Lake Ormstrup, located in Denmark (lat 56.326∘, long 9.639∘) (Fig. 1, depth map with GHG sampling locations), is an 11 ha shallow lake (average depth of 3.4 m), with a maximum depth of 5.5 m and a relatively long hydraulic retention time (> 1 year). The lake is eutrophic with high total phosphorus (TP) and chlorophyll a (Table 1; Søndergaard et al., 2023b) and very sparse occurrences of submerged plants.

In June 2020, a NexSens (NexSens Technology, Fairborn, OH, USA) CB-450 data buoy system was deployed at the deepest point of the lake equipped with a NexSens TS210 thermistor string with temperature nodes measuring at four levels: one sensor “in air”, ca. 5 cm above the water surface (but shielded from direct light), and three sensors at -1, -2 and -3 m relative to the water surface. In addition, two Aqua TROLL 500 (In-Situ, Fort Collins, CO, USA) multi-sondes were mounted near the surface (-1.0 m) and at a deeper-water depth (-3.8 m). The near-surface and deeper-water sondes were configured with sensors to measure dissolved oxygen (DO) and water temperature (Tw). The optical sensors were calibrated according to the manufacturer's guidelines and checked on a weekly basis.

The optical sensors of the Aqua TROLL 500 have a built-in wiper mechanism to clean sensor heads to hamper bio-fouling. The wiper function was enabled to perform cleaning in sync with sensor measurements, hence every 15 min. In addition, manual cleaning of sensor heads was done every week, whilst routine manual field monitoring was carried out at the lake. Prior to the deployment of the buoy, and as a validation exercise for the buoy data, weekly manual profiles of DO and Tw were collected at the deepest point.

Periods of stratification and the depth of the thermocline were defined using the R package rLakeAnalyzer (Winslow et al., 2019) based on the density gradient of the water column from the weekly manual profiling of the system. During periods defined as stratified, there were partial mixing events where the depth of the thermocline changed, and there was some mixing of the sub-epilimnetic water and the surface waters, whilst the bottom waters below 3.5 m remained undisturbed.

Water samples for the analysis of chlorophyll a were collected weekly from 20 April 2020 from surface (-0.5 m) water at station 3 (Fig. 1). A volume of water ranging from 0.2 to 1 L was filtered; the GFC papers were preserved for chlorophyll a analysis, which were determined spectrophotometrically after ethanol extraction (Jespersen and Christoffersen, 1987); and alkalinity was measured weekly by Gran titration (Søndergaard et al., 2005). Depth profiles of temperature, electrical conductivity (EC) and dissolved oxygen (DO) were measured manually with an Aqua TROLL 500 probe from every -0.5 or -1 m down to a -5 m depth).

To test for a significant difference among the emissions from the stratified and mixed phase, we used generalised least squares (GLSs) with a variance function to account for the heterogeneity of variance between the phases. In the case of the ebullitive flux, as the collected phase often covered periods including both mixed and stratified phases, there were three categories, namely mixed, stratified, and both mixed and stratified. All analyses were carried out in R version 4.2.1 (R Development Core Team, 2022), and the GLS used the nlme package (Pinheiro and Bates, 2023).

Depth profiles measured weekly from April show that stratification was initiated by 26 May 2020. This may have broken down briefly and may have been established again, visible from the temperature sensors for the buoy on 5 June 2020 (Fig. 2). There was then 12 d of mixing followed by a stable period of stratification with the onset of 14 June 2020 for a duration of 16 d until a mixing event around 30 June 2020. The following 2 weeks had cooler water and a mixed water column and thereafter a ca. 6 d period of stratification from 15 to 21 July 2020. A mixed phase of 2 weeks then followed until stratification was reestablished on 4 August 2020 and persisted until the end of August. Partial mixing is indicated by the buoy data from 21 August 2020, but the weekly manual profile of deeper water indicates that full mixing did not occur until after 25 August 2020. The effects of the stratification and mixing events on the high-frequency DO data measured at -3.8 m are clear, with rapid deoxygenation occurring after the onset of stratification and oxic bottom waters returning when the lake mixed (Fig. 2). The patterns in chlorophyll a also follow, to some degree, those of stratification, with the exception of early spring. Chlorophyll a values were extremely high in spring, peaking at the start of June 2020 and falling gradually (Fig. 2, Søndergaard et al., 2023b). During the periods of stratification, chlorophyll biomass was lower, and when a mixing event occurred the values increased, which is particularly evident in the July mixing periods (Fig. 2).

The concentrations of the dissolved gases showed great variation between near- and below-surface atmospheric concentrations in some cases, with up to an extremely high concentration (over 5 mg CH4 C L-1) in the bottom waters on 30 June 2020. There was some spatial heterogeneity in the surface waters, with the more littoral locations showing the greatest variation and the highest values (Figs. 3, 4, 5). In particular, the most littoral zone, where the water was shallower around 1 m in depth, showed the highest values just prior to, or coincident with, the stratification turnover. Table 2 shows the mean diffusive flux of each gas over the sampling period along with the mean flux in mixed and stratified phases. For CO2 there was a lot of temporal variation in flux dynamics, though not a large difference between mixed and stratified phases in terms of mean values (Table 2). There were some periods of CO2 influx in spring and late summer, and these tended to coincide with the end of a mixed phase and the start of the stratification phase. Nitrous oxide concentrations were generally low (Figs. 4 and 5), with the lake being a source of N2O in the spring period and a sink or a very small source thereafter. The CH4 concentration in the surface waters (Fig. 3) and the calculated diffusive emissions were relatively low but did increase in the stratification periods with higher average values (Table 2 and Fig. 6). There was also some spatial variation with higher CO2 and CH4 diffusive emissions in the shallower sampling locations, both in stratified and mixed conditions (Fig. 6).

The most significant patterns in the GHG concentration were evident in the bottom waters sampled at -4.5 m, which accumulated to very large concentrations of CO2 but particularly CH4 in the periods of stratification (Figs. 3 and 4). The ratio of CO2 to CH4 is illustrative in highlighting how stratification has altered the biogeochemical processes in the hypolimnion with CH4 production becoming more prevalent. For example, on 30 June 2020 after 16 d of stratification, the CO2 : CH4 ratio in the bottom waters was 0.8, whereas 7 d later, after the mixing event, it was 187 at the same depth.

The CH4 bubble flux, presented here as mean values for each of the four transects, ranged from 0.303 to 81.1 mg CH4 C m2 d-1 for the individual transects over the growing-season measurement. There is a very clear, statistically significant impact of stratification on the ebullitive efflux of CH4, with stratified periods showing significantly higher levels of emission (Fig. 7 and Table 2). In addition, there was a difference in average emissions among the different transects, with those with a lower average water depth (T2 and T4) having lower emissions than the transects with chambers over deeper water (T1 and T3) (Fig. 7). The samples collected from the chambers reflect 2 weeks of bubble and diffusion collection, and the quantification of the flux is therefore an average of the period of chamber deployment, which was 2 weeks, or in one case a single week (Fig. 7). This 2-week period on occasion covered both stratified and mixed phases, and on these occasions efflux was intermediate between purely mixed and stratified periods (Table 2 and Fig. 7).

Scaling up the results to total flux of gases from the whole lake over the period of study and including the estimated emissions from two turnover events show a very different effect of stratification on the balance of the types of emissions for the three gases. The majority of CH4 emissions (56 %) result from the two short-lived turnover events (Fig. 8), whereas their contribution to CO2 and N2O emissions was 5 % and 1 %, respectively.

Fluxes of CO2 and N2O were mostly diffusive, which represented 95 % of emissions of both gases. The methane diffusive flux was 14 % of the total emission, whereas CH4 ebullition was more than twice as much at 29 % of the total CH4 emission. In terms of global warming potential, CO2 and CH4 emissions were comparable, but the contribution of the turnover efflux was the dominant factor for CH4 emissions.

Diffusive emissions did not, on average, show a strong stratification effect (Table 2). In particular, variation in N2O emissions did not match patterns of stratification, with emissions more directly related to nitrate concentrations (Audet et al., 2020), as reflected by the fact that the lake is a sink of N2O in late summer when nitrate was below detection limits for several weeks. There were peaks in emissions of CH4 and CO2 at the end of stratification periods, particularly in the shallower-water sampling points (Fig. 6). There were periods of influx of CO2, which coincided somewhat with periods of stratification, but the pattern was not consistent as other factors, for example, chlorophyll a concentration, also play a role.

Littoral zones can have markedly different GHG dynamics than deeper zones due to shallower water having lower pressure (Wik et al., 2013), less time for CH4 oxidation (Bastviken et al., 2008) or abundant plants, which influence a range of biogeochemical processes (Davidson et al., 2018; Esposito et al., 2023). It is therefore possible that littoral zone dynamics could cause these differences. However, the increase occurred at all three sampling points at the end of June 2020, which indicates a lakewide driver, and the peak may represent the start of mixing after stratification. Strong winds were measured on 29 and 30 June 2020 (Søndergaard et al., 2023b), coincident with these increased littoral emissions. These winds would have caused lateral movement of the surface water, causing an upwelling of bottom water, rich in CH4 and CO2, in the littoral margins at the opposite end of the lake. Thus, whilst we do not have direct evidence, it seems more likely that these increased emissions in the littoral zone were driven, at least in part, by the upwelling of GHG-rich bottom waters.

In contrast to the diffusive flux, the ebullitive emission of CH4 shows a very clear response to stratification with an order of magnitude difference in emissions between periods where the sampling reflected purely mixed or stratified periods (Table 2 and Fig. 7). The 2-week resolution of the sampling meant that some samples covered both stratified and mixed phases, and these samples had intermediate fluxes, as they covered both low- (mixed) and high-emission (stratified) periods. The spatial variation in ebullition is also illustrative of the impacts of stratification and the role of anoxia in shaping CH4 fluxes. The two transects with the largest mean and maximum depths (T1 and T3) had the largest emissions, with the deeper of the two (T1) having the highest emissions and showing the largest relative increase during the stratification phases. This pattern is different from that found by some other studies where bubble emissions were larger in shallower water (Wik et al., 2013), although in this and another study (Sø et al., 2023), there was an increase in bubble flux in deeper water in late summer. The deeper water at Ormstrup experienced anoxia early in the season, resulting in locations with deeper water having higher ebullition rates than shallower areas. This is at odds with ideas stemming from the metabolic theory of ecology stating that temperature (Yvon-Durocher et al., 2014), in particular at the sediment surface (Bartosiewicz et al., 2019), can be used to predict CH4 efflux. Whilst CH4 production is temperature dependent at the cellular level, CH4 emissions are rather independent of the sediment temperature; for example, in the first 2 weeks of July 2020 emissions were low, and the sediment surface temperature was relatively high. Thus, temperature alone is a poor predictor of ecosystem-scale CH4 emissions.

It should be noted that the method used to estimate bubble flux here, where floating chambers are sampled every 2 weeks, is a “less-than-perfect” method, which in nearly all cases will underestimate ebullitive flux. Logistical and financial constraints make continual sampling difficult, and here we balanced these constraints against the greater time required to apply more accurate methods, such as bubble traps (Wik et al., 2013) and automatic-flushing chambers (Bastviken et al., 2015). The variability of bubble flux in space and time is such that using measurement from a shorter period of 1–2 d can result in a larger error in the estimation of emissions than results from the longer-term deployment of a static chamber (see Tables S1, S2 and Fig. S1). The results in Fig. 7 show that sampling a single week a year or even more regular monthly sampling of a shorter duration would be unlikely to accurately characterise ebullition. Bubble traps have been used on longer terms, but in eutrophic systems they can suffer extensive biofouling, which can impede their use. Thus, the continuous monitoring of ebullition using static chambers with known biases was deemed the least-worst method available, but we acknowledge that ebullitive emissions are underestimated. We further acknowledge that this approach of static chambers should, where possible, be replaced by other methods to estimate ebullition, such as automatic-flushing chambers. It is difficult to compare the mean values of emission with other studies as there are different scales of measurement both in space and time. However, comparing the values for ebullition recorded here with other longer-term studies carried out in lakes using bubble traps (Burke et al., 2019; DelSontro et al., 2016) shows higher values recorded at Lake Ormstrup compared with other lakes but lower values that have been measured in ponds (Ray and Holgerson, 2023; DelSontro et al., 2016), the latter being known to have higher emissions of CH4 (Holgerson and Raymond, 2016).

In addition to the diffusive and ebullitive emissions, the turnover flux, which consists of the gases accumulated in the hypolimnion being released upon turnover, was also estimated, with a correction of CH4 oxidation applied. There were two major turnover events at the end of June and in late August 2020, which were preceded by 16 and 22 d of stratification, respectively. It was not possible to directly measure turnover fluxes, as they are relatively discrete events where the efflux likely occurs over the course of a few hours or a few days (Søndergaard et al., 2023b). Thus, the efflux estimation is based on a series of assumptions and thus must be treated with caution. Notwithstanding this uncertainty, we can be confident that the turnover flux represents a very large proportion of the total emission of CH4 from Lake Ormstrup over the growing season. We estimate it contributed more than 50 % of growing-season CH4 emissions and 5 % of CO2 emissions. This highlights a very significant, and difficult to measure, contribution to GHG emissions from lakes undergoing temporary stratification, which are among the most common lake type in Denmark (Søndergaard et al., 2023a).

The results here suggest that GHG dynamics were driven both directly and indirectly by the stratification patterns and the anoxia it induced in the bottom waters. At Lake Ormstrup the thermal stratification of the water column quickly led to anoxia, with only a matter of hours to days for the oxygen to be consumed once the bottom waters were isolated (Fig. 2). The ratio of CO2 : CH4 shows how this promotes CH4 production over CO2 production in the stratification phase (see Fig. 9). In addition to promoting CH4 production, such conditions would preclude, or severely limit, oxic CH4 oxidation, which has the potential to consume a large proportion of CH4 produced in the anoxic sediments (Bastviken et al., 2008), though anoxic consumption of CH4 can still occur (Blees et al., 2014). The raw emission data do not provide any direct information on the balance of production versus oxidation, but the CO2 : CH4 ratio suggests there was a marked shift to conditions where methanogenesis was the dominant process and where there was reduced CO2 production. Studies have shown that CH4 oxidation can consume large proportions of the CH4 produced under hypoxia (Saarela et al., 2019), and it is possible that intense CH4 oxidation occurs at the thermocline during the periods of stratification at Lake Ormstrup, but this was not directly measured at the lake. In addition to the more direct effects of anoxia, there may be some indirect effects of the patterns of stratification and mixing that promote greater GHG emissions. It has recently been reported how nutrient dynamics at Lake Ormstrup were altered by the lake stratification (see Søndergaard et al., 2023b, for full details); of relevance here is the impact on chlorophyll a, which saw a large spring peak after which the abundance tracked the stratification and mixing regime, with a time lag. There was a general reduction, or at least no increase, as the stratification period progressed, perhaps due to nutrient limitation in the epilimnion. Upon mixing there was generally an increase in chlorophyll a, though the weekly sampling resolution makes this difficult to assess. Chlorophyll a and the labile dissolved organic carbon (DOC) that result from abundant chlorophyll a have been shown to be associated with higher diffusive and ebullitive CH4 emissions (Davidson et al., 2015; Beaulieu et al., 2019; West et al., 2012; Zhou et al., 2019). It is not possible to say here whether a stable summer-long stratification would have led to decreased chlorophyll a as nutrients became limited due to their isolation in the bottom waters, and reliable high-frequency chlorophyll a data are required to convincingly demonstrate this phenomenon. Notwithstanding these uncertainties, it may be the case that the temporary stratification observed here, interspersed with mixing events, represents a “sweet spot”, providing both the resources, i.e. chlorophyll a and the labile DOC it produces, and the optimal conditions (anoxia) for CH4 production.

Predicting climate change effects on GHG emissions in a future warmer world is not straightforward, as there are multiple interacting drivers which combine to shape the GHG emissions of lakes. However, this study suggests that temporary stratification, which is increasingly recognised as prevalent in ponds and shallow lakes (Holgerson et al., 2022) and is likely to become more common with continued climate change impacts (Woolway and Merchant, 2019), is likely to increase GHG emissions. This will in particular be the case in more eutrophic systems where abundant algal-derived dissolved organic matter can fuel CH4 production (Zhou et al., 2019).

The combination of high-frequency data on water temperature and dissolved oxygen with weekly measurements of GHGs increases the reliability of the findings presented here. Up until relatively recently it has been assumed that for shallow lakes, such as Lake Ormstrup, stratification is not an important feature. Sampling has therefore focused on the surface layers of water bodies, using dissolved concentrations of gases or floating chambers to characterise flux (e.g. Davidson et al., 2015; Audet et al., 2020; Peacock et al., 2021). Thus, most studies have overlooked bottom waters and do not have the temporal resolution required to capture turnover flux emissions from surface measurements. Furthermore, whilst many studies now include estimates of bubble emissions of e.g. CH4 (Bergen et al., 2019), the necessary temporal resolution to accurately characterise ebullitive emission is not well established. The finding here indicates that in such dynamic systems, near-continuous measurement is desirable, and short-term collection over 1 or 2 d could cause massive over- or underestimates of CH4 ebullition.