Inland waters are an active component of the carbon cycle where
transformations and transports are associated with carbon dioxide (
Along the main stem of the Seine River, simulations of DIC, total alkalinity, pH and
Results from metabolism analysis in the Seine hydrographic network
highlighted the importance of benthic activities in headwaters while
planktonic activities occurred mainly downstream in larger rivers. The net
ecosystem productivity remained negative throughout the 4 simulated years
and over the entire drainage network, highlighting the heterotrophy of the
basin.
Rivers have been demonstrated to be active pipes for transport, transformation, storage and outgassing of inorganic and organic carbon (Cole et al., 2007). Although there are large uncertainties in the quantification of flux from inland waters, carbon dioxide (
Direct measurements of
The Seine River (northwestern France) has long been studied using the
biogeochemical riverine Riverstrahler model (Billen et al., 1994; Garnier et al., 1995), a generic model of water quality and biogeochemical functioning of large river systems. For example, the model has made it possible to quantify deliveries to the coastal zone and understand eutrophication phenomena (Billen and Garnier, 1999; Billen et al., 2001; Passy et al., 2016; Garnier et al., 2019), nitrogen transformation and
The purpose of the present study was to quantify the sources, transformations, sinks and gaseous emissions of inorganic carbon using the
Riverstrahler modeling approach (Billen et al., 1994; Garnier et al., 2002;
Thieu et al., 2009). A further aim in newly implementing this
Situated in northwestern France, 46
Characteristics of the Seine basin:
The maximum water discharge of these tributaries occurs during winter with the lowest temperature and rate of evapotranspiration; the opposite behavior is observed during summer (Guerrini et al., 1998).
Except for the crystalline rocks in the north and from the highland of the Morvan (south), the Seine basin is for the most part located in the lowland Parisian basin with sedimentary rocks (Mégnien, 1980; Pomerol and Feugueur, 1986; Guerrini et al., 1998). The largest aquifers are in carbonate rock (mainly limestone and chalk) or detrital (sand and sandstone) material separated by impermeable or less permeable layers.
The concept of Strahler stream order (SO) (Strahler, 1957) was adopted for describing the geomorphology of a drainage network in the Riverstrahler model (Billen et al., 1994). The smaller perennial streams are in order 1. Only confluences between two river stretches with the same SO produce an increase in Strahler ordination (SO
Hydro-morphological characteristics of the Seine drainage network,
The Seine basin is characterized by intensive agriculture (more than 50 %
of the basin; EEA, 2012) and is densely populated. The population is mostly concentrated in the Paris conurbation, which had 12.4 million inhabitants in 2015 (Fig. 1) (INSEE, 2015). Located 70 km downstream of Paris, the largest wastewater treatment plant in Europe (Seine Aval, SAV WWTP) can treat up to
The core of the biogeochemical calculation of the pyNuts-Riverstrahler
model (described hereafter) is the RIVE model (e.g., Billen et al., 1994; Garnier et al., 1995, 2002; Servais et al., 2007)
(
The model also describes benthic processes (erosion, organic matter degradation, denitrification, etc.) and exchanges with the water column with the explicit description of benthic organic matter, inorganic particulate P and benthic biogenic Si state variables. The benthic component does not explicitly represent all the anaerobic reduction chains, denitrification being the major anaerobic microbial process.
A detailed list of the state variables of the RIVE model is provided in Sect. S1 in the Supplement. Most of the kinetic parameters involved in this description have been previously determined through field or laboratory experiments under controlled conditions and are fixed a priori (see detailed description of all kinetics and parameter values in Garnier et al., 2002). To date, there has been no explicit representation of inorganic carbon in the RIVE model (see this new input in Sect. S1).
Riverstrahler allows for the calculation of water quality variables at any point in the aquatic continuum based on a number of constraints characterizing the watershed, namely, the geomorphology and hydrology of the river system and the point and diffuse sources of nutrients.
A drainage network can be described as subbasins (tributaries) connected to one or several main axes that define a number of modeling units. The modeling approach considers the drainage network as a set of river axes with a spatial resolution of 1 km (axis object), or they can be aggregated to form subbasins that are idealized as a regular scheme of tributary confluences where each stream order is described by mean characteristics (basin object). Here, the Seine drainage network starts from its headwater, ends at its fluvial outlet (Poses), and was divided into 69 modeling units, including six axes (axis object) and 63 upstream basins (basin object). A map and a table introducing the main characteristics of the modeling units are provided in Sect. S2.
Runoffs were calculated over the whole Seine basin using water discharge
measurements at 48 gauged stations (source: Banque Hydro database,
Water temperature was calculated according to an empirical relationship, adjusted on inter-annual averaged observations (2006–2016), and describes seasonal variation of water temperature in each Strahler order with a 10 d time step (see Sect. S2).
Riverstrahler manages the calculation of the RIVE model according to a Lagrangian routing of water masses along the hydrographic network (Billen et al., 1994) and is a generic model of water quality and biogeochemical functioning of large drainage networks that simulates water quality. PyNuts is a modeling environment that can calculate the constraints (diffuse and point sources) on the Riverstrahler model at a multiregional scale (Desmit et al., 2018, for the Atlantic façade).
The carbonate system was described by a set of equations (named the
Schematic representation of the ecological RIVE model (initially developed by Billen et al., 1994, and Garnier and Billen, 1994), with gray lines indicating the main processes simulated in the water column and at the interface with sediment (oxygen not shown), and implementation of the new inorganic module, based on total alkalinity (TA, maroon) and dissolved inorganic carbon (DIC, blue).
The exchange of
Stoichiometry of the biogeochemical processes, influencing dissolved inorganic carbon (DIC) and total alkalinity (TA) in freshwater, as taken into account in the new inorganic carbon module. TA and DIC expressed in mol : mol of the main substrate (either C or N).
These processes affecting TA and DIC result in equations governing inorganic
carbon dynamics as follows:
The different values of constants and parameters used in the inorganic carbon module are introduced in Table 1 of Sect. S3.6. The full inorganic carbon module is described in Sect. S3 (Sects. 3.1 to 3.6).
Diffuse sources are calculated at the scale of each modeling unit, based on several spatially explicit databases describing natural and anthropogenic constraints on the Seine River basin. Diffuse sources are taken into account by assigning a yearly mean concentration of carbon and nutrients to subsurface and groundwater flow components, respectively. These concentrations are then combined with a 10 d time step description of surface and base flows to simulate the seasonal contribution of diffuse emissions to the river system. For nutrients, several applications of the Riverstrahler on the Seine River basin refined the quantification of diffuse sources: e.g., Billen and Garnier (1999) and Billen et al. (2018) for nitrogen; Aissa-Grouz et al. (2018) for phosphorus; Billen et al. (2007), Sferratore et al. (2008) and Thieu et al. (2009) for N, P and Si. In this study we revised our estimates for diffuse organic carbon sources and propose TA and DIC values for the Seine basin. The summary of all the carbon-related inputs of the model is provided in Table 3.
Summary of the carbon related inputs of the pyNuts-Riverstrahler model.
Dissolved organic carbon (DOC) input concentrations were extracted from the
AESN database (
Total POC inputs were calculated based on estimated total suspended solid (TSS) fluxes, associated with soil organic carbon (SOC) content provided by the LUCAS Project (samples from agricultural soil; Tóth et al., 2013), the BioSoil Project (samples from European forest soil; Lacarce et al., 2009) and the Soil Transformations in European Catchments (SoilTrEC) project (samples from local soil data from five different critical zone observatories (CZOs) in Europe; Menon et al., 2014; Aksoy et al., 2016). TSS concentrations were calculated using fluxes of TSS provided by
WaTEM-SEDEM (Borrelli et al., 2018) and runoffs averaged over the 1970–2000 period (SAFRAN-ISBA-MODCOU, SIM; Habets et al., 2008). The POC mean was 8.2 mg C L
DIC and TA are brought by subsurface and groundwater discharges (Venkiteswaran et al., 2014). DIC is defined by the sum of bicarbonates (
To calculate DIC concentrations in groundwater, we therefore used our own
Boxplots of total alkalinity (
Documenting TA and DIC diffuse sources based on MESO units ensures a representation of their spatial heterogeneity in the Seine River basin. Carbonate waters showed higher TA and DIC mean concentrations while crystalline waters had the lowest mean concentrations in TA and DIC (primary and anterior basements from the Devonian; Fig. 3). Aquifers from the Tertiary and alluvium from the Quaternary had a more heterogeneous distribution of their concentrations (Fig. 3). TA and DIC by MESO units were then spatially averaged at the scale of each modeling unit of the pyNuts-Riverstrahler model (69 modeling units, subdivided according to Strahler ordination; Sect. S2), thus forming a semi-distributed estimate of groundwater concentrations.
TA and DIC measurements in lower-order streams cannot be considered as
representative of subsurface concentrations because lower-order streams are
expected to degas strongly in a few hundred meters, as shown for
The pyNuts-Riverstrahler model integrates carbon and nutrient raw emissions from the local population starting from the collection of household emissions into sewage networks until their release after specific treatments in WWTPs. In the Seine River basin, most of these releases are adequately treated before being discharged to the drainage network. DOC discharge from WWTPs was described according to treatment type, ranging from 2.9 to 9.4 g C per inhabitant per day while POC discharge ranged from 0.9 to 24 g C per inhabitant per day based on the sample of water purification treatment observed in the Seine basin (Garnier et al., 2006; Servais et al., 1999).
TA and DIC were measured at eight WWTPs selected to reflect various
treatment capacities (from
Nutrients and organic carbon cycling within the three reservoirs of the Seine River network were simulated using the biogeochemical RIVE model adapted for stagnant aquatic systems (Garnier et al., 1999). Owing to the absence of an inorganic carbon module in the modeling of reservoirs, we used mean measurements of TA and DIC in reservoirs as forcing variables to the river network. The Der lake reservoir was sampled 3 times (24 May 2016, 12 September 2016, 16 March 2017) and, among others, TA and DIC were measured (see Table 3). Recent sampling campaigns showed that TA and DIC are similar for the three reservoirs (Xingcheng Yan, personal communication, 2019).
We selected the 2010–2013 timeframe for setting up and validating the new
inorganic module. This period includes the year 2011, which was particularly
dry in summer (mean annual water discharge at Poses, 366 m
The
All data were processed using R (R Core team, 2015) and QGIS (QGIS Development Team, 2016). Kruskal–Wallis tests were used to compare simulated and measured
Root mean square errors normalized to the range of the observed data (NRMSEs) were used to evaluate the pyNuts-Riverstrahler model including the inorganic module, indicating the variability of the model results with respect to the observations, normalized to the variability of the observations. NRMSE analyses were performed on inter-annual variations once every 10 d for the 2010–2013 period, combining observations and simulations at four main monitoring stations along the longitudinal profile of the Seine River: Poses, Poissy (downstream of Paris), Paris and Ferté-sous-Jouarre (upstream of Paris).
Simulations of
Carbon dioxide concentrations in the Seine waters (
In the same period (2010–2013), a focus on the main stem from the Marne
River (SO6) until the outlet of the Seine River (Poses, SO7) showed that the
model correctly reproduced longitudinal variations. Higher concentrations of
Observed (dots) and simulated (line) mean carbon dioxide concentrations (
The 10-day simulated (lines) and observed (dots) water discharges over
the 2010–2013 period (
Upstream, within Paris, and downstream of Paris, the model provides
simulations in the right order of magnitude of the observed
For DIC, simulations upstream from Paris (Fig. 6, right) seemed lower than the observations (but summer data are missing); however, downstream at the
other three stations selected, simulations accurately represented the
observations (Fig. 6, NRMSE
Although the level of phytoplankton biomass was adequately simulated, the
summer bloom observed at the outlet was not reproduced, whereas the early
spring bloom observed in the lower Seine was simulated with a time lag compared to the observations (Fig. 6, bottom, NRMSE
The way of taking into account the gas transfer velocity in the modeling
approach could explain these discrepancies in SO6 and SO7 (Fig. 4). Different values of
Influence of the gas transfer velocity formalisms along the main stem of the Seine River basin (Marne–lower Seine River) impacted riverine
Therefore, for river widths greater than 100 m, a
Although these results can be improved, organic and inorganic carbon and total alkalinity budgets can be calculated at the scale of a whole drainage basin for the first time.
We established an average inorganic and organic budget for the period studied (2010–2013) (Table 4). The budget of inorganic and organic carbon (IC and OC) of the entire Seine River basin (from headwater streams to the beginning of the estuary) showed the high contribution of external inputs (sum of point and diffuse sources accounted for 92 % and 68 % of IC and OC inputs, respectively) and riverine exports (68 % and 66 % of IC and OC outputs, respectively). These exports were at least 1 order of magnitude higher for the IC budget (Table 4). The substantial contribution of the Seine aquifer water flow led the IC flux brought by groundwater to dominate over those from the subsurface (57.5 % vs. 34 % of total IC inputs, respectively), while for OC, the subsurface contributions were higher than the groundwater contributions (54 % vs. 14 % of the total OC fluxes).
Budget of the Seine hydrosystem for inorganic and organic carbon
(kg C km
Interestingly, the relative contributions of point sources to OC inputs were higher than for IC (23 % and 7 % of the OC and IC inputs, respectively) (Table 4).
Heterotrophic respiration by microorganisms accounted for only 1.5 % of the IC inputs. Similarly, IC losses by net primary production also accounted for a small proportion, i.e., 0.6 %, of the IC inputs. For the OC budget, despite a contribution of autochthonous inputs from instream biological metabolisms (net primary production, NPP, and nitrification: 9 % of inputs; heterotrophic respiration: 7 %), which was relatively high compared with their proportion in IC fluxes (2.3 %), allochthonous terrestrial inputs still dominated the OC budget (Table 4).
The Seine River, at the outlet, exported 68 % of the IC entering or produced in the drainage network, and 66 % of the OC brought to the river
(including both particulate and dissolved forms) (Table 4). Instream OC
losses were related to heterotrophic respiration (7 %) and to a net transfer to the benthic sediment compartment, including sedimentation and
erosion processes (estimated at 28 % of losses). In the IC budget,
A similar calculation was performed for the TA budget. As
for inorganic carbon, the contribution of internal processes remained
relatively low compared with the high levels of TA in lateral inputs
(diffuse sources: 93 %; point sources: 6 %) and flows exported to the basin outlet (97 %). Indeed, instream production mostly relied on
heterotrophic respiration (
Whereas IC and OC budgets of the Seine hydrosystem were clearly dominated by external terrestrial inputs and outputs through deliveries at the coast, an attempt was made here to analyze instream processes involved in the IC and OC cycles (Figs. 8 and 9).
Instream processes involved in the inorganic carbon cycle simulated by pyNuts-Riverstrahler and averaged over the 2010–2013 period for the Seine River network until its fluvial outlet at Poses.
Metabolism for small, intermediate and large stream orders (SO) (here represented by SO1, SO5 and SO7, respectively) of the Seine basin simulated by pyNuts-Riverstrahler and averaged over the 2010–2013 period: net primary production (NPP, g C m
The average spatial distribution of IC processes, as calculated by the model, was mapped for the 2010–2013 period (Fig. 8). Benthic activities were the greatest in smaller streams. By contrast, net primary production and heterotrophic planktonic respiration, which both followed a similar spatial pattern, increased as Strahler order increased, reaching their highest values in the lower Seine River. All these biological processes involved in the IC cycle were therefore highly active in the main stem of the river, while on the other hand
Regarding the OC processes, mostly linked to biological activity, they were
analyzed in terms of ecosystem metabolism (Fig. 9). The net ecosystem
production (NEP, g C m
Simulations showed that NEP would remain negative in the entire drainage
network (Fig. 9). However, NEP must be analyzed with caution since the
phytoplankton pattern was not adequately represented (see Fig. 6). In SO1,
this negative NEP was associated with almost no NPP, and heterotrophic
respiration was dominated by benthic activities (see Fig. 8). In SO5, NEP was less negative than in SO1 (Fig. 9), and heterotrophic respiration was lower than in SO1 while NPP was higher. In the lower Seine River (SO7), NPP increased as did heterotrophic respiration, which reached its highest value
in this downstream stretch receiving treated effluents from WWTPs. Therefore, the increase in NPP did not result in positive NEP. The entire drainage network was thus supersaturated in
Simulated
Regarding gas transfer velocity values, an equation for large rivers with no tidal influence using wind speed could be more appropriate (Alin et al., 2011) and could decrease NRMSE in these downstream sections of the river. However, the Riverstrahler model does not consider wind as an input variable, which would have required the model to have a much higher spatiotemporal resolution to reflect its spatiotemporal heterogeneity in the Seine basin, with for example the diurnal cycle affected by phenomena such as breezes (Quintana-Seguí et al., 2008).
Future work with direct
Regarding seasonal patterns, DIC and alkalinity amplitudes were suitably
captured and the level of the values was correct. DIC and TA observations
showed a strong decrease from June–July to November (maximum amplitude
decrease, 10 mg C L
The model showed a weak performance in representing
The new implementation of an inorganic carbon module in the
pyNuts-Riverstrahler model allows us to estimate
The outgassing found for the Seine River by the surface area of the river of
Regarding organic carbon, Meybeck (1993) estimated the DOC export to the ocean for a temperate climate at 1.5 g C m
We estimated the DIC export of the Seine River at
Model simulations with the new inorganic carbon module can be used to
analyze spatial variations of
The model highlights the importance of benthic activities in headwater streams (Fig. 8) that decreased downstream as heterotrophic planktonic activities increased in larger rivers, a typical pattern described by the
river continuum concept (RCC; Vannote et al., 1980) and quantified for the Seine River (Billen et al., 1994; Garnier et al., 1995; Garnier and Billen, 2007). These results are also in agreement with those reported by Hotchkiss et al. (2015), who suggested that the percentage of
Mean NEP would remain negative in the entire basin, resulting from
heterotrophic conditions producing
During the recent 2010–2013 period studied herein, and in all SOs, the NPP
never exceeded heterotrophic respiration (ratio of NPP to het. resp less than 1 or
The pyNuts-Riverstrahler model of biogeochemical river functioning now
includes the processes involved in the inorganic carbon cycle in order to
represent the spatial dynamics and seasonal variations of
Our Riverstrahler modeling has shown that there are many factors that control
The datasets generated during the current study are available from the corresponding author on reasonable request.
The supplement related to this article is available online at:
All the authors contributed to the design of the study. JG and VT are co-supervisors of the PhD. AM participated as a PhD student in the field campaigns, lab chemical analyses and implementation of the new inorganic carbon module. NG and MS provided technical and scientific support for the modeling. AM wrote the first draft of the manuscript, and all the co-authors helped to interpret the data and write the article.
The authors declare that they have no conflict of interest.
The project leading to this paper received funding from the European Union's Horizon 2020 research and innovation program under the Marie Sklodowska-Curie grant agreement no. 643052. A PhD grant was attributed to Audrey Marescaux. Many thanks are due to Sébastien Bosc, Anunciacion Martinez Serrano and Benjamin Mercier for their kind participation in the fieldwork and for their assistance with chemical analyses in the lab. We thank Emmanuel Soyeux (Veolia Water, France), Muriel Chagniot (Veolia Water, France), and the operators of the Veolia WWTPs for their precious help in organizing the field campaigns. The SIAAP (Vincent Rocher) is also sincerely acknowledged for their contribution to sampling the largest WWTP of the Paris conurbation and the long-term view on treatments in the SIAAP WWTPs provided by their recent book (Rocher and Azimi, 2017). Vincent Thieu (assistant professor at Sorbonne University, Paris) and Josette Garnier (Research Director at the Centre National de la Recherche Scientifique, France) are co-supervisors of the PhD. Nathalie Gypens is Professor at the Université Libre de Bruxelles (Belgium). Marie Silvestre is GIS Engineer at the Centre National de la Recherche Scientifique (France).
This research has been supported by the Marie Sklodowska-Curie grant (grant no. 643052).
This paper was edited by Anas Ghadouani and reviewed by three anonymous referees.