Sediment respiration drives circulation and production of CO2 in ice‐covered Alaskan arctic lakes

The goals of our study were to (1) quantify production of CO2 during winter ice‐cover in arctic lakes, (2) develop methodologies which would enable prediction of CO2 production from readily measured variables, and (3) improve understanding of under‐ice circulation as it influences the distribution of dissolved gases under the ice. To that end, we combined in situ measurements with profile data. CO2 production averaged 20 mg C m−2 d−1 in a 3 m deep lake and ∼ 45 mg C m−2 d−1 in four larger lakes, similar to experimental observations at temperatures below 4°C. CO2 production was predicted by the initial rate of loss of oxygen near the sediments at ice‐on and by the full water column loss of oxygen throughout the winter. The time series data also showed the lake‐size and time dependent contribution of sediment respiration to under‐ice circulation and the decreased near‐bottom flows enabling anoxia and CH4 accumulation.

Rates of ecological processes in lakes are often disproportionately high compared to their area thus these ecosystems represent biogeochemical hotspots in the landscape. In particular, processing of carbon by inland waters is now recognized as a significant component of regional and global carbon dynamics (Cole et al. 2007;Raymond et al. 2013;Wik et al. 2016). Despite long periods of ice cover in arctic and sub-arctic lakes and the potential for significant accumulation of CO 2 and CH 4 , annual budgets that include the contribution from winter are rare and primarily from shallow lakes (Karlsson et al. 2008(Karlsson et al. , 2013Jammet et al. 2015;Denfeld et al. 2018). Under-ice transformations of organic matter may be considerable in arctic lakes. Emissions of CO 2 near ice-off can be five times those during the subsequent summer period, and the near-surface supersaturation observed during summer may result, in part, from partial retention of the gases produced during winter (Kling et al. 1991(Kling et al. , 1992S. MacIntyre unpubl.). In addition to the need for quantifying winter respiration and accumulation of CO 2 and CH 4 , extrapolating to other lakes across the landscape requires: (1) relating emissions to organic substrates (Wik et al. 2016;Gudasz et al. 2017), (2) approaches to compute CO 2 production when access during winter is difficult, and (3) extending prior understanding of how respiration and sediment heat fluxes structure the water column under the ice to moderate accumulation and emissions of CO 2 and CH 4 . Mortimer and Mackereth's (1958) pioneering under-ice study that linked hydrodynamics and biogeochemistry provides a framework for developing a mechanistic understanding of controls on winter production and distribution of CO 2 . Heat fluxes from the sediments and solutes, such as dissolved inorganic carbon, produced by sediment respiration increase the density of the water at the sediment-water interface. This water flows down slope as a density current with slow upward flows expected to balance the downward ones such that an overturning circulation develops (Fig. 1). While depletion of dissolved oxygen at deep locations due to the downslope movement has been described (Terzhevik et al. 2009), links with accumulation and vertical distribution of CO 2 and CH 4 and with potential for evasion at ice-off have not been made.
We used autonomous instruments deployed throughout the winter in five arctic lakes and profile data obtained before ice-on, within a month of ice-on, and under the ice in spring to quantify respiration, CO 2 production, and determine under-ice circulation and extent of anoxia. We tested the hypothesis that winter CO 2 production can be estimated from the rate of drawdown of dissolved oxygen near the sediment-water interface during the initial period of ice formation and from our results infer that CO 2 is produced primarily from sediment respiration. We illustrate under-ice thermal and density structure as well as the time scales for upward flows in a 150 ha and a 1.6 ha lake. Via detailed analysis, we show that the increase in near-bottom density which drives the under-ice overturning circulation is initially from heat flux in the larger lake and solute flux from respiration in the smaller lake. The rapid near bottom increase in density in the smaller lake created conditions favorable for the onset of anoxia and CH 4 accumulation.

Study sites
Five oligotrophic and net heterotrophic kettle lakes in the northern foothills of the Brooks Range, Alaska, were instrumented and sampled between September 2012 and July 2016 ( Table 1). Attributes of lakes in the region including surface area, depths, landscape age, water chemistry, and light attenuation are described in Luecke et al. (2014). The surface area of the basins we studied ranges from 1 ha to 150 ha and their depths from 3 m to 24 m; the depth of the upper mixed layer and thickness of the thermocline during the ice free period are larger for Toolik Lake with its greater wind exposure than for smaller lakes (Luecke et al. 2014). Ice-on and ice-off dates vary with lake size and by year, but typically the lakes are ice-covered by early October and become ice free in June. Snowmelt typically begins in early May, a month prior to ice-off. Ice thickness is spatially variable in each lake and for all lakes the maximum tended to be between 1.0 m and 1.4 m. Fig. 1. Hydrodynamic processes and stratification in lakes under the ice. (a) Sediment heating of the overlying water at temperatures below 48C creates density currents which flow downslope; respiration at the sediment-water interface contributes by producing solutes which also increase density (Mortimer and Mackereth 1958). Upward flows balance the downslope flows resulting in an overturning circulation (Welch and Bergmann 1985;Malm 1998;Rizk et al. 2014). (b) Heat and solutes become stratified as warm water enriched with the products of respiration flows to the deepest points of lakes. (c) As sediment heat fluxes and respiration rates decrease over the winter, later intrusions flow at progressively shallower depths. With reduced flow into the lower layer, anoxia develops and CH 4 may accumulate. Table 1. Characteristics of the lakes, their sub-basins, and watersheds: WA is watershed area; Area is lake or sub-basin surface area. Z max and Z mean are maximum or mean lake and sub-basin depths, respectively; OC shallow and OC deep describe the organic carbon content by weight percent of lake sediments corresponding to shallow and deep sediment core sites; S-Resp are respiration rates measured on sediment cores collected from shallow and deep sites in the autumn; O 2 change-lake is the rate of drawdown of O 2 measured immediately following ice on using data from the deepest logger in each lake or sub-basin.
Analysis was conducted in two of Toolik Lake's sub-basins, Toolik Main (TM) and Toolik Inlet Bay (TIB), and in the larger of Lake E1's basins (East).  * From Luecke et al. (2014) and ARC LTER database. Landscapes were glaciated over 250,000 yr ago (E5 and E6), 60,000 yr ago (E1), 12,000 yr ago (N2), and 12,000 yr and 60,000 yr ago (Toolik). Bathymetry and watershed area from Jason Stuckey and Randy Fulweber with bathymetry of smaller lakes done in 2013 and Toolik Lake in 2008. Watershed area for Toolik Lake from Kling et al. (2000). † Standard deviation and number of years in which the measurement was made are in parentheses. ‡ Cornwell and Kipphut (1992). Other data personal communication, Anne Giblin. § Lake was experimentally fertilized: N2 (1985-1991), E5 and E6 (2001-2013).

Latitude
k Cores were from 5 m sites. Lower value is from control site, higher is mean from fertilized side of the lake.

Instrumentation and sampling
Meteorological data and high sensitivity biogeochemical arrays used below and through the ice in Toolik Lake are described in Cort es et al. (2017). One array in all lakes was at the deepest part with one or more additional arrays near the margin. Each array included instruments to measure temperature (RBR 1050, 1060, and SoloT: 6 0.0028C), specific conductance (SC) (Onset U24: 6 5 lS cm 21 ), and dissolved oxygen (DO) (PME MiniDOT: 6 0.3 mg L 21 ) deployed at 4-7 depths, including a DO sensor within a meter of the bottom (MacIntyre and Cort es 2017). Profile data were obtained using a Hydrolab DS5, JFE Advantech conductivity, temperature, depth (CTD) profiler, and Yellow Springs Instruments Castaway CTD with the latter used monthly over the winter . Water samples for CO 2 and CH 4 were collected in all lakes in mid-September prior to ice-on, in November about one month after ice-on, and in late April or early May prior to snowmelt. CO 2 and CH 4 concentrations were measured with the headspace method and gas chromatography (Kling et al. 1991). Gas samples from 2012 to 2013 were injected into glass vials after displacing a saline solution; those in subsequent years were injected into evacuated exetainers. Sediment respiration rates were measured on cores taken from the deepest locations in the lakes and sites along the lake margins in fall. Sediment incubations were conducted in the laboratory under ambient lake temperatures following Daniels et al. (2015).

Computations
Density was computed based on temperature and salinity, with salinity computed from measurements of major ions for all the lakes and regressed against specific conductance . Data from fixed instrument moorings were supplemented with profile data to improve accuracy and create a more detailed grid of SC for density calculations. Isotherms and isopycnals, lines of constant density, were computed from linear interpolation of time series temperatures and densities. Upward velocities were computed from the rate of rise of isopycnals. We created a grid of vertical points throughout the water column, calculated the distance of each grid point from the upper density discontinuity caused by the presence of ice or appreciably colder water, and computed the time scales of rise by dividing each distance by velocity (Fig. 2). We computed the CO 2 produced over the winter from the sum of the moles in volume strata computed from the mid-point between measurements, less the moles present just prior to ice-on in September and divided by lake volume when we sampled in late April/early May. CO 2 production was also estimated from the initial rate of drawdown in dissolved oxygen obtained from the oxygen sensors closest to the sediments (Supporting Information Fig. S1). We multiplied this rate by the sensor height, the area of sediments below 3 m, and the time from ice-on until our first sampling in spring. We selected 3 m as rocky substrata decrease below 3 m. Full lake oxygen loss was computed in Toolik Main, Lake N2, and in Lake E1, in years when we had at least five DO sondes in the smaller lakes and six in Toolik Lake, based on the mass present at ice-on and that present on our first sampling date in spring. The latter was calculated from the product of concentration and volume in the depth intervals between DO sensors. To convert these oxygen-based calculations to CO 2 , we assumed a respiratory quotient of 1. Production was normalized to mean concentration by dividing the moles produced by lake volume in late April/early May prior to snowmelt. Production on an areal basis was obtained by dividing the mass produced by lake surface area during ice-free conditions.

Results
Immediately after ice-on, near bottom temperatures increased (Fig. 2). This increase provides evidence for downward flowing gravity currents induced by the sediments heating overlying water (Fig. 1). Specific conductivity increased (not shown) and oxygen concentrations decreased (Supporting Information Fig. S1) indicating the further modification of near-bottom waters by respiration. The rising isotherms throughout the winter in Toolik Lake support the inference of upward flow which balanced the downward flows (Figs. 1, 2a). In Lake N2 the water column warmed for a short period and then cooled. Isopycnals rose from the lower to the upper water column in both lakes until arrested by increasing ice thickness and intensified temperature gradients below the ice (Fig. 2b,d). In Toolik Lake their rate of upward rise at times was more rapid than that of the isotherms, and in Lake N2, their upward rise occurred despite falling isotherms. These differences occurred because solutes, when present, make an appreciable contribution to density at temperatures near 48C in freshwater (Supporting Information Fig. S2), and continuing sediment respiration produced solutes which contributed to flows (Fig. 1).
Rising isopycnals are indicative of the upward flows that are a component of the overturning circulation (Figs. 1, 2). Initially, the upward flow was primarily driven by the increased temperatures from sediment heat flux in Toolik Lake and by the increased SC from respiration in Lake N2 (Fig. 3, Supporting Information Fig. S2). In both lakes, the later rise in isopycnals resulted from the solutes produced by respiration. Observations were similar in Lake E1 where we also had sufficient SC sensors for the analysis (data not shown). Thus, ongoing respiration modified the density structure independently of heat fluxes from the sediments. Sediment respiration drove the overturning circulation in the two smaller lakes whose sediments, based on the sediment respiration rates from cores, were more organic rich than those in Toolik Lake (Table 1).
Near the sediment-water interface, the rate of upward movement of isopycnals decreased after ice-on as the nearbottom density gradient increased, and the time scale of upward movement near the bottom increased to over 9 months (Fig. 2b,d). Anoxia developed within a month in Lake N2 (Figs. 2d, 3g). Mid-water column in both lakes, the rising time tapered from weeks to 6 months (Fig. 2b,d). The depth dependent changes in rise time indicate reduced ventilation of the lower water column and a larger volume of water intruding above the stronger near-bottom stratification (Fig. 1c). The rise times remained higher for a longer period in winter in Toolik Lake than in Lake N2. The between-lake difference results, in part, from continued introduction of heat and solutes mid-water column in Toolik Lake as opposed to only solutes in Lake N2 (Supporting Information Fig. S2). Continued sediment heat fluxes and persistent downslope flow provide an explanation for the delayed onset of anoxia in Toolik Lake in 2013-2014. However, anoxia was observed to occur over a month earlier in the following winter. The step-like shapes of the DO and CO 2 profiles by late April 2014 provide support for temporally varying intrusion depths (Fig. 3). The level isopycnals in Lake N2 imply lack of exchange with the water above. The continued near bottom rise of isopycnals in Toolik Lake indicates continued upward flow in the full lake.
During spring, additional hydrodynamic processes became operative and modified the rising times. These include penetrative convection ($ day 490), snowmelt ($ day 500), and subsequent penetrative convection ($ day 525). Mixing was not complete by ice-off and solar heating intensified stratification immediately thereafter. The incomplete mixing implies that not all the gases produced over the winter would ventilate, with mixing near the bottom impeded by the increased solute stratification from respiration.
The increased near-bottom SC and the decreased DO are accompanied by increased CO 2 (Fig. 3). For all the lakes, CO 2 produced over the winter was similar to or somewhat less than predictions based on the initial rate of DO weeks, and months) in Toolik Lake and neighboring Lake N2 from fall 2013 to spring 2014. Black dots on the y-axis (a, c) indicate the depth of temperature loggers and red dots (b, d) the depth of SC loggers and typically paired DO sensors. Red arrows mark the onset of anoxia. Gray shapes above the time series data indicate ice thickness and duration. Isotherms and isopycnals were interpolated over 4-d intervals. Isopycnals every 0.0025 kg m 23 and 0.005 kg m 23 in Toolik Lake and Lake N2, respectively. Density above 1000 kg m 23 results from the additional component from solutes. Isotherms are computed at 0.18C and 18C intervals below and above 3.58C, respectively. drawdown at the sediment water interface (Fig. 4). These results imply that the CO 2 primarily was produced by sediment respiration as opposed to respiration in the water column. Loss of oxygen could occur by oxidative processes, such as nitrification (Powers et al. 2017), and confound our interpretation. However, concentrations of both NH 4 and NO 3 are near the limits of detection or at most a few lM in Toolik Lake and nearby lakes indicating that appreciable nitrification over the winter is unlikely to increase the loss of oxygen (MacIntyre et al. 2006;Luecke et al. 2014; ARC LTER database). Rather than the CO 2 that flowed downslope remaining near the sediment water interface, however, the increases higher in the water column in all the lakes resulted from the upward fluxes which balanced the downward flowing gravity currents and from mid-water column intrusions (Fig. 3d,h). The similarity in our calculation of CO 2 produced based on the full water column decrease in oxygen (Fig. 4, loss bars, Supporting Information Fig. S1) relative to that measured and that computed from the initial loss of DO at the sediment water interface (Fig. 4, rate bars), provides further evidence for the upward movement of solutes produced by respiration as opposed to local respiration. We infer that the CO 2 maxima above the bottom in Lake N2 was induced by intrusions above the sharp near-bottom discontinuity (Fig. 3f,h) with similar observations in the other small lakes. Near the bottom, where anoxia was present, CH 4 accumulated (Fig. 3).
Accumulation of CO 2 over the winter was similar in each lake over the 4 yr (Fig. 4). As measured accumulation was similar to or less than that calculated from the initial rate of DO drawdown, we infer that the CO 2 is largely produced by respiration in the sediments and that the computation based on drawdown of DO provides an assessment integrated for the full lake. The inverse relation between mean CO 2 concentration and mean depth of the lakes results, in part, from the larger volume of water in the deeper lakes. Sediment respiration rates, based on sediment cores, tended to be higher from deeper sites than sites along the lake margins (Table 1). Results from Toolik Inlet Basin likely differ from this pattern due to their proximity to Toolik Inlet. With the exception of the main basin in Toolik Lake, where the margin site was at 10 m rather than $ 2 m as in the other basins, rates at margin sites were 1.4-3 times higher than rates computed based on the initial drawdown in DO (Table 1). As the values of CO 2 production from the rate measurements are similar or higher than measured values (Fig. 4), these comparisons Fig. 3. Profiles in Toolik Lake (TM, upper panels) and Lake N2 (lower panels) of (a, e) temperature (T), (b, f) specific conductance (SC); (c, g) dissolved oxygen (O 2 ); (d, h) concentration of CO 2 in September 2013 prior to ice-on, in November 2013 approximately one month after ice-on, and in late April 2014 prior to snowmelt. CH 4 did not accumulate in Toolik Lake over the winter. In Lake N2, the concentration of CH 4 in late April was 0.05 lM from the surface to 8 m whereas it increased to 164 lM in the anoxic water at 10 m. We measured T, SC, and O 2 at meter intervals with a Hydrolab DS5. The increases in SC and CO 2 and decreases in O 2 are indicative of respiration, and the resulting CO 2 production is predicted from the initial rate of decrease in oxygen measured above the bottom and the full water column decrease in O 2 using the time series data (Rate, Loss, respectively in Fig. 4).
imply that whole lake estimates of CO 2 production from sediment cores are too high, likely because of high spatial variability in respiration rates. The initial rate of SC increase was $ 6 times lower at Toolik Main and in Lake E5 with their lower sediment respiration rates in incubations than in the other lakes (Table 1). Thus, measurements from sediment cores are indicative of respiration potential.

Discussion
Our measurements of CO 2 production and changing hydrodynamics under the ice in five Alaskan arctic lakes indicate that sediment respiration is the primary source of CO 2 , provide a basis for extrapolating to other arctic lakes, and illustrate approaches using high resolution time series data for quantifying production in lakes where access is difficult during the ice-covered period. On an areal basis, accumulation of CO 2 was least for the shallow lake, 4 g C m 22 , and ranged from 8 g C m 22 to 10 g C m 22 for the deeper basins (Fig. 4, inset). These values are in the range of those obtained from temperate and boreal lakes (Ducharme-Riel et al. 2015) and a shallow arctic lake (Ramlal et al. 1994). With these limited observations, it is not clear whether production of CO 2 on an areal basis scales with mean depth as observed by Ducharme-Riel et al. (2015). The lower areal production in shallow lakes or lakes with shallow mean depth may be biased low when weighted by lake surface area in the ice-free period. It may also result from colder temperatures above the sediments and resulting lower respiration rates or a shift to anaerobic respiration once the sediments become anoxic with anoxic area relatively larger due to their small volume.
Daily averaged respiration rates, $ 45 mg C m 22 d 21 for the deeper lakes, were near the mean value predicted for respiration for water temperatures between 08C and 48C (Fig. 4, inset) (Gudasz et al. 2010). Although rates were half as much for Lake E6, they were similar to predictions for oligotrophic lakes at cold temperatures (Gudasz et al. 2010). Areal production of CO 2 was independent of watershed area and implied loading of organic matter and appeared to be independent of prior fertilization as accumulation in recently fertilized Lake E5 was similar to that in unfertilized Toolik Lake (Table 1; Fig.  4). Data are unavailable to assess the contribution from benthic algae, macrophytes, and attached periphyton. These contribute considerably to primary production in summer (Ramlal et al. 1994;Whalen et al. 2008) and respiration by benthic algae accounted for 55% of total CO 2 production in winter in shallow subarctic lakes (Karlsson et al. 2008). Studies of benthic communities and their contribution to sediment organic matter may be warranted to understand controls on CO 2 production in arctic lakes during winter.
The contribution of winter time production of CO 2 relative to that in summer is unknown in these lakes. Prior Fig. 4. Volume weighted mean concentrations of CO 2 produced in four winters (2012-2013, 2013-2014, 2014-2015, 2015-2016) for the six basins studied (blue, Measured), and values predicted from rate of oxygen loss near the sediment-water interface immediately after ice-on (red, Rate O 2 ) and from overall oxygen loss from ice-on until the first day of sampling (green, Loss O 2 ) (only 1 yr of data was available for this calculation in L. E1). Data are arranged according to mean depth (Table 1). Inset: Daily averaged CO 2 production rates (left axis, mg C m 22 d 21 ) and CO 2 accumulation over 220 ice covered days (g C m 22 ), where 220 d is approximate time from ice-on until the first sampling date, vs. mean lake depth. Standard deviations are indicated by vertical bars.
emissions of CO 2 from Toolik Lake during the ice-free period averaged 30 mmol m 22 d 21 over a 9-yr period, which gives an annual average of 3 mol m 22 (Kling et al. 1992). Over a 200-d period under the ice, we found, on average, that 0.8 mol m 22 accumulated (Fig. 4, inset, units converted). Interpretation of these differences is confounded as summer emissions include the ice-off period when fluxes were up to seven times higher than the mean. Our areal values from the rate of DO drawdown were 10 mmol m 22 d 21 . As the ice covered period is three times longer than the ice-free period, and using the above rates, the amount of CO 2 produced over the winter is equivalent to the average emissions reported by Kling et al. (1992). Winter production ranged from 11% to 100% of that in summer in 12 shallow subarctic lakes (Karlsson et al. 2013). The scarcity of data in winter to complete annual budgets highlights the need for more measurements in arctic lakes (Denfeld et al. 2018).
The circulation under the ice is an overturning circulation initiated by descending density currents and balanced by slow upward flows (Figs. 1, 2). In lakes a few ha in size and $ 10 m deep, solutes produced by respiration drove the overturning circulation. In larger lakes, the initial flow was primarily due to the density increases from sediment heat flux and later sustained by solute fluxes from respiration. We extend these observations to take into account respiration potential as measured by respiration rates in intact sediment cores (Table 1). As discussed previously, high respiration potential is correlated with a rapid flux of solutes from the sediments. Depending on bottom slopes, the increased solute flux increases the likelihood that sediment respiration will drive the overturning circulation. It will also cause a more rapid increase in near bottom density with accompanying onset of anoxia and accumulation of CH 4 . Sediment respiration also led to or contributed to the sustained rise of isopycnals above the bottom layer within a month of ice-on, with velocities similar to those computed by Malm (1998) and Rizk et al. (2014). These indicate intrusions, which we expect to be depleted in oxygen and enriched in CO 2 , flowed into the overlying less stratified water (Fig. 1c). We do not have evidence for two overturning circulation layers rotating in different directions as predicted for lakes with sloping margins (Phillips 1970;Rahm 1985). The overturning circulation may not be maintained in shallow lakes where the strong temperature-induced density gradient below the ice constrains upward flow.
Combined hydrodynamic and biogeochemical processes cause an overturning circulation in ice-covered lakes. Controls on sediment heat content at the onset of ice-on are not yet known, but this analysis suggests greater retention of heat in the larger of our study sites (Fig. 2, Supporting Information Fig. S2). The extent to which internal wave motions, as proposed by Kirillin et al. (2012), or other hydrodynamic processes operative at ice-on moderate fluxes from the sediments and resultant rates of circulation are also not known.
We do know that a microbially mediated process, sediment respiration, contributes to circulation for much of the icecovered period. Our observations provide a basis for further modeling which takes into account spatial and temporal variability in sediment heat fluxes, factors regulating sediment organic carbon content, temperature dependent respiration rates, and within lake hydrodynamics as they vary with basin morphometry. Such effort will establish controls on CO 2 production and on conditions conducive to anoxia and accumulation of CH 4 . Increased near-bottom stratification associated with the latter will moderate the extent of mixing at ice-off and the potential for evasion of CO 2 and CH 4 produced over the winter.