Surface water inundation in the boreal-Arctic: potential impacts on regional methane emissions

The land area considered in this analysis was determined using boreal-Arctic peatland maps (i.e. Gunnarsson and Löfroth 2009, Yu et al 2010, Franzén et al 2012), and the Regional Carbon Cycle and Assessment Processes (RECCAP) tundra domain (McGuire et al 2012). To coincide with the spatial extent of AMSR-E Fw coverage, we removed grid cells having ⩾50% permanent ice or open water cover using the UMD Moderate Resolution Imaging Spectroradiometer (MODIS) land cover product (described in Jones et al 2010), and applied a conservative 25 km coastline buffer to minimize Fw retrieval contamination by ocean water. The resulting study region spans approximately 1.6 × 10⁷ km² (figure 1), and contains 70% of northern continuous and discontinuous permafrost affected landscapes (Brown et al 1998).

Zoom In Zoom Out Reset image size

The JULES model approach (Clark et al 2011, Bartsch et al 2012) accounts for the major factors (i.e. temperature, carbon substrate availability, landscape wetness) that control global methane emissions (Bloom et al 2010, Olefeldt et al 2013). Albeit relatively simple and lacking in detailed physical processes, this method is useful for pan-Arctic simulations because it avoids extensive parameterization requirements that can substantially increase estimate uncertainty (Riley et al 2011). The model regulates methane emissions according to available carbon substrate (C, kg m⁻²) and an efflux rate constant (${{k}_{{\rm C}{{{\rm H}}_{4}}}}$, d⁻¹) that is modified by a temperature dependent Q₁₀ factor (Gedney et al 2004, Clark et al 2011). The temperature effects on methane production are controlled using daily input Modern Era Retrospective-analysis for Research and Applications (MERRA) surface soil temperature (T_s, in kelvin) and a thermal reference state (T₀, 273.15 K):

$$\begin{eqnarray}&&{{F}_{{\rm C}{{{\rm H}}_{4}}}}=C\times {{k}_{{\rm C}{{{\rm H}}_{4}}}}\times {{Q}_{10}}^{\left( {{T}_{s}}-{{T}_{0}} \right)/10}\times FThw.\end{eqnarray} \tag{ 1 }$$

For this analysis, we limit our investigation to non-frozen surface conditions defined using daily satellite passive microwave sensor derived binary (0 or 1) freeze/thaw (FThw) constraints (Kim et al 2013). The resulting daily fluxes (${{{\rm F}}_{{\rm C}{{{\rm H}}_{4}}}}$, CH₄ m⁻²) were averaged over a 15 day time step and scaled to the 25 km grid cell domain (tonne CH₄ cell⁻¹) using Fw and volumetric soil moisture (θ, m³/m³) information to regulate landscape methane emissions. Methane efflux from inundated portions of the grid-cell were assumed to be non-inhibited, whereas non-inundated cell fractions $\left( 1-Fw \right)$ were weighted by θ to account for reduced methane loss due to oxidation. In this study, the emissions weighting process was applied at a 15 day time step to address potential delays in methanogen response following surface wetting or drying (Blodau and Moore 2003, Turetsky et al 2014).

The daily input T_s and θ (⩽10 cm soil depth) records were obtained from the NASA GEOS-5 MERRA Land reanalysis archive with native 0.5° × 0.6° resolution (Reichle et al 2011) and posted to a 25 km resolution polar equal-area scalable earth grid consistent with the AMSR-E Fw data. The MERRA land parameters have been evaluated for high latitude regions, with favorable correspondence in relation to independent satellite microwave and in situ observations (Yi et al 2011, Watts et al 2014). Soil metabolic carbon (C_met) pools obtained from a Terrestrial Carbon Flux (TCF) model (Kimball et al 2009, Yi et al 2013) were used as the substrate for methanogenesis. The TCF carbon estimates reflect daily changes in labile plant residues and root exudates, and have been evaluated against existing soil organic carbon inventory records for the high latitude regions (described in Yi et al 2013). The C_met inputs (kg C m⁻² d⁻¹) were generated for the study region by a 1000 yr spin-up of the model using a 10 yr (2000–09) record of MODIS 1 km resolution Normalized Difference Vegetation Index, MERRA daily surface meteorology and soil moisture inputs.

The JULES model ${{k}_{{\rm C}{{{\rm H}}_{4}}}}$ and Q₁₀ parameters were calibrated using mean monthly eddy covariance methane fluxes (mg CH₄ m⁻¹ d⁻¹) from five northern wetland tower sites (figure 1) that are described in the published literature (i.e. Rinne et al 2007, Sachs et al 2008, Wille et al 2008, Zona et al 2009, Long et al 2010, Parmentier et al 2011), in conjunction with mean MERRA reanalysis C_met and T_s climatology over the 2003–11 summer (May through September) period. A resulting Q₁₀ value of 3.7 and a ${{k}_{{\rm C}{{{\rm H}}_{4}}}}$ rate of 3.7 × 10⁻⁵ d⁻¹ minimized the root mean square error (RMSE) differences between the model and flux tower observations at 17.62 mg CH₄ m⁻² d⁻¹. A Q₁₀ of 3.7 was also used by Clark et al (2011) and is similar to those reported in other studies (Ringeval et al 2010, Waldrop et al 2010, Lupascu et al 2012). Further model verification was also obtained by evaluating summer flux chamber measurements (see supplementary table S1) from tundra (n = 15 site records), boreal wetland (n = 11) and lake (n = 17) locations.

Grid-scale (25 km) wetland methane emissions were obtained using dynamic 15 day, monthly and annual summer AMSR-E Fw means or maximums from 2003 to 2011, to examine the potential impact of temporal Fw scaling on methane emission estimates. Methane simulations were also examined using a static mean summer Fw map derived from the 2003–11 record. The regional simulations were evaluated against NOAA ESRL atmospheric methane flask measurements (Dlugokencky et al 2013) from Barrow, AK, Lac LaBiche, CN, and Pallas Sammaltunturi, FI, to assess the ability of the model to capture between-year changes in methane concentrations that may correspond with fluctuations in wetland methane emissions (Lelieveld et al 1998). For Barrow and Sammaltunturi, the dry air mole fractions were available from 2003 through 2011; the Lac LaBiche data were available from  2008 onward. A Hybrid Single Particle Lagrangian Integrated Trajectory (HYSPLIT; Draxler and Rolph 2013, Rolph 2013) model, with a 100 m receptor point altitude and input GDAS-1 meteorology (Rodell et al 2004), was used to obtain backward (30 day) atmospheric trajectories for each flask site, and showed the dominant source contributions at Barrow to originate primarily from northern Alaska, the Yukon River basin eastward to the Northwest Territories, and eastern Siberia. For the respective Lac LaBiche and Sammaltunturi locations, the major source regions were from northern Canada, or extending from Scandinavia eastward into western Russia. To determine the relative correspondence between modeled annual methane emission contributions and observed mean summer dry air mole fractions, Pearson product-moment correlation coefficients (r) were derived using spatial means from a 3 × 3 grid cell window centered on each flask location. Regional point correlation maps (Ding and Wang 2005) were also obtained by evaluating r (e_j , a_k ) for each grid cell within the methane source regions, where e_j is the modeled mean summer emissions time series at a given cell location and a_k is the atmospheric methane concentration time series at a flask sampling site.

Regional changes in surface water coverage, soil moisture and temperature were evaluated using a non-parametric Mann-Kendall trend analysis that accounts for serial correlation prior to determining trend significance (Yue et al 2002, Watts et al 2012). The Kendall rank correlations were applied to the mean summer AMSR-E Fw, and MERRA T_s and θ records on a per-grid cell basis from 2003 to 2011. Trend significance was determined at a minimum 95% (p < 0.05) probability level. The Kendall trend was also applied to the modeled cumulative annual methane emissions to identify regions that may be vulnerable to increasing anaerobic carbon losses. A linear regression analysis was then used to determine the rate of change in the annual emission estimates.

The model simulations captured overall temporal variability (r² = 0.65, p < 0.05) observed in the monthly tower eddy covariance records, with a RMSE value of 17.6 mg CH₄ m⁻² d⁻¹ that is similar to other regional studies (Meng et al 2012, Zhu et al 2013). Significant differences (α = 0.05; two-sample t-test with unequal variance) were not observed (figure S1) between the model estimates and mean monthly tower eddy covariance (t = 1.45, p = 0.15), boreal chamber (t = 0.05, p = 0.96), and northern lake (t = 0.79, p = 0.45) fluxes. However, the modeled fluxes were significantly smaller (t = 3.67, p < 0.01) than the tundra chamber observations and did not adequately capture larger (>140 mg CH₄ m⁻² d⁻¹) eddy covariance fluxes from a peatland site in northern Sweden (Jackowicz-Korczyński et al 2010). These discrepancies may reflect the presence of tall sedges (e.g. E. angustifolium), which can substantially increase emission rates through aerenchymateous tissue pathways (Joabsson et al 1999), or the limited representation of landscape scale emissions by chamber measurements given the potentially large contrasts in methane fluxes from dry and wet vegetation communities (Parmentier et al 2011) and functional groups (Kao-Kniffin et al 2010). The modeled methane fluxes were within the 5–140 mg CH₄ m⁻² d⁻¹ range observed in the lake measurements (Zimov et al 1997, Laurion et al 2010, Desyatkin et al 2009, Sabrekov et al 2012), although these observations primarily reflect diffusive gas release and background bubbling instead of episodic ebullition events. As a result, the model simulations may underestimate ebullition release from open water bodies, particularly in carbon-rich thermokarst regions characterized by methane seeps (Walter et al 2006). However, the fraction of lake bodies exhibiting this seep behavior is not well quantified, and a recent analysis of sub-Arctic lakes reported that summer ebullition events averaged only 13 mg CH₄ m⁻² d⁻¹, with a low probability of bubble fluxes exceeding 200 mg m⁻² d⁻¹ (Wik et al 2013).

3.2.1. Wetland inundation characteristics

Approximately 5% (8.4 × 10⁵ km² ± 6% SE) of the boreal-Arctic domain was inundated with surface water during the non-frozen summer season, as indicated by the 2003–11 AMSR-E Fw retrieval means. Over 63% of the wetlands were located in North America, primarily within the Canadian Shield region, and the majority of inundation occurred above 59° N within major wetland complexes, including the Ob-Yenisei and Kolyma Lowlands in Siberia (figure 2). A strong seasonal pattern in surface water was observed across the high latitudes, with an abrupt increase in May or early June following surface ice and snow melt, and the onset of spring precipitation (figure 3). In Eurasia, peak inundation occurred in June, followed by a gradual decline with summer drought and increased evaporative demand (Rawlins et al 2009, Schroeder et al 2010, Bartsch et al 2012, Watts et al 2012). In North America, the seasonal expansion of surface water continued through July, before beginning to subside with the onset of surface freezing.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The influence of wet/dry cycles on surface water extent was evident throughout the boreal-Arctic region. The summer of 2004 was the driest observed over the AMSR-E Fw record, with a 6% decrease in inundation from the long-term mean that coincided with drought conditions across the Arctic Basin and Alaska (Rinsland et al 2007, Zhang et al 2008, Jones et al 2013). Summer 2008 was the wettest in North America, with a 3% increase in Fw coverage that was also reflected in positive drainage anomalies observed in the Mackenzie River basin (Watts et al 2012). The wettest summer in Eurasia occurred in 2007, particularly within the Ob, Lena and Kolyma drainage basins, with a 7% increase in surface water that coincided with regionally high summer temperatures (Dlugokencky et al 2009), snow melt and summer precipitation, and record river discharge (Rawlins et al 2009, Zhang et al 2013).

3.2.2. Regional summer methane simulations

Summer methane emissions estimated for non-inundated land fractions averaged 47.8 ± 1.8 Tg CH₄ yr⁻¹ over the northern wetlands. This increased to 53.2 ± 1.9 Tg CH₄ yr⁻¹ when also considering contributions from inundated landscapes based on the 15 day AMSR-E Fw means. These results are within the range of emissions (39 to 89 Tg CH₄ yr⁻¹) reported from previous modeling studies using other satellite-based Fw retrievals (table 1; Petrescu et al 2010, Ringeval et al 2010, Riley et al 2011, Spahni et al 2011, Wania et al 2013), but are higher than those from atmospheric inversion analyses of northern peatlands (approximately 16–30 Tg CH₄ yr⁻¹; Spahni et al 2011, Bruhwiler et al 2014). The coarse resolution (0.5° × 0.6°) reanalysis meteorology used in the model simulations do not well represent sub-grid variability in soil wetness and temperature controls (von Fischer et al 2010, Sachs et al 2010, Sturtevant and Oechel 2013), which may lead to systematic biases when evaluating methane emissions at larger scales (Bohn and Lettenmaier 2010). However, we recognize that top-down inversion analyses are also prone to uncertainties from atmospheric transport conditions and the limited number of observation sites within high latitude regions (Berchet et al 2013, Nisbet et al 2014).

Table 1.  Wetland methane (CH₄) emissions and associated surface inundation extent determined by regional modeling studies using satellite microwave based surface water (Fw) retrievals to define the spatial extent of methane producing area. The Fw inputs include those scaled using 15 day, monthly and annual Fw means and maximums, or a static multi-summer Fw mean climatology. The methane emissions determined in this study are reported for inundated and combined inundated/non-inundated wetland landscape fractions.

Study	Model	Domain	Fw source	Fw period	Fw scaling	Fw area (km2)	Simulation period (CH4)	Emissions (Tg ${\rm C}{{{\rm H}}_{4}}{\rm y}{{{\rm r}}^{-1}})$ ± Std. Dev.
Petrescu et al (2010)	PEATLAND-VU	55°–70° N	Prigent et al (2007)	1993–2000	Monthly Clim. (avg.)	1.6 × 106	2001–2006
					Adjusted area	4.4 × 106		89
Ringeval et al (2010)	ORCHIDEE	>50° N	Prigent et al (2007)	1993–2000	Month avg.	—	1993–2000	41
Riley et al (2011)	CLM4Me	45°–70° N	Prigent et al (2007)	1993–2000	Month avg.	2 to 3 × 106	1995–1999	70
Spahni et al (2011)	LPJ-WHyMe	45°–90° N	Prigent et al (2007)	1993–2000	Month avg.	2.1 × 106	2004	38.5–51.1
Wania et al (2013)	LPJ-WHyMe	>45° N	Prigent et al (2007), Papa et al (2010)	1993–2004	Annual clim. (avg.)	—	1993–2004	40
Wania et al (2013), Melton et al (2013)	LPJ-Bern	35°–90° N	Prigent et al (2007), Papa et al (2010)	1993–2004	Monthly clim. (avg.)	—	2004	81
This study (all areas)	JULES-TCF	45°–80° N	Jones et al (2010), Watts et al (2012)	2003–2011	15 day avg.	8.4 × 105	2003–2011	53.2 ± 1.9
					15 day avg.	—		5.4 ± 0.3
					15 day max.	1.1 × 106		7.5 ± 0.3
					Month avg.	8.9 × 105		5.5 ± 0.3
This study (inundated only)	JULES-TCF	45°–80° N	Jones et al (2010), Watts et al (2012)	2003–2011	Month max.	1.3 × 106	2003–2011	8.4 ± 0.3
					Annual avg.	9.7 × 105		5.8 ± 0.2
					Annual max.	1.6 × 106		10.8 ± 0.4
					Annual clim. (avg.)	1.5 × 106		5.9 ± 0.3

In northern wetlands, 80–98% of annual methane emissions occur during the summer (Alm et al 1999, Jackowicz-Korczyński et al 2010, Song et al 2012) due to strong thermal controls on methane production, carbon substrate and water availability (Ström et al 2003, Christensen et al 2003, Wagner et al 2009). The influence of summer warming on regional methane emissions was apparent in the model simulations, with peak efflux occurring in June and July (figure S2). This seasonal pattern has been observed in atmospheric methane mixing ratios across the Arctic (Aalto et al 2007, Fisher et al 2011, Pickett-Heaps et al 2011). Also evident was the impact of wet/dry cycles on regional methane contributions, with annual summer emission budgets fluctuating by ±4%, relative to the 2003–11 mean. The modeled emissions were lowest in 2004 despite anomalously high temperatures throughout the boreal-Arctic region (Chapin et al 2005), due to drought conditions in Alaska and northern Canada. In contrast, higher emissions in 2005 resulted from warm and wet weather in North America.

Surface moisture variability also influenced the correspondence between modeled emissions and summer atmosphere methane concentrations from the regional flask measurements. Regions showing a positive correspondence between modeled methane emissions and atmosphere concentrations largely reflected transport trajectories indicated in the HYSPLIT simulations (figure S3), with stronger agreement (r > 0.7, p ⩽ 0.05) occurring in areas characterized as open water or prone to periodic inundation (figure 4). Immediate to the flask sites, mean summer inundation varied from 2 to 10%, with moist soil fractions accounting for >85% of simulated emissions. At Lac LaBiche, annual emissions variability corresponding to wet soil fractions agreed well (r = 0.96, p = 0.02) with the flask observations. In contrast, relatively poor agreement was observed at Barrow and Sammaltunturi where emission patterns for inundated portions of the landscape corresponded more closely with atmospheric methane concentrations (table 2). At Barrow, the correspondence was similar (r > 0.43, p ⩽ 0.12) for model simulations using dynamic 15 day or annual Fw inputs, reflecting methane source contributions from thermokarst lakes and inundated tundra in the surrounding landscape (Dlugokencky et al 1995). In contrast, the modeled emissions at Sammaltunturi corresponded closely (r = 0.86, p < 0.01) with flask observations when accounting for 15 day variability in Fw extent, but showed minimal agreement when using annual Fw inputs. This discrepancy may be attributed to less open water cover in the surrounding region and a tendency for summer precipitation events to produce intermittent flooding due to shallow soil layers and limited drainage (Aalto et al 2007). These results differ from the Lac LaBiche site, where nearby peatlands are characterized by deeper layers of surface litter and moss (Dlugokencky et al 2011) that can substantially reduce surface water coverage.

Zoom In Zoom Out Reset image size

Table 2.  Mean summer fractional water (Fw) inundation and Pearson correspondence (r, with associated significance) between flask station dry air mole fractions (nmol CH₄ mol⁻¹) and cumulative methane emission estimates (tonne CH₄ grid cell⁻¹) within a 3 × 3 window centered at Barrow (BRW), Lac LaBiche (LLB) and Pallas Sammaltunturi (PAL). The model simulations incorporate dynamic 15 day or mean annual Fw; non-inundated grid cell fractions are regulated by surface soil moisture content (θ).

		Dynamic Fw	Annual Fw	θ	Fw + θ
Location	Fw inundation (%)	r
BRW	5–15%	0.46 (p = 0.11)	0.43 (p = 0.12)	−0.14 (p = 0.36)	0.05 (p = 0.45)
LLB	3–4%	0.65 (p = 0.24)	0.74 (p = 0.18)	0.94 (p = 0.03)	0.96 (p = 0.02)
PAL	1–3%	0.86 (p < 0.01)	0.02 (p = 0.48)	0.10 (p = 0.4)	0.13 (p = 0.37)

Wetland studies have increasingly used satellite microwave remote sensing to quantify the extent of methane emitting area, given the strong microwave sensitivity to surface moisture and relative insensitivity to solar illumination constraints and atmospheric signal attenuation. Regional inundation information has been incorporated into model simulations using monthly, annual, or static multi-year Fw means (Ringeval et al 2010, Petrescu et al 2010, Hodson et al 2011, Riley et al 2011, Spahni et al 2011, Meng et al 2012, Wania et al 2013). However, our simulation results show that temporal Fw scaling can lead to substantial differences in methane emission estimates (table 1).

In this analysis, inundation extent within the boreal-Arctic wetlands increased by 6–15% and 13–31% when using respective mean monthly or annual AMSR-E Fw inputs instead of finer (15 day) temporal intervals. The coarser Fw temporal inputs resulted in respective increases in estimated methane emission budgets by 6% (t = 3.5, p < 0.01) and 17% (t = 8.7, p < 0.01) in Eurasia, relative to simulations using finer 15 day Fw temporal inputs. The impacts of Fw temporal scaling in North America were not significant (t ⩽ 0.7, p > 0.5), with corresponding increases of 0.7% (Fw monthly) and 2% (Fw annual) in estimated annual methane emissions. The observed emissions sensitivity to Fw scaling in Eurasia primarily results from precipitation and flooding events in early summer, followed by mid-summer drying (Serreze and Etringer 2003). As a result, Fw means considered over longer time intervals in these regions may be biased towards spring inundation conditions, and may not reflect regional decreases in surface wetness occurring during the warmer mid-summer months. Directly incorporating Fw maximums, sometimes used to quantify multi-year surface hydrology trends (Bartsch et al 2012, Watts et al 2012), also led to substantial increases (t >9.66, p < 0.01) in estimated methane emissions by 23–38% in North America and 21–54% in Eurasia for 15 day to annual time intervals relative to simulations using static Fw means.

Significant (p < 0.05) increases in surface inundation were observed over 4% (7.1 × 10⁵ km²) of the high latitude wetlands domain from 2003 to 2011, with substantial Fw wetting occurring within northern tundra and permafrost affected landscapes (figure 5). The extent of wetting increased to 6% (9.7 × 10⁵ km²) when including regions with slightly weaker trends (p < 0.1). While the regional wetting patterns may correspond with shifts in northward atmospheric moisture transport (Rawlins et al 2009, Skific et al 2009, Dorigo et al 2012, Screen 2013), trends within the Arctic Rim may be more closely influenced by thermokarst expansion, reductions in seasonal ice cover (Smith et al 2005, Rowland et al 2010, Watts et al 2012), and summer warming (figure 6(a)). In portions of western Siberia, localized cooling and residual winter snow melt (Cohen et al 2012) may also contribute to surface wetting. Regional drying was also observed across 3% (5.3 × 10⁵ km², p < 0.05) and 5% (7.6 × 10⁵ km², p < 0.1) of the northern wetland domain, particularly in northern boreal Alaska, eastern Canada and Siberia (figure 5). These declines in surface water extent may result from an increase in summer evaporative demand (Arp et al 2011) and the terrestrialization of open water environments following lake drainage (Payette et al 2004, Jones et al 2011b, Roach et al 2011, Helbig et al 2013).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The combined influence of warming and wetting in the AMSR-E Fw and reanalysis surface meteorology records contributed to an increase in methane emissions across 6% (p < 0.05, figure 6(b)) to 21% (p < 0.1) of the boreal-Arctic domain. This finding is similar to a projected 15% increase in methane emitting area with continued climate change in the northern wetland regions (Gao et al 2013). The corresponding mean rates of methane increase from 2003 to 2011 were 0.07 and 0.11 ± 0.02 Tg CH₄ yr⁻¹, respectfully, and occurred primarily in Canada and eastern Siberia where summer warming has been observed in both in situ measurements and reanalysis records (figure S4, Screen, Simmonds (2010), Smith et al 2010, Walsh et al 2011). A decrease in modeled methane emissions, associated with surface drying and cooling patterns, was also observed across 7% (p < 0.05; −0.12 ± 0.03 Tg CH₄ yr⁻¹) and 15% (p < 0.1; −0.15 ± Tg CH₄ yr⁻¹) of the study area, and offset regional gains in the methane emissions. When including all regions showing significant change (|t| > 3.6; p < 0.05), as indicated by the linear regression analysis (without grid cell screening using the more conservative Mann-Kendall tau), the respective rates of increase and decrease in boreal-Arctic wetland emissions were 0.56 ± 0.16 Tg CH₄ yr⁻¹ and −0.38 ± 0.07 Tg CH₄ yr¹.