitable
climate zone. It can be assumed that this approach can be effective for
analysis of spatial variability in methane emission, which appears to be
higher than the temporal variability (Treat et al., 2007; Olefeldt et al., 2013; Sabrekov
et al., 2014). Additionally, controls of spatial variability seem to have
lower predictive ability (for example in terms of
For global modeling it is important to know which lakes, and with what kind
of ecological features and in what season, there exists SOC behavior. These
lakes can emit significantly more methane because methane bypasses the
oxidation filter through ebullition. The most interesting question in this
concern is about the limits of environmental controls in time and space that
define the switch between ebullitive and non-ebullitive regimes. Because of
the variability in MMPR and diffusivity of lake sediments, the presence of
such methane emission hot spots as small shallow lakes is expected in
any climate zone. However, because of their great extent in the taiga and
tundra regions, small lakes in those zones are particularly relevant for the
global CH
The code for the methane emission model and the full data set we used are available upon request from the author (Aleksandr F. Sabrekov, sabrekovaf@gmail.com).
The functional forms of process controls were chosen in order to obtain reliable estimates of the governing parameters using publically available information from the appropriate climatic zone. There was no calibration of any model parameters.
Oxygen and methane dynamics in the water column from the water–atmosphere
border to the lower boundary of sediments were modeled using the following
equations according to Tang and Riley (2014):
Oxygen and methane diffusion can be written as (Stepanenko et al., 2011; Tan
et al., 2015)
List of the model parameters.
Methane production in lake sediments was taken into account by multiplying
maximal methane production rate (MMPR) pH ( temperature ( substrate availability obtained using DOC (g m
Methane oxidation within the profile was calculated based on oxygen and
methane concentrations (Michaelis–Menten kinetics) and temperature. The
maximal intensity of methane oxidation
Oxygen is consumed not only by methane oxidation but also by the plankton
respiration in the lake water
Photosynthesis is the only process that produces O
Ebullition was calculated under the assumption that emitted methane bubbles
immediately reach the surface (Stepanenko et al., 2011; Tan et al., 2015).
As for the boundary conditions, we specify zero flux for both gases at the lower
bound:
At the upper bound, we specified diffusive methane flux calculated according
to Riera et al. (1999), Bastviken et al. (2004) and Rasilo et al. (2015):
Since data about the temperature and pH sensitivity of methane production are highly variable (Dunfield et al., 1993; Segers, 1998; Meng et al., 2012), special consideration is required for these important controls.
To constrain the
Empirical function of relative methane production dependence on
pH. The function from Meng et al. (2012) is given for comparison. Whiskers
denote
Empirical function of relative methane emission
The authors declare that they have no conflict of interest.
This article is part of the special issue “Changing Permafrost in the Arctic and its Global Effects in the 21st Century (PAGE21) (BG/ESSD/GMD/TC inter-journal SI)”. It is not associated with a conference.
Support from BIO-GEO-CLIM grant no. 14.B25.31.0001, RFBR grants 15-35-50740, 15-05-07622, and 15-44-00091 and from the European Union FP7-ENVIRONMENT project PAGE21 under contract no. GA282700 is acknowledged. Shamil S. Maksyutov is supported by the Environment Research and Technology Development Fund (2A1202) of the Ministry of the Environment, Japan. We thank all participants in the 2014 summer field campaign. Aleksandr F. Sabrekov thanks Jonathan R. G. Greenwood for his contribution. Edited by: Paul Stoy Reviewed by: two anonymous referees