Processes affecting greenhouse gas production in experimental boreal reservoirs



[1] Flooding land for water reservoir creation has many environmental impacts including the production of the greenhouse gases (GHG) carbon dioxide (CO2) and methane (CH4). To assess processes governing GHG emissions from the flooding of terrestrial carbon, three experimental reservoirs were constructed in upland boreal forest areas of differing carbon stores as part of the Flooded Upland Dynamics Experiment (FLUDEX). We calculated process-based GHG budgets for these reservoirs over 5 years following the onset of flooding. Stable isotopic budgets of carbon were necessary to separate community respiration (CR), which produces CO2, from net primary production (NPP), which consumes CO2, and to separate CH4 production from CH4 consumption via oxidation. NPP removed up to 44% of the CO2 produced from CR. CR and NPP exhibited different year-after-year trends. CH4 flux to the atmosphere increased about twofold over 3 years, yet isotopic budgets showed CH4 production in flooded soils increased nearly tenfold. CH4 oxidation near the flooded soil-water interface greatly decreased the CH4 flux from the water column to the atmosphere. Ebullition was the most important conduit of CH4 to the atmosphere after 3 years. Although CH4 production increased with time, the total GHG flux, in CO2 equivalents, declined. Contrary to expectations, neither CR nor total GHG fluxes were directly related to the quantity of organic carbon flooded. Instead, these reservoirs produced a strikingly similar amount of CO2 equivalents over 5 years.


[2] Water reservoirs are constructed for irrigation, hydroelectricity generation, drinking water supply, flood control, low flow augmentation, and recreation [Tremblay et al., 2005]. Globally, reservoirs are ubiquitous, and reservoir surface area estimates range between 0.26 × 106 km2 and 1.5 × 106 km2 [Downing et al., 2006; St. Louis et al., 2000]. A large number are small- and medium-sized impoundments that contribute to a large increase in the extent of affected rivers [Lehner et al., 2011; Wisser et al., 2010]. Reservoir creation kills vegetation on landscapes and causes flooded vegetation and soils to decompose. As a result, reservoir creation results in net emissions of the greenhouse gases (GHGs) carbon dioxide (CO2) and methane (CH4) to the atmosphere for decades and possibly centuries following flooding [Duchemin et al., 1995; Kelly et al., 1997; St. Louis et al., 2000; Barros et al., 2011]. However, to predict the magnitude of GHG fluxes from reservoirs, the rates and relative importance of biological processes driving carbon (C) cycling on flooded landscapes must be quantified.

[3] Much reservoir research has focused on the fluxes of CO2 and CH4 from reservoir surfaces [e.g., Duchemin et al., 1995; Huttunen et al., 2002]. Surface fluxes require accurate measurements of both CO2 and CH4 surface-water concentrations and gas exchange rates, which are the net result of a number of processes. In addition, the source of the CO2 and CH4 cannot be directly identified. Recently, more nuanced approaches have expanded temporal resolution of CO2 and CH4 measurements in reservoirs [e.g., Teodoru et al., 2012] and measured fluxes of carbon GHGs from landscapes before and after flooding [Rudd et al., 1993] instead of simply comparing reservoirs to nearby lakes. Flooded soils and wetlands have been indirectly identified as the source of carbon GHGs produced in reservoirs via various types of measurements [Duchemin et al., 2006; Huttunen et al., 2004, 2006; Rudd et al., 1993]. Lake sediments are disparate sources of CH4 to the atmosphere [Bastviken et al., 2004, 2008] as are flooded soils and wetlands in reservoirs [Bodaly et al., 2004; Huttunen et al., 2003, 2006; Kelly et al., 1997]. However, the relative contribution of the important processes to the net surface fluxes of GHGs has not been determined at the ecosystem scale.

[4] In the Flooded Upland Dynamics Experiment (FLUDEX), three boreal upland areas with differing amounts of organic C (OC) stores in vegetation and soils were experimentally flooded for five consecutive years at the Experimental Lakes Area (ELA) in NW Ontario, Canada, to test the hypothesis that GHG production from reservoirs is positively related to the quantity of OC stores [Bodaly et al., 2004; Kelly et al., 1997; St. Louis et al., 2004]. During the first 3 years of flooding, Matthews et al. [2005] found that forests and soils that had been net sinks of CO2 and CH4 became net sources of both GHGs to the atmosphere. Net dissolved inorganic C (DIC) production declined markedly from the first to third years of flooding, while net CH4 production increased [Matthews et al., 2005] (Figure 1).

Figure 1.

Net DIC production, net CH4 production, and CH4 ebullition in the three FLUDEX reservoirs for 1999–2003. Values for 1999–2001 are reported in Matthews et al. [2005].

[5] Ebullition added an increasing amount to net CH4 production, from 0% to 5% in the first year to 50% to 145% and 130% to 175% in the second and third years (Figure 1). CO2 ebullition was very small relative to net DIC production [Matthews et al., 2005].

[6] To be able to extend the results from these experimental reservoirs to other landscapes around the world, reservoir modelers require knowledge and quantification of the relative importance of these internal biological processes that control the net flux of GHGs to the atmosphere from reservoirs. Further, without knowledge of relative rates, the importance of carbon sources or processes can be misinterpreted. As examples of these competing biological processes, both primary production and CH4 oxidation consume the products of aerobic and anaerobic OC decomposition, causing net flux of GHGs to the atmosphere to be lower than gross GHG production.

[7] Our rates pertain to the shallow zones of boreal reservoirs, the most important for GHG release. Boreal reservoirs cover about 1% to 5% of the 13.7 × 106 km2 of the boreal forest [Barros et al., 2011; St. Louis et al., 2000; Intergovernmental Panel on Climate Change, 2000] with the potential for at least doubling this coverage [Sims et al., 2007; Canadian Hydropower Association, 2002]. However, the relative rates of internal process observed in our study are important for all reservoirs.

[8] In this paper, we extend the record of net GHG emissions from the 3 years [Matthews et al., 2005] to the full 5 years of the experiment. More importantly, though, we use natural abundance stable carbon isotope ratios (δ13C values) to quantify the four most important processes affecting GHG production over the 5 years of the experiment: community respiration (CR), net primary production (NPP), CH4 production, and CH4 oxidation. Although these processes are universal, this is the first time they have been directly measured at the ecosystem scale in reservoirs. The key benefit of using δ13C is that each process produces or consumes C with a different and constrainable isotopic fractionation, where isotope fractionation is defined as the quotient of the heavy to light isotope ratios minus 1 (ε [Coplen, 2011]). The δ13C-DIC enables the separation of the contribution of DIC production via CR from DIC loss via NPP [e.g., Quay et al., 1986]. The δ13C-CH4 provides assessment of methanogenic pathways and the role of CH4 oxidation [Whiticar et al., 1986; Whiticar, 1999].

[9] Although a handful of studies report δ13C budgets for DIC only in lakes (e.g., Quay et al. [1986] at natural abundance and Cole et al. [2002] with an isotope spike), this research constitutes the first δ13C annual budgets for any reservoir and employs a novel method of separating and quantifying processes at the whole ecosystem scale. Our findings are critical for understanding why certain reservoirs emit more GHGs than others: it is needed by modelers of reservoir GHG emissions and policy makers evaluating the efficacy of global GHG emissions reductions from the energy production sector.

Material and Methods

Study Site

[10] The three experimental FLUDEX reservoirs were constructed on the Canadian Shield at the ELA (49°40′N 94°45′W). The reservoir sites were chosen based on the quantity of OC in vegetation and soil and designated High, Medium, or Low C [Hall et al., 2005]. These ranged from 30,900 kg C ha−1 to 45,900 kg C ha−1 with large differences in both soil and vegetation OC between reservoirs (Table 1). All vegetation and organic soil were severely burned during a forest fire in 1980, and regrowth was dominated by a dense, even-aged jack pine (Pinus banksiana Lamb.) and paper birch (Betula papyrifera Marsh.) forest, common where the fire frequency is 50–100 years [Stocks et al., 2002; Bergeron et al., 2004]. These soil values are comparable to those measured in Boreal Ecosystem-Atmosphere Study (BOREAS), 3800 kg C ha−1 to 30,000 kg C ha−1 [Rapalee et al., 1998; Trumbore and Harden, 1997]. At the time of initial flooding in 1999, overburden within the FLUDEX reservoirs ranged from nil on exposed bedrock to >1 m of soil. These reservoirs were sized to be manageable ecosystems for very detailed research, allowing accurate estimation of isotope-mass budgets and quantification of preflood OC stores, and yet be not so windy that gas exchange with the atmosphere decreases our sensitivity to calculate processes. Additionally, these reservoirs resemble the hundreds of thousands of small and medium-sized reservoirs [Downing et al. 2006; Lehner et al., 2011; Wisser et al., 2010] as well as the warm, shallow zones of large reservoirs.

Table 1. FLUDEX Reservoir Physical Characteristics, Forest Type, Dominant Vegetation and Soil Communities, and Carbon Stores
 High CMedium CLow C
Water surface (m2)740050006300
Upland catchment (m2)47,8007300900
Volume (m3)687042707170
Mean depth (m)
Total soil C (kg C ha−1)18,300720011,100
Total aboveground vegetation C (kg C ha−1)27,60027,70019,800
Total C (kg C ha−1)45,90034,90030,900
Dominant vegetation and soil communitiesWet forest (53%)Dry forest (100%)Dry forest (73%)
Dry forest (47) Lichen and 0bedrock (22%)
Lichen-moss pillow (5%)

[11] To determine the mass and δ13C value of the soil and vegetation carbon, 20–25 soil cores were collected at 20 m intervals along sampling transects in each reservoir, separated into horizons (litter, fungal-humic, charcoal, and mineral), and analyzed for C and N concentration and δ13C (Table S1 in the supporting information). Biomass of low shrubs, herbs, mosses, and lichens was estimated from material in five to 10 quadrats per reservoir [Heubert, 1999]. Woody biomass in tall shrubs and trees was estimated from trunk diameter measurements [cf. Dyck and Shay, 1999]. Foliar biomass was estimated from monthly collections of litter-fall from three locations per reservoir [Matthews et al., 2005]. Representative samples of vegetation were collected from each reservoir and analyzed for C and N concentration and δ13C.

Reservoir Construction and Hydrology

[12] Reservoirs were built against natural slopes on the landscape, and water was contained on the lower sides by wooden walls and gravel dikes to a maximum depth of 2 m. Where water depth was expected to be >1 m, polyethylene-lined wooden walls were constructed, anchored, and sealed directly to bedrock. Gravel dikes with a polyethylene liner were constructed where water depth was expected to be less than <1 m (see aerial photos in the supporting information and Bodaly et al. [2004]). Areas above the height of flooding were not diked, leaving regions open to input from the upslope catchment (Table 1). Water was continuously pumped into the reservoirs for 15–16 weeks each year from nearby oligotrophic Roddy Lake (total phosphorus, 6 ± 3 µg L−1, mean ± 1SD, n = 47, Table S1) beginning in May or June each year from 1999 to 2003. Yearly hydrologic budgets (Figure S1 and Table S2) were calculated for each reservoir by summing the daily water inputs (pump inputs, precipitation, and catchment input) and outputs (outflow at gauged weir, seepage, and evaporation) [Matthews et al., 2005]. Pumped input volumes were quantified by an in-line flow meter. Continuous stage-level recorders and V-notch weirs were used to calculate outflow volumes. Seepage and fracture flow were channelized, and their rates were measured once per season. Precipitation was recorded at the ELA meteorological station (<1 km from the reservoirs), and evaporation was estimated with a Class A evaporation pan in each reservoir. Upslope catchment input to each FLUDEX reservoir was determined by scaling down water flow data from the well-studied northwest inflow stream of ELA Lake 239 (L239 NWIF). Storage was the small volume of water left standing at the end of each flood season; each autumn, the reservoirs were drained to simulate seasonal water level fluctuations characteristic of the shallow areas of boreal and temperate hydroelectric reservoirs.

Field Sampling

[13] Matthews et al. [2005] presented detailed field sampling procedures for DIC, CO2, and CH4 concentrations. Pumped inputs and gauged reservoir outflows were sampled weekly for DIC, CO2, and CH4 concentrations and δ13C in glass serum bottles. Samples were sealed with a prebaked (60°C for 12 h) Vacutainer stopper punctured by a needle to eliminate any air trapped in the stopper. Saturated sodium azide solution (0.3% of bottle volume) was added within 2 h of collection to preserve samples.

[14] To determine DIC, CH4, and δ13C values in catchment input from the undiked upslope areas of the reservoirs, samples were collected from small runoff channels formed during and after large rain events. Lake 114 inflow stream (L114 IF; 2 km from the reservoirs) was also sampled as a proxy for the δ13C of catchment input since the L114 IF could be regularly sampled throughout the experiment due to its relatively larger catchment area and water storage capacity. Vegetation and soils were similar in FLUDEX and L114 catchments (logged in the 1970s, extensively burned in 1980) so the δ13C values should be similar.

[15] To determine ebullition fluxes of GHGs from the reservoirs, five to 10 inverted-funnel bubble traps (30 cm diameter) were continuously deployed in each reservoir. Fluxes were determined by multiplying weekly accumulated bubble volumes by CO2 and CH4 concentrations in fresh gas bubbles collected by probing flooded reservoir soils and trapping the bubbles in an inverted funnel. Fresh bubbles were collected to obtain gas representative of ebullition rather than gas sitting in a trap for a week. The gas was transferred to serum bottles for concentration analysis or to pre-evacuated tubes (12 mL of gas in a 10 mL tube) for δ13C analysis.

[16] To independently assess the diffusive GHG flux across the flooded soil-water interface (SWI) due to decomposition of soil organic matter, benthic chambers (~13 L) were deployed on the flooded soil. Before flooding, 15 plastic collars (30 cm diameter) per reservoir, in groups of three, were inserted into the soil. After flooding, chambers with a sampling port, motor, and propeller were deployed from a boat onto the collars for 4 to 6 h every 2 weeks. Darkened chambers covered with foil were deployed alongside the light-exposed chambers to assess the effects of photosynthetically derived O2 on DIC and CH4 flux across the SWI. Typically, one clear and one dark chamber were deployed at four or five sites, depending on site accessibility. Chambers were sampled via tubing attached to the chambers. In a second method of assessing the GHG flux across the SWI, soil pore water at 2–5 cm intervals and reservoir water columns at 20 cm intervals were sampled with two permanently emplaced profilers per reservoir.

[17] Periphyton (attached algae and associated organisms) growth was the major form of NPP in the reservoirs. Independent sampling indicated that phytoplankton were less than 1% of NPP (D. Findlay, unpublished data, 2001). Emergent macrophytes were not observed in the reservoirs until the third year of flooding and then numbered fewer than 20 plants. To assess the δ13C-NPP of periphyton, pairs of pine dowels were hung at five sites in each reservoir as a surrogate substrate. This periphyton was sampled monthly into Whirl-Pak bags and freeze-dried before analyses. After each flood season, large periphyton sheets that formed between the branches of the flooded trees were collected from each reservoir. Both types of periphyton samples were analyzed for δ13C and C/N ratios.


[18] To quantify DIC, CO2, and CH4 concentrations, liquid samples were analyzed via headspace equilibration, and gas samples were analyzed directly on the same equipment as in Matthews et al. [2005]. Briefly, a subsample of the headspace was injected into a Varian 3800 gas chromatograph. A Hayesep D column separated the CO2 from CH4 gases. A ruthenium methanizer converted CO2 to CH4 before detection via a flame ionization detector.

[19] To quantify δ13C values of DIC, CO2, and CH4, liquid samples were analyzed via headspace equilibration, and gas samples were analyzed directly with a Micromass Isochrom GC-IRMS. Headspace was created in liquid samples by displacing sample water with 5 mL of helium. A subsample of the equilibrated headspace was injected into a HP 6890 gas chromatograph [Venkiteswaran and Schiff, 2005]. Precision of δ13C-DIC and δ13C-CO2 analyses was ±0.3‰ and of δ13C-CH4 analysis was ±0.5‰.

[20] Solid samples were freeze-dried and ball milled. Filters were freeze-dried. Dissolved OC (DOC) samples were filtered through precombusted Whatman GF/F filters, acidified to pH < 4 (37% hydrochloric acid), and freeze-dried. Analyses of δ13C and C/N ratios were performed on an EA-IRMS (Carlo Erba EA1108 and Micromass Isochrom). Precision of analyses was ±0.2‰ for δ13C and < ±1% for C and N. Reported C/N ratios are molar ratios.

δ13C-Mass Budgets and Processes

[21] To calculate δ13C-mass budgets, concentration and δ13C values were multiplied by hydrologic volumes. All major components of the δ13C-mass budgets were measured directly (pumped input, gauged outflow, and storage). The δ13C-mass budgets were calculated daily because concentrations, δ13C values, and water flow rates changed at different rates over the flood season. Daily concentrations and δ13C values were linearly interpolated from weekly measurements. Pumped input and gauged outflow rates were measured daily. Daily values were summed into seasonal inputs and outputs. End-of-season storage was calculated by measuring vertical profiles of DIC and CH4 concentrations and δ13C in the reservoirs on the day before the reservoirs were drawn down in autumn and multiplying these data by reservoir-specific volume-to-depth curves. Net DIC and CH4 production and their δ13C values were determined as follows:

display math(1)
display math(2)

where net is net result, Q the water flow rates, C the concentration, δ the δ13C, i day, and j the number of flood days, and the subscripts net, out, and in denote the net, outputs, and inputs.

[22] The δ13C of some budget items (gas exchange, seepage, and precipitation) had to be estimated because they could not be directly measured. The δ13C-CO2 lost via gas exchange was calculated from δ13C-DIC, pH, reservoir surface-water temperature, and isotopic fractionation factors [Clark and Fritz, 1997]. Reservoir pH was relatively constant in all years (6.2 ± 0.4, mean ± SD, n = 140). Gas exchange coefficients were determined experimentally in each reservoir with SF6 [Matthews et al., 2003]. To calculate the δ13C-DIC in precipitation, the pH of precipitation was measured and combined with the average atmospheric CO2 concentration and δ13C-CO2 [Battle et al., 2000]. Isotopic fractionation for CH4 dissolution was smaller than measurement precision (ε = −1‰ [Knox et al., 1992]). CH4 concentration and δ13C-CH4 of precipitation were assumed to be equilibrated with atmospheric CH4 [Quay et al., 1999].

[23] Net DIC production in the reservoirs was the balance between CR and NPP because a portion of the CO2 produced via CR was consumed by NPP within the reservoirs. CR depends largely on the lability of the flooded organic matter and temperature [Wetzel, 2001], while NPP depends on many diverse factors such as nutrient concentrations, mixing, DOC, depth, and light attenuation [Falkowski and Raven, 1997]. The δ13C-DIC mass budgets were used to quantify CR and NPP via a two end-member mixing model. End-member δ13C values of the CO2 produced by CR and the CO2 fixed by NPP were assigned a priori based on supporting data because they could not be measured directly. CR and NPP were calculated as

display math(3)
display math(4)

where net DIC production is from equation ((1)) and δnet DIC production is net δ13C-DIC production from equation ((2)).

[24] The δ13C-CR was assigned based on three lines of evidence. First, laboratory incubations of FLUDEX vegetation and soils indicated that early stages of soil decomposition were associated with isotopic fractionation (εCR) of +2‰ to +3‰ [Baril, 2001; Boudreau, 2000; Oelbermann and Schiff, 2008, J. J. Venkiteswaran, unpublished data, 2000]. Second, long-term field studies show isotopic fractionation during decomposition of vegetation and soils has εCR values around +1‰ to +2‰ [Nadelhoffer and Fry, 1988; Fogel and Cifuentes, 1993]. Third, the DOC intermediates produced during decomposition, which can be subsequently converted to DIC, have the same δ13C value as the vegetation or soils [Ferguson, 2000] (δ13C-DOC measured in reservoirs was −27.4 ± 0.4‰, mean ± SD, n = 94, all reservoirs, all years). Thus, the δ13C-CR was assigned a value of −27‰, slightly higher than measured values of vegetation and soil δ13C (−30‰ to −27‰; Table S3). If the δ13C-CR assigned is marginally overestimated, then estimates of CR are maxima and NPP are minima.

[25] NPP is associated with an isotopic fractionation (εPP) of about −19‰ between newly fixed C and the source C where CO2 is not limiting [Hecky and Hesslein, 1995] such as in these reservoirs. The δ13C-CO2 used by NPP was calculated from measured DIC concentration, pH, temperature, and δ13C-DIC. The calculated δ13C-NPP values were compared with measured δ13C of periphyton (Table S1).

[26] Different microbial pathways produce CH4 with different δ13C-CH4 values. Methyl-type fermentation has isotope fractionation of −35‰ to −25‰ between source compounds and CH4 [Whiticar, 1999]. Whereas CH4 produced via CO2 reduction has isotopic fractionation more negative than −55‰. Thus, δ13C-CH4 of unoxidized CH4 in the flooded soils can be used to infer the dominant microbial pathway. Furthermore, the amount of CH4 oxidation that occurred subsequent to production can be calculated using a Rayleigh equation [Clark and Fritz, 1997] and the observed δ13C-CH4 of the remaining CH4 (δresidual) if both the δ13C-CH4 prior to oxidation (δunoxidized) can be measured and the CH4 oxidation isotopic fractionation (εMO) is known

display math(5)

[27] The measured water column δ13C-CH4 values and calculated annual net δ13C-CH4 production values from isotope-mass budgets were used as values for the partially oxidized CH4 (δresidual). Anoxic pore water provided the most representative source of unoxidized CH4 (δunoxidized) and was measured throughout the flood season with the permanently installed profilers. Although a wide range for εMO has been reported [cf. Whiticar, 1999] (on average about −10‰, but as low as −30‰), incubations of methanotrophs from the FLUDEX reservoirs yielded εMO within a narrow range (High C reservoir, −13.9‰ at 22°C and −12.5‰ at 30°C; Medium C reservoir, −14.3‰ at 22°C and −14.0‰ at 30°C [Venkiteswaran and Schiff, 2005]).

[28] While there are non-Rayleigh methods of estimating CH4 oxidation that use open systems, steady state supply of production, and branch point assumptions [e.g., Chanton et al., 2009; Kankaala et al., 2007; Monson and Hayes, 1980], given the fully oxic water column and evidence of CH4 oxidation in the shallow subsurface, the Rayleigh approach is a good, though not perfect, approximation of flooded soil processes in FLUDEX.

Error Assessment

[29] Potential error on the rates of CR and NPP was assessed by changing the δ13C values assigned to CR and NPP by ±1‰ (a range >2SD of δ13C-DIC measurements and comparable to reasonable estimates of uncertainty in the end-member values). These adjustments yielded an average of ±70 kg C ha−1 in both CR and NPP. This error was comparable to the errors associated with other budget items such as concentrations, gas exchange, and volumetric measurements of inflow and outflow water [cf. Matthews et al., 2005] and about half of average NPP values. Similarly, potential error on the fraction of CH4 oxidized was assessed by adjusting the δ13C-CH4 value of unoxidized CH4 by ±2‰ (2SD of the measured pore water values) and ±1‰ on εMO (2SD uncertainties). This produced a range of 5 percentage points around the estimates of the fraction of CH4 oxidized.

Results and Discussion

[30] DIC and CH4 were produced in all three reservoirs in all 5 years as a result of the flooding of upland forest soils. Net DIC and CH4 production varied over the 5 years, as did the relative importance of processes contributing to net DIC and CH4 production (Figures 13 and Tables S2S4). DIC concentrations in the inflow from Roddy Lake were similar every year (134 ± 26 µmol L−1, mean ± 1SD, n = 80; Table S1 and Figure S2). Catchment input was only important in the High C reservoir (8–40% of total DIC inputs depending on year) because only this reservoir had a significant area of upslope catchment. DIC concentration of High C catchment input was 690 ± 140 µmol L−1 (geometric mean ± 1SD, n = 16). Reservoir outflow DIC concentrations varied greatly (110–590 µmol L−1, all reservoirs, all years) and were always greater than inflow DIC concentrations measured on the same day even though the residence time of the water in the reservoirs was short (6–9 days). Therefore, DIC production was substantial within all reservoirs in all 5 years.

Figure 2.

DIC concentrations and δ13C-DIC of FLUDEX reservoir inflow and outflows. Flooding began on 22 June 1999, 30 May 2000, 29 May 2001, 02 June 2002, and 04 June 2003.

Figure 3.

CH4 concentrations and δ13C-CH4 of FLUDEX reservoir inflow and outflows. Flooding began on 22 June 1999, 30 May 2000, 29 May 2001, 02 June 2002, and 04 June 2003. CH4 concentrations were insufficient for δ13C-CH4 analysis in reservoir inflow from Roddy Lake and reservoir outflow prior to the seventh week of flooding in 1999.

[31] Over the 5 years of flooding, net DIC production declined markedly, while both net CH4 production and CH4 ebullition peaked in the third year (Figure 1). In the fourth and fifth years, this remained high in one reservoir but declined in the others (Figure 1). Similar year-to-year trends in surface fluxes were observed in a northern Québec reservoir flooded in 2006 [Teodoru et al., 2011, 2012].

[32] Although the DIC concentrations of the inflow were constant with time, the δ13C-DIC values ranged from −18.0‰ to −2.2‰ (Table S1 and Figure S2). During summer, gas exchange and photosynthesis increased the δ13C-DIC of Roddy Lake from the low δ13C-DIC values characteristic of vernal turnover [Chomicki, 2009; Herczeg, 1987]. The δ13C-DIC of the L114 IF stream, used as a proxy for catchment input, was much lower than Roddy Lake input at −25.7 ± 0.8‰ (mean ± 1SD, n = 4). Outflow δ13C-DIC peaked between mid-July and early August each year, and the total range was −21.7‰ to −10.3‰. The δ13C-DIC was always more negative than the inflow (Figure 2). Therefore, the net DIC produced in the reservoirs had a δ13C-DIC value more negative than reservoir outflows. Additionally, these δ13C-DIC values were quite different than the δ13C-CR end-member, making the calculation of CR and NPP rates possible.

[33] CH4 was also produced in all reservoirs in all years (Figure 3 and Table S5). CH4 concentrations in the inflow from the Roddy Lake epilimnion and the catchment input were very low (<0.2 µmol L−1). Outflow CH4 concentrations peaked during August each year and, on average, increased through the first 3 years and declined modestly in the fourth and fifth years (Figure 3). A meta-analysis of 89 reservoirs indicates they were all sources of CH4 to the atmosphere and suggests a year-after-year exponential decline in CH4 fluxes [Barros et al., 2011], whereas Teodoru et al. [2012] suggest CH4 fluxes peak in the fifth year after reservoir construction.

[34] Outflow δ13C-CH4 exhibited different trends in each year. In the first year, outflow δ13C-CH4 (−89.3‰ to −65.9‰) in all reservoirs remained more negative than CH4 in most freshwater systems [Whiticar, 1999]. In the second year, outflow δ13C-CH4 became more positive as the flood season progressed (−99.4‰ to −35.9‰). In the third to fifth years, the δ13C-CH4 in the Medium and Low C reservoir outflow generally remained steady at −50‰ to −40‰; however, δ13C-CH4 in the High C reservoir peaked at −17.0‰, −11.4‰, and −12.7‰ in sequential years (Figure 3). The increase in the High C reservoir δ13C-CH4 occurred concomitantly with a decline in CH4 concentration.

Community Respiration and Net Primary Production

[35] CR rates generally declined year after year in all reservoirs (Figure 4 and Tables 2 and S3). CR rates were greatest (876–965 kg C ha−1) in the first year and declined thereafter. CR over 5 years declined by about half in the High C (51%) and Medium C (47%) reservoirs and by 75% in the Low C reservoir.

Figure 4.

Rates of CR and NPP in FLUDEX reservoirs from 1999 to 2003 calculated from δ13C-DIC budgets.

Table 2. Summary of CR, NPP, Net DIC Production, Net CH4 Production, and CH4 Ebullition Rates for FLUDEX Reservoirs for 1999–2003 (kg C ha−1)
High C Reservoir
Net DIC production702478444481388
Net CH4 production4.611.
CH4 ebullition0.05.619.114.16.5
Medium C Reservoir
Net DIC production797408478325432
Net CH4 production3.08.411.69.08.2
CH4 ebullition0.14.418.13.83.2
Low C Reservoir
Net DIC production703470415306244
Net CH4 production3.410.
CH4 ebullition0.014.719.94.02.6

[36] There was sufficient stored OC in the forested catchments, even in the sparsely treed areas, to support GHG fluxes to the atmosphere for the 5 year span of this experiment. Mineralization of DOC from reservoir inflow from Roddy Lake could not have been a significant source of DIC for several reasons. Reservoir residence times were short (6–9 days). ELA lakes similar to Roddy Lake have residence times greater than 15–20 years, and thus Roddy Lake DOC had already been subjected to extensive microbial and photolytic degradation prior to entering the reservoirs. Additionally, light penetration in the reservoirs was much lower than in ELA lakes due to high reservoir DOC concentrations (mean of 660 µmol C L−1, n = 128, all reservoirs, all years) and reduced photolytic loss rates of the refractory lake DOC even further. In the short reservoir residence times, microbial degradation of recalcitrant DOC would also have been small. Thus, although Roddy Lake DOC could not be distinguished from vegetation and soil decomposition in the reservoirs based on δ13C (δ13C-DOC of inflow water was −26.9 ± 0.2‰, mean ± SE, n = 30), the contribution of this refractory DOC must have been small. DOC concentrations were greater in the reservoirs than the inflow (mean of 440 ± 40 µmol C L−1, n = 43) so DOC produced and subsequently converted to DIC within the reservoirs is included in CR. In these experimental reservoirs, CR resulted primarily from the decomposition of internal OC stores.

[37] NPP rates were sufficiently large to alter the δ13C-DIC values measured in the FLUDEX reservoirs. Although flooded vegetation and soils were the main source of OC for decomposition, net δ13C-DIC production values (mean of −22.4‰, range of −23.9‰ to −18.8‰, n = 15, for all reservoirs, all years) were not similar to δ13C of vegetation or soil (−30‰ to −27‰; Table 2). Although net DIC production declined in each reservoir by 55%, 54%, and 35% over 5 years in the High, Medium, and Low C reservoirs, the δ13C did not exhibit a trend during the same time (Table S3). NPP varied between reservoirs and was likely dependent on mean depth and flow, as well as differences in nutrients supplied by different decomposing vegetation and soil communities.

[38] NPP exhibited similar temporal patterns in the High C (48–195 kg C ha−1) and Medium C (87–207 kg C ha−1) reservoirs where NPP was similar for 3 years and then declined in the final 2 years of flooding (Table S4). NPP was most variable in the Low C reservoir (0–322 kg C ha−1) increasing from the first to third years and then declining to our limit of detection by the fifth year (Figure 4).

[39] Large periphyton sheets, visible during flooding and following drawdown, were not present after snowmelt the following spring. Since a significant portion of periphyton biomass was likely decomposed between flood seasons, periphyton may have been only a temporary C sink. Thus, net DIC production and gas exchange during the flood season may underestimate the total loss of GHGs to the atmosphere as a result of flooding upland landscapes.

[40] When CR and NPP are considered together, 21% ± 10% (mean ± 1SD, n = 15, range 0% to 44%) of CR was removed by NPP. Thus, NPP significantly reduced CO2 loss to the atmosphere, and as such, these processes need to be separated to fully understand the drivers behind CO2 emissions from reservoirs. δ13C allowed us to do this because the δ13C-CR end-member was tightly constrained and changes in δ13C-DIC were large.

CH4 Production and CH4 Oxidation

[41] Net CH4 production in the reservoir is the CH4 remaining after partial oxidation that is released from soils to the reservoir water column. Net CH4 production increased from 3.0–4.6 kg C ha−1 in the first year to 11.4–15.4 kg C ha−1 in the third year and then subsequently declined to 5.2–8.0 kg C ha−1 in the fifth year (Figure 1 and Tables 2 and S4). These rates are comparable to those measured for 4 years in a northern Québec reservoir [Teodoru et al., 2012].

[42] The δ13C-CH4 values of the net CH4 production were very low in the first year, −73‰ to −93‰, and much higher after that, −48‰ ± 10‰ (mean ± 1SD, n = 12, range −65‰ to −26‰) (Figure 3 and Table S4), indicative of either an increased fraction of CH4 being oxidized despite the increase in net CH4 production or a change in CH4 production pathway.

[43] In the first year of flooding, measured water column and pore water δ13C-CH4 values (more negative than −72‰) were similar to expected values of −75‰ to −85‰ for CO2 reduction (calculated from measured vegetation, soil, pore water δ13C-DIC, pH, and temperature). CH4 oxidation or a shift to fermentation can increase pore water δ13C-CH4 values so pore water provides information about processes and end-member values for CH4 production. Measured pore water at 2 cm below the SWI was always anoxic. In the second and third years of flooding, CH4 was again produced by CO2 reduction (water column and pore water more negative than −80‰) in the first month of flooding. Thereafter, most CH4 was produced by fermentation because the δ13C-CH4 of anoxic pore water was more positive than −55‰. In the fourth and fifth years of flooding, most CH4 was produced by fermentation because the δ13C-CH4 of anoxic pore water was always more positive than −55‰.

[44] Having the end-member δ13C-CH4 value in anoxic pore water allows us to assess the role of oxidation in CH4 production. In these reservoirs, the oxic-anoxic boundary was at the SWI and not in the water column, unlike in some reservoirs or many lakes [e.g., Bastviken et al., 2008; Duchemin et al., 1995; Huttunen et al., 2004, 2006; Rudd and Hamilton, 1975]. To separate CH4 production from oxidation, the water column (residual CH4 after partial oxidation) was compared to anoxic pore water (before oxidation). The water columns exhibited little or no variation in δ13C-CH4. Water column δ13C-CH4 in the lowermost 20 cm also matched the δ13C-CH4 of the CH4 flux entering light-exposed benthic chambers, suggesting that CH4 oxidation occurred at or below the SWI. In the absence of light, the δ13C-CH4 of the CH4 flux into the darkened chambers matched the δ13C-CH4 of the anoxic pore water measured with the profiler even though the water in the benthic chamber was not anoxic. The diffusion of photosynthetically derived O2 into the anoxic flooded soil must have played a significant role in CH4 oxidation.

[45] On average, three quarters (75% ± 17%) of the CH4 produced in the anoxic soils was oxidized before being released to the water column as net CH4 production in years 2 to 5. The fraction of CH4 oxidized increased from nil in the first year to about 80% in year 3 and remained high (83% ± 7%) in the final 2 years of flooding. For net CH4 production to have increased from the first to third years while the fraction of CH4 that was oxidized also increased, CH4 production in the flooded soils (methanogenic activity) must have increased by about tenfold.

[46] CH4, produced by either fermentation or CO2 reduction, is much more negative than the organic matter source but is accompanied by DIC that is more positive than the organic matter source. Within the reservoir, any CO2 produced as a result of CH4 oxidation would have a δ13C value more negative than the δ13C-DIC from organic matter decomposition, but this would be partly offset by the more positive DIC contemporaneous with CH4 production. More importantly, the ratio of DIC:CH4 is very high in pore waters. Median soil pore water DIC:CH4 ratio was about 50:1 (25th and 75th percentiles were 30:1 and 100:1) throughout the experiment. Thus, oxidation of the relatively small amount of CH4 to CO2 was a negligible contribution to the DIC pool and could not have been observed in the reservoir water column or mass budgets.

Contribution of Ebullition to Total GHG Flux

[47] CH4 ebullition rates were consistently low in all reservoirs in the first year. CH4 ebullition increased sharply 8–10 weeks after flooding in each of the second to fifth years. The δ13C-CH4 of ebullition increased during each flood season and followed outflow δ13C-CH4 values. Flux-weighted δ13C-CH4 values of ebullition were between −56‰ and −49‰ in years 2 to 5 of flooding (Table S4) and were generally similar to or less than net CH4 production δ13C values. This was pronounced in the High C reservoir in the fourth year of flooding when the flux-weighted ebullition δ13C-CH4 value was −57‰ and the net production value was −26‰. Although ebullitive δ13C-CH4 was more negative than water column δ13C-CH4, ebullitive CH4 was more positive than anoxic pore water and thus affected by oxidation. CH4 captured within the bubbles was the residual after 50% to 75% of the original CH4 had been oxidized. Therefore, bubble formation must have occurred within the top of the soil because CH4 in bubbles was partially oxidized, and pore water was always anoxic with much lower δ13C-CH4.

[48] The depth to which O2 from benthic photoautotrophs diffused into flooded soils must have been severely limited by O2 consumption due to CR. Because CH4 in bubbles was subject to less oxidation than the diffusive flux across the SWI, in general, an increasing fraction of CH4 production was released to the atmosphere from the first to fifth years of flooding in all reservoirs. Thus, total (or gross) CH4 production must have been highest in the third or fourth year given that a greater fraction was oxidized and net CH4 production was greater than other years.

[49] CO2 ebullition was very small in all years and all reservoirs, less than 0.5% of DIC output (Table S4). Ebullitive δ13C-CO2 values were similar to δ13C-CO2 pore water values calculated from measured δ13C-DIC, pH, and temperature.

Organic Carbon Stores as Predictors of GHG Production

[50] Decomposition of OC stores, as measured by CR, ultimately differed little between reservoirs: the 5 year sum of CR for each reservoir was within 5% (Figure 5). The initial hypothesis of the Experimental Lakes Area Reservoir Project [Kelly et al., 1997; St. Louis et al., 2004] and FLUDEX [Bodaly et al., 2004] was that long-term GHG production was positively related to the quantity of OC stored in vegetation and soils. Although the differences between CR in each reservoir increased in each year (9%, 16%, 23%, 24%, and 58%), the reservoirs with the greatest and least CR in each year were seldom the High and Low C reservoir. Within FLUDEX reservoirs, total OC stores were not good short-term predictors of carbon lability.

Figure 5.

Community respiration, net CH4 production, and CH4 ebullition in FLUDEX reservoirs for 1999–2003 expressed as kg CO2 ha−1 of global warming potential (GWP). The 100 year time horizon GWP of CH4 is 23 times that of CO2. For each year, bars are grouped as High, Medium, and Low C reservoirs.

[51] All reservoirs contained a dry forest community, but the High C reservoir also had a moist forest community and the Low C reservoir also had a moss-lichen and an exposed bedrock community. The 5 year decline in CR was greatest in the Low C reservoir but similar in the High and Medium C reservoirs. Litterbag and laboratory incubations of vegetation and soil indicated that the dry forest communities have a greater short- and medium-term OC decomposition rates than the moist forest community [Baril, 2001; Hall and St. Louis, 2004], suggesting that OC lability was different in each community. Further, DOC leachability (ability to leach DOC from parent material) and DOC lability from vegetation and soils likely differed between community types, also suggesting that the soil and vegetation decomposability likely differed between communities [Ferguson, 2000]. Nevertheless, the 5 year total CR was very similar among reservoirs suggesting that, over 5 years, OC was not limiting decomposition. This leaves the role of variables such as temperature and nutrients to be explored as potential limits on CR. Long-term CR will become more related to OC stores as leachability and lability become less important with time as more easily decomposable components are consumed. Detailed year-after-year rates of processes such as these and detailed surfaces fluxes such as those of Teodoru et al. [2012] are required to build process-based scalable predictive models of reservoirs.

[52] Increases in CH4 production and the shift from gas exchange (CH4 more susceptible to oxidation) to ebullition (CH4 less susceptible to oxidation) compensated for much of the decline in CR because CH4 has 23 times the GWP of CO2 [Ehhalt et al., 2001]. The 5 year decline in GHG GWP was 42%, 40%, and 68% in the High, Medium, and Low C reservoirs (Figure 5). Five-year GHG GWP values for each reservoir were very similar: within 1% of each other. Thus, after 5 years of FLUDEX data, the hypothesis that GHG emissions would be positively correlated with preflood OC stores did not hold true.

[53] Rudd et al. [1993] and St. Louis et al. [2000] suggested that GHG emissions from reservoirs should be directly related to OC stores not including tree boles. This may be the case if the following confounding factors are constant between reservoirs: lability of OC stores and factors controlling CR; photosynthetic consumption of DIC and factors controlling NPP; gas exchange coefficient; and water residence time. Comparing CR, rather than surface GHG fluxes, to OC stores not including tree boles eliminates these confounding factors. GHG emissions from the surfaces of different reservoirs can only be easily predicted and compared from measures of OC stores if these other factors are similar. Having demonstrated the temporal evolution of CR and NPP in shallow boreal reservoirs, we therefore recommend a monitoring program designed to capture intra-annual and interannual variability of GHG fluxes from reservoirs [e.g., Demarty et al., 2009; Teodoru et al., 2011, 2012] while noting the usefulness of isotopes in determining within-reservoir processes. Such a program, in addition to providing GHG flux data, can also be used to build predictive relationships between reservoir characteristics and GHG fluxes.

[54] These results confirm the hypothesized much greater longevity of GHG fluxes from flooding large OC such as wetlands [Kelly et al., 1997; St. Louis et al., 2004] than those from flooding lower OC upland forests. Single annual estimates of CO2 or CH4 emissions may not be indicative of longer-term trends in decomposition, and thus comparisons between reservoirs and between reservoirs and their preflood OC stores must be undertaken with caution [cf. Barros et al., 2011]. Furthermore, consideration of the relative importance of each of the processes that, when combined, control GHG emissions must accompany attempts to translate results from one reservoir study to the prediction of other sites or scenarios. This research has shown that 20 year old forests contained enough OC to sustain GHG fluxes for at least five flooding seasons—a clear and large change from the C fixation rates of a young, regenerating forest.


[55] R. J. Elgood provided important logistical support. We thank the many people who provided field assistance, particularly J. English, K. T. Maurice, M. A. M. Pinsonneault, M. K. Puchniak, M. A. M. Saquet, and A. Wojtyniak. W. A. Mark provided EA-IRMS analysis and IRMS assistance. This research was funded by the Canadian Foundation for Climate and Atmospheric Sciences, the Centre for Research in Earth and Space Technology, the Climate Change Action Fund, and the Natural Sciences and Engineering Research Council of Canada. Environment Canada's Science Horizons Youth Internship Program funded a laboratory assistant. FLUDEX was funded by Fisheries and Oceans Canada, Manitoba Hydro, and Hydro-Québec.