The PIREN-Seine program (https://www.piren-seine.fr), originating in 1989 as an applied inter-disciplinary program devoted to the Seine river system and co-constructed with regional water managers [Billen 2001], turned out to be a very fruitful source of inspiration for many research studies going well beyond the regional framework. The initial wide and long-term vision and the extreme open-mindedness of its founder, Professor Ghislain de Marsily, are certainly not unrelated to the unprecedented longevity of this still ongoing program.
The local stakes of river management, which were at the origin of the creation of the PIREN-Seine program were mainly related to water quality, endangered by pollution through untreated urban wastewater point release, particularly those of the huge Paris agglomeration (12 million inhabitants), as well as by diffuse sources of agricultural nutrients, pesticides and other toxic compounds [Meybeck et al. 1998; Garnier et al. 2022a]. Large sectors of the Seine River at that time were still oxygen deficient in summer, algal blooms were observed in the main tributaries, groundwater resources were threatened by nitrate and pesticides originating from one of the most intensive agricultural region of Europe, development of toxic algal species occurred in the Seine river plume in the English Channel [Cugier et al. 2005; Garnier et al. 2019b]. The PIREN Seine program, in close connection with the Seine Water Agency and other stakeholders involved in the application of National and European directives, contributed to bring about efficient responses to several of these problems, although those linked to agriculture are still far from being solved.
At the same time, the awareness of global environmental issues has grown up. The concept of planetary boundaries [Rockström et al. 2009] defining the safe operating space for humanity with respect to the global Earth System, lead to identify the three major issues at stakes globally: increased greenhouse gas emission leading to climate change, loss of biodiversity and perturbation of the nitrogen cycle. Looking closely, the local stakes of river water management and the major planetary challenges appears as two sides of a same coin.
Here we aim to illustrate the development of the concept of water-agro-food system that gradually emerged from the diverse researches in the PIREN-Seine program. This concept offers a general framework for understanding the major challenges of environmental transition in an integrated view [Garnier et al. 2015; Billen and Garnier 2021]. The essence of this approach is that of territorial biogeochemistry, derived from the territorial ecology concept [Buclet 2015; Barles et al. 2011]. It is based on the conviction that most traits of the function of a territory (a geographical space appropriated by a society) are revealed by the analysis of the matter fluxes that cross it and ensure its operation [Billen et al. 2007]. Understanding and modeling these flows requires identifying their main control mechanisms and formalizing them. This can only be done based on a continuous confrontation of the calculated results with the observed field reality, which in turn allows increasing the predictive capacity of the model and its generalization to other situations and its applicability on larger scales. The Seine River system and its watershed have often served as the “demo-case” for conceiving original concepts and methods and to confront them to field reality, before being further applied to other systems and possibly enlarged in scope.
We report the gradual extension of our approach, starting from a renewed vision of river pollution issues established for understanding nutrient flows through the continuum of ecosystems from soil and agricultural systems (including human food requirement and wastes) to river systems and the coastal seas (Land-to-Sea continuums). This approach, although developed on the basis of confrontation with concrete observations at the local or watershed scale, could also be extended to regional and global scales. Further, we will illustrate how such an approach is of considerable help to conceive and explore future visions for the agro-food systems of France, Europe and the World, and specifically one scenario combining changes towards more healthy human diet and a sustainable agro-ecological agriculture, allowing to preserve the quality of water resources and halving greenhouse gas emissions.
2. Understanding river water quality
A river can be viewed from different perspectives, among which are (i) a receptacle and an easy way of disposal of wastewater; (ii) an aquatic ecosystem and habitat for fish; (iii) an active conveyor of water and nutrients originating from the watershed to the sea. The challenge of the early PIREN-Seine program was to consider these three perspectives together in a same modelling approach. The water quality models of the 1980’s were dominated by the issue of oxygen deficits, a long lasting problem resulting from point wastewater release from Paris city [Gérardin 1875; Streeter and Phelps 1925; Lesouëf and André 1982]. Taking into account the kinetics of microorganisms in the representation of the self-purification processes marked a great progress in the capacity to explain and predict oxygen deficits in the Seine river, not only immediately downstream from the large wastewater discharge from Paris agglomeration, but also 150–200 km downstream, in the Rouen estuarine sector, where ammonia was nitrified after a considerable lag required for the development of a large enough population of nitrifying bacteria [Chestérikoff et al. 1992; Brion et al. 2000; Brion and Billen 2000; Garnier et al. 2007]. Also, in the 1990’s algal development in large rivers became an issue for drinking water production [Garnier et al. 1995, 2005]. Taking into account phytoplankton and bacterial metabolisms in river water quality models opened the way to their transformation into complete aquatic ecosystem models, considering the dynamics of autotrophic and heterotrophic microorganisms, and the related major forms of carbon and nutrients. The RIVE module of the microbiological processes involved in the ecological functioning of aquatic systems since then has grown considerably. It consists now of a generic representation of biogeochemical aquatic processes (Figure 1a). It has recently been fully coded in Python and published under GNU general public license (https://gitlab.in2p3.fr/rive/pyrive). RIVE is shared by a large number of modeling tools such as Riverstrahler [Billen et al. 1994; Garnier et al. 2002], Prose [Even et al. 1998], QualNET [Minaudo et al. 2018], and Barman [Garnier et al. 2000]. With a limited number of kinetic parameters, all of which have been experimentally determined in a variety of aquatic environments, subject to various hydrological and climatic constraints as well as to material inputs from the terrestrial systems, the biogeochemical functioning of these aquatic systems could be satisfactorily modelled.
A river system cannot be considered independently of its relationships with the terrestrial watershed, which controls not only the water flows into the tributaries, but also the quality of its surface water, defining the diffuse sources to the river system. The Riverstrahler model was developed within the PIREN-Seine program as a framework to formalize the relationships between an entire drainage network and the hydrological properties and land use of its watershed [Billen et al. 1994; Garnier et al. 2002] (Figure 1b). The same framework was still adopted in a Python version of the model, called PyNuts-Riverstrahler [Raimonet et al. 2018; Marescaux et al. 2020].
This Riverstrahler model is useful to describe quantitatively how nutrients brought from the watershed either as point or diffuse sources, are transferred, transformed, retained or eliminated during their travel through the drainage network down to the river outlet. Before reaching surface water however, flows of dissolved substances from the surface, sub-surface or groundwater runoff have to cross the riparian system, an often biogeochemically very active zone at the interface between land and river. The processes occurring in this riparian zone, including denitrification and N2O emissions have been explicitly taken into account in a recent version of the Riverstrahler model [Billen et al. 2018a, 2020a].
While most developments of the Riverstrahler model have been achieved and validated based on observations on the Seine River system, and often motivated by water management concerns in the Seine Basin, the modelling approach was applied to a large number of river systems in France [Garnier et al. 2018a, b, 2019b], Europe [Garnier et al. 2002; Billen et al. 2005; Thieu et al. 2009; Desmit et al. 2018] and other rivers in the world with differing climate and land cover [Nordic rivers, Sferratore et al. 2008; the subtropical Red River, Le et al. 2015].
3. Modelling agro-food systems at territorial scale
Understanding the diffuse losses of nutrients to ground- and surface water requires description of how biological material is transformed within the mosaic of terrestrial systems that make up the watershed. Agriculture from that respect plays a major role. Very detailed agronomical models, such as the STICS or EPIC models, are able to calculate the growth of major crops of conventional systems and the associated flows of water and nutrients, including nitrogen leaching, at the plot scale [Williams et al. 1989; Brisson et al. 2003]. Application of such models at larger regional scale has been achieved, and coupled with hydrological models to predict the level of groundwater contamination from a detailed description of farming practices [Ledoux et al. 2007; Beaudoin et al. 2021]. Such a modelling approach, involving the coupling of the EPIC model to a river system model, is also at the core of the SWAT model of water quality in regional watersheds [Neitsch et al. 2001].
There is a need to understand the flows of nutrients involved in agriculture at territorial scale, in relation with structural features of the agro-food system, such as crop- and livestock farming orientations, human diet and trade exchanges of food and feed across the territory. For this purpose, the GRAFS approach (Generalized Representation of agro-food systems) was designed [Billen et al. 2013, 2021; Le Noë et al. 2017, 2018]. GRAFS considers cropland, permanent grassland, livestock and human as the four main compartments where fluxes of nitrogen, carbon and phosphorus are transformed and channelized through the system (Figure 1c). With regards to nitrogen, cropland receives inputs as synthetic fertilizers, manure, atmospheric deposition and symbiotic fixation and transforms them into harvested crops with a surplus at the origin of environmental losses. Livestock is fed partly from harvested crops, from grassland and/or from imported feed. Human diet, and its share of animal and vegetable food products defines the fate of crop and livestock production, as well as the amount of human excretion. GRAFS can be used as a simple framework for establishing a comprehensive budget of these material flows (N, P, C) within the agro-food system in a given territory based on available agricultural statistics such as fertilizer application, crop areas and production, livestock numbers and production, etc. Alternatively, once such a full budget of material flows has been established, a set of functional relationships linking those fluxes can be calibrated, which then allows GRAFS to be used as a predictive model for the construction of scenarios based on some hypotheses regarding possible changes in the agro-food system.
The most important of these relationships is the one linking harvested yield of the cropping system (Y , expressed in nitrogen, kgN/ha/yr) to the total N soil input (F, kgN/ha/yr) whatever its form (synthetic fertilizers, manure, atmospheric deposition or symbiotic fixation by legume crops). It has been shown that a robust relationship exists between Y and F, integrated over the full crop rotation cycle under given pedo-climatic conditions, whatever the cropping system, either organic or conventional [Lassaletta et al. 2014; Anglade et al. 2015; Le Noë et al. 2017; Billen et al. 2018b]. This yield-fertilization relationship follows a hyperbolic function with a single parameter (Ymax, in kgN/ha/yr):
The second important set of relationships in the GRAFS model link the flow of material through livestock, relating edible production (edProd, in ktN/yr), excretion (Excr, in ktN/yr) and ingestion (Ingest, in ktN/yr), with only two dimensionless parameters, calibrated from observed data in real livestock system, the conversion efficiency of vegetal feed into edible animal products (cveff) and the non-edible to edible production ratio (ned:ed):
GRAFS also calculates the C flows through the agro-food system, including those associated with the non-harvested aerial or underground residues that are returned to the soil and feed the organic carbon pool of the soil, making it possible to address the issue of long term soil C sequestration by coupling GRAFS with a soil C dynamics model such as the AMG model [Le Noë et al. 2019a, b; Garnier et al. 2022a].
For gaseous N losses, empirical relationships with farming practices and climate variables have been established for calculating NH3 volatilization [Sanz-Cobena et al. 2014] and N2O emission [Garnier et al. 2019a]. As GRAFS calculates N surplus and leaching in all land use of a territory, it can be coupled with the Riverstrahler model for calculating the water quality resulting from a given operation of the agro-food system, allowing retrospective reconstruction of the past water quality [Billen et al. 2007].
4. Predicting the risk of coastal water eutrophication
Coastal phytoplanktonic production is controlled by the interplay of several factors including its morphology and hydrological properties as well as nutrient delivery from land base sources or direct discharge. An unbalanced availability of nutrients with nitrogen and phosphorus in excess over silica can lead to the preferential development of non-siliceous, often harmful, algal species rather than of diatoms which initiate linear trophic chains leading to fish [Billen and Garnier 2007; Garnier et al. 2010, 2021]. Such new production of non-siliceous algae, normally restricted to regenerated production, is the common characteristics of coastal eutrophication which can results in very diverse manifestations such as the accumulation of mucilaginous material, toxin production, or oxygen deficits caused by degradation of accumulated unpalatable algal biomass.
The ICEP has been proposed by Billen and Garnier  as an indicator of the eutrophication potential of nutrient delivery by river watersheds. It is defined as the excess of N or P over Si with respect to the requirements of balanced diatoms growth (according to Redfield ratios) and is expressed in kgC/day/km2 (i.e. an algal biomass potentially produced per km2 of watershed) for a direct comparison of N-ICEP and P-ICEP among river systems:
N- or P-ICEP is preferred according to the most limiting element in the marine environment.
The ICEP permits to assess the potential for non-siliceous algal growth to be sustained by N and P in excess over silica in the nutrient load of a given river and can easily be calculated by the Riverstrahler model, at the outlet of large rivers discharging into coastal seas (Figure 3a).
The ICEP indicator has recently been included as an indicator for marine eutrophication in the UN Sustainable Development Goals (SDG14: https://sdg.iisd.org/news/ioc-unesco-provides-update-on-sdg-14-indicator-development/).
The ICEP indicator does not take into account any characteristics of the receiving coastal zone. Yet, according to their morphological and hydrological peculiarities, the pelagic ecosystem of different bays can react quite differently to a given loading of nutrients. For this reason, a derived indicator, the B_ICEP (for Bay integrated indicator of coastal eutrophication potential) was further developed, combining ICEP with some characteristics of the specific receiving Bay [Garnier et al. 2021]. It is defined as the ratio of the riverine flux of nutrient in excess over silica to the volume of the receiving bay multiplied by its flushing rate.
So defined, the B-ICEP represents the maximum concentration of non-siliceous algae which can be developed based on excess riverine N and P over Si in a particular marine bay. The pertinence of the B_ICEP approach, i.e., the ability of this indicator to predict the order of magnitude of undesirable algal blooms in particular bays, has been demonstrated by using an idealized model of marine algal development [Garnier et al. 2021].
5. Integrated agro-ecological scenarios of future water-agro-food systems
The modelling chain GRAFS—Riverstrahler—B_ICEP has the potential for an integrated assessment of water quality over the whole continuum from headwaters to coastal sea resulting from a given scenario of the agro-food system. Here, as an example of the approach, we will explore some results from an extreme scenario which makes the assumption of a generalization of agro-ecological practices in agriculture. The purpose is to assess the capacity of such a scenario, which would consider organic farming without recourse to synthetic fertilizer, to feed the human population, as well as its effect on water quality and coastal eutrophication, and its ability to reduce greenhouse gas emission.
Three major levers have been combined in conceiving the scenario. The first one concerns human diet. WHO  considers a protein intake of 3.4 kgN-protein per capita per year, as the vital minimum requirement. Values twice as high are observed in western countries, and have been associated to important health problems [Willet et al. 2019]. Importantly, the share of animal-based protein consumption varies between less than 10% in some African countries to as much as 70% in rich western countries in Europe and the USA. The EAT-Lancet Commission recently recommended a Reference Healthy Diet which, compared to the current European diet, implies a strong reduction not only of total protein intake but also of the share of animal products Willet et al.  The assumption of the agro-ecological scenario presented here is very close to the EAT-Lancet one (Table 1).
Current average European diet per day (d) or per year (yr) in terms of apparent consumption [Westhoek et al. 2015] and prospective diet considered in the agro-ecological scenario exposed in this paper at the horizon 2050, close to the EAT-Lancet reference healthy diet
|Average current European diet||Prospective healthy diet 2050|
|Roots and tubers||0.8||0.25||100||80||0.09||80||64||0.07|
|Animal products (excluding fish and seafood)||3.15||54||1.25||25|
|Fish & seafood||1||2.9||27||27||0.29||5||23.5||24||0.25||5|
|Products without N||[kgNeq/cap/yr]∗||[kgNeq/cap/yr]∗|
Europe refers here to the EU28 plus UK, non EU ex-Yugoslavian countries, Norway and Switzerland.
∗Nequivalent in corresponding harvested product before extraction of oil or sugar (0.075 equN/kg oil and 0.013 kgN/kg sugar).
The second lever of the agro-ecological scenario is the adoption of organic farming practices, strictly banning the use of synthetic inputs (fertilizers and pesticides). There is large diversity of such practices world-wide, often adjusted by the experience of farmers to territorial climatic peculiarities [Altieri 2002; Altieri and Nicholls 2017; Compagnone et al. 2018; Garnier et al. 2016; Billen et al. 2021]. All of them use long and diversified crop rotations leaving ample room to grain and forage legumes, the symbiotic N2 fixation of which constitutes the main source of new fertilizing nitrogen input to the whole rotation. An extensive review of organic systems effectively in use in Europe led to a typology of these crop rotations, and to an estimate of their N2 fixation capacity (Figure 4).
The third lever to be operated in such a scenario concerns the reconnection of crop and livestock farming. Animals have not only the function of providing meat and milk to human: they are also the agent able to convey nutrients from legumes to other crops and from grassland to cropland. In our scenario, livestock density in each region is determined by the available resources of grass and forage crop production, with no recourse to feed imports from outside.
The results of the application of this scenario to the two contrasting regions taken as examples above, the Seine basin and the Great West, are illustrated in Figure 3b. The scenario would lead to reintroduction of livestock in the center of the Paris basin, while its density would be strongly reduced in Brittany. In the scenario, both regions would feed their human and livestock populations without import of feed and would still export ample food surpluses. Similarly, at the scale of France, this scenario predicts that French exportation could still amount 220 ktN/yr as vegetable products (40% of the current value) and 3 ktonN/yr as animal products (10% of current exports), without importing any soybean based feed [Billen et al. 2018b]. For Europe the conclusions are similar: the agro-ecological scenario would be able to feed the population without synthetic N fertilizers while still exporting cereals and animal products at rates similar to current levels to the rest of the world (Figure 5).
At the global scale, the possibility of feeding a population of 10 billion people by agro-ecological agriculture, with no deforestation, and less recourse to international trade (hence more food sovereignty) was demonstrated by several authors [Billen et al. 2015; Lassaletta et al. 2016; Erb et al. 2016], provided that a diet not exceeding 30–40% of animal products in the total protein ingestion is adopted everywhere. This work thus demonstrate that, contrarily to what is sometimes reported [Connor 2013, 2018; Barbieri et al. 2021], agro-ecological practices can feed the world provided they are combined with dietary changes and structural reorganization of the relationships between crop and livestock farming, allowing to close the nitrogen cycle.
The effect of this scenario on water quality can be assessed by comparing the leaching water nitrate concentration calculated by the GRAFS model between the current and the scenarized situation (Figure 6). Further, coupling GRAFS and Riverstrahler models shows how nitrate concentration would be reduced in the drainage network of the different rivers of the Western Atlantic façade and English Channel [Desmit et al. 2018; Garnier et al. 2019b, 2022b], reducing the risks of coastal eutrophication (Figure 4b).
Greenhouse gas emissions by agriculture are also predicted to be drastically reduced in this agro-ecological scenario, owing to reduced livestock numbers (hence reduction of methane emissions) and suppression of synthetic N fertilizers (hence reduction of CO2 emissions for fertilizer synthesis and N2O emissions from cropland (Figure 7).
For the Seine Basin, the Great West and France as a whole, the contribution of agriculture to total greenhouse gas emissions in terms of CO2 equivalent is estimated to have reduced by 33%, 74% and 50% respectively [Garnier et al. 2019a, 2022a].
The water-agro-food continuum concept, gradually developed in the past 30 years, was a guide for dealing with the tremendous complexity of the intertwined challenges of food production, international trade, environmental quality and climate changes. It stresses the central role that agriculture and food supply play in a globalized world. It has helped conceiving a vision of our future where these challenges would be met altogether. These challenges are global in nature as they ultimately concern the whole planet. However, the solutions, just like the research on which they are based, needs to be rooted in territories. The Seine watershed, from that perspective was a very fruitful case study, because it is a paradigmatic example of a territory organized around a huge metropolis of global significance, ans well as a highly specialized agricultural area with extremely intensive farming practices. This is likely to be one of the main reasons for the exceptional longevity and productivity of the PIREN-Seine program.
Conflicts of interest
Authors have no conflict of interest to declare.
The authors wish to express their deep gratitude to all their students and collaborators who, since three decades, were involved in the gradual elaboration of the approaches exposed in this paper. Ghislain de Marsily welcomed both of us in his lab where he ensured the best scientific environment for developing creative and interdisciplinary research. We are extremely grateful to him for his openness and attention. His catchphrase was: “I let the researchers do the research they want, because that is what they do best!”.