Introduction
Global warming is expected to accelerate permafrost thaw, which may trigger
the release of the large amount of frozen soil organic matter (SOM) stored
in the Arctic as carbon dioxide (CO2) and methane (CH4) into the
atmosphere, possibly forming a positive feedback to climate change (Treat
et al., 2015; Knoblauch et al., 2013; Elberling et al., 2013). Permafrost
thawing leads to significant changes in soil water saturation, creating
favorable conditions for anaerobic respiration and methanogenesis (Lawrence
et al., 2015).
Current biogeochemical models predominantly represent SOM decomposition
under aerobic conditions (Manzoni and Porporato, 2009). They are
modified for use under anaerobic conditions. For example, the Community Land Model with coupled
carbon and nitrogen (CLM-CN) decomposition cascade is used to implicitly
represent anaerobic decomposition with a moisture response function that
approaches unity at saturation and an oxygen scalar that has a large
unresolved uncertainty (Oleson et al., 2013). In a recent
permafrost carbon–climate feedback modeling study, the carbon release rate
from permafrost soils after thawing under aerobic conditions was assumed to
be 3.4 times higher than the release rate under anaerobic conditions
(Koven et al., 2015; Schädel et al., 2016). However, in incubations
with soils from Alaska and Siberia, carbon release under aerobic conditions
was 3.9–10 times greater than under anaerobic conditions
(Lee et al., 2012), and CO2 production
appeared ceased at late times in anaerobic microcosms (Xu et al., 2015; Roy
Chowdhury et al., 2015), indicating that these existing models do not
adequately represent the anaerobic processes for accurate prediction of SOM
turnover and heterotrophic respiration.
In addition, it is important to accurately represent methanogenesis in the
context of competing anaerobic processes because CH4 has a 100-year
global warming potential that is about 26 times greater than CO2
(Forster et al., 2007; IPCC, 2013) and an atmospheric residence time of
approximately 10 years (IPCC, 2013), and methanogenesis rate can be high
under favorable conditions. Methanogenesis is carried out by a group of
strictly anaerobic archaea. The free energy of methanogenesis reactions is
less favorable than the reduction of O2, NO3-, Mn (IV),
Fe(III), and SO42- along the redox ladder
(Conrad, 1996; Bethke et al., 2011). The
accumulation of CH4 has been widely observed to lag behind CO2 for
periods ranging from days to years in incubations (Knoblauch et al.,
2013; Roy Chowdhury et al., 2015; Cui et al., 2015; Hoj et al., 2007; Fey et
al., 2004; Jerman et al., 2009; Tang et al., 2013c). The implication is that
a first-order representation (including constant CO2 / CH4 ratio
parameterization) normally overpredicts CH4 production rate before
methanogenesis initiation and underpredicts CH4 production rate
afterwards, and the uncertain lag time introduces large bias in CH4
production prediction.
Besides temperature (Fey and Conrad, 2003; Hoj et al., 2007; Jerman et
al., 2009; Cui et al., 2015) and initial methanogen abundance (Conrad,
1996; Knoblauch et al., 2013), the wide range of redox buffers provided by
the alternative electron acceptors is likely a cause of the wide range of
observed lag times (Estop-Aragonés and Blodau, 2012; Fey et al.,
2004; Jerman et al., 2009; Yao et al., 1999; Conrad, 1996; Knorr and Blodau,
2009). As a result, the ratio of CH4 to CO2 ranges from 0.00001 to
0.5 (Wania et al., 2010; Drake et al., 2009; Bridgham et al., 2013),
highlighting the limitation of the CH4 / CO2 ratio approach.
Nevertheless, some land surface models (LSMs) parameterize methanogenesis as
a fraction of carbon mineralization (Wania et al., 2013; Oleson et al.,
2013; Koven et al., 2015; Cheng et al., 2013). While methanogenesis is
explicitly represented in some models (Xu et al., 2015; Grant, 1998) and
the reduction of alternative electron acceptors is explicitly represented in
others (Fumoto et al., 2008; Segers and Kengen, 1998; van Bodegom et al.,
2000, 2001), these models do not have an aqueous phase
that is essential for explicit biogeochemical calculations, e.g., pH, Eh,
and thermodynamic calculations. Because methanogenesis is sensitive to redox
conditions, the lack of explicit biogeochemical representation of the redox
processes contributes to the prediction uncertainty of CH4 emission.
Anaerobic bacteria and archaea usually depend on simple substrates such as
sugars, alcohols, organic acids, and H2 as carbon and energy sources
that are rarely simulated in ecosystem models (Manzoni and Porporato, 2009;
Xu et al., 2015). Instead, they are typically lumped together as dissolved
organic matter (DOM) or low-molecular-weight organic carbon (LMWOC) (e.g.,
Tian et al., 2010). The abundance and importance of DOM and LMWOC in SOM
turnover in the Arctic soils are becoming increasingly recognized
(Hodgkins et al., 2014). The DOM concentration in
water flowing from collapsing permafrost (thermokarsts) on the North Slope
of Alaska ranges from 0.2 to 8 mM, with biodegradable (degrading in 40 days) DOM
accounting for 10–60 % (Abbott et al., 2014; Arnosti, 1998, 2000;
Arnosti et al., 1998). Ancient LMWOC was found to fuel rapid CO2
production upon thawing (Drake et al., 2015). On the other
hand, new SOM consists of mostly macromolecules of plant and microbial
residues such as carbohydrates (polysaccharides, including cellulose and
hemicellulose), lipids, nucleic acids, and proteins
(Kögel-Knabner, 2002). While conceptual models and measurements
connecting SOM with LMWOC have long existed (Drake et al., 2009; Tveit et
al., 2013, 2015; Bridgham et al., 2013), the hydrolysis and
fermentation reactions have been poorly represented and quantified in the
Arctic as well as temperate and tropical soils. Among over 250 SOM
decomposition models that have been developed in the past 80 years
(Manzoni and Porporato, 2009), only a few models explicitly
simulate simple substrates (Xu et al., 2016b).
Either a simple carbon pool (Cao et al., 1995, 1998;
Kettunen, 2003) or a DOM pool (Tian et al., 2010; Xu and Tian, 2012) has
been assumed for methanogenesis. The production of acetate and H2 has
been parameterized as a function of carbon mineralization (van Bodegom et
al., 2000, 2001; Grant, 1998; Xu et al., 2015). It is
not surprising that CH4 production prediction is sensitive to simple
substrate production (Kettunen, 2003; Weedon et al., 2013). While
detailed SOM decomposition models include depolymerization to produce
monomers under aerobic conditions (Riley et al.,
2014), production and consumption of simple measurable substrates, such as
acetate, H2, and formate, are not explicitly represented under
anaerobic conditions.
In addition to electron acceptors and substrates, SOM turnover is also
sensitive to soil pH. Most methanogens grow over a relatively narrow pH range
(6–8), while some adapt to acidic or basic environments (Garcia et al.,
2000; Van Kessel and Russell, 1996; Wang et al., 1993; Sowers et al., 1984;
Rivkina et al., 2007; Hao et al., 2012; Kotsyurbenko et al., 2004, 2007).
Soil pH can change by 1–2 logarithmic units in laboratory incubations (Xu et
al., 2015; Roy Chowdhury et al., 2015; Peters and Conrad, 1996; Drake et al.,
2015) and it can vary significantly through the soil profile and along
topographic and vegetation gradients in the field (Cao et al., 1995; van
Bodegom et al., 2001; Lipson et al., 2013b). The pH feedback on
methanogenesis could be up to 30 % (Xu et al., 2015). However, soil pH is
often fixed into LSMs (Oleson et al., 2013; Tian et al., 2010). pH is
calculated using soil acidity and soil buffer capacity (van Bodegom et al.,
2001) or as a function of acetate concentration (Xu et al., 2015). It is
desirable to use a geochemical model to describe pH evolution
mechanistically. The pH response functions (reaction rate adjustment factor
as a function of pH) in LSMs are empirical and vary substantially (Xu et al.,
2016b). Assessing the efficacy of these functions is needed to better
represent pH impacts on carbon mineralization and methanogenesis.
Temperature is another critical factor controlling SOM turnover to CO2
and CH4. The reported Q10 values for methanogen temperature
response vary from 1.5 to 4 (Xu et al., 2016b). Methanogenesis has been
widely observed to diminish when the temperature decreases toward
0 ∘C (Dunfield et al., 1993; Fey et al., 2004; Hoj et al., 2007;
Sowers et al., 1984), predicting little CH4 production from the surface
layers of frozen Arctic soils. However, recent observations suggest that
CH4 emissions during the winter season account for ≥ 50 % of
the annual emission in the Arctic (Zona et al., 2016). The cold-season
CH4 production is among the most uncertain processes for predicting
seasonal CH4 cycle in northern wetlands (Xu et al., 2016a). The
temperature response functions (reaction rate adjustment factor as a function
of temperature) need to be assessed as well.
Overall, anaerobic SOM turnover is controlled by the hydrolysis of the
macromolecules to produce simple substrates and the sequential microbial
reduction of electron acceptors along the redox ladder. Because SOM turnover
and CO2 and CH4 productions are sensitive to redox potential, pH,
and temperature, it is desirable to simulate the redox and pH explicitly with
geochemical models. With the accumulation of new data on metabolic
intermediates, electron acceptors, greenhouse gases, and pH from incubations
with Arctic soils at various temperatures (Drake et al., 2015; Herndon et
al., 2015a, b; Yang et al., 2016; Mann et al., 2015), our objectives are to
integrate these new data into geochemical models to (1) extend the CLM-CN
decomposition cascade to include simple substrates such as sugars and organic
acids and add Fe(III) reduction and methanogenesis processes; (2) account for
gas-, aqueous-, and adsorbed-phase speciation; (3) describe pH mechanistically;
and (4) assess the existing temperature and pH response functions. Unlike
with previous LSMs, we simulate speciation of CO2 and CH4 in the gas,
aqueous, and solid phases, and represent sugars, organic acids, Fe(II),
Fe(III), Fe reducers, and methanogens, and account for both thermodynamic and
kinetic control. Our results provide predictive understanding and mechanistic
representations that can be incorporated in LSMs, e.g., CLM-PFLOTRAN (Tang et
al., 2016), to improve climate model predictions.
The carbon cycle involves coupled hydrological, geochemical, and biological
processes interacting from molecular to global scales. The implicit
empirical first-order approach used in existing LSMs limits our
understanding of the land atmosphere interaction and is a source of
prediction uncertainty. To improve our understanding and reduce prediction
uncertainty, we attempt to use relatively more explicit mechanistic
representations developed in the reactive transport model literature (Tang
et al., 2016). Even though explicit representation does not necessarily
improve the match between the predictions and observations over well-tuned
existing models immediately (e.g., Wieder et al., 2015; Steven et al., 2006),
our approach provides a systematic means to incorporate ongoing
process-rich investigations to improve mechanistic representations in LSMs
across scales. For a preliminary study, we constrain our scope to extending
CLM-CN with minimum revision to describe anaerobic CO2 and CH4
production from several recent microcosm studies in this work. We discuss the
next steps briefly in the “Results and discussion” section.
Results and discussion
Experimental observations
The experimental results of anoxic soil incubation experiments were published
elsewhere (Herndon et al., 2015a; Roy Chowdhury et al., 2015), so we briefly
describe the original observed headspace CO2 and CH4 concentration,
soil Fe(II) and organic acid concentration, and pH (Fig. 2). The variations
in the overall observations appear to be better explained by the differences
between the soil horizons (organic vs. mineral soils) than among the
microtopographic locations (center, ridge, and trough) of ice-wedge polygons.
Up to 20 % CO2 was observed in the headspace by the end of the
2-month incubations, with higher concentrations in the organic soils than in
the mineral soils (Fig. 2a1–3 vs. Figs. 4–6). This can be attributed to the
higher organic content of the organic soils compared to that of the mineral
soils (Tables 2, S1).
CO2 in the headspace increased rapidly in the beginning and then the
increase slowed (Fig. 2). The initial rapid increase can be attributed to
fast decomposition of the easily degradable substrates such as sugars and
alcohols (Yang et al., 2016; Fey and Conrad, 2003; Glissmann and
Conrad, 2002; Kotsyurbenko et al., 1993). As the easily degradable substrates
were exhausted, the CO2 production rate decreased. These observations
are similar to those for the anaerobic incubations with soils from a trough
location in a high-center polygon at the same site (Yang et al., 2016) and
deep Siberian permafrost soils (Knoblauch et al., 2013). However, CO2
continued to increase well beyond 2 months in these previous studies,
and the CO2 production rates stabilized, probably reaching a rate
limited by the slow rate of hydrolysis in the Siberian soil microcosms. These
observations are different from the observed CO2 level-off in the
current microcosms (Fig. 2a2, a4, a5).
CH4 in the headspace increased slowly at the beginning and then
accelerated (Fig. 2b1–5), except in the center organic soils. CH4
accumulation lagged behind CO2 for about 10 days in most of the microcosms
and by a few days for the center organic soil microcosms at 4 and
8 ∘C. These lag times are shorter than those observed in microcosms
with deep Siberian permafrost soils (average 960 ± 300 days) (Knoblauch
et al., 2013). This is probably because of the initial abundance of
substrates such as organic acids in the Barrow soils (Fig. 2c1–6). In
addition, the shallow Barrow soils experience freezing and thawing, and so
does microbial activity every year, while the deep Siberian permafrost soils
were frozen for extended periods; as a result, the amount of initial biomass
in the shallow Barrow soils is probably much higher than that in the deep
Siberian soils.
Comparison of observed and modeled CO2 (a1–6) and CH4
(b1–6) in the headspace, organic acid (Ac, c1–6), extractable Fe(II) (d1–6),
and pH (e1–6) in the incubation tests with soils from an Arctic lower-center
polygon. Symbols represent observations with blue, green, and red for -2, 4,
and 8 ∘C. For CO2 and CH4, different symbols of the
same color represent duplicates. The organic acids, such as formate,
acetate, propionate, and butyrate, reported by Herndon et
al. (2015) are combined as Ac in (c1–6). The rest of the data were taken from
Roy Chowdhury et al. (2015). The curves are calculations
based on model parameter values listed in Table 1 and experimental parameter
values listed in Table 2. Trough, ridge, and center denote the
microtopographic locations in the polygon, and mineral and organic denote
soil horizons. Increasing the initial bioavailable Fe(III) fFe
from 0.005 (continuous line) to 0.01 (dashed line) and 0.02 (dash-dotted line) brings the
predictions close to the observations for Fe(II) and pH for center and ridge
organic soils.
![]()
Organic acids generally accumulated at the beginning, decreased as CH4
concentration increased, and exhausted in the mineral soil microcosms
(Fig. 2c1–6). In contrast, organic acids were not exhausted in the center
organic soil microcosms (Fig. 2c6). In comparison with similar tests with
soils from the high-center polygon trough, organic acids accumulated for over
5 months in the organic soils and were not exhausted in the mineral soils
(Yang et al., 2016). The accumulation and disappearance of organic acids have
been widely observed in the literature (van Bodegom and Stams, 1999; Fey et
al., 2004; Glissmann and Conrad, 2002; Jerman et al., 2009; Kotsyurbenko et
al., 1993; Lu et al., 2015; Peters and Conrad, 1996; Yao and Conrad, 1999).
Fe(II) concentrations increased and leveled off (Fig. 2d1–6), with similar
trends for pH (Fig. 2e1–6). The increase in pH concurred with Fe(III)
reduction, which released hydroxides from Fe(OH)3a dissolution. The
pH increase is in contrast to the observed pH decrease when Fe(III) reduction
was absent (Xu et al., 2015). While Fe(III) reduction was reported to inhibit
methanogenesis through direct inhibition (van Bodegom et al., 2004) or
substrate competition (Miller et al., 2015; Reiche et al., 2008), the impact
appears less significant than expected in these incubations, as well as
incubations with the high-center polygon trough soils (Yang et al., 2016).
This is consistent with the observation that methane production initiated in
the presence of oxidants (Roy et al., 1997). In addition, Fe(III) reduction
can both inhibit and promote methanogenesis (Zhuang et al., 2015). In the
Barrow soils, the initial abundance of organic acids probably mitigates the
competition between Fe(III) reducing and methanogenic populations, decreasing
the lag time between CH4 and CO2 accumulation.
Substantial microbial activity was observed at -2 ∘C, which is
above the soil water freezing point due to osmotic and matric potentials.
These incubations led to an increase in CO2 (Fig. 2a1–6), organic acids
(Fig. 2c1–6), Fe(II) (Fig. 2d1–6), and pH (Fig. 2e1–6). CH4
concentrations were low but detectable in the headspace at -2 ∘C.
The lag time between CH4 and CO2 increases with decreasing
temperature, which was widely observed in the literature as well (Fey and
Conrad, 2003; Hoj et al., 2007; Jerman et al., 2009; van Bodegom and
Scholten, 2001; Fey et al., 2004; Kotsyurbenko et al., 1993; Lu et al.,
2015). The transition from -2 to 4 and 8 ∘C appears to be gradual,
except for the center organic soils, where CH4 increases were drastic
from -2 to 4 ∘C (Fig. 2a1 vs. b1). The observed overall
temperature responses are diverse, as manifested by Q10 values from 1.6
to 22 (Roy Chowdhury et al., 2015).
Modeling results
Overall
With the same model parameter values given in Table 1 and Table S2 and
different experimental parameter values listed in Table 2, the model roughly
predicts the observed trends for different soils at the three temperatures
(Fig. 2): CO2 and CH4 accumulate in the headspace; CO2
accumulation slows down, while CH4 speeds up at later times; CH4
lags behind CO2; organic acids accumulate and then decrease; Fe(II)
accumulates and levels off; pH increases and levels off; and carbon
mineralization and methanogenesis rates increase with temperature.
While the model predicts little CO2 and CH4 in the headspace at -2 ∘C, which is similar to what was observed, it predicts little
change in Fe(II) and pH as well, which is not consistent with the
observations. To improve the prediction at -2 ∘C, which can be
important (Zona et al., 2016; Xu et al., 2016a), it is necessary to
understand why little CO2 or CH4 was observed to occur with Fe(III)
reduction, which was indicated by the increase in Fe(II) and pH.
The same model parameter values describe the observed differences in the
mineral soils better than in the organic soils. For the mineral soils, the
model overpredicts the increasing trend for CO2 in the headspace at late
times because the observations leveled off (Fig. 2a1–3). The initial rapid
CO2 increases lasted for over 2 months in the 3-year incubations with
Siberian permafrost soils under 4 ∘C and anaerobic conditions
(Knoblauch et al., 2013). In these long-term tests, CO2 increased
rapidly at the beginning and the rate stabilized as the carbon release became
limited likely by hydrolysis of polymers. The observed sustained CO2
accumulation in these closed microcosms indicates that the observed trends in
Fig. 2a1–6 at later times are probably uncertain. Except for these
mismatches, the model predictions generally agree with the observations for
the mineral soils reasonably well.
In contrast, the predictions do not agree as well with the observations for
the organic soils. For the trough organic soils, the model underpredicts
CO2 in the headspace (Fig. 2a4) but describes the rest of the
observations reasonably well. In addition to CO2 (Fig. 2a5), the model
underpredicts Fe(II) and pH increase in the ridge organic soils (Fig. 2d5,
e5). The prediction of the center organic soils differs from the observations
the most (last column in Fig. 2). These mismatches might be explained by
model biases in initial Fe(III) content, labile DOC, and biomasses.
Fe(III) reduction
Agreement between predictions and observations for the Fe(II) and pH increase
can be improved for the ridge and center organic soils by increasing the
Fe(III) content from fFe3=0.005 to 0.01 and 0.02 (Fig. 2d5–6,
e5–6). This also increases the predicted CO2 and CH4 for the
center organic soils (Fig. 2a6, b6) because of the predicted pH increase
(Fig. 2e6), which increases the reaction rates as the pH response function
increases when the calculated pH increases toward an optimal pH of 6.2 in
Eq. (3). For the ridge organic soils, fFe3= 0.01 increases the
predicted CH4 like the center organic soils, but fFe3= 0.02
decreases CH4 prediction because of the competition between methanogens
and iron reducers and limited availability of substrates (Fig. 2b5). This
provides an explanation as to why Fe(III) reduction can both suppress and
promote methanogenesis (rather than strict thermodynamic control, e.g.,
Bethke et al., 2011; direct inhibition, e.g., van Bodegom et al., 2004; or
indirect inhibition through substrate competition, e.g., Mill et al., 2015;
Reiche et al., 2008).
As the bioavailable Fe(III) in the organic soils is reported to range from
0.2 to 1 % of dry soil mass (Lipson et al., 2013a), the short-term tests
are not expected to be Fe(III)-limited for the mineral soils. Increasing
bioavailable Fe(III) makes the model overpredict Fe(II) and pH increases at
later times for the mineral soils (Fig. 2d1–5, e1–4), and Fe(III) reduction
and methanogenesis at later times are predicted to be limited by organic
substrate availability at 4 and 8 ∘C (Fig. 2b1–4). The latter is
consistent with the observed very low organic acid concentrations at the end
(Fig. 2c1–5). As a result, the model underpredicts CH4 accumulation,
indicating the current parameterizations, in particular the half-saturation
and growth rate constants, may overpredict the ability of iron-reducing
bacteria to outcompete methanogens.
CO<inline-formula><mml:math display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> distribution among gas, aqueous, and adsorbed phases
While increasing Fe(III) slightly increases the predicted CO2 for ridge
mineral soils (Fig. 2a2), it decreases the predicted CO2 in the
headspace for trough and center mineral soils (Fig. 2a1 and a3). This is
because CO2 solubility is predicted to increase significantly as pH
increases, resulting in the dissolution of CO2 from the headspace into
the aqueous phase (Fig. S1 in the Supplement). To examine this impact, we
conduct numerical simulations with a 45 mL headspace with an initial 1 atm
N2 gas and 10 mL solution with 10 mM total inorganic carbon at various
temperature and pH values. CO2(g) and CO2(aq) or carbonic acid
dominate at a pH lower than 5 (Fig. 3). As the pH increases above the
carbonic acid pKa (around 6.3 under standard conditions), CO2(g) in the
headspace and CO2(aq) decrease as HCO3- becomes dominant
in the aqueous phase, and the gas-phase fraction decreases dramatically. The
gas-phase fraction also decreases with decreasing temperatures (Fig. 3).
Partition of CO2 among gas- and aqueous-phase species under
various temperatures. The calculations are conducted with 45 mL of headspace
with N2 and 10 mL of solution with 10 mM total inorganic carbon using
PHREEQC. Gas phase dominates at lower pH and high temperature. As pH
increases, the gas-phase CO2 fraction is very low after pH 7, implying
potential underestimation of carbon mineralization based on headspace
CO2 concentration measurement only.
![]()
Increasing initial labile DOC better describes the observed initial
rapid CO2 increase in the headspace for the organic soils. See Fig. 2
caption for more information.
![]()
In addition, CO2 was reported to adsorb to surface sites (Appelo et al.,
2002; van Geen et al., 1994; Villalobos and Leckie, 2000). With the surface
complexation reactions between Fe(OH)3a and carbonate species, we
add 1 mmol Fe(OH)3a (about the mean values in Fig. 2 for the case
fFe3= 0.02) to the abovementioned numerical experiments. The
calculations show that the adsorption phase can dominate at low pH (Fig. S2),
with the total amount dependent on the abundance of surface sites. For the
high-temperature high-Fe(III) initial content cases in Fig. 2, adding
CO2 sorption reactions provides a substantial buffer against the early
increase in CO2 in the headspace (Fig. S3). As the Fe(OH)3a is
reduced and dissolved, the adsorbed CO2 is predicted to be released,
contributing to an increase in headspace CO2 increase later on.
In addition to pressure, these calculations suggest the need to
appropriately account for pH and its impact on the gas, aqueous, and
adsorbed phases CO2 partition when we use headspace concentration
measurements from anaerobic incubations to estimate CO2 emission.
Otherwise, substantial uncertainties can be introduced. A geochemical model
with accurate thermodynamic data and accounting for CO2 sorption can be
useful in accurately quantifying CO2 production in these closed
microcosms.
Initial CO<inline-formula><mml:math display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> accumulation in the organic soil microcosms
The model underpredicts the early CO2 increase in the headspace for the
organic soil microcosms (Fig. 2a4–6), which is mostly apparent in the center
organic soil microcosms. The reason is that the organic soil microcosms
contain more labile organic carbon than the mineral soil microcosms, as
evidenced by water-extractable organic carbon (Table 2). In particular, the
center organic soil microcosms contain about half total organic carbon of the
other microcosms, double the water volume, and 3 to 5 times water-extractable carbon (Table 2). As a result, it produces the most CO2 and
CH4 and has a very short lag time between CH4 and CO2. If we
increase the initial labile DOC content fLabileDOC from 0.0005, as
shown in Fig. 2, to 0.01, and 0.02 for the organic soil microcosms, the
underprediction of the early CO2 increase in the headspace is more or
less mitigated (Fig. 4).
The predicted rapid initial CO2 increase is due to the fast fermentation
reactions (Fig. S4a1–6, e1–6). The predicted steep transition in CO2
concentration increases appears reasonable for the center and trough soil
microcosms, but less so for the ridge soil microcosms. In addition to the
20 h and 14-day turnover time differences, fermentation reactions decrease
the pH, and further inhibit the predicted SOM1 decomposition reactions, Fe(III)
reduction, and methanogenesis, making the predicted transition steeper. The
fast fermentation is consistent with the observed rapid disappearance of
glucose and increase in CO2 after glucose addition in similar
experiments with soils from a high-center polygon trough from the same site
(Yang et al., 2016). However, the observed decrease in natural free reducing
sugars was gradual, with about one-third of the original reducing sugars left
over after 150 days of incubations. Along with the predicted rapid initial
labile DOC decrease and CO2 increase, the model predicts a rapid initial
increase in organic acids, which is close to the observations for the center
soil microcosms but much greater than the observations for the trough and
ridge soil microcosms. The latter indicates that the ratio of organic acids
to CO2 of 2:1 from the fermentation Reaction (R1) may not be
accurately representative of the experiments.
Detailed measurements showed a rapid initial increase and then a quick
decrease in organic acids in the mineral soil microcosms and a gradual
increase and slow decrease in the organic soil microcosms from a trough
location in a high-center polygon in the first 144 days of anaerobic incubation
mineral and organic soil microcosms for ethanol, and were generally more
gradual for organic acids than for ethanol (Yang et al., 2016). To explain
the various observations for the organic soil microcosms and for accurate
predictions, the diversity of the hydrolysis products (Feng and Simpson,
2008) and the subsequent pathways (Tveit et al., 2015) may need to be
accounted for. Additional detailed data are needed to support increasingly
mechanistic models, e.g., with reducing sugars to represent less rapid
fermentation, and additional specific organic acids such as propionate and
butyrate to better describe diverse observations in the incubations.
Carbon mineralization
Less than 1 % of the total initial carbon turned over to CO2 and
CH4 in about 2 months, which is attributed mostly to decomposition of
labile SOM (SOM1), labile DOC, and organic acids (Fig. S4). Few changes are
predicted in the slow pools (SOM3, and SOM4, not shown) even though they
comprise a large portion of the soil carbon pool. The small amount of
respired carbon is similar to the incubation tests conducted with Siberian
permafrost soils under 4 ∘C, which was estimated to be 3.1 % and
0.55 % under aerobic and anaerobic conditions for 1200 d (Knoblauch et
al., 2013), the 1-year aerobic incubation tests (Feng and Simpson, 2008),
and the incubations from a wide range of Arctic soils (Schädel et al.,
2014). All of these results suggest that the hydrolysis of macromolecular
organics by extracellular enzymes could be a rate-limiting step at late
times. To predict the long-term vulnerability of the organic carbons, it is
important to understand and describe the hydrolysis of macromolecular
components in SOM.
CH<inline-formula><mml:math display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">4</mml:mn></mml:msub></mml:math></inline-formula> accumulation
Besides Fe(III) reduction, the predicted CH4 production is dependent on
the substrate production. With slabile= 0.2, the model generally
predicts less CH4 and more CO2 than the case with
slabile= 0.4 because less SOM is assumed to respire through the
anaerobic pathway in the slabile= 0.2 case (Fig. S5). With
increased slabile= 0.6, the model predicts more CH4 and less
CO2. The impact on the mineral soils is generally more pronounced than
the organic soils because the former is more substrate limiting than the
latter. Unlike CO2, CH4 solubility and adsorption are much lower.
Gas-phase CH4 in the headspace dominates over aqueous and adsorbed
phases. The model predicts the general exponential increase trend with a lag
time behind CO2 (Fig. 2). However, the prediction is sensitive to
Fe(III) reduction, pH, temperature (Fig. 2), and labile substrates (Fig. 4).
The model substantially underpredicts early fast CH4 production for the
center organic soil microcosms (Fig. 4b3). While the cell count for the
center organic soils is not available for day 0, the data did show that the
center organic soils had the highest amount of biomass after 100-day
incubations (Roy Chowdhury et al., 2015). The disagreement between the
predictions and the observations can be mitigated by increasing the initial
biomass fbio from 10-6 to 10-5 and 2 × 10-5
for the center organic soil microcosms (Fig. 5). With increased initial
biomass, Fe(III) reduction and methanogenesis are predicted to speed up the
recovery of the initial pH drop caused by organic acid accumulation so that
the model predicts a fast CH4 increase that is comparable to the
observed increase. However, the model overpredicts the CH4 increase at late times,
indicating alternative inhibition mechanisms rather than substrate limitation
on methanogenesis at late times or CH4 consumption such as
anaerobic oxidation (Caldwell et al., 2008; Smemo and Yavitt, 2011).
Increasing the initial biomass predicts rapid CH4
accumulation at early times that is close to the observations but misses the
level-off trend at late times for the center organic soils. See Fig. 2
caption for more information.
![]()
pH
With the complexation reactions involving proton or hydroxide anion with
carbonate species, ferrihydrite surface, and SOM, the geochemical model
describes the observed pH evolution reasonably well (Fig. 2). The initial pH
was lower in the mineral soils than in the organic soils (Fig. 2), probably
because of less buffering capacity due to less organic matter in the mineral
soils and/or more reducing condition in the organic soils as reduction
reactions typically consume protons. Because the ridge mineral soils have the
lowest initial pH, the CLM4Me pH factor is the lowest (Table S1),
contributing to the underprediction of CH4 (Fig. 2b2). With high organic
content, the organic matter dominates the aqueous geochemistry, and the
predicted pH is sensitive to the surface sites specified for WHAM. If the
specified WHAM organic matter is reduced by 25 %, then the pH buffering
capacity is decreased and the predicted pH increases substantially
(Fig. S6e1–6) even though the predicted changes in organic acids and Fe(II)
are small. For the trough soils, the predicted pH surpasses the optimal of
6.2, and f(pH) (Eq. 3) decreases (Fig. S6e1, e4). As a result, predicted
CO2 and CH4 are decreased. The pH impact becomes complex around the
optimal pH. If we increase the specified WHAM organic matter by 25 %, the
predicted pH is lower due to larger pH buffering and the reaction rates are
generally smaller. Setting the WHAM sites at measured total organic carbon
works reasonably well for the experiments with the CLM4Me pH response
function.
Comparison of pH response functions used in CLM4Me
(Riley et al., 2011), TEM (Raich et al., 1991), and
DLEM (Tian et al., 2010) as described by Eq. (3), (A1–3).
Reaction rates are sensitive to pH and pH response functions vary
substantially, introducing prediction uncertainty.
![]()
Comparison of temperature response functions used in (a) land surface models CLM-CN (Thornton and Rosenbloom, 2005),
CENTURY (Parton et al., 2010), (b) Q10 (Oleson et al., 2013), (c) Ratkowsky
equation (Ratkowsky et al., 1982), and (d) Arrhenius equation (Wang et al., 2013) described by Eqs. (4,
B1–B4). Reaction rates are sensitive to temperature and temperature response
functions vary substantially, introducing prediction uncertainty.
![]()
Comparing the CLM4Me pH response function with these used in TEM and DLEM,
all three response functions show that the reaction rates are sensitive to pH
(Fig. 6), which is expected to influence the predictions for these incubation
tests as the pH increases from about 5.5 to 7. In this range, CLM4Me and DLEM
have a similar slope, but the latter has a greater rate reduction effect.
While CLM4Me and TEM have a similar rate reduction effect, CLM4Me has a
steeper curve than TEM. These differences translate to substantial
differences in model predictions (Fig. S7). All calculated f(pH) values
increase during the tests (Fig. S7f1–f6). As the f(pH) calculated by DLEM
is the lowest, the predicted changes are the smallest. The f(pH) calculated
by TEM is slightly greater than CLM4Me at the beginning and is the opposite
at late times (Fig. 6). As a result, TEM generally predicts slightly faster
evolution than CLM4Me as the reaction rates at the late times are limited by
substrates rather than pH. While the pH ranges from 3.3 to 8.6 in the Arctic
soils (Schädel et al., 2014), the range and the variability in the data
are limited in the evaluation of these pH response functions. Nevertheless,
model predictions are sensitive to pH response functions; the microbes are
likely adapted to the site pH conditions such that the response functions are
expected to vary among sites and functional groups. Therefore, pH response
function can be an important source of prediction uncertainty.
Temperature response
Temperature effects on reactions between inorganic aqueous species, and the
aqueous and gas species, are taken into account in the established reaction
database. The temperature impact on surface complexation reactions with
ferric hydrous oxides, and with SOM in WHAM, is not quantified, which can be a
potential source of uncertainty. LSMs generally use empirical (e.g., CLM-CN,
CENTURY), Q10, or Arrhenius equations. The CLM-CN temperature
response function is compared with the CENTURY, Q10 equation, Arrhenius
equation, and Ratkowsky equation in Figs. 7 and S8. All of these temperature
response functions describe increasing rate with increasing temperature. When
the temperature response functions f(T) are plotted in arithmetical
scale, the shapes are similar except for CENTURY, which approaches 1 when the
temperature increases above 20 ∘C; CLM-CN is close to Q10 with
Q10= 2.5, the Arrhenius equation with Ea=60 kJ mol-1 and the Ratkowsky equation with Tm=260 K. When f(T) is plotted in log scale (Fig. 7), Q10 and
Arrhenius equations are approximately linear, while the rest have a similar
shape; CLM-CN appears close to the Ratkowsky equation with Tm=260 K. At our temperatures -2, 4, and 8 ∘C, CLM-CN is very close
to CENTURY, Q10= 2.5, Ea= 60 kJ mol-1, and
Tm= 260 K (Figs. 7, S8). Despite their consistency, the
predictions can be different for the different response functions (Figs. S9,
S10), reflecting the sensitivity of the temperature effect on the
biogeochemical reaction rates. The difference is amplified when different
Q10, Ea, or Tm is used (not shown),
introducing potentially large uncertainty in model predictions. Because the
temperature response functions are expected to vary for different
microorganisms, extracellular vs. intracellular enzymes, and geochemical
reactions in the soil environment, improved quantification is needed.
Predicted impact of headspace gas accumulation
The accumulation of gases in the headspace may impact the soil carbon
mineralization and methanogenesis. Knoblauch et al. (2013) and Yang et
al. (2016) flushed the headspace of the microcosms, while Roy Chowdhury et
al. (2015) and Herndon et al. (2015) did not. The field conditions are likely
somewhere between an open system and a closed system because neither the
atmospheric pressure nor the hydrostatic pressure is constant, and the
produced CO2 and CH4 are not always free to release to the
atmosphere. To assess the impact of CO2 accumulation in the headspace on
the soil carbon mineralization and methanogenesis, we conduct numerical
experiments with 10 and 100 times the headspace volume of the experimental
values. With increased headspace volume, the headspace and aqueous CO2
concentrations are predicted to decrease (Fig. S11f1–6, g1–6), and the pH
increase is predicted to slow down. As a result, the biogeochemical reaction
rates are generally slower (Fig. S11e1–6). Eventually, the predicted total
CO2 and CH4 production generally decrease with lower headspace
CO2 concentration (Fig. S11a1–6, nb1–6). However, the impact on
CO2 production is very small for the organic soils in the trough and
ridge location, and the CO2 production is predicted to increase with
decrease in headspace CO2 concentration for the organic center soils.
Because of the complicated nonlinear relationships in the biogeochemical
processes, the impact of headspace gas accumulation on carbon mineralization
and methanogenesis is not linear. While it is debatable which
experimental conditions (flush the headspace or not) reflect the field
conditions, biogeochemical models like ours provide a mechanistic method to
account for this impact by using boundary conditions that reflect the
reality. Additional targeted experiments and mechanistic models are necessary
to better understand the impact under different conditions, and develop
representations that reflect field conditions.
Summary and conclusion
Soil organic carbon turnover and CO2 and CH4 production are
sensitive to redox potential and pH. However, land surface models typically
do not explicitly simulate the redox or pH, particularly in the aqueous
phase, introducing uncertainty in greenhouse gas predictions. To account for
the impact of availability of electron acceptors other than O2 on soil
organic matter (SOM) decomposition and methanogenesis, we extend the CLM-CN
decomposition cascade to link complex polymers with simple substrates and
add Fe(III) reduction and methanogenesis reactions. Because pH was observed
to change substantially in the laboratory incubation tests and in the field
and is a sensitive environmental variable for biogeochemical processes, we
use the Windermere Humic Aqueous Model (WHAM) to simulate pH buffering by
SOM. To account for the speciation of CO2 among gas, aqueous, and solid
(adsorbed) phases under varying pH, temperature, and pressure values, as well
as the impact on typically measured headspace concentration, we use a
geochemical model and an established reaction database to describe
observations in recent anaerobic microcosms. Our results demonstrate the
efficacy of using geochemical models to mechanistically represent the soil
biogeochemical processes for Earth system models.
Together with the speciation reactions from the established geochemical
database and surface complexation reactions for ferric hydrous oxides, WHAM
enables us to approximately buffer an initial pH drop due to organic acid
accumulation caused by fermentation and then a pH increase due to Fe(III)
reduction and methanogenesis. The single input parameter for WHAM is total
organic carbon content, which is available in any SOM decomposition model.
Therefore, adding WHAM does not necessitate any additional characterization.
However, the temperature effects on surface complexation reactions with
ferric hydrous oxides and organic matter may need to be further quantified.
The equilibrium geochemical speciation reactions predict a substantial
increase in CO2 solubility as the pH increases above 6.3 because the
aqueous dominant species shifts from CO2 to HCO3-. Adding
CO2 adsorption to surface sites of metal oxides further increases
predicted solubility at low pH. Without taking
speciation, pH, and the temperature and pressure impact
into consideration, the carbon mineralization rate can be substantially underestimated
from anaerobic microcosms based on headspace CO2 measurements.
Because various microbes respond to the temperature and pH change
differently, it is challenging to describe observed diverse responses with
any single one of the existing response functions. As the microbes adapt to
the low temperature and pH conditions in the Arctic, the optimal growth
temperature and pH value in these response functions may need to be adjusted
to account for biological acclimation.
We demonstrate that a geochemical model can mechanistically predict pH
evolution and accounts for the impact of pH on biogeochemical reactions,
which enhances our understanding of and ability to quantify the experimental
observations. Because pH is an important environmental variable in the
ecosystems and land surface models either specify a fixed pH or use simple
empirical equations, a geochemical model has the potential to improve model
predictability for greenhouse emissions by mechanistically representing the
soil biogeochemical processes.
Another follow-up task could be assessing this new framework of anaerobic
SOM decomposition in field studies with CLM-PFLOTRAN. This can be done
incrementally, i.e., adding/removing reactions one at a time without source code
modifications. CLM-PFLOTRAN currently uses the CLM4.5 vertically resolved grid.
The resolution can be adjusted, possibly in three dimensions, to reflect the
heterogeneity of any structural soil column to account for the limitation of
electron donors and electron acceptors at individual locations. As we
gradually implement more and more processes, such as gas and aqueous
transport through soils and aerenchyma, explicitly representing microbial
processes for carbon decomposition, we hope the new framework will be useful
for future investigation and model developments.