of official agricultural statistics available at the département scale (typically
6,000 km
2 ). It provides the required data for running the Riverstrahler model
[6, 7] (www.fire.upmc.fr/rive), which calculates the nutrient transfers and the ecological functioning of each tributary of the river system, given the diffuse and point
sources of nutrient and organic matter from the watershed. The calculated nutrient
fluxes at the outlet of the river system can then be used by a coastal marine model
such as ECO-MARS 3D to assess the eutrophication generated by these fluxes [8–
11].
The ARSeine database [12] offers a spatially detailed and distributed description
of the Seine-Normandie cropping systems over the 1970–2015 period, including
land use, crop rotations and detailed management techniques at the Petites Régions
Agricoles scale (typically 1,000 km
2 ). It has been designed to provide the inputs to a
2D-distributed version [13] of the STICS model [14–18]. STICS is an agronomical
crop model simulating crop production and the components of the N cycle at the
same space and time resolution. Input soil parameters have been defined for each soil
unit of the Soil Geographic Database of France at the 1:1,000,000 scale [19], using
local pedotransfer functions [20]. Daily values of nitrate leaching predicted by
STICS are used as an input to the hydrogeological MODCOU model [21], which
calculates the recharge and nitrate contamination of the basin’s main aquifer formations [13, 22].
Evaluating the uncertainty on the results of such long-term reconstruction of
environmental data is a critical task. As far as modelling approaches are concerned,
two types of uncertainty can be distinguished: structural uncertainties related to the
adequacy of the model’s representation of the system and operational uncertainties
related to the accuracy in the data and parameters used [23]. The latter can be
evaluated using Monte Carlo methods to assess how uncertainty on the raw data
propagates to final model results; this approach shows typical uncertainties of
approximately 25% for the GRAFS approach [4]. Structural uncertainties are by
essence much more difficult to assess. They have been roughly estimated at 15% for
the STICS model [20].
3 Trajectory and Biogeochemical Functioning
of the Agricultural System
3.1 Long-Term Changes in the Structure of the Northern
France Agricultural System
Until the beginning of the twentieth century, mixed crop and livestock farming
systems dominated everywhere in France (Fig. 2a). Manure and symbiotic N fixation
by grassland and legume crops inserted in rotations were the only sources of
cropland fertilisation. Specialisation into stockless cropping systems, relying on
The Seine Watershed Water-Agro-Food System: Long-Term Trajectories of C. . .
95
6,000 km
2 ). It provides the required data for running the Riverstrahler model
[6, 7] (www.fire.upmc.fr/rive), which calculates the nutrient transfers and the ecological functioning of each tributary of the river system, given the diffuse and point
sources of nutrient and organic matter from the watershed. The calculated nutrient
fluxes at the outlet of the river system can then be used by a coastal marine model
such as ECO-MARS 3D to assess the eutrophication generated by these fluxes [8–
11].
The ARSeine database [12] offers a spatially detailed and distributed description
of the Seine-Normandie cropping systems over the 1970–2015 period, including
land use, crop rotations and detailed management techniques at the Petites Régions
Agricoles scale (typically 1,000 km
2 ). It has been designed to provide the inputs to a
2D-distributed version [13] of the STICS model [14–18]. STICS is an agronomical
crop model simulating crop production and the components of the N cycle at the
same space and time resolution. Input soil parameters have been defined for each soil
unit of the Soil Geographic Database of France at the 1:1,000,000 scale [19], using
local pedotransfer functions [20]. Daily values of nitrate leaching predicted by
STICS are used as an input to the hydrogeological MODCOU model [21], which
calculates the recharge and nitrate contamination of the basin’s main aquifer formations [13, 22].
Evaluating the uncertainty on the results of such long-term reconstruction of
environmental data is a critical task. As far as modelling approaches are concerned,
two types of uncertainty can be distinguished: structural uncertainties related to the
adequacy of the model’s representation of the system and operational uncertainties
related to the accuracy in the data and parameters used [23]. The latter can be
evaluated using Monte Carlo methods to assess how uncertainty on the raw data
propagates to final model results; this approach shows typical uncertainties of
approximately 25% for the GRAFS approach [4]. Structural uncertainties are by
essence much more difficult to assess. They have been roughly estimated at 15% for
the STICS model [20].
3 Trajectory and Biogeochemical Functioning
of the Agricultural System
3.1 Long-Term Changes in the Structure of the Northern
France Agricultural System
Until the beginning of the twentieth century, mixed crop and livestock farming
systems dominated everywhere in France (Fig. 2a). Manure and symbiotic N fixation
by grassland and legume crops inserted in rotations were the only sources of
cropland fertilisation. Specialisation into stockless cropping systems, relying on
The Seine Watershed Water-Agro-Food System: Long-Term Trajectories of C. . .
95