Influence of changes in wetland inundation extent on net fluxes of carbon dioxide and methane in northern high latitudes from 1993 to 2004

Below we briefly describe our methods consisting of: (1) a description of TEM and its modifications to better represent carbon dynamics in wetland ecosystems; (2) our representation of the variations in wetland extent in this study and how it differs from earlier efforts; (3) our representation of the spatial and temporal variation of other environmental factors that influence carbon dynamics in the northern high latitudes; (4) our calculations of regional CO₂ and CH₄ fluxes from the region; and (5) our evaluation of the importance of considering changes in wetland extent on these fluxes. More detailed descriptions of the methods can be found in the supplementary material.

The TEM uses spatially-referenced information on climate, elevation, soils, and vegetation to make monthly estimates of vegetation and soil C and nitrogen (N) fluxes and pool sizes. The model is well documented and has been used to examine high latitude terrestrial CO₂ and CH₄ dynamics (e.g., Zhuang et al 2003, 2004, 2006, 2007, Euskirchen et al 2006, Balshi et al 2007, Sitch et al 2007, McGuire et al 2010). The model explicitly couples biogeochemical processes with soil thermal dynamics of permafrost and non-permafrost soils (figure S1(a)), and is able to simulate the soil temperature profile and the seasonal freeze-thaw depth to define the active layer (Zhuang et al 2001, 2002, 2003). The freeze-thaw depth determines the lower boundary of microbial activity in the soil. The density of soil carbon decays exponentially through the soil profile. The biogeochemistry has also been coupled with a sophisticated Hydrological Module (HM, figure S1(a)), which is able to simulate the soil moisture profile and water table depth for both upland and wetland ecosystems. TEM also couples carbon/nitrogen dynamics to the production of methane (figure S1(a)). The Methane Dynamics Module explicitly simulates the processes of CH₄ production (methanogenesis) and CH₄ oxidation (methanotrophy), as well as the transport of the gas between the soil and the atmosphere (figure S1(b), Zhuang et al 2004). Various versions of TEM have been applied for quantifying carbon dynamics in boreal regions (e.g., He et al 2014, Zhu et al 2014). However, these previous model simulations have not explicitly treated the carbon stoichiometry associated with both aerobic and anaerobic decomposition occurring in dynamically changing extent of upland and wetland areas of boreal ecosystems.

In this study, we define net ecosystem exchange (NEE) as the net exchange of CO₂ with the atmosphere, which is consistent with the definition of NEE by Chapin et al (2006). For upland ecosystems, NEE is calculated as the difference between heterotrophic respiration (R_H) and net primary production (NPP). For wetland ecosystems, we extend this definition of NEE to also consider CO₂ emissions due to wetland methane oxidation (R_CWM), and CO₂ release accompanied with the production of CH₄ under anaerobic condition (R_CM):

$$\begin{eqnarray}&&{\rm{NEE}}={R}_{{\rm{H}}}+{R}_{{\rm{CWM}}}+{R}_{{\rm{CM}}}-{\rm{NPP}}.\end{eqnarray} \tag{ 1 }$$

A negative NEE represents a land sink of atmospheric CO₂ whereas a positive value represents a land source.

To make spatially and temporally explicit estimates of net CO₂ and CH₄ exchanges between the land surface and the atmosphere due to changes in wetland inundation extent in northern high latitudes (north of 45°N), we apply TEM to two datasets of static wetland inundation extent and one time-series dataset where the maximum wetland inundation extent changes annually from 1993 to 2004. The first static wetland inundation extent dataset is extracted from the NASA/GISS global natural wetland dataset at a spatial resolution of 1° × 1° (Matthews and Fung 1987). This dataset is resampled to a spatial resolution of 0.5° × 0.5° (referred to as Wet1). The second static wetland inundation extent dataset is from the Global Lakes and Wetlands Database (GLWD, Lehner and Döll 2004) where we resample a 30 s raster wetland distribution map to a spatial resolution of 0.5° × 0.5° (referred to as Wet2). The time-series dataset of dynamic inundation extent (Papa et al 2010) contains the monthly distribution of surface water extent at a ∼25 km sampling resolution and we resampled to 0.5° × 0.5° raster maps, which are geographically consistent with Wet1 and Wet2. To associate the inundation data with our model parameterizations for wetland ecosystems, we overlay the monthly changes of inundation extent from 1993 to 2004 to the Wet1 dataset to create 144 monthly maps. The new wetland inundation extent data set (hereafter referred to as Wet3) contains the wetland types defined in Matthews and Fung (1987), but with annual maximum inundation extents (figure S2). Permanent inland water bodies are excluded in the original satellite data (Prigent et al 2007). Specifically, for any particular year within the period 1993–2004, an annual map of maximum inundation extent is developed. The annual map consists of the maximum inundation extent for every pixel, which is determined based on its monthly inundation data. These annual maximum inundation extent maps with a mean annual inundation area of 3.15 Mkm² are used in Wet3 simulations (figure 1(b)). Note that inundation area for some months could be larger than the static wetland area in Wet1 for some pixels. The extra inundated-area for a particular pixel in the Wet3 data is assumed to be covered with the same wetland types as in the Wet1 data for that pixel.

Zoom In Zoom Out Reset image size

To drive TEM, we also use spatially-explicit data of monthly climate, land cover, soil texture, soil-water pH, elevation, and leaf area index (LAI). The static data sets include potential natural vegetation (Melillo et al 1993), soil texture (Zhuang et al 2003), and soil-water pH (Carter and Scholes 2000). The historical monthly air temperature, precipitation, vapor pressure, and cloudiness data sets are from the National Centers for Environmental Prediction (http://www.ncep.noaa.gov/). LAI is internally estimated as a function of vegetation carbon within TEM for each 0.5° latitude × 0.5° longitude grid cell. More details about the input data can be found in supplementary material.

To conduct the study for the region, which is spatially heterogeneous with respect to land ecosystem types, soils, and climate, we apply the parameterizations for both tundra and boreal forest ecosystems described in previous studies (Zhuang et al 2003, 2004) to simulate CH₄ consumption as well as net CO₂ exchanges in upland ecosystems for the corresponding classes in each wetland map. To simulate methane emissions from wetland ecosystems, we use the parameterization of methane dynamics from Zhuang et al (2004), in which the simulation of CH₄ emissions was optimized to field measurements at several sites (table S1). To simulate net CO₂ exchanges in wetland ecosystems, we calibrate the revised TEM for three wetland ecosystem types (table S1) including wet tundra wetlands, alpine tundra wetlands, and forested wetlands using a Bayesian inference technique (Tang and Zhuang 2009). Field-based estimates of NEE (table S1) are used to compare with model simulations to optimize TEM parameters.

To examine the effect of inter-annual changes in wetland inundation extent on carbon fluxes, we conduct a 'core' regional simulation with the organized wetland (Wet3) and upland data for the period of 1993–2004. In this simulation, the annual maximum inundation area detected with satellite is treated as wetlands, i.e., the wetland distribution and area change from year to year. To illustrate the effect of the interannual changes of wetland inundation extent, we conduct two other simulations with the 'static' inundation extent datasets Wet1 and Wet2. Each simulation was spun-up for 100 years. For each simulation, the model is run twice for each pixel, once for the upland parameterization and once for the wetland parameterization. In the cases where a grid cell goes from wetland to upland or from upland to wetland, we multiply the fluxes for each land cover type by the new fraction of wetland and upland.

With the 'core' simulation, we first examine the carbon balance in the region with an explicit consideration of both wetland and upland ecosystems. We then analyze how CO₂ and CH₄ fluxes are related to changes of climate and wetland area extent at an annual time step. In addition, we examine the effects of soil moisture, soil temperature, water table depth, and active layer depth on CO₂ and CH₄ dynamics in the region.

To quantify the sensitivity of both CO₂ and CH₄ fluxes to the uncertain wetland inundation extent in the region, we conduct two additional simulations with slight modifications of the Wet3 dataset. One simulation assumes a 10% increase of the wetland inundation extent for those pixels having wetland distribution in Wet3 whereas the other assumes a 10% decrease in wetland inundation extent. We chose 10% because this value was the approximate percentage change of inundation area between 1993 and 2004.

Driven with the dynamic wetland inundation data, our simulation indicates that inter-annual variations of CH₄ emissions from the land ecosystems in the region range from 56.9 Tg CH₄ yr⁻¹ in 2004 to 81.3 Tg CH₄ yr⁻¹ in 1994 (table 1). The region acts as a sink of −1.28 ± 0.03 Pg C yr⁻¹ for the period of 1993–2004. Wetland ecosystems are persistently a net sink for CO₂ that ranges from −0.38 to −0.41 Pg C yr⁻¹ during the period of 1993–2004 for the total area ranging from 2.93 to 3.46 million km². The upland ecosystems are also a sink from −0.6 to −1.1 Pg C yr⁻¹ for the total area ranging from 37.18 to 37.71 million km² over the region (figure 2). The inter-annual variation in net CH₄ emissions of the entire region is significantly correlated with annual maximum wetland inundation extent and inversely correlated with annual precipitation, but there is no significant correlation with the annual air temperature (table 2). It is not clear why annual precipitation and annual inundation extent are not correlated. It is possible that the variability in annual inundation extent is determined by seasonal patterns in precipitation (e.g., rain in late summer or high snow pack in winter), which may not necessarily correlate with annual precipitation. There are also no significant correlations of annual CH₄ emissions with the mean annual soil temperature at depth 20 cm, active layer depth, and water table depth. Inter-annual variation in NEE is not significantly correlated with these environmental variables except wetlands area at the annual temporal resolution.

Zoom In Zoom Out Reset image size

The spatial patterns of exchanges of CO₂ and CH₄ (figure S3) vary depending on the wetland inundation extent, which fluctuates in the region (figure S2). Both CO₂ and net CH₄ emissions exhibit a significant spatial variability as demonstrated with two inundation extent extreme years 1994 (the highest) and 2004 (the lowest). During 1994, a larger wetland inundation area estimates a higher methane production and emissions (81.3 Tg CH₄ yr⁻¹). For CO₂ dynamics, the region acts as a carbon sink of −1.26 Tg C yr⁻¹ as a balance of a wetland sink of −0.41 Pg C yr⁻¹ and an upland sink of −0.85 Pg C yr⁻¹ (figure 2). In contrast, the smallest wetland inundation extent in 2004 estimates the regional net methane emissions as small as 56.9 Tg CH₄ yr⁻¹ (figure 1, table 1). The NEE is a sink of −1.29 Pg C yr⁻¹ as a result of both uplands and wetlands sinks (figure 2).

There are large differences in net methane emissions and net carbon exchanges under static and dynamical wetland inundation conditions (table 1). Under the 'static' wetland inundation extent conditions Wet1 and Wet2, the regional methane emissions range from 51.3 ± 2.6 to 73.0 ± 3.6 Tg CH₄ yr⁻¹ during the period of 1993–2004 (figure S4(a)). For NEE, the region acts as a persistent sink ranging from −1.29 ± 0.04 to −1.40 ± 0.05 Pg C yr⁻¹ under static wetland inundation conditions (table 1).

All three 'core' simulations show a similar seasonal CH₄ emission pattern, peaking in July (figure S4(b)). In wetland areas, the carbon sink corresponds to the size of wetlands in three simulations in summer months (figure S5(a)). The sink estimates range from −0.1 to −0.2 Pg C month⁻¹ in June and July. Wetland NEE is similar between the three simulations from January through May, and September through December (figure S5(a)). In contrast, changes in the extent of uplands do not result in substantial differences in NEE between the three simulations (figure S5(b)). As a result, changes of annual inundation extent result in seasonal NEE differences up to 0.04 Pg C month⁻¹ in the region (figure S5(c)). Overall, the inundated area changes result in higher summer CH₄ emissions, but lower summer NEE sinks (figures S4 and S5).

For a 10% change in annual maximum wetland inundation extent, our simulations indicate that the regional net exchanges of CO₂ and CH₄ are affected by subsequent changes of wetland and upland areas (table 1). Net exchanges of CH₄ are more sensitive to wetland inundation extent changes than NEE. Specifically, if wetland inundation extent increases or decreases by 10% in each wetland grid cell, the regional source of methane increases 13% or decreases 12%, respectively. In contrast, the regional NEE responds with only 7–9% changes to the changes in wetland inundation extent (table 1). The response of net CH₄ emissions is mainly due to the stronger effect of the changes in inundated areas on methane production than on methane consumption as methane consumption in uplands is an order of magnitude less than wetland methane production (see figure 3). The response of regional NEE to changes in wetland inundation area is dominated by the response of aerobic decomposition to changing wetland area. Upland area changes due to changing inundation area are relatively small given that the upland area is much larger than wetland area, the upland NEE thus does not change significantly.

Zoom In Zoom Out Reset image size

Nine inversion studies estimated that net CH₄ emissions of boreal region wetlands are 34 ± 13 Tg CH₄ yr⁻¹ (Chen and Prinn 2006). Another inversion constrained by atmospheric observations of methane and its ¹³C/¹²C isotopic ratio data indicates that the boreal region releases 30–64 Tg CH₄ yr⁻¹, presumably mainly from wetlands (Mikaloff Fletcher et al 2004, McGuire et al 2012). A more recent study using satellite data estimated that mean summer emissions are 53 Tg CH₄ yr⁻¹ from boreal-Arctic wetlands (Watts et al 2014). The 'static' wetland inundation dataset Wet1 estimates CH₄ emissions that are comparable with the various inversions, while both Wet2 and dynamic wetland inundation extent dataset Wet3 estimates much higher emissions due to the larger wetland or inundated area in the latter two datasets (table 1). Annual emissions correlate well with annual wetland inundation area. In particular, the highest emissions occur in 1994, which is primarily explained by the largest inundation area in the period. Both static and dynamic wetland simulations estimate higher emissions in 1998 than in 1999, which is consistent with the inversion results (e.g., Mikaloff Fletcher et al 2004, Chen and Prinn 2006). While annual maximum wetland inundation area is larger in 1998 than in 1999, so are the annual precipitation and air temperature (figures 1(c) and (d)), suggesting that climatic conditions in 1998 also favor the methane emissions. A number of studies have suggested that the difference in CH₄ emissions between 1998 and 1999 is the responses to temperature changes associated with the warm conditions in 1998 due to the strong El Niño phenomena (e.g., Dlugokencky et al 2001, Cunnold et al 2002). Indeed, 1998 was marked with elevated temperatures in the region from June to August (Bell et al 1999), and its annual temperature was higher than the 12-year average annual temperature over the region and the annual temperature of 1999. In addition, some high northern latitudes regions experienced elevated precipitation from April to September, which might have also led to the elevated wetland emissions for the year 1998 (Curtis et al 2001, Dlugokencky et al 2001). Our analysis indicates that precipitation in 1998 is indeed higher than in year 1999 while air temperature and inundated land area is also above the average of the study period (figure 1(d)). The temperature and wetland inundation responses together dominate the positive anomaly of net methane emissions for the year 1998 in the region.

Our simulations further indicate that the three wetland datasets estimate similar seasonal dynamics of methane emissions with different magnitudes over the region (figure S4(b)). The peak emissions occur in July. These seasonal dynamics are dominated by climatic factors, rather than wetland inundation extent. To consider the seasonal rather than annual wetland inundation extent in a future analysis will improve the estimates of seasonal dynamics. To date, most atmospheric transport chemistry and inversion models have primarily based their analyses on methane fluxes estimated without considering changes of wetland inundation extent (e.g., Houweling et al 1999, Cunnold et al 2002, Dentener et al 2003, Wang et al 2004, Chen and Prinn 2005, 2006). Coupling estimates of methane emissions considering both seasonal and interannual wetland inundation extent dynamics with these atmospheric models should improve modeling atmospheric methane concentrations.

Our estimates of NEE account for the CO₂ emissions from both aerobic decomposition and anaerobic decomposition of the soils (see equation (1)). Based on the static wetland inundation extent of the Wet1 and Wet2 datasets, the region is estimated to be a consistent sink (table 1). The annual NEE ranges from −1.23 to −1.50 Pg C yr⁻¹ for the period 1993–2004. The dynamic wetland inundation extent estimates the region to be a similar sink at −1.28 ± 0.03 Pg C yr⁻¹. The larger regional sink under static wetland inundation conditions is due to the larger wetland area prescribed by the Wet2 dataset that enhances the carbon sink. The smaller regional sink is mainly due to the decrease of the regional wetland carbon sink as wetlands area decreases under the dynamic wetland inundation condition. Although the wetland area is much smaller than the upland area in this region and both NPP and R_H of upland are much higher than that of wetlands, the strength of the carbon sink of the wetlands account for about one-third of the total regional sink. For instance, the sink of wetlands (−0.42 Pg C yr⁻¹ or 126.4 g C m⁻² yr⁻¹) of 2002 is about half of the upland sink (figure 2(c)). This suggests that the wetland ecosystems play a disproportional role in affecting regional NEEs. Breakdown of the wetland emissions of CO₂ shows that the majority of emissions are from the unsaturated zone of the wetland soils, which is created with the fluctuation of water table depth (figure 3). This aerobic decomposition accounts for 80% of the total wetland CO₂ emissions while the release of carbon associated with the methane oxidation and production varies between 40 and 60 Tg C yr⁻¹ with wetland inundation extents.

In comparison with an earlier estimate of a sink of 51.1 Tg C yr⁻¹ over the 1997–2006 time period (McGuire et al 2010), our estimated regional NEE magnitudes based on the satellite-detected dynamic wetland extent dataset are much larger. If we account for fire emissions (246.5 Tg C yr⁻¹) and the decomposition of agricultural and forestry products (13.5 Tg C yr⁻¹) estimated by McGuire et al (2010) for the same region in our analysis, our estimates of NEE will be a large sink of −1.0 Pg C yr⁻¹ under the dynamic wetland conditions. A recent inter-model comparison shows that the multi-model mean annual NEE for Alaska is largely carbon neutral (Fisher et al 2014) while some observations suggest that portions of the region have been a source of CO₂ to the atmosphere (Oechel et al 2000, 2014). The differences of carbon budget estimates between various models and our estimates are at least due to: (1) different processes being considered by the models; (2) different models parameterizations; (3) different sensitivity of the models to the same forcing data; and (4) different treatment of land ecosystems (e.g., the inundated area changing) among the models (Melton et al 2012, Bohn et al 2015). The TransCom model means for the period of 1995–2006 for the region range from a source of 0.03 Pg C yr⁻¹ to a sink of −0.89 Pg C yr⁻¹ with a large uncertainty (Gurney et al 2004, 2008). Using the estimates of NEE that consider the effects of wetland inundation extent changes and both aerobic and anaerobic decomposition as priors for atmospheric inversion modeling may reduce the current discrepancy between process-based biogeochemical modeling and inversion estimates.

Previous evaluation of regional radiative forcing estimated as GWPs was based on regional methane emissions and NEEs that have not considered changes in wetland inundation extent (e.g., Zhuang et al 2007). Consideration of wetland inundation dynamics reduced the estimated regional negative climate forcing (figure 4, table 1). Overall, the estimated regional GWP sink using either static or dynamic inundated area data increases during the study period (figure 4). The increase estimated using dynamic inundation data is mainly due to the decrease of CH₄ emissions resulting from the declining inundated area (figure 1). The slight increase estimated using two static wetland datasets is due to the slight increase of the carbon sink (table 1). Our sensitivity analysis indicates that the regional GWP under the conditions of either a 10% increase or a 10% decrease in inundation area of the Wet3 dataset differs little (less than 6%) from the estimates with Wet3 (table 1). Thus, changes in CH₄ fluxes associated with a 10% change in wetland inundation extent appear to compensate for changes in CO₂ fluxes. This suggests that rather large changes in wetland inundation extent would be required to significantly alter the balance of NEE and methane emissions and to influence the GWP in the region. The heterogeneity of NEE and methane emissions as wetland and upland fractions change inter-annually across the landscape as well as climatic changes results in a large spatial variability in GWP across the pan-Arctic (figure S3).

Zoom In Zoom Out Reset image size

The regional annual GWP is significantly correlated with interannual wetland inundation extent (table 2). Net methane emissions correlate with wetland inundation extent more significantly than NEE, which suggests that future changes of wetland inundation extent will greatly affect regional net ecosystem carbon and CH₄ exchanges at annual scales. If wetland inundation extent increases, there will be more regional CH₄ emissions, but less regional CO₂ emissions from a relatively smaller upland area and a stronger carbon sink in a relatively larger wetland area. As a result, the regional GWP source will be weaker (figure 4). Future climate predictions with general circulation models therefore should consider the feedback that is induced by these CO₂ and CH₄ dynamics under changing wetland areas in addition to the physical feedbacks of wetland distribution (e.g., Gutowski et al 2007, Prigent et al 2011).

It is important to note that the greatest limitation of this analysis is the use of relatively coarse resolution (i.e., 0.5° × 0.5°) wetland inundation datasets, which do not capture small features within wetlands that create CH₄ hotspots. An average wetness across these large pixels is also likely to underestimate actual CH₄ fluxes (e.g. Becker et al 2008, Baird et al 2009). Additional studies are needed to analyze how these small-scale features of wetlands affect our analysis.

This study considers the influence of water table depth and the stoichiometry of CO₂ and CH₄ during soil anaerobic and aerobic decomposition processes in developing regional estimates of CO₂ and CH₄ fluxes. The study shows that the use of satellite-detected wetland inundation extent essentially reduces the regional GWP sink estimated using static wetland inundation extent data sets and the regional dynamics of CO₂ and CH₄ respond differently to changes of wetland inundation extent with CH₄ fluxes being more sensitive than CO₂ fluxes. Future studies should thus consider the dynamics of wetland inundation extent in quantifying both net CO₂ and CH₄ fluxes and their feedbacks to the atmospheric climate and chemistry at regional and global scales. Inundation area data developed from satellite sensors such as NASA Soil Moisture Active Passive will be valuable for making additional progress on understanding how inundation area influences greenhouse gas exchange in the study region.