The Danube River is the second largest river in Europe and its international catchment of 817 000 km2 covers parts of 19 European countries. Originating in the Black Forest Mountains in Germany, the Danube flows southeast for over 2857 km before discharging into the Black Sea. The long-term mean discharge is 6360 m3s-1 and ranged from 2930–11 300 m3s-1 in the period 1921–2015 at the Ukrainian station of Reni (Romanova et al., 2019). Receiving meltwater from the Alps and the Carpathian Mountains, the hydrology of the Danube River shows a pronounced seasonality. In general, peak discharge lasts from April–June and low flow conditions prevail between September and November. Driven by rainfall in the lower catchment, a secondary discharge maximum usually occurs from December through January. During our study, seasonal air temperature changes varied from around 0 °C in winter to 20–25 °C in summer. The probability of cloud cover was higher in winter than in summer (Fig. 1a).

Close to the mouth, the river splits into three arms that form the Danube Delta: Chilia in the north constitutes the border with Ukraine, Sulina in the middle was modified for maritime commercial navigation while Sfântu Gheorghe (St. George) limits the delta area in the south (Fig. 1b). The surface area of the Danube Delta between these river reaches is approximately 4150 km2, of which more than 80 % lies within Romania. As the youngest and the longest among the three arms, the 120 km long Chilia branch carries more than 50 % of the Danube water, the 64 km long Sulina channel about 27 % and the 70 km long river reach of St. George contributes about 20 % to the total water discharge. Travel-time analysis based on our sensor data resulted in average flow velocities in the range of 0.7 and 1.0 ms-1.

An intricate network of approximately 470 lakes, swamps, shallow water pools and about 3500 km of channels connects the delta's wetlands with the Danube River (Gomez-Baggethun et al., 2019). Although considered the most pristine delta in Europe, a series of hydrotechnical works in the Danube Delta, mostly for agriculture and fishery, led to doubling the length of the internal canals after 1980 (Gatescu, 1993). This, in turn, gradually increased the water flow through the delta from 260 m3s-1 between 1951 and 1960 to 620 m3s-1 in the period 1981–1990 (Bondar, 1994). Today, approximately 10 % of the total Danube discharge is estimated to enter the delta, of which about 20 % (∼120m3s-1) may be lost via evapotranspiration in the wetlands. Reed (Phargmites australis) is the dominant species in the Delta covering ∼1600km2 or almost 40 % of the study area (Oosterberg et al., 2000). Large stretches of the channels are bordered by riparian forests composed mainly of black alder (Alnus glutinosa) and silkvine (Periploca graeca) (Oprea et al., 2024). Most of the lakes are eutrophic and classified into three types: (1) lakes at short distance to the river with high flushing rates and dense submerged vegetation, (2) large, relatively deep lakes like Puiu and Rosu with longer residence time and high Potamogeton-cover that collapses during winter, and (3) small, isolated lakes receiving “black-water” inflows from floating reedbeds and dominated by Ceratophyllum demersum and Nitellopsis obtusa (Coops et al., 2008). A high diversity of submerged and floating macrophytes is present in the canals and old meanders (Oosterberg et al., 2000).

In the first phase of this study, we deployed four sensor packages (EXO2 Multiparameter Water Quality Sondes, YSI) along the main branches of the Danube at the stations labelled Tulcea, Chilia, Sulina, and St. George (Fig. 1b). Tulcea served as the upstream reference station for water entering the delta. The Chilia branch receives additional input from a Ukrainian catchment with shallow lakes to the north, the Sulina channel serves as the main shipping route with lateral inputs from adjacent wetlands, and the St. George branch collects inflow mainly from wetlands on its left side. The monitoring stations recorded data every 15 min between November 2015 and February 2018. For consistency, all timestamps were recorded in Easter European Time (EET) without a switch to summertime (EEST). For further details on time intervals and locations see Fig. A1 and Table A1. One multiprobe was damaged by floating ice in December 2016 which limited the time series at Chilia station to one year. In the second phase from February 2018–December 2018, the remaining three EXO2 probes were installed in channels within the delta at Puiu-Rosu, Balanova and Busurca. Puiu-Rosu is influenced by outflow of a large and relatively deep lake complex, Balanova receives mixed input from lakes wetlands while Busurca is mainly connected to reedbeds with negligible lake influence. Note that Balanova refers to a local name along the Central Channel that connects the Sulina and St. George branches (Fig. 1b).

The EXO2 multiprobe sensors recorded water temperature, pH, specific conductivity, and dissolved oxygen every 15 min. Data of sensors for turbidity and fluorescent dissolved organic matter was not analyzed for this study. Before each measurement, the sensor caps were automatically cleaned by a wiper. The EXO probes were moored at approximately 1 m below the water surface. Data gaps occurred sporadically due to unexplained energy drain from batteries and when probes were removed to prevent damage by floating ice sheets during cold spells in winter. Once a month, the probes were removed from the water for cleaning and pH and O2 sensors were calibrated using standard buffer solution and air-saturated water, respectively. Post-deployment, the data was drift-corrected using the calibration data. Crosschecks with discrete in situ measurements using the YSI Optical Dissolved Oxygen and Professional Plus Multiparameter instruments (Maier et al., 2021) showed good agreement. We used a low conductivity threshold (<200µScm-1) to detect and delete invalid data points recorded at times when the sensor was out of the water for servicing.

We derived the alkalinity time series from specific conductivity data using the linear correlation between laboratory analysis of alkalinity and specific conductivity (SpCond) measurement as: alkalinity [mmolL-1]=0.0057× SpCond [µScm-1] + 0.60 (R2=0.70, Fig. B1). The same approach has been applied to derive multidecadal trends for CO2 evasion in the Loire River (Nguyen et al., 2025) and to estimate global riverine CO2 emission rates (Raymond et al., 2013). To calculate CO2 from pH and alkalinity, we used the CO2SYS MATLAB code version 1.1 with Matlab R2017a and 2022b (van Heuven et al., 2011). These calculations were based on dissociation constants for a temperature range of 2–35 °C (Cai and Wang, 1998). To obtain an upper and lower estimate for the derived CO2 time-series, we propagated the uncertainty of the alkalinity-conductivity regression estimates and obtained an error of ±13%. The calculated CO2 values correlated well with discrete headspace measurements taken at the same time and location (R2=0.82, Fig. B1). Finally, we used solubilities (Weiss, 1974) and the temperature recordings from the EXO2 probes to calculate the deviation from atmospheric equilibrium in µmolL-1 of the O2–CO2 pairs. We defined the frequent excess of CO2 and the typical deficit of O2 with regards to atmospheric equilibrium as exCO2 and exO2 [µM] with a positive or negative sign, respectively. For the correlation of daily O2–CO2 covariance data with potential physical drivers, we used averaged daily water temperature from the EXO2 sensors and cloud cover from a meteorological station in Tulcea (Fig. 1a). The time-shift of one hour during daylight saving time was removed to synchronize the meteorological data with the continuous recording of EXO probes.

Some data-filtering was necessary for the time series of Chilia (2016) which was sporadically influenced by water from an adjacent wetland and for Sulina (2017) which recorded disturbances from the Black Sea, respectively. We flagged these spikes in conductivity by comparing the time series with upstream Tulcea station and removed data if they deviated more than 40 µScm-1 from the baseline. This approach missed the beginning and the end of the spike events. To refine the cleanup, we took 5 h before and after an identified spike event and eliminated periods with changes in specific conductivity larger than 2 µScm-1 and completed the task with few manual adjustments.

Vachon et al. (2020) proposed a set of parameters obtained from covariance analysis of paired exO2–exCO2 measurements to gain insights into aquatic ecosystem metabolism. Because buffer effects may distort the exO2–exCO2 data clouds at low CO2 concentrations (Diamond et al., 2025; Nguyen et al., 2025), we included the DIC timeseries and paired exO2–DIC measurements in our analysis. We developed a workflow in Mathematica 14.3 to analyze the Danube Delta time series at monthly and daily timescales. Days with fewer than 85 data pairs were excluded from analyses, as typically 96 pairs were recorded daily at 15 min intervals. Monthly analyses usually involved more than 2800 pairs and months with less than 500 pairs were omitted. Results of the covariance calculations included the centroids which represent the mean of the exO2 and exCO2 data over the period of days or months and the offset of the mean from an ideal line with slope -1 for photosynthesis and respiration (Fig. 2a). Stretch and width were obtained from the Eigenvalues of the covariance matrix as the major and minor axes length of the 95 % confidence ellipse and we obtained the slope scov of the stretch-line from the Eigenvectors of the covariance matrix. Parameters for the 95 %-ellipses obtained from covariance analysis provide valuable insights into the intensity of photosynthesis and respiration (stretch), the ecosystem quotient (EQ=1/slope), and the impact of lateral inflow (centroid position, offset).

To analyse monthly day-night dynamics, we calculated the average 24 h day-night cycles for each month of sensor deployment. For instance, monthly averages for 08:00 am in June were obtained by taking the mean of all June measurements at 08:00, 08:15, 08:30 and 08:45 EET. This approach resulted in average timing and amplitudes of the exO2, exCO2 and DIC cycles. The time-of-day averages reduced the influence of outliers and resulted in slopes s24 that may reflect EQs more closely than the tilt of the monthly covariance ellipse (scov) (Fig. 2b). The following analysis, however, focuses on amplitudes and stretch as indicators of metabolic intensity and on the offset and centroid positions as indicators of wetland discharge.

For a more detailed analysis, boxplots of the amplitudes of exO2, exCO2, DIC and stretch parameters were based on all available monthly data at the seven stations (Table A1). The boxes show the median and quartiles (25 % and 75 %), whiskers indicate the range of values (excluding outliers). The statistical difference of these parameters for metabolic intensity between river and delta stations was evaluated by Kruskal–Wallis tests. To assess the influence of temperature and daylight on the aquatic ecosystem metabolism, stretch parameters from daily covariance analyses were correlated by linear regression and covariance analysis with average daily cloud cover data from Tulcea Airport and water temperature measured in-situ by the EXO2 probes.

The intensity of photosynthesis and respiration, as recorded by CO2 oversaturation with respect to atmospheric equilibrium, O2 deficits and DIC concentrations, varied strongly in space and time during the observation period. The results allow testing of the hypothesis that aquatic metabolism is more intense within the delta compared to the main river reaches. The inflowing Danube at Tulcea showed DIC minima of 2.6 mM in the warm season and maxima of 3.5 mM during the cold months (Fig. C1). Averaging the 24 h cycles for O2, CO2 and DIC at the monthly scale provided insights into the intensity and seasonality of river metabolism (Figs. 3 and C2). In 2017 the sensors at Tulcea recorded small and rather constant supersaturation (exCO2) and undersaturation (negative exO2) in the range of ±40µM with diel amplitudes from ∼1µM in the cold months and below 10 µM in the warm season. Within the delta, the 24 h cycles at Balanova channel showed maximum daily fluctuations of more than 100 µM for exO2 in August. Overall, the 24-h amplitudes spanned two orders of magnitude between the extremes of 1 µM for the cold season in the upstream Tulcea station and 100 µM for warm season amplitudes within the delta.

The Balanova observations were characterized by maxima in the early morning (06:00–08:00 am EET) and minima in the afternoon (04:00–06:00 pm EET) between April and November (Fig. 3). Compared to exCO2, the oxygen cycles represented a mirror image where the strong undersaturation in the morning hours relaxed to near-equilibrium conditions in the late afternoon. Compared to CO2 and O2, the 24 h DIC curves showed more stochastic deviations from the sinusoidal pattern, likely reflecting the rapidly changing DIC concentration in the inflow (Figs. C1 and C2).

A comparison of median 24 h amplitudes based on four indicators of metabolic intensity (amplitudes of exO2, exCO2, DIC and the stretch parameter from covariance analysis) revealed significant differences between sites (Fig. 4). The two downstream sites of St. George and Chilia showed consistently higher median amplitudes compared to Tulcea. Amplitudes of exCO2 were generally lower than those of exO2, whereas the 24-h DIC variability was typically higher. Overall, the daily metabolic cycles intensified on the final ∼60-120km of river flow between Tulcea and the three near shore stations.

Compared to the river stations, amplitude and stretch values recorded in the delta were significantly larger (p<0.01 for exO2, exCO2 and stretch, p∼0.01 for DIC). Overall, the average daily fluctuations revealed increasing median values in the amplitude of exO2 from the apex to the inner delta, with low values from 6.5 and 4.5 µM at Tulcea and Sulina, respectively, to higher amplitudes of 29 and 9.3 µM at St. George and Chilia and consistently high values within the delta (28, 50, and 29 µM in Puiu-Rosu, Busurca and Balanova). In terms of median amplitudes, metabolic activity per unit volume was thus about 4–8 times higher at stations within the delta than at the upstream station of Tulcea. A similar pattern was observed for the median DIC amplitudes with 13 and 12 µM in Tulcea and Sulina, 50 and 23 in St. George and Chilia and elevated values within the delta: 23, 32 and 40 µM at Puiu-Rosu, Balanova and Busurca, respectively. Using DIC as an indicator, metabolic activity within the delta was 2–3 times higher than at the upstream Danube station. The results demonstrate how river metabolism intensified in the slow-flowing waters of the delta.

Plotting the 24 h cycles in the exO2–exCO2 plane allowed for a closer look at diel dynamics and the intensities of autotrophic versus heterotrophic ecosystem metabolism over the course of a year (Fig. 5). In the following, we focus on differences between the cold season (October–March) and the warm season (April–September). At Tulcea and Sulina, these plots illustrate minimal metabolic activity during cold months with exO2 amplitudes in the range of 1–2 µM and comparatively small signals of 7–9 µM during the warm months. In contrast, Chilia and St. George, were characterized by significantly stronger daily cycles with mean warm season exO2 values of 17 and 32 µM, respectively.

The delta stations reveal different types of dominant metabolic activities and a broad dynamic range of O2 and CO2. At the lake-outlet of Puiu-Rosu, autotrophic activity significantly exceeded heterotrophic metabolism with a pronounced buildup of dissolved oxygen and positive daytime exO2 values during the warm season. In contrast, the wetland discharge defining the water characteristics at Busurca was dominated by heterotrophic signatures with extremely high exCO2 concentrations of 300–500 µM during the warm season, while exO2 oscillated between -150 and -250µM and approached anoxic conditions. Balanova, showed an intermediate signature between the lake and wetland waters with warm-season exO2-values ranging between 0 and -100µM.

At high CO2 supersaturation observed for instance at Busurca, Balanova and St. George, the slopes of linear regression lines often converged towards idealized values of -1 for the O2:CO2 stoichiometry of photosynthesis and respiration (Fig. 5). At Puiu-Rosu, however, steeper slopes were observed due to the buffer effect of the carbonate system when CO2 values approach atmospheric equilibrium (Shangguan et al., 2025). The Busurca data followed a heuristic trajectory along a slope of -0.5, extending towards exO2≈-200µM, and exCO2≈+400µM.

To assess the effect of potential drivers for the intensity of river metabolism such as temperature and cloud cover, we related the daily stretch parameters with mean water temperature measured by the in-situ probes and the mean daily cloud cover observed at Tulcea Airport (Table 1). Cloud cover blocking sunlight was negatively correlated to the diel stretch parameters for the seven analysed time series. The Pearson's coefficients ρ ranged from -0.26 at Busurca to -0.46 at Chilia. On average, the cloud-cover sensitivity of the stretch factor was ∼3 times larger at the delta stations compared to the river sections. The R2 values were typically <0.2, but the correlations were significant at the p<0.05 level.

Water temperature has been identified as a key variable for respiration rates in streams (Perkins et al., 2012). In this study, the correlation between the stretch parameters and mean daily water temperatures was rather strong (range 0.37<ρ<0.77) and significant (p<0.05) for the seven stations (Table 1). On average, increasing water temperature enhanced the indicator of metabolic activity by 0.7 µM°C-1 at the river stations and by 2.8 µM°C-1 within the delta. For a seasonal shift in water temperature of 20 °C, the corresponding change in the stretch parameter amounted to 14 and 56 µM for typical Danube River reaches and delta channels, respectively. This factor 4 difference in the temperature-sensitivity of the daily stretch parameter between river and delta sites mirrors the metabolic response to cloud cover and is in line with the higher metabolic amplitudes in the delta.

The following analyses address our second hypothesis that lateral inputs modify the downstream O2–CO2 dynamics. The exO2–exCO2 patterns revealed significant differences in the stretch (elongation) of the 95 % covariance ellipses between the main reaches of the Danube in 2016 and the stations within the delta in 2018 (Fig. 6). In November, data from the river reaches were enclosed by rather circular shapes with minimal offset from the 1:1 diagonal. The June records, however, resulted in elongated ellipses characteristic for higher amplitudes. The data patterns within the covariance ellipses formed distinct monthly fingerprints which were only slightly distorted after moving from Tulcea to the downstream stations of Sulina and Chilia but appeared randomized at St. George.

The delta stations showed smaller seasonal differences in stretch between June and November but marked shifts in the position within the exO2–exCO2 diagram. While the Puiu-Rosu data indicated intense photosynthesis with O2 supersaturation and equilibrium-level CO2, Busurca represented an extreme case of wetland discharge from the surrounding reed stands with CO2-rich and O2-depleted water masses. Balanova located downstream of Puiu-Rosu reflected a mixture of lake- and wetland water. Its monthly fingerprints resembled those of Puiu-Rosu but with a downward shift along the theoretical line of slope -1.

In the downstream Danube River reaches, wetland discharge led to increasing offsets and a shift along the heuristic slope -0.5 (Busurca, Fig. 5). Centroids were displaced towards O2 undersaturation by -40µM and CO2 supersaturation of +35 to +85µM. While offsets in June remained similar between Tulcea and Chilia, the two other downstream stations showed offset increases of 21 and 16 µM at Sulina and St. George, respectively. In November, covariance clouds remained compact and close to the -1:1 line except for St. George where the offset exceeded that of Tulcea by more than 20 µM. Overall, these examples from two contrasting months provide clear evidence of wetland discharge affecting dissolved gas dynamics between Tulcea and the stations near the Black Sea coast.

A more systematic analysis of the centroids allows tracking how the monthly means of exO2–exCO2 evolved over time. The results reveal distinct annual journeys at the four Danube stations in 2016 and 2017 (Fig. 7). The trajectories at Tulcea were confined to a narrow quadrant between 0 and ±40µM for both exO2 and exCO2. At the downstream stations, however, the warm-season centroids were typically more depleted in O2 and enriched in CO2 compared to Tulcea. The trajectories often moved along a heuristic slope of -0.5 for wetland discharge (Busurca, Fig. 7). The increase in average exCO2 between Tulcea and downstream stations reached factors as high as 4 (June 2016, Sulina) and 18 (June 2017, St. George). In contrast, the downstream depletion of exO2 at Sulina and St. George was less pronounced than the exCO2 loading and maximum exO2-differences compared to Tulcea were limited to factors of ∼3 (Sulina, July 2017) and 2 (St. George, October 2016). The Chilia station also showed consistently lower exO2 and higher exCO2 values compared to Tulcea, but extreme additions of CO2 rich waters were not observed in this more natural river reach which remains less affected by dredged channels.

Flood dynamics appear to play an important role at the stations within the delta. The largest flood peak during the three years of our study occurred in early April 2018 (Fig. 1) and was followed by a strong flood recession until mid-June. All three delta stations showed the highest CO2 loads with the lowest O2 levels in May 2018, when the flooded reed stands were likely exporting carbon-rich water into the adjacent channels and lakes. Higher discharge in July–August reversed the situation and moved the Busurca centroids towards photosynthesis-respiration line with slope -1. However, when exceptionally dry conditions re-established the wetland discharge in October–November, the Busurca centroids moved back towards the slope -0.5 line (Fig. 7).

Finally, we address the third hypothesis that a boost in aquatic metabolism in the delta and additional wetland discharge will increase the average aquatic CO2 emissions on the final stretch before a river reaches the sea. To compare CO2 emission rates across the delta, we calculated average exCO2 values at the seven stations. To avoid sampling bias caused by interruptions during winter, we interpolated missing months by average cold season values and obtained mean emission rates over two years for Tulcea, Sulina and St. George and over one year for the other stations (Table 2). The average exCO2 at the three downstream stations was 54 µM corresponding to a 1.9 times higher supersaturation level than at Tulcea.

In 2018 the lake outflow at Puiu-Rosu showed only 31 % of the exCO2 supersaturation observed at Tulcea in the two preceding years, whereas the wetland discharge at Busurca reached almost an order of magnitude higher exCO2 levels. In a previous study based on discrete samples and floating chamber measurements, Maier et al. (2021) determined the median values of the gas transfer coefficients k600 [ms-1] for the different waterscapes of river reaches, lakes and channels. These results allowed the estimation of updated CO2 emission rates based on the continuous sensor observations. For the river reaches, we assumed an emission scenario of a linear increasing between Tulcea and the three downstream stations. For comparison with an earlier study (Maier et al., 2021), the Puiu-Rosu station was used as representative estimate for lakes and the average between Balanova and Busurca was assumed to represent the mean channel emission rate (Table 3).

Despite substantially higher CO2 concentration in channels compared to the river reaches and lakes, the waterways of the delta contribute a similar amount as lakes to total emissions because of their small surface area. According to data from global meta-analyses, the vast area of reed stands holds the potential for significant CO2 drawdown (Li et al., 2025; Lu et al., 2017).

Why is it relevant to study the differences in river metabolism between lowland river reaches and wetland channels? Our study provides evidence for significantly faster metabolic processes in the slowly moving waters of the delta wetlands compared to the main river reaches of the Danube. Monthly averages of 24 h amplitudes and the stretch of exO2–exCO2 covariance ellipses consistently indicate a 4–8 times more intense aquatic ecosystem metabolism within the delta. This distinction is particularly relevant in the context of ongoing wetland restoration efforts across the Lower Danube (Csagoly et al., 2018), where changes in hydrological connectivity may substantially alter carbon processing, nutrient retention, and greenhouse-gas dynamics (Hemes et al., 2018; Tschikof et al., 2022).

Additional analyses of daily DIC variability support the evidence for more intense carbon processing within the delta, although the DIC patterns exhibit higher stochasticity due to hydrological fluctuations. In contrast the direct response of O2 to photosynthesis and respiration, DIC concentrations are influenced by a broader range of interacting processes, like groundwater inputs, sediment-water exchange, and episodic flooding events. These hydrological perturbations can alter both the concentration and residence time of dissolved carbon, partially masking the metabolic signal at daily time scales.

The mean daily exO2 amplitude as an indicator for metabolic activity covered two orders of magnitude from cold season values in the upstream station of the Danube with cold-season values around 1 µM at Tulcea to a warm season maximum of 100 µM observed at Balanova station in August 2018. The dense aquatic vegetation in the delta lakes and channels including floating reed, riparian vegetation, macrophytes and high chlorophyll concentration translate into high substrate availability for ecosystem respiration (Coops et al., 2008; Oprea et al., 2024). A hydrological model for the Delta estimated travel times across the network of channels and lakes on the order of 2–3 weeks for the Puiu-Rosu and Balanova stations (Oosterberg et al., 2000), facilitating a high biomass concentration in prolonged contact with moving water. Average exCO2 concentrations close to 0.5 mM at Busurca in May and July 2018 may therefore result from the accumulation of mineralization products over timespans of weeks in water parcels traversing the reed stands (Fig. 5). By contrast, water residence times in the main river reaches are much shorter: sensor data revealed travel times between Tulcea and in the three downstream stations on the order of one day for Sulina and St. George and two days for Chilia essentially limiting the time for accumulating changes in O2 and CO2.

The significant differences between the three river reaches, can likely be attributed to their morphology and hydrological connectivity. Patterns exO2–exCO2 pairs within covariance ellipses retained close similarity between Tulcea and the downstream stations of Sulina and Chilia, whereas they appeared “randomized” at St. George (Fig. 6). The Sulina channel has been strongly modified for navigation with straight geometry and dykes that reduce lateral exchange, while the Chilia branch retains a more natural meandering geometry. The St. George reach consist of natural meanders and straight cuts for navigation, resulting in a broader distribution of water travel times and more intense exchange with adjacent wetlands and aquifers. This close interaction with wetlands at St. George likely superimposes multiple water masses with different metabolic histories, thereby disrupting the coherent diel covariance patterns that were preserved at Sulina and Chilia. Consequently, the 24 h amplitudes of exO2, exCO2 and DIC all follow the sequence St. George > Chilia > Sulina consistent with decreasing lateral connectivity (Fig. 4).

We used the daily stretch factors to test hypotheses that water temperature and cloud cover would influence the intensity of aquatic metabolism (Table 1). Overall, the correlation between cloud cover and covariance stretch was weaker than for water temperature as a forcing factor. The relations between cloud cover, photosynthetically active radiation, PAR, (Kathilankal et al., 2014) and metabolic response are inherently complex in multiyear (Dodds et al., 2013) or multisite analyses (Mulholland et al., 2001). As cloud cover does not directly translate to in-situ light conditions due factors like the seasonality of daylight and water transparency, further assessments require local PAR analyses at the monitoring stations. Such local data were available for determining the temperature sensitivity of metabolic amplitudes. The regression of daily mean in-situ temperatures with diel stretch factors resulted in a temperature sensitivity of the aquatic metabolism of 0.7 µM°C-1 in river reaches and 2.8 µM°C-1 in delta channels. These results suggest that a warming climate would affect the wetland's aquatic metabolism about four times more strongly than main river branches. Previous studies have shown that biofilm respiration exhibits complex temperature sensitivity (Dybdahl et al., 2024) while ecosystems with higher quantities of organic substrate are subject to stronger temperature sensitivity of ecosystem respiration (Jankowski et al., 2014).

The timing of daily cycles indicates a preferred sampling window around noon for obtaining representative values of O2, CO2 and DIC, whereas the early hours of 05:00–08:00 am EET and the late afternoon of 03:00–06:00 pm would introduce the largest sampling biases in the case of the lower Danube aquatic ecosystems (Figs. 3 and D1). For logistical reasons, field work is often scheduled between the late morning and early afternoon hours, which likely explains, why the CO2 emission rates based the continuous monitoring of this study (Table 3) agrees very with results from grab samples collected during field campaigns during the same years (Maier et al., 2021).

What are the effects of wetland discharge on river biogeochemistry and how can we identify its origin and magnitude? Large-scale approaches may analyse the CO2 dynamics within an entire river system: sampling transects across the Congo River a study demonstrated that lateral connectivity with riparian wetlands was driving spatial and temporal variability of CO2 concentrations and the river's greenhouse gas emissions (Borges et al., 2019). More targeted study designs focus directly on wetland drainage pathways: recent studies revealed disproportionally high contributions of ditches to CO2 emissions from Dutch peatlands (van der Knaap et al., 2025) and from a Phragmites dominated Chinese wetland (Xue et al., 2025). In the present study, we combined both approaches, although logistical constraints prevented synchronous observations of river reaches and delta channels.

The order of magnitude of exCO2 found in the delta channels compares well with a global meta-analysis of wetland ecosystem metabolism (Richardson et al., 2024). These authors reported mean values of dissolved pCO2 for fens and bogs of 3400 and 4100 ppm, respectively, which corresponds to approximately 100 µM exCO2 and falls within the range of downstream Danube stations with exCO2 of ∼50µM and the extreme case of Busurca with 500 µM (Fig. 7). The mixing ration of wetland discharge in the Danube branches can be estimated based on the mean warm-season offset at Busurca (∼100µM). If we interpret offsets of 0–100 µM as a proxy for lateral inflow of 0 %–100 %, then the warm-season offsets observed in 2016 at Chilia, Sulina, and St. George would correspond to contributions of wetland discharge of 13, 22 % and 25 % to total flow in these River reaches. These estimates should be considered as upper limits considering that the diversion of Danube water through the delta was estimated at 10 % (Oosterberg et al., 2000). The discrepancy can be attributed to incomplete horizontal mixing, and the positioning of the sensors close to the riverbanks. At the St. George station the inflow from the Tartaru Channel probably contributed the large offsets observed between May and July 2017 (Fig. 7).

A quantitative assessment of the different lateral interactions affecting the three Danube River reaches was based on monitoring the average monthly difference in DIC concentrations between the downstream stations and the inflow at Tulcea (Fig. C3). Integrating these differences from April–December 2016 reveals an annual DIC addition of 0.5 and 1.2 molm-3yr-1 at Sulina and St. George, respectively. The observations in 2017 provided a similar result for St. George, but a higher accumulation rate for the Sulina stations. The seasonal balance for the Chilia branch, however, resulted in a negative value (-0.1molm-3yr-1). These contrasting observations can be explained by differences in hydrological connectivity and the origin of lateral inflows. Sulina and St. George are receiving carbon-rich wetland water. The important northern inflow to Chilia, however, originates from Lake Yuluph, the largest freshwater lake in Ukraine. Water from this system is transported via Lake Kuhurlui and connecting channels to the Chilia branch and appears to deliver DIC-depleted water.

A final question remains: how can we explain the absolute stoichiometry of up to 1:2 for exO2 : exCO2 during the warm season at the station downstream of the delta and within the Balanova and Busurca channels receiving wetland discharge? A mapping campaign in October 2017 with in-situ CH4 sensors reported median CH4 concentrations of 0.54, 0.70 and 2.2 µM for the Danube reaches, lakes and channels and an extreme value of 16 µM at a hotspot site near Busurca station (Canning et al., 2021). These data provide indirect evidence for an important source of CO2 co-generated by methanogens in anaerobic sediments of the reed stands. Although lateral water fluxes were not measured directly, the combination of CO2 enrichment, O2 depletion and deviations from 1:1 stoichiometry provides a consistent biogeochemical signature of wetland-derived inputs (Borges et al., 2019; Maier et al., 2021; Zuijdgeest et al., 2016). This anaerobically generated CO2 contributes to the offset towards a diagonal of slope -0.5 (Fig. 7). A portable membrane inlet mass spectrometer for O2, N2, He and Ar used during the same campaign in May and October 2017 revealed additional processes affecting O2 : CO2 ratios including excess air formation in riverbanks and root ventilation in Phragmites stands. These processes may fuel methane oxidation but will remain “invisible” to sensors in the water column (Maier et al., 2022). Therefore, anaerobically generated CO2 accumulating in the water column of reed beds, provided a distinct biogeochemical signature for identifying wetland discharge to channels and river reaches in the delta.

In their perspective paper, Abril and Borges (2019) argued that riparian wetlands should be included in an expanded version of the reactive pipe concept for rivers (Cole et al., 2007). With their high productivity and slow gas-exchange at the air–water interface, riparian wetlands release large amounts of dissolved CO2. Additional fluxes of plant litter, and DOC add to the function of wetland discharge as a carbon pump. There is now increasing evidence that the wetland CO2 pump is increasing CO2 emissions of inland waters across different climatic zones from the tropical Congo River (Borges et al., 2019) to the arctic River Ob (Vorobyev et al., 2024).

The exCO2 records (Fig. 5 and Table 2) allow constraining magnitude and timing of the wetland pump in the Danube Delta. Overall, the downstream stations exhibited exCO2 values and emission rates which were 1.7–2.1 times higher than measured at Tulcea (Fig. 8). Within the delta the high productivity of the Lake Puiu-Rosu complex reduced the average CO2 supersaturation to 30 % with reference to the inflowing Danube water indicating substantial CO2 uptake. In contrast, wetland discharge increased the average exCO2 by factors of 1.6 and 8.6 at Balanova and Busurca. The excess CO2 levels remained relatively constant at Tulcea and in the near natural Chilia branch receiving mainly lake outflow. Within the delta channels and at the Sulina and St. George stations, however, the wetland pump was most active during the warm season with exCO2 ratios between the warm and cold season of 1.3–2.1 (Table 2). This pronounced seasonality of the carbon pump is likely driven by the significant temperature dependence of the aquatic metabolism (Table 1) and the hydrological dynamics with flood peaks in spring an early summer (Fig. 1a).

The day-night cycles of exCO2 reached their maximum typically in the morning hours and showed a minim in the late afternoon and early evening (Fig. 3). Individual water samples were taken across the delta in 2016 and 2017 during daytime (Maier et al., 2021). The analysis of the three river reaches revealed no significant difference between individual water samples and the average values from the high-frequency records in this study (Table 3). This result seems to contradict a global study (Gomez-Gener et al., 2021) which reported evidence for higher night-time CO2 emissions from rivers based on an extensive compilation of day and night CO2 data. Analysing the 24 h cycles at the four Danube River stations in more detail (Fig. D1) reveals that sampling around midday captured values relatively close to mean daily conditions because exCO2 maxima and minima were cantered around the early morning and late afternoon hours, respectively. For systems affected by the wetland pump, these results indicate that the day-night sampling bias will depend on factors like lateral connectivity and water exchange rates.

Finally, we must address the “elephant in the delta” – namely the role of CO2 uptake by wetland vegetation and specifically by Phragmites stands. Direct carbon uptake by emergent wetland plants remains largely invisible to aquatic sensing, yet meta-analyses of net ecosystem production (NEP) report average global values of 168 and 208 gCm-2yr-1 for marshes (Li et al., 2025) and coastal wetlands (Lu et al., 2017), respectively. Applied to a 1600 km2 area of Phragmites dominated reed stands in the Danube Delta, these rates correspond to a carbon uptake on the order of 270–330 GgCyr-1. This estimated uptake is roughly five time higher than the diffuse CO2 emissions from water surfaces in the delta region estimated as 49 GgCyr-1 (this study, Table 3) and 60 GgCyr-1 (Maier et al., 2021). This scenario estimates indicate that the wetland CO2 pump observed in the Danube delta is driven by the highly productive reed stands which act as strong net sink for CO2 at the landscape scale. Much of the CO2 released into channels and river branches originates from the respiration of recently fixed organic matter within reed stands and wetland sediments. At the same time, the sustained primary production of Phragmites seasonally removes atmospheric CO2 and transfers carbon into submerged biomass and accumulating sediments. The high CO2 emissions from water surfaces in the delta are therefore part of an efficient carbon processing ecosystem. A complete assessment of the carbon budget would require ecosystem-scale CO2 and CH4 exchange measurements, for example by eddy-covariance techniques, which have provided the basis for many wetland carbon budgets (Zou et al., 2022).