Flowing over 2,850 km and receiving water from 19 different European countries, the Danube River discharges into the Black Sea through a three-arm delta, the Danube Delta. It is Europe’s second largest river delta after the Volga wetlands on the Caspian Sea and located at the border of Ukraine to eastern Romania. Runoff from the Carpathian Mountains and the Alps determines the seasonality of the river’s hydrology, where peak discharge in spring is followed by low discharge in autumn (Supplementary Figure S1). These changes in hydrology directly translate into changes in water level in the delta, leading to flooding of the wetland area in spring and subsequent drainage in autumn. Between the main river branches, the delta consists of vast reed areas and shallow flow-through lakes, which are connected to the main branches via small natural and artificial channels (Figure 1). The river water generally enters the delta at the West and flows back into the river in the East close to the Black Sea. However, depending on the water level fluctuations, direction of flow is also known to reverse in some of the eastern channels (Irimus, 2006). With respect to the 20-years average, the Danube River showed below average discharge throughout most of the monitoring year 2017 (Supplementary Figure S1; ICPDR (2019)).

The Danube Delta is a Ramsar Wetland Site and a World heritage site (UNEP-WCMC (UN Environment Programme World Conservation Monitoring Centre), 2021). Since 1998, it is part of the transboundary UNESCO Biosphere Reserve (UNESCO, 2021). The Danube Delta Biosphere Reserve (DDBR, Figure 1) aims to protect the deltas’ diversity of flora and fauna. To this end, about 9% of the deltas’ area classifies as strictly protected core zone with another 48% representing buffer zones surrounding the core areas (UNESCO, 2021), where tourism and reed harvesting are restricted.

About 23% of the 220,000 ha reed area represent potential harvesting areas (Covaliov et al., 2010). The reed, predominantly consisting of Phragmites australis (Hanganu et al., 2002), is either used as cattle food by locals or professionally harvested by private companies during winter. Burning of the remaining crop parts increases future yield and prevents succession by other plants (Covaliov et al., 2010). Reed grows on reed peat, consisting of a network of viable rhizomes. Initially, this reed peat forms an about 1 m thick floating layer with no contact to the underlaying mineral soil. In Romanian, this is called “plaur”. Bottom contact establishes as the peat grows thicker or during low water levels (Hanganu et al., 2002). Both harvesting and burning increases the buoyancy of the plaur and the reed density (Covaliov et al., 2010), yet reduce the abundance of old and broken reed culms, which play an important role in oxygenating the rhizome (Sorrell and Brix, 2013).

As road infrastructure through the delta is lacking, ships and boats are the most important means of transport there. This boat traffic likely enhances gas exchange in some of the small, highly frequented channels, as well as in the main branches.

For this study, we complemented CO₂ and CH₄ measurements with measurements of dissolved noble gases (He and Ar), O₂ and N₂. Noble gases such as He and Ar help to understand the physical changes affecting all dissolved gas concentrations. As most poorly soluble gas in the water (Figure 2), He is affected most strongly by gas bubbles rising through the water column. On their way through to the surface, less soluble gases from the surrounding water replace more soluble gases initially present in the bubble (McGinnis et al., 2006), which is called stripping. We expected to see such a stripping effect for He during ebullition, i.e., enhanced gas transfer to the atmosphere via bubbles.

The solubility properties of Ar are very similar to those of O₂ and comparable to N₂ (about 2:1). It is therefore often used to account for physical changes affecting these gases or to improve measurement accuracy by employing O₂:Ar or N₂:Ar ratios (Craig and Hayward, 1987; Groffman et al., 2006). O₂:Ar ratios are frequently used in the ocean for quantifying net community production (Craig and Hayward, 1987; Tortell et al., 2015) but also in ice cores (Zhou et al., 2014).

In combination, noble gases can indicate the presence of excess air, i.e., an observed surplus of dissolved gas in excess to atmospheric equilibrium concentrations with air-like gas composition (Heaton and Vogel, 1981). This phenomenon is traditionally observed in unconfined groundwater bodies, where air entrapment during a rising groundwater table leads to an increase in hydraulic pressure, allowing the water to take up more gas and thus partially or fully dissolving the trapped gas bubble (Aeschbach-Hertig et al., 1999). This presence of excess air can raise the O₂ and N₂ saturation beyond expectation in river banks and groundwater (Mächler et al., 2013).

The concentration of O₂ represents an indicator for the ecosystem metabolism, i.e., autotrophic versus net heterotrophic conditions. In-situ O₂ concentrations are often used as a measure for net primary productivity, i.e., primary production–respiration. Different methods include the measurement of diel O₂ changes in streams and lakes (Staehr et al., 2012; Hall et al., 2016; Bernhardt et al., 2018).

Since our measurement system was not sensitive enough to resolve small biogenic changes in N₂ from the comparably large N₂ background concentration of air-saturated water, we used molecular nitrogen as a third conservative gas species in this study.

We continuously measured the concentrations of dissolved gases (CO₂, CH₄, He, Ar, O₂, N₂) in the southern part of the Danube Delta during two field campaigns in May and October 2017. The journey of our sensor-equipped houseboat covered the different waterscapes of the delta, i.e., main river branches, small channels and large flow-through lakes (Figure 1). The size of our houseboat prevented accessibility of shallower lakes or narrower channels (Supplemenatry Figure S2). Due to time constraints, the paths in May and October deviate slightly in the northern part of the mapping area.

For the analysis of dissolved He, Ar, O₂ and N₂, we used a membrane-inlet mass spectrometer (MIMS) from Gasometrix (Brennwald et al., 2016) in combination with a submergible pump and a membrane-equilibrator. The total number of datapoints per gas species amounted to ∼8,600 in May and ∼8,900 in October. We filled gasbags with a known composition of pressurized air enriched with CO₂ and CH₄ and used those as standard gas mixture during the field trip. This standard gas and ambient air were measured in regular intervals (every fifth and 10th sample batch, respectively, with 1 sample batch representing five individual measurements) to ensure data accuracy. The partial pressures and concentrations of the gases dissolved in water were quantified from the raw MIMS data following the procedures described in Brennwald et al. (2016). Given the large concentration gradients in the delta, we performed the calculations for individual measurements. Where necessary, e.g., to compare gases measured by different detectors, averages and standard deviations were calculated for the closest five measurements (i.e., 1 sample batch). Standard/ambient air measurements were performed along the journey, therefore introducing regular gaps in the dissolved concentration maps. Ambient air measurements of He, Ar, N₂ and O₂ were processed the same way as other samples and used as quality control. The standard deviation varied <1.7% for Ar and O₂, ∼0.5% for N₂ and <7% for He. Average ambient air measurements were up to about 2% (N₂, Ar, O₂) or 3.5% (He) smaller than global values (see Supplementary Table S1).

In addition to the MIMS, we used a set-up with combined flow-through sensors, which included CONTROS HydroC^® CO₂ FT and CONTROS HydroC^® CH₄ FT sensors (formerly Kongsberg Maritime Contros GmbH, Kiel, Germany; now -4H- JENA Engineering GmbH, Jena, Germany) to measure CO₂, and CH₄, respectively. The setup also included a SBE 45 micro thermosalinograph (Sea-Bird Scientific, Bellevue, WA, United States) to measure temperature and conductivity on-board. For details please refer to Canning A. R et al. (2021). This part of the data is available via the Pangaea database (Canning et al., 2020) and details on CH₄ concentrations were previously published by Canning A et al. (2021). We recorded GPS position, depth and in-situ temperature with a Lowrance HDS5 sonar.

Both CO₂ and CH₄ showed very large concentration ranges and steep gradients across the delta with a hotspot area in the East that persisted across seasons (Figure 3). As the green colors indicate, CO₂ was undersaturated in many parts of the lakes in both seasons with a median across all lakes of 96% saturation in May and 65% in October. We encountered exceptions for the western lake areas close to the inflow, which receive water from wetlands and channels. These channels, as well as the river main branches, were oversaturated in CO₂ across seasons (median in May and October: 279 and 283%, respectively, in the river branches and 1,270%, respectively, 270% in the channels). Compared to the river reaches, the oversaturation in the channels showed a larger range (Supplementary Table S3). The highest concentrations were found in Canal Vatafu-Imputita (median: 3,000% and 4,400% in May and October, respectively, Supplementary Table S4) and the adjoining channels, Canal Busurca (leading north) and Canal Rosu-Imputita (leading south to Lake Rosu, Figure 1). While Canal Busurca showed high concentrations along the whole channel stretch in May, concentration dropped along the channel in October, indicating the reversal of flow direction along the first part of the channel.

CH₄ was oversaturated in all parts of the delta, with the lowest saturations occurring in the main branches of the Danube (median: 10,700 and 18,000% in May and October, Figure 3 and Supplementary Table S3) followed by the lakes (median: 16,600 and 22,300% in May and October, respectively). Channels showed the highest CH₄ saturations (median: 41,200 and 57,700% in May and October, respectively). In contrast to CO₂, the highest CH₄ saturations were not found in Canal Vatafu-Imputita directly, but rather in Canal Rosu-Imputita. Canal Cordon-Litoral showed elevated CH₄ partial pressures that covered a long stretch especially in October. CO₂ was only mildly oversaturated in this channel.

The noble gases, He and Ar (Figure 4), generally showed undersaturation conditions in lakes, with the strongest undersaturation occurring in Lake Isac for both gases in May (median: 77 and 91%, respectively). The strongest oversaturation, in contrast, was found in small channels. While the highest He saturations were observed in the Northern Channels in October (median: 108%), the eastern channels around Canal Vatafu-Imputita showed the maximum Ar saturations in both seasons (median: 108% in May and 112% in October). Along the stretch of this channel, we observed both under and oversaturated conditions with respect to He.

O₂ saturations covered a large range from 19% to 162% in May and 5.2% to 139% in October (Figure 5). The main river branches were slightly undersaturated in both seasons with a median of 90 and 92% in May and October, respectively. In contrast to the other gases, O₂ saturations were highest in the lakes, especially in Lake Isac (median: 147% in May). The CO₂ hotspot, Canal Vatafu-Imputita, had the lowest O₂ saturations. The saturations were as low as 19% in May and 5% in October at that spot and adjoining channels, which is less than 2 mg/L and therefore too low for the survival of most fish species.

The N₂ saturations we observed ranged between 74 and 111% in May and 87–108% in October (Figure 5). N₂ was generally undersaturated in the lakes, while some of the channels showed oversaturation. N₂ saturations were elevated in the channels downstream of small settlements and lodging houses, for example in the channel leading from the main river to Lake Uzlina (Canal Uzlina) and the channel entering Lake Puiu from the west (Canal Caraorman). The strongest oversaturation in both May and October (median: 109 and 108%, respectively) was found in the CO₂ hotspot, Canal Vatafu-Imputita.

The simultaneous analysis of different dissolved gas concentrations at high spatial resolution allows us to identify systematic patterns of concentration anomalies across the waterscapes of the Danube Delta. Comparing reactive with non-reactive gas concentrations, we identify relevant physical and biological processes such as excess air formation (Kipfer et al., 2002), ebullition (Brennwald et al., 2005; McGinnis et al., 2006) and ecosystem metabolism (Staehr et al., 2012). Two features in the concentration maps (Figures 3–5) deserve special attention. In the Eastern channels (Canal Busurca, Canal Rosu-Imputita and Canal Vatafu-Imputita; Figure 1) we observed hotspots of CH₄ and CO₂ concentrations (Figure 3). The clear oversaturation of non-reactive gases such as Ar and N₂ in this area could help us to identify potential drivers of these emission hotspots (Figures 4, 5). By contrast, the survey revealed a general under-saturation of dissolved CO₂ in lakes (Figure 3). Here, changes in diel O₂ concentrations could allow estimating the intensity of photosynthesis (Staehr et al., 2012). However, the observed under-saturation of He, Ar and N₂ (Figures 4, 5) points to a non-biological sink of dissolved gases that could affect primary production estimates based on O₂ concentrations.

A statistical display of the individual measurements in May and October reveals a broad distribution of measured values (Figure 6). The plot summarizes all data grouped according to waterscapes in the left panel, while the right-hand side presents only data from “boxed” areas in Figure 1: the Northern Channels, the hotspot channel Canal Vatafu-Imputita in the East and Lake Isac in the center of the study area. Part of the variability is caused by dramatic concentration changes at the local level (Figures 3, 5). The small channels in the vicinity of Canal Vatafu-Imputita are a case in point: here, the hydrology is characterized by stagnant conditions with very slow flow velocities. Sudden changes in discharge in the main river can even reverse the flow direction and introduce water from the Sulina main branch to an area that otherwise receives water from the delta. In contrast to the main branch, the water from the delta has a strong signature from the reed and elevated greenhouse gas concentrations. Therefore, dissolved gas concentrations of CO₂, CH₄ and O₂ change abruptly along channel junctions. In the Northern Channels, the gradients were more gradual. There, larger concentration changes show a relation to channel junctions. We observed similar broad concentrations shifts in lakes such as Lake Isac. Here, greenhouse gas concentrations were highest in the East, where water was entering from a small connecting channel with Canal Litcov and from riparian reed beds. O₂ showed the opposite trend with lowest concentrations in the East. Wind-driven gas exchange drove re-equilibration and therefore, the dissolved gas signatures supplied by the lakes to the outflow channels changed gradually with increasing distance from the lake source.

The comparison of dissolved gas patterns across waterscapes yields valuable diagnostic insights (Figure 6). Focusing on non-reactive gases, the reaches of the Danube River remained close to saturation, whereas parts of the channels and most prominently Canal Vatafu-Imputita remained clearly oversaturated with respect to Ar and N₂ during both seasons. Therefore, an effective mechanism of excess air formation seems to be present in some of the small channels that form hot spots of potential greenhouse gas emissions. By contrast, lakes in general, and especially Lake Isac, showed a tendency towards undersaturation in their non-reactive gas concentrations. This undersaturation in He and N₂ in Lake Isac was more pronounced in May (77 and 81% for He and N₂, respectively) compared to October (∼90% for both He and N₂) and indicates that a stripping mechanism is active in the transfer of unreactive gases to the atmosphere.

As expected, the O₂ and CO₂ patterns revealed heterotrophic characteristics of the main branches of the Danube River (Figure 6) with oversaturation of CO₂ and general O₂ deficits. The channels showed the same heterotrophic pattern but with a very broad distribution that could potentially be traced to their interaction with the littoral zone and the reed beds. Lake Isac, by contrast, showed evidence for intense photosynthesis in the summer, with around 150% O₂ saturation and lower autotrophy in October with about 110% O₂.

To discuss possible sources and sinks of unreactive gases, we will address a few hypotheses that could help identifying the role of different physical processes in generating the observed patterns in the Danube Delta (Figure 7). In the context of excess air in channels, several studies documented the role of water table fluctuations in unconfined groundwater bodies (e.g., Kipfer et al., 2002; Mächler et al., 2013). In the Eastern Danube Delta, however, floating reed vegetation is a much more prominent feature of the littoral zone than sand bars or riparian soils, which calls for another supersaturation formation hypothesis. In lakes, the release of methane bubbles at the sediment-water interface could drive gas ebullition to the atmosphere and result in dissolved gas deficits (McGinnis et al., 2006). Alternatively, accumulating oxygen bubbles at the surface of submerged macrophytes would provide an additional pathway for the loss of unreactive gases in the water column (Long et al., 2020).

The solubility of gases decreases with increasing temperature. The temperature sensitivity of this equilibrium process varies between different gases (Figure 8). While the solubility of He decreases by about 6% when the water temperature increases from 10 to 30°C, Ar solubility will diminish by 33%. This contrasting behavior can be used to estimate the amount of excess air, i.e., the excess in dissolved atmospheric gases, caused by dissolving air bubbles in water. Plotting the data from our surveys in Figure 8 reveals significant excess air components in the samples of Canal Vatafu-Imputita during both seasons, while the Northern Channels only show consistent excess air in October. The data from Lake Isac, by contrast, indicate significant gas loss (Figure 8).

Knowledge regarding the geomorphological and ecological context of the different sites helps in the identification of processes that cause the observed patterns in Figure 8. The Northern Channels are framed by riverbanks representing a potential setting for excess air formation via water table fluctuations (Supplementary Figure S8). During rising water table, entrapped bubbles in the unsaturated zone dissolve and increase local concentrations above saturation (Kipfer et al., 2002). Both campaigns in spring and autumn took place during falling water level (Supplementary Figure S1), when we expect previously enriched water to exfiltrate from the riverbanks back into the channels (Figure 7A). Once in the channel, gas exchange ensures re-equilibration of the oversaturated water with the atmosphere. In addition, tourist boats navigate the Northern Channels extensively from spring through summer and autumn, enhancing gas exchange, but defining a causal link would require time-resolved monitoring of excess air and boat traffic.

Canal Vatafu-Imputita, on the other hand, is situated in a depression zone framed by permanently inundated reed beds (Supplementary Figure S8). We were thus initially surprised to find supersaturated conditions at this location, especially since it is one of very few locations showing excess air in both seasons (Figure 9). While the excess air signal in October was comparable to the Northern Channels, the equilibration temperature was about 5°C lower (Figure 8; Supplementary Figure S6). As there are no adjacent riverbanks and rather little traffic at Canal Vatafu-Imputita, we found only two potential drivers for the excess air signal at this location: rapid heating or convective gas addition via reed vegetation.

As indicated by the difference between equilibrium temperature obtained from the UA-model and the in-situ measured temperature (Supplementary Figure S7), the equilibration could have taken place at an approximately 5°C lower temperature in the sun sheltered reed, with the water parcel heating rapidly without changing its gas content once reaching the channel (Supplementary Figure S4). The channels within reed beds are rather wind sheltered environments as long as the main wind direction is perpendicular to the channel orientation. However, during sporadic measurements of daily temperature cycles in the channel, we only observed temperature fluctuations up to 2°C over the course of 1 day. Therefore, it seems unlikely that rapid heating would lead to the observed excess air signals.

We can therefore hypothesize that the plant physiology of the dominating reed species was the relevant cause of the oversaturation (Figure 7B). Equipped with aerenchyma (i.e., tissue with air channels, e.g., Baldantoni et al. (2009)), Phragmites australis transports O₂ in order to aerate its root zone (e.g., Brix et al., 1996). Aerating the root zone is vital for plants living in water logged or inundated soils, since anoxic sediment environments promote several adverse redox reactions (DeLaune and Reddy, 2008). The transportation mechanisms affect all gases present in air and wetland sediments (Sorrell and Brix, 2013). This way, He, Ar and N₂ are pumped into the root zone together with O₂ and could be used as tracers for plant-mediated oxygen supply.

Convective flow in plants is driven by a pressure gradient within the plant. The two most important gas transport processes in Phragmites australis are venturi-induced convection and humidity-induced convection (Armstrong et al., 1992). Venturi-induced flow is driven by wind blowing across the culms, thereby creating a pressure differential that sucks air out of tall dead culms. Short broken culms act as inflow and lead air along the underground root system to tall culms. The effectiveness of this process increases with wind speed (Armstrong et al., 1992). Humidity-induced flow works best during hot and dry days, since it relies on the pressure difference created by moistening of the air inside the plants living green shoots. Dead and broken culms act as efflux sites in this case (Armstrong et al., 1996). According to field measurements shortly above the water table, the pressure inside live culms is up to 2% higher than ambient air pressure (Arkebauer et al., 2001). The exchange of gases between plant and rhizome is governed by diffusion (Armstrong et al., 1996).

Assuming the plants provide enough air to their root system to saturate the surrounding water with dissolved gases, we can estimate the level of noble gas oversaturation to be expected. The plaur which Phragmites australis grows on is about 0.8–1.3 m thick, so it is reasonable to assume that the root systems reach a depth of 1 m below water level. At this depth, the local hydrostatic pressure is about 10⁴ Pa higher than atmospheric pressure. If the plants would provide enough air to this depth to saturate the water phase at this increased pressure, we would expect 10% oversaturation when comparing it to water equilibrated at local atmospheric pressure. This is approximately the amount of oversaturation we observed for He, Ar and N₂ in May and October in Canal Vatafu-Imputita (Supplementary Figure S5).

Based on field measurements, Brix et al. (1996) estimated the total gas exchange by Phragmites australis to 9–11 L m⁻² h⁻¹. Assuming an air-like gas composition in the gas transport system of the plant and a dissolution of 10% of the gas transported by the total flux observed by Brix et al. (1996), would mean that 1.5–3.5 h were sufficient to supply the amount of N₂ and Ar to the root systems needed to create the observed oversaturation.

If reed beds play a significant role in the injection of excess air and high concentrations of CH₄ and CO₂, then how can we explain the different patterns observed in the GHG and excess air maps (Figures 3, 9)? Canal Vatafu-Imputita drains a core protection zone of the Danube Delta Biosphere Reserve, which is populated by Phragmites australis, while other core zones generally protect other vegetation types. In the core zone, harvesting and burning of the reed is not permitted, thus we expect more dead and broken culms that can function as efflux culms for the convective flow in this area, together with more degradable organic matter. Brix et al. (1996) report 280 dead and 85 living culms per square meter in their study area. At sites that are harvested and burnt we would expect the ratio to be shifted towards living culms as dead culms are removed regularly, which might restrict sediment aeration. A functioning soil aeration results in the oxidation of reduced species near the root zone, e.g., oxidation of CH₄ produced during organic matter degradation to CO₂ (DeLaune and Reddy, 2008). Looking at the maps in Figure 3, the high ratio of CO₂ to CH₄ indicates that CH₄ oxidation could be an important process in Canal Vatafu-Imputita. This seems plausible, considering that biofilms inhabiting the root surfaces of Phragmites australis consist to about one-third of bacteria capable of oxidizing CH₄ (Faußer et al., 2012). This could explain the buildup of high CO₂ concentrations in the water phase below the floating plaur which subsequently drain into Canal Vatafu-Imputita. The adjacent Canal Rosu-Imputita is draining the buffer area, where reed harvesting is allowed within limits. It shows higher CH₄ and lower CO₂ concentrations than Canal Vatafu-Imputita, especially in spring. The oversaturation of Ar, N₂ and He is also less in this channel during the early growing season, which may indicate that nearby reed beds are managed and O₂ transfer during the early growing season is hampered, resulting in a reduced oxidation capacity in the root zone.

Once the supersaturated water reaches the channel, diffusive re-equilibration with the atmosphere would decrease the oversaturation towards equilibrium. Intense boat traffic would enhance gas exchange and ebullition could strip the gases even below equilibrium. A combination of these processes might explain the comparably low He saturations observed at Canal Vatafu-Imputita (Figure 6).

In contrast to the channels, we observed undersaturation of He, Ar and N₂ in the lakes, especially in Lake Isac. This coincided with a CO₂ undersaturation and an O₂ oversaturation. The undersaturation of He, Ar and N₂ indicates that gas stripping by ebullition affects the dissolved gas concentrations, since a rising bubble exchanges gas with the surrounding water column and thereby strips dissolved gases depending on their solubility (Brennwald et al., 2005; McGinnis et al., 2006). Accordingly, in Lake Isac the hardly soluble He showed the strongest undersaturation, followed by the more soluble N₂ and Ar (compare Figures 2 and 6). As O₂ and Ar exhibit similar solubilities, undersaturation of Ar indicates that O₂ must be affected by gas stripping. We calculated the effect on O₂ concentration via O₂:Ar ratios (Eq. 4 and Figure 10). For oversaturated O₂ conditions in May, we found a linear relation with missing O_2,dif down to about 2.5 mg/L at 150% saturation.

Gas ebullition is the most likely cause for the observed oxygen deficit and there are two potential drivers of such a process: CH₄ ebullition from the sediment (Figure 7C) and O₂ ebullition from macrophyte leaves (Figure 7D). Evaluating the dataset from Maier et al. (2020) showed that CH₄ ebullition was most often present in lakes and channels, where it occurred in 43 and 42% cases of flux measurements around the year (Maier et al., 2020). However, with a depth of only about 1.5–4 m, the sampled delta lakes are comparably shallow (Oosterberg et al., 2000), which limits the contact time between water column and the rising gas bubble, thus restricting gas exchange (McGinnis et al., 2006). A pure CH₄ bubble with an initial size of 2 mm would still contain 70% of the initial CH₄ if released from 4 m depth. Larger bubble diameter or shallower release depth would increase the percentage of initially CH₄ transported (Greinert and McGinnis, 2009). CH₄ bubbles released from the sediment usually contain large amounts of N₂. Assuming the minimum amount of CH₄ needed to trigger the rise of a bubble through the sediment would result in an approximate CH₄:N₂ composition of 35:65% (Langenegger et al., 2019). A CH₄/N₂ bubble released in a lake with 2 m depth would transport on average about 5.3 μg O₂ per bubble to the atmosphere (Supplementary Figure S10). Thus, about 0.8 million methane bubbles per m² would have to be released to the atmosphere from a fully mixed water column of 2 m depth to account for the buildup of a daily oxygen deficit of 2 g m⁻³. This would equal a CH₄ flux of 3.1 mol m⁻² day⁻¹, which is about 60 times higher than the maximum total CH₄ flux measured in lakes and more than 2,000 times higher than the median total CH₄ flux from lakes in the Danube Delta (Maier et al., 2021). In the case of CH₄ ebullition being the sole cause of O₂ reduction, we would expect gas stripping to be strongest after macrophyte die-off or in autumn when large amounts of easily degradable material are available. Considering the large fluxes needed and the unusual timing, it is unlikely for CH₄ ebullition to be the dominant process reducing the saturation of dissolved gases in May and we therefore turn our attention to photosynthesis.

In spring, the bottom of Lake Isac was covered by macrophytes, many of them reaching close to the water surface (Supplementary Figure S11). If photosynthesis is very productive, bubbles containing N₂ and a lot of O₂ form on the leave surface of macrophytes and can be released via ebullition (Koschorreck et al., 2017). In contrast to CH₄ bubbles rising from the sediment, these O₂ bubbles spent considerably longer time in contact with the water column until they detach and rise to the surface. This process would also explain the quasi-linear relationship between observed oversaturation and estimated amount of missing O₂, assuming that stronger oversaturation coincided with more O₂ bubble formation and increased release to the surface. Strong oversaturation together with missing O₂ was almost exclusively observed in the lakes (Supplementary Figure S9), especially in Lake Isac and Lake Puiu in spring where median O₂ saturation was 147 and 115%, respectively. Both lakes are also the sites where He and Ar saturations were lowest with median saturations of 76.9 and 91.4% in Lake Isac, and 87.5 and 96.9% in Lake Puiu, respectively. Temperature-corrected, specific conductivity was lower in these lakes than in the main river (Supplementary Figure S3). This may also indicate high primary productivity because intense photosynthesis takes up CO₂, thus increasing pH, shifting the carbonate equilibrium, and leading to calcite precipitation and a decrease of specific conductivity (Dittrich et al., 2004; Ostrofsky and Miller, 2017).

The combined measurements of non-reactive gases and O₂ thus indicate that photosynthesis was even higher than indicated by in-situ O₂ oversaturation. Whether the inter-lake differences are caused by differences in submerged macrophyte abundance (Coops et al., 1999) or caused by differences in macrophyte species (Niculescu et al., 2020) cannot be concluded without further research. In any case, the current findings support several recent studies that highlight the importance of considering ebullitive gas exchange when estimating metabolism from diel O₂ concentration measurements in shallow ecosystems and propose to constrain ebullition rates using funnels or modelling the gas exchange from noble gas measurements (Koschorreck et al., 2017; Howard et al., 2018; Long et al., 2020). The reduction of daily net O₂ production estimates by ebullitive O₂ fluxes is estimated to about 1–21% (Howard et al., 2018).

High-resolution mapping of biogenic and inert dissolved gases across the waterscapes of the Danube Delta revealed a broad range in the saturation state of the greenhouse gases CO₂ and CH₄ (Figure 3). The main river reaches showed the expected behavior of a net heterotrophic ecosystem metabolism with slight supersaturation in both gases (Figure 6).

The channels in a delta system represent the reactive pipes transferring water from lakes and wetlands back into the main river and finally to the sea. Measuring the patterns of unreactive gases in channels provided additional insights: Signals of excess air in a canal close to floating natural Phragmites stands provided evidence for convective gas transfer into the root zone and for intensified CH₄ oxidation resulting in elevated levels of CO₂ at comparatively low CH₄ concentration. This preliminary assignment of most plausible processes to a set of dissolved gas patterns may help in guiding further research into the “black-box” of terrestrial-aquatic linkages at densely vegetated aquatic-terrestrial boundaries.

In comparison to the river reaches and channels, the shallow lakes were dominated by photosynthesis with clear CO₂ deficits and O₂ oversaturation (Figures 3, 5). In these highly productive systems (Durisch-Kaiser et al., 2011; Maier et al., 2021), non-reactive gases gave clear evidence for O₂ ebullition from macrophytes, a process that deserves further attention. Ebullitive loss of O₂ to the atmosphere complicates the estimation of net ecosystem production in shallow lakes (Koschorreck et al., 2017), beds of macrophytes and with floating vegetation (Kleinschroth et al., 2021). Such plant communities in freshwater systems remain a significant challenge for efforts to upscale net CO₂ exchange.

The current discussion and global synthesis of the carbon transfer along the land-ocean aquatic continuum (LOAC, Regnier et al., 2013) is still dominated by active pipe models (Cole et al., 2007; Abril and Borges, 2019). Such a river engineering approach falls short of capturing the dramatic changes in the biomorphology of large river corridors of the last century. Surveys identified a loss of around 80% of the riparian wetlands in the lower Danube River corridor (Csagoly et al., 2016). On a global scale, preliminary estimates arrive at a loss of wetland area in the range of 17–50% between 1970 and 2008 depending on the region (Dixon et al., 2016). Loss or restoration of riparian wetland areas will certainly affect the greenhouse gas budgets of large river systems, but adequate measurements and models are needed to derive reliable estimates.

Based on this case study, we conclude, that adding non-reactive gas tracers to the toolbox of biogeochemical analysis offers significant potential for disentangling physical and biological processes in deltaic systems. Our analysis identifies root aeration of Phragmites australis reed beds as a plausible process leading to the oversaturation of small channels with non-reactive gas species as well as greenhouse gases. Furthermore, we highlight the effect of O₂ ebullition on dissolved gas concentrations in shallow lakes and the risk of underestimating metabolic rates derived from O₂ concentrations.