Seasonal and spatial variability of the partial pressure of carbon dioxide in the human-impacted Seine River in France

Carbon evasion from rivers is an important component of the global carbon cycle. The intensification of anthropogenic pressures on hydrosystems requires studies of human-impacted rivers to identify and quantify the main drivers of carbon evasion. In 2016 and 2017, four field campaigns were conducted in the Seine River network characterized by an intensively cropped and highly populated basin. We measured partial pressures of carbon dioxide (pCO2) in streams or rivers draining land under different uses at different seasons. We also computed pCO2 from an existing data set (pH, water temperature and total alkalinity) going back until 1970. Here we report factors controlling pCO2 that operate at different time and space scales. In our study, the Seine River was shown to be supersaturated in CO2 with respect to the atmospheric equilibrium, as well as a source of CO2. Our results suggest an increase in pCO2 from winter to summer in small streams draining forests (from 1670 to 2480 ppm), croplands (from 1010 to 1550 ppm), and at the outlet of the basin (from 2490 to 3630 ppm). The main driver of pCO2 was shown to be dissolved organic carbon (DOC) concentrations (R2 = 0.56, n = 119, p < 0.05) that are modulated by hydro-climatic conditions and groundwater discharges. DOC sources were linked to land use and soil, mainly leaching into small upstream streams, but also to organic pollution, mainly found downstream in larger rivers. Our long-term analysis of the main stream suggests that pCO2 closely mirrors the pattern of urban water pollution over time. These results suggest that factors controlling pCO2 operate differently upstream and downstream depending on the physical characteristics of the river basin and on the intensity and location of the main anthropogenic pressures. The influence of these controlling factors may also differ over time, according to the seasons, and mirror long term changes in these anthropogenic pressures.

around 284 ppm were measured in the Loire (croplands) 12 , from 1604 to 6546 ppm in the podzolized Arcachon catchment's streams, with higher values when discharge is low 13 , and 2292 ppm in the carbonate-rock-dominated Meuse watershed, which is mostly covered by forests, grasslands and croplands 12 . A recent study of the Meuse River 14 revealed marked variations in pCO 2 (34 to 10,033 ppm) the higher values being associated with watersheds dominated by agriculture and lower values with forested watersheds. CO 2 undersaturation with respect to the atmospheric equilibrium has been demonstrated in the upstream part of the Danube River basin related to photosynthetic uptake in summer 15 .
In the Seine River basin, previous carbon investigations focused on organic carbon 16 , methane emissions from soils, livestock and the river network 17 or on benthic respiration 18 and ecological status based on the production/ respiration ratio 4 . These studies did not specifically address CO 2 concentrations. Our objective here was to quantify pCO 2 in the Seine River, using both recent in situ measurements and calculations based on long time series of existing data, in order to evaluate the distribution of pCO 2 and CO 2 evasion in the drainage network, and to identify the major factors controlling pCO 2 .

Materials and Methods
Study site. The Seine River watershed, located in northern France, covers an area of 76,750 km 2 with a median slope of 2.2° and 89.5% of its area is less than 300 m A.S.L. It has a pluvio-oceanic regime and its annual water flow in the period 2013-2016 averaged 550 m 3 s −1 at the river outlet at Poses (Fig. 1). The Poses monitoring station, located at a navigation dam, is the most downstream station not subject to the dynamic influence of the tidal estuary. Low water flows (<300 m 3 s −1 ) are generally observed from March to November, while high flows (>800 m 3 s −1 ) occur in winter, from December to February (data provided by the HYDRO database, http://www. hydro.eaufrance.fr, 2018).
Three major diverted reservoirs (Fig. 1a) located in the upstream part of the basin (the Marne reservoir, the Aube reservoir and the Seine reservoir) were built to reduce high water events in winter, and to sustain the flow in late summer. They have a combined storage capacity of 800 10 6 m 3 and a surface area of 65 km² 19 .
The basin is densely populated (~230 inhabitants km −2 ) mostly concentrated in the Paris conurbation (12.4 million inhabitants in 2015) in the central part of the basin. The largest wastewater treatment plant in Europe: Seine Aval, (French acronym SAV) WWTP with a dry weather capacity of 1,500,000 m 3 d −1 ) is located 50 km downstream of the center of Paris (Fig. 1b). The corresponding effluents account for more than 30% of the organic carbon load of all the WWTPs in the basin 20 . Major upgrading of wastewater treatments of the SAV followed the Urban Wastewater Treatment Directive (1991, 91/271/CEE) among which the addition of flocculation (2000)(2001)(2002)(2003), nitrification (2007) and denitrification (30% in 2007 and 70% in 2012) 21 . The Seine counts about 1700 smaller capacity WWTPs spread throughout the basin (Fig. 1b). The basin comprises 56.8% arable land (mainly under intensive agriculture), 25.8% forests, 9.7% grasslands and 7.0% urban areas (CLC database, IFEN 2012, Fig. 1c). Wetlands have been estimated at between 10.9% and 15.6% of the surface area of the basin 22 . The Strahler stream order 23 of the main stream of the basin is 6 th order for the Marne River and 7 th order for the Seine River downstream of Paris (Fig. 1a). The sedimentary basin of the Seine River is characterized by geological formations with low slope gradients resulting in concentric lithology dominated by carbonate and limestone in the central part of the basin, a wide band of Cretaceous chalk and a narrow band of clay followed by Jurassic limestone at the periphery (Fig. 1d).
Sampling strategy, physical-chemical analysis and direct measurements of pCO2. We sampled 30 sites in streams chosen because they mainly drain grasslands, forests and wetlands, croplands, and along the main streams of the Marne River (including in its reservoir) and of the lower Seine (Fig. 1a, exact locations in Supplementary Material 1). Sampling campaigns were carried out in four contrasting hydro-climatological periods. Water discharges were measured at the outlet of the basin (Poses) and temperatures were measured at each sampling site in winter from February 22 to March 10, 2016, (1030 m 3 s −1 , 6.9 °C on average), in summer/autumn from September 7 to 14, 2016, (270 m 3 s −1 , 18.8 °C), spring from March 14 to 23, 2017, (580 m 3 s −1 , 9.9 °C) as well as during a spring flood event that was exceptional in its timing, from May 23 to June 2, 2016, (1500 m 3 s −1 , 13.0 °C, at sampling time with a maximum discharge reaching 2000 m 3 s −1 at the river outlet, at Poses). The field campaigns were assumed to be key seasonal and hydrological periods and were conducted in areas representing the main types of land use in the Seine River basin.
Direct pCO 2 measurements were based on the syringe headspace technique 12,24 combined with non-dispersive infrared gas analysis (IRGA) (Li-cor ® models 820 or 840; accuracy <3% of reading). Calibration was performed using CO 2 concentration of 799 ppm and CO 2 -free dinitrogen. Four syringes coupled with three-way valves were filled directly in the stream or river, each replicate containing 30 mL of river water and 30 mL of atmospheric air. Closed syringes were continuously shaken for 10 min to equilibrate CO 2 concentrations of gas and water. The equilibrated gas was injected into the IRGA and water temperature inside the syringe was measured. The first injection served as a purge and the other three were used for pCO 2 measurements. The initial pCO 2 in water was computed based on the pCO 2 measured in the equilibrated air of the syringe and in the atmospheric air, and Henry's law accounting for the water temperature in the syringe and in situ.
Simultaneously, 2 L water chemistry high-density polyethylene sampling bottles were used to collect samples from bridges over the main stream, and along the banks of smaller streams. Water temperature, pH, dissolved oxygen and conductivity were measured in the field using a multi-parameter probe (YSI ® 6600 V2, accuracy ± 0.2 units). Calibrations of the probe were completed with pH 7 and pH 4 buffers for pH (NBS Scale), potassium chloride (KCl) electrolyte solution for dissolved oxygen and 10 mS cm −1 standard for conductivity. In the laboratory, water subsamples were filtered on combusted filters for 4 h at 500 °C: GF/F 0.7 µm, 25 mm) to analyze particulate inorganic and organic carbon (PIC and POC, respectively). Filtrates enabled measurement of dissolved inorganic and organic carbon (DIC and DOC) concentrations and total alkalinity (TA). One milliliter of sulfuric acid (3 M) was added to the DOC samples to stop biological reactions. Dissolved inorganic and organic carbon were analyzed with a TOC analyzer (Aurora 1030). Nongaseous DIC analyses required acidification of the filtrated sample by adding sodium persulfate reagents (100 g L −1 ) to dissociate the carbonates in the CO 2 that were detected by an IRGA. The inorganic free sample was used for DOC measurements. DOC was measured by wet oxidation by adding 10% phosphoric acid oxide followed by high temperature (680 °C) catalytic combustion, and then detected using an NDIR technique. TA (µmol kg −1 ) was analyzed using an automatic titrator (TitroLine ® 5000) on three 20 mL replicates of filtered water (GF/F: 0.7 μm), with hydrochloric acid (0.1 M).
Values of total suspended solids (TSS) were determined as the weight of material retained on a Whatman GF/F membrane per volume unit after drying the filter for 2 h at 120 °C. Chlorophyll a concentrations (Chl. a) were determined according to Lorenzen 25 .
Aquifer waters were also sampled during the same periods. Groundwater was pumped from the piezometers using a peristaltic pump. Before the samples were collected, the piezometers were emptied by flushing to remove the standing water (5-10 L in each piezometer) 26 . The same variables were measured or analyzed, except Chl. a. pCO 2 calculations from existing data. pCO 2 were computed with the CO 2 SYS software 27 using the water temperature and two of the three following measurements: pH, TA and DIC. In contrast to DIC, TA is often measured by the French water authorities Agence de l'Eau Seine Normandie (French acronym AESN, http://www. eau-seine-normandie.fr/, 2018), and thus were preferred to compute pCO 2 in combination with pH and water temperatures. The carbonate dissociation constants (K1 and K2) applied were from Millero 28 with zero salinity.
During our field campaigns (winter 2016, spring 2017, spring flood 2016, summer/autumn 2016, see previous section), we systematically combined direct measurements of pCO 2 with measurements of water temperature, pH and TA (130 samples). We found a positive relationship between the pCO 2 values directly measured during our field campaigns and those calculated using water temperatures, pH and TA (Fig. 2). This relationship was then used to correct possible bias 12 of pCO 2 values calculated with CO2SYS program.
We also used the database (42,108 data with simultaneous water temperatures, pH and TA measured between 1971 and 2015) provided by the French water authorities (AESN) to compute and analyze the pCO 2 dynamics in the Seine basin in space and over time since the 1970s. These pCO 2 data series were corrected by the relationship previously established and then averaged by months and years at each monitoring station. Within this timeframe , two periods of interest (ii) 1989-1991 (92 monitoring stations) and (ii) 2013-2015 (234 monitoring stations) were defined as representative of the changes that occurred recently in the Seine river basin. The former period (1989)(1990)(1991) represents the period of highest organic pollution from WWTPs, only treated by activated sludge. The most recent period (2013-2015) illustrates the state after a full implementation of the Urban Wastewater Treatment Directive (1991, 91/271/CEE), including the reduction of point sources of organic carbon (industrial and domestic) discharge into the river, as well as phosphorus and nitrogen 20,21,29 . In addition, we assessed the spatial variability of pCO 2 along the main stream of the Seine River, comparing the concentrations of the most important effluents up-and downstream (Paris and Poses stations, Fig. 1) of the SAV WWTP. We also computed pCO 2 at a constant temperature of 10 °C (pCO2@10 °C) downstream of the SAV WWTP to show the impact of solubility on pCO 2 (see results section: "Long-term pCO 2 variability (1970-2015)"). Raymond et al. 30 pointed out that gas transfer velocity equations including slope and water velocity enable easy measurements and recommended the use of these equations at large spatial scales. We selected the equation requiring only the slope and the water velocity and that had the highest squared-R in the Raymond et al. 30 Table 2 in Raymond et al. (2012)) as we wanted to compare the variability of CO 2 evasion in the basin in space and over time. We used kinematic water viscosity coefficients and Schmidt numbers calculated according to Wanninkhof et al. 31 .  The slopes of the streams and rivers were provided by the French water authorities (AESN). Water velocities were estimated from discharge records available at the scale of the whole drainage network for the period 2012-2014. Water temperatures were averaged by season of interest based on our field campaigns (winter, spring and summer/autumn). The k-values were calculated by stream order and then aggregated by small streams (Strahler orders (SOs) 1-4) and along the main stream (SOs 5-7). Calculated k-values for the spring flood event were based on averaged spring water temperatures associated with measurements of high water flow collected during the exceptional spring flood (May 2016).

study (Equation 5 in
According to Wanninkhof 32 , Wilke and Chang 33 and Raymond et al. 30 , the gas transfer velocity k CO 2 (m d −1 ) under negligible wind conditions in rivers can be calculated as: The flux (fCO 2 , mgC-CO 2 m −2 d −1 ) at the interface of the river and the atmosphere can be calculated as: Statistical tests. All statistical tests were performed using R software 34 .
Wilcoxon signed-rank tests were used to compare measured pCO 2 in the four periods, and Kruskal-Wallis tests were used to compare measured pCO 2 averages for different land uses during each period. A Shapiro-Wilk test was applied to test the normal distribution before performing the linear regression between measured pCO 2 and calculated pCO 2 . Linear regressions were then performed between pCO 2 and water quality variables.

Results
Measured versus calculated pCO 2 . The streams and rivers sampled during our field campaigns were neutral or basic and carbonate-buffered (Fig. 2a,c), excluding the overestimation of calculated pCO 2 already shown to be linked to the low buffering capacity of the carbonate system 12 . A logarithmic transformation was performed on both measured and calculated pCO 2 to obtain normal distribution (Shapiro-Wilk test, p > 0.01) to calculate a linear regression. A positive relationship was established (R 2 = 0.56, n = 130, p < 0.01). Field campaign dataset overview. Average water temperatures ranged between 6.9 °C and 18 °C which corresponds to the expected seasonal range for the Seine basin (Table 1). pH values were generally neutral to basic, with median values of pH and TA in all streams and rivers ranging respectively from 7.75 to 8.25 and from 3150 µmole l −1 to 4350 µmole l −1 (see Table 1). Only two acidic pH values measured during the winter in streams draining forests (data not shown). The high total alkalinity measured in all the streams and rivers (Table 1) indicated that waters were carbonate-buffered due to the lithology of the basin, which is dominated by carbonate rocks (Fig. 1d) 35 . Indeed, dissolved inorganic carbon (DIC) concentrations were high (min.: 19.3 mgC l −1 , Table 1 Variability in pCO 2 . Spatial and seasonal variability of pCO 2 . All samples were supersaturated in CO 2 with respect to the atmosphere, regardless of river characteristics (small stream or main stream), the associated dominant land use and the season (Fig. 3a). pCO 2 increased significantly i.e., by an average of 49% and 62% from winter to summer/autumn, in streams draining forests and croplands (p < 0.05, Wilcoxon signed-rank test, Fig. 3a). Values in grasslands did not typically follow this pattern (p > 0.05, Wilcoxon signed-rank test, Fig. 3a), and pCO 2 remained rather stable at 2,900 ppm, whereas pCO 2 was the highest in wetlands (p < 0.05, Kruskal-Wallis test), especially in spring and summer/autumn (>4500 ppm).
For each of the four seasons monitored, pCO 2 average decreases ranked as follows: wetlands > grasslands > forests > croplands (p < 0.05, Kruskal-Wallis tests, Fig. 3a). At the outlet of the Seine River, the main stream drains composite land uses where pCO 2 averages were found to be equivalent to those measured in small streams draining grasslands (Fig. 3b).
In the lower Seine River, which is highly impacted by urbanization and associated treated WWTP effluents, a 69% increase in pCO 2 was observed from winter (December to February), to spring (March to June) to summer/autumn (June to November) with limited dilution by the discharge (mean discharge in winter: 1030 m 3 s −1 ; spring: 580 m 3 s −1 ; summer/autumn: 270 m 3 s −1 ) (Fig. 3b, Table 1).
In the groundwater, pCO 2 averaged 27,000 ppm at a yearly scale, but reached up to 65,000 ppm (summer/ autumn 2016), i.e., a factor of 5 to 10 compared to surface waters.
According to the k 600 equation selected (see Materials and Methods, equations 1-3), gas transfer velocity (k-values) was estimated from the slopes. k-value was higher for small streams (0.006 m m −1 ) than larger rivers (0.0004 m m −1 ) (Fig. 4a). Seasonal variations in water temperature increased from winter to summer/autumn (Fig. 4b), and velocities decreased from winter to summer/autumn (Fig. 4c). The resulting k-values ranged from 0.08 to 0.24 m h −1 with a decrease from small streams (k annual average = 0.19 m h −1 ) to larger rivers (k annual average = 0.09 m h −1 ) (Fig. 4d).
Using equation (2), the higher slopes found in small streams led to higher k-values ( Fig. 4a and d). Additionally, in the small streams in the Seine River basin, the water velocity effect prevails over the seasonal k dynamics, while control by water temperature is greater in higher stream orders (Fig. 4b-d). During the spring flood event, the increase in the water discharge (and velocity) led to a greater increase in k in small streams than in larger rivers, respectively +26% (spring flood: 0.24 m h −1 ) and +11% (spring flood: 0.10 m h −1 ) compared to average spring k-values (small streams: 0.19 m h −1 , larger rivers: 0.09 m h −1 ). CO 2 fluxes at the water-atmosphere interface (Fig. 4e) were estimated using the pCO 2 measurements we made during our field campaigns (Fig. 3), CO 2 saturation values that depend on water temperatures (see Table 1), atmospheric pCO 2 , and k-values estimations (Fig. 4d). The same seasonal pattern was observed for pCO 2 and CO 2 fluxes. 2 variability (1970-2015). Long-term analysis of French water authority (AESN) databases showed supersaturation of CO 2 of the Seine River dating back to 1970 (98.5% data suggested supersaturation with respect to atmospheric equilibrium -pCO 2 median = 3030 ppm; mean = 4765 ppm). From that period on, the Seine River has been a source of CO 2 to the atmosphere even when frequent phytoplankton blooms occurred before wastewater treatment was improved. However, focusing on the bloom events (Chl. a >50 µg l −1 , Fig. 5), we observed the opposite pattern between Chl. a and pCO 2 dynamics, with depletion of pCO 2 concomitantly with peaks of phytoplankton. This consumption of CO 2 was not sufficient to cause undersaturation of CO 2 in the river (Fig. 5).

Long-term pCO
When we compared the two contrasted periods with respect to water sanitation and associated organic carbon releases, we found a similar range of temperature and discharge values and a seasonal pattern typical of temperate oceanic hydro-climatology regimes, i.e., high temperatures and low water in summer/autumn. The first of these two periods was however slightly drier than the second (on average 400 m 3 s −1 vs. 545 m 3 s −1 , respectively) with no notable change in temperature (averaging 13.9 °C vs. 14.2 °C, respectively. We observed that pCO 2 computed at both 10 °C and at water temperature were similar during winter but values at 10 °C were slightly lower than at water temperature during summer (Fig. 6). However, general trends of pCO 2 did not change.
In contrast, pCO 2 was reduced by a factor of 2.7 between the two periods (average 8250 ppm for the period 1989-1991 versus 3020 ppm for the period 2013-2015). These weak hydro-climatologic changes cannot explain the marked decrease in pCO 2 at the river outlet between the two periods. We found no relationship between pCO 2 and discharge at this time scale, despite a clear antiparallel trend for these two variables (Fig. 6).
To further explore the recent decrease in pCO 2 , we assessed spatial variations in pCO 2 at the scale of the whole Seine drainage network (Fig. 7). Although fewer measurements were available in the earlier period (1989)(1990)(1991), the decrease in pCO 2 between the two periods was obvious along the lower reach of the main stream of the Seine River, downstream of the Paris conurbation. In the recent period, the pCO 2 of both the upstream parts of the drainage network and the main stream of the Seine River appear to be equally supersaturated (p > 0.05, Kruskal-Wallis tests, Fig. 7).
Since the 1970s, upstream of the discharge of treated effluent from the SAV WWTP, the long-term trend of pCO 2 values in Paris has varied around 5000 ppm (Fig. 8a). A few kilometers downstream, at the outlet of the Seine River at Poses (strongly influenced by Parisian wastewater discharges) pCO 2 progressively increased to  reach a maximum of 12,000 ppm in the 1990s, and then slowly decreased to present values of 3000-4000 ppm (Fig. 8a). This decrease in pCO 2 was concomitant with changes in the fluxes of biodegradable total organic carbon (BTOC) discharged by the WWTPs of the Parisian conurbation managed and operated by the Greater Paris sanitation authority (French acronym SIAAP) after treatment (Fig. 8a). From the 1990s to 2015, the BTOC load decreased by 80% (from 13.8 to 2.8 kt BTOC yr −1 ) following treatment improvements on the SAV WWTP site, the construction of a new WWTP in 1991 and of three new WWTPs between 2005 and 2008, conjointly with improvement in treatment at existing plants. A positive linear relationship (R 2 = 0.52, n = 29, p < 0.05) was found between annual pCO 2 at the outlet of the Seine River (Poses) and BTOC fluxes from the SIAAP WWTPs (Fig. 8b). pCO 2 environmental controls. We found a positive linear relationship between pCO 2 and DOC (R 2 = 0.56, n = 119) (Fig. 9a). DOC measured in grasslands (DOC average: 10.3 mg L −1 ; SD: 5.8 mg L −1 ) and wetlands (DOC average: 21.0 mg L −1 ; SD: 14.6 mg L −1 ) showed wider and higher ranges of concentration compared to arable lands (DOC average: 3.8 mg L −1 ; SD: 2.7 mg L −1 ) (Fig. 9a). Generally, the ranges of DOC and pCO 2 were lower in winter and higher in summer/autumn and during the spring flood. No relationship was found between pCO 2 and DIC or nutrients (Fig. 9b, see Supplementary Material 2). The highest pCO 2 and the lowest oxygen concentrations were measured in anoxic wetlands, whereas the opposite was found in the Marne River reservoir, and overall, a negative relationship between pCO 2 and dissolved oxygen was observed for all land uses (R 2 = 0.22, n = 120, p < 0.05, Fig. 9c). We also found a positive relationship between pCO 2 concentrations and Chl. a concentrations (R 2 = 0.26, n = 113, p < 0.05, Fig. 9d).    Interestingly, pCO 2 values measured during the field campaigns in the Marne reservoir showed CO 2 undersaturation with respect to the atmospheric equilibrium, averaging 360 ppm in the reservoir and 413 ppm in the air; slight supersaturation (457 ppm (reservoir) and 402 ppm (air)) was only observed in the late summer/autumn samples. Chl. a concentrations in the reservoir were low (mean: 2.6 µg l −1 ) and DOC level was around 3.2 mgC l −1 , i.e., with no sign of eutrophication (Fig. 9a,d).
Discussion pCO 2 supersaturation of the Seine hydrosystem. Since the 1970s, the whole drainage network of the Seine basin has been supersaturated in pCO 2 with respect to the atmospheric equilibrium. Supersaturation was observed for 98.5% of computed pCO 2 as well as for the direct field measurements. These results are in agreement with those obtained in the lower reaches of other temperate rivers. In comparison to the mean of 3000 ppm at the outlet of the Seine basin, average pCO 2 in the Meuse River (Belgium) on the period 2011-2014 was found equaling 2004 ± 912 ppm and all samples were also supersaturated in CO 2 (min. 971 ppm) 14 . Such variations can be found within the Scheldt River estuary (Belgium) and measurements in five of its tributaries (Dender: 8300 ppm, Zenne: 5700 ppm, Dijle: 7252 ppm and Nete River: 6700 ppm, and an average of 9500 ppm for the lower Scheldt) 36 . Other Rivers as the Leyre (France) showed the same range of values (average: 4429 ppm, min.: 901 ppm max.: 23,047 ppm) 12 . At the global scale, pCO 2 in streams and rivers have been averaged at 1600 ppm in a range of 132 to 11,770 ppm 2 . The wide range of pCO 2 values in rivers were already mentioned with variations from 10 to 100 times the saturation value 37 .
Conversely, we measured undersaturation in the Marne reservoir with pCO 2 below or near atmospheric equilibrium, in agreement with the results reported by Crawford et al. 38 for river basins containing dam reservoirs. Riverine reservoirs have a higher residence time than the river itself, leading to particle sedimentation and a decrease in turbidity, conditions that favor primary production, i.e., consumption of CO 2 and production of oxygen. During our field campaigns, we did not observe eutrophication conditions (Fig. 9d) or relationships between pCO 2 and nutrients (see Supplementary Material 2). Without eutrophication of the reservoir, the biomass produced does not form an organic load that would -paradoxically -consume O 2 and release CO 2 39 . pCO 2 is known to be affected by metabolic processes related to nutrient availability 40 . In the Seine River, we could have expected a relationship with ammonium (NH 4 + ) as activated sludge treatment releases dissolved organic carbon and high ammonium load 21 . However, no direct relationship was found with NH 4 + or other nutrients, which shows the complexity of the controls on pCO 2 mentioned below.
Because the main stream of the Seine River was known for its phytoplankton blooms before domestic wastewater was efficiently treated 21,29 , we analyzed bloom events (Fig. 5) to try and identify possible short periods of undersaturation. Despite the fact we found evidence for the opposite pattern between phytoplankton (Chl. a) and pCO 2 for phytoplankton blooms above 50 µg Chl. a l −1 , the Seine River waters remained supersaturated.
This result supports the assumption that other environmental variables actively control pCO 2 in the Seine River.
Hydro-climatic controls on pCO 2 . As shown in Fig. 6, seasonal pCO 2 concentrations (in the long term) varied in parallel with temperature (i.e., with the highest values in summer/autumn) and opposite to hydrology. Such dynamics are typical for the temperate oceanic regime of the Seine River, with high discharge in winter and low discharge in summer/autumn 41 .
Hydro-climatic effects resulted from a combination of water temperature and hydrology leading to a seasonal increase in pCO 2 and CO 2 evasion fluxes (fCO 2 ) from winter to summer/autumn 6,14 . Indeed, the hypothesis of control by water temperature is strengthened by the results of the field campaigns for different land uses with increasing pCO 2 according to the season (pCO 2 in winter < spring < summer/autumn), which can be interpreted as an enhancement of DOC mineralization whatever the land use. However calculating pCO 2 at 10 °C, revealed that temperature effect on solubility is rather low. In addition, long term seasonal variations in pCO 2 suggest possible control by hydrological regimes (high pCO 2 in low flow periods). In fact, both water temperature and hydrological regimes (water velocity) contributed to the variations in the gas transfer velocity (k-values), and the associated fCO 2 . Moreover, for both k-values and fCO 2 we demonstrated opposite seasonal patterns in the upstream and downstream parts of the Seine River system, differences that could be more attributed to water velocities in small streams and water temperatures in higher stream orders (see equation 4, Fig. 4b-d). Higher k-values upstream, caused by higher turbulence, logically led to important CO 2 outgassing compared to the lower k-values of the lower Seine River, down to its outlet.
The highest pCO 2 values were measured during the exceptional flood when groundwater overflows may have reinforced pCO 2 in the surface water. These high in-stream pCO 2 levels were found concomitantly with high k-values (10-55% higher than levels measured in small streams in the other seasons) and would be expected to enhance CO 2 evasion from rivers to the atmosphere. Similar effects of hydro-climatic conditions have also been observed in the tropics, e.g., in a large Amazonian river with a 20% higher outgassing of CO 2 during extreme flood years than in other years 42 , and in the Zambezi River, with pCO 2 up to twofold higher during the wet season 43 . Polsenaere and Abril (2012) 44 compared two French streams and one river and observed that the stream with the highest concentration of CO 2 also had the highest CO 2 degassing flux.
Several authors have already suggested that climate change may alter the frequency and amplitude of flood events in the Seine River basin, with more extreme hydrological conditions [45][46][47] , so that pCO 2 and CO 2 evasion could increase in the future.  (Figs 3 and 9a). According to Arrouays et al. 48 , the organic carbon stocks in croplands are less than 4.5 kg C m −2 , but reach nearly 7.0 kgC m −2 in grasslands and forests and around 9.0 kgC m −2 in wetlands. These values are consistent with the higher carbon sequestration rate of grasslands and wetlands:104 ± 73 gC m 2 year −1 on average in Europe 49 . Thus, differences in pCO 2 according to land use here can be explained by the drainage of different organic soils and subsequent POC and DOC mineralization depending on water circulation and temperature. This result is clearly illustrated by the flood event flushes during the growing season when DOC (spring flood DOC median: 11.44 mgC l −1 ) and pCO 2 (spring flood median: 3297 ppm) reached their highest values.
In addition to the carbon leached from riparian zones and sediments, organic carbon can be leached from soils where spring biological activity had already built up a large quantity of biomass that is potentially subject to mineralization.
Organic carbon quality has also been shown to influence pCO 2 in streams in the North Central European plains in Germany and Poland 50 , and Belgium 14,51 . The biodegradable fraction of DOC is usually around 25% in upstream waters but may decrease to 5% in winter, and may be 50% in treated effluents 16 (Garnier et al., unpublished data). This supports lower observed pCO 2 in winter and higher values linked to WWTP effluents. Increasing biological mineralization of land-based organic matter (OM) in response to a rise in temperature 52 during the growing season, or increasing biodegradable DOC exports during high water or flood events 53 appear to be two major driving factors of pCO 2 .
DOC and pCO 2 inputs originating from land runoff and/or aquifer base flow (i.e., diffuse sources), are added to inputs from wastewater effluents (i.e., point sources, as wastewaters treated in specific plants are well localized).
Control of pCO 2 by urban effluents: long term evidence. Whereas hydro-climatic conditions and diffuse pCO 2 and DOC inputs appeared to determine the seasonal variations in pCO 2 , long term changes in pCO 2 observed over 1970-2015 suggested control by point sources, which are known to dominate observed changes in the Seine River 20,21,29,54 . Indeed, the long-term annual pCO 2 values in the urbanized main stream of the Seine River strictly mirror variations in releases of urban OM by the largest WWTP of the Paris conurbation. Until 1990, the wastewater collection rate was intensified but wastewater treatment was not improved 55,56 . Later on, the OM from discharged effluent decreased, with a stepwise increase in the number of WWTPs within the Parisian conurbation 55 and improved treatment processes, in response to both the Urban Wastewater Directive (1991/271/ EC) and the Water Framework Directive (WFD, 2000/60/EC). In 2012, the technique changed from activated sludge to fully operational tertiary treatment (nitrification in 2007 followed by 70% denitrification in 2012) 21 , and improved water quality in terms of organic pollution and nutrients. Subsequently, that helped in reducing pCO 2 concentrations and enabled to recover acceptable levels of dissolved oxygen 54 downstream of major urban releases in the lower Seine an estuary. Because CO 2 evasion pattern is likely to follow the pattern of pCO 2 (see Figs 3 and 4), our results would support those reported by Prasad et al. 57 . Indeed, they compared the urbanized Anacostia waters to the lower Potomac waters flowing into the Chesapeake Bay (U.S.A.), and showed similar effect of organic matter and nutrients from urbanized landscapes on CO 2 evasion.
Whatever the period studied during the last 45 years, point source organic pollution appeared to be the main determinant of pCO 2 downstream of the treated effluents discharged into the lower Seine River. However, hydro-climatic conditions also influence pCO 2 . For example, with no significant seasonal variations in OM fluxes discharged as point sources, higher pCO 2 concentrations in summer are explained by a low OM dilution rate during low waters and high temperatures.
Limits of the approach. The measurements we took in 2016 and 2017 showed neutral or basic carbonate buffered waters and DOC average seasonal concentrations of 3.5 to 11.4 mgC l −1 , excluding overestimation of calculated pCO 2 linked to the contribution of organic acids to TA 12 . Abril et al. 12 also emphasized the importance of accurate pH measurements. We think that the variability we found when establishing the relationship between measured pCO 2 and computed pCO 2 (Fig. 2) could be linked to the accuracy of pH measurements. As a result, the long-term pCO 2 analyses were subject to similar variability (see the 95% confidence intervals in Fig. 8a). Nevertheless, the amplitude of pCO 2 variations over the 45 years period enabled a robust analysis.
The choice of computing k using one of the equations provided by Raymond et al. 30 could lead to bias. Indeed the equation was proposed based on measurements made on small streams (median depth, 0.28 m) and during low flow (median discharge, 0.54 m 3 s −1 ). We took into account slope, water velocity -discharge divided by the wetted cross section-and water temperature, but not other physical or environmental factors causing turbulence in streams, e.g., water turbidity, bed frictions, the direction and the intensity of wind, and chemical or bio-films 30,44 . Although there is need for direct measurements of k in higher stream orders to reduce uncertainties in flux calculations, k-values calculated for the Seine River range between 0.08 m h −1 (in winter for the main stem) and 0.21 m h −1 (in winter for streams). At this stage, it is not possible to quantify the apportionment of pCO 2 originating from carbonated groundwater from that resulting from carbon mineralization or WWTP inputs. The modelling approach in progress should provide quantitative insights and δ 13 C-DOC/POC analysis could also be useful to identify the different sources of pCO 2 44 .

Conclusions
This study showed that since 1970, both small-order streams and urbanized downstream rivers in the Seine River basin have been supersaturated in CO 2 and a source of CO 2 to the atmosphere. CO 2 supersaturation with respect to the atmospheric equilibrium appeared to be controlled differently in space (depending on land uses or on the location of the main WWTP effluent discharge) and over time (seasonal or interannual). CO 2 supersaturation depended on complex interactions between land based and groundwater discharges (upstream diffuse sources), and urban pressures (downstream point sources) modulated by hydro-climatic factors.
In the small streams of the drainage network, in sparsely populated zones, the highest pCO 2 in summer was shown to originate from mineralization (increasing with water temperatures) of organic carbon from diffuse sources including in-stream bottom sediments, riparian and/or terrestrial soils varying according to land uses. Hydro-climatic variations, especially water velocity in small streams greatly affected gas transfer velocity, and helped determine in-stream pCO 2 (and evasion). During the exceptional flood event, high water discharges following a period of growth probably increased the DOC flushed from soils, leading to higher pCO 2 , especially in streams draining wetlands and grasslands. High pCO 2 in streams may be also linked to high pCO 2 of groundwaters that feed the surface water during low flow, and to the overflow of aquifers during floods, with particularly high pCO 2 .
Based on the 1970-2015 time series, point source organic pollution appeared to be the main driver of pCO 2 in the lower Seine River, downstream of the main outlet of WWTP effluents, and whatever the period studied. pCO 2 was highest in summer during low waters and high temperatures, and lower in winter when the discharged effluents were diluted. Despite the notable decrease in organic pollution following improvements in WWTPs since the 1990s, pCO 2 has remained higher than atmospheric values, strongly suggesting the influence of carbonated groundwater.
In the next step, a CO 2 budget of the Seine drainage network will help (i) quantify the role played by temperate human-impacted rivers in the global carbon budget, and (ii) estimate the amount of pCO 2 point sources vs. diffuse sources. The present study also points to the need for high frequency and more spatially resolved pCO 2 values and direct measurements of k. In addition, to anticipate the impact of climate change with the expected extreme hydrological conditions, further research is needed to understand the interactions between the terrestrial (soils and their land-use), and aquatic (hydrosystems 59 , groundwater discharges) compartments of watersheds.

Data Availability
The datasets generated during the current study are available from the corresponding author on reasonable request.