Page 106 - Modelisation du devenir des pesticides...

Chapitre 3 Evaluation du modèle STICS version pesticide sur différents sites
expérimentaux
the collect of in-situ data is a difficult and almost impossible task when considering
the large amount of pesticide applied in the field and their associated metabolites.
For instance, in the Orgeval basin, a 100 km2 basin in the Northern France, 84
chemical pesticide families were used during the last 20 years 87 .
Determining the soil properties according to the soil texture and classical pedo-
transfer function can double the flux, while taking into account some soil variabi-
lities can lead to a 30% variation. Soil properties do not only affect the pesticide
transfer, but any solute. Thus, the evolution of the leaching of pesticide was com-
pared to the one of Bromure. The differences between passive solute and pesticide
fluxes are strong, and vary according to the pesticide. Indeed, change of the field
capacity and wilting point affects about 100 times more the Bentazone accumula-
ted leachate at Vredepeel than the bromure one, while it is almost the opposite for
the Atrazine in Grignon. Quite similar impacts are found for the Isoproturon and
bromure in Kerlavic. Such large differences are difficult to anticipate, but are lin-
ked to the adsorption and degradation properties of the pesticide. Such sensitivity
underlines again the need for in-situ determination of the soil properties.
To complete these sensitivity tests, the comparison with the observed bentazone
profiles at Vredepeel was also performed and the statistical results summarized in
Table 3.11 can be compared to the reference simulation.
Although Koc and DT50 have rather similar impact on the leaching flux, the
Koc has a dominant impact on the quality of the model prediction, all tested value
leading to a larger overestimation of the pesticide content (liquid + linearly ad-
probed bentazone) and a negative Nash. In the opposite, the three tested values of
DT50 lead to larger degradation, which tends to reduce the error on the reaming
part of pesticide in the soil profile. The increase of model efficiency suggested that
DT50 values measured in the laboratory do not reflected the transformation dy-
namics in the field, in accordance with the results already presented in part 3 .
Surprisingly, the use of constant organic profile, although having weak impact on
the leaching flux, provides the best Nash and lower NRMSE. As organic matter
is mainly located in the top soil, such results might be surprising. The improve-
ments are linked to a more homogeneous sorption, and may compensate for error
on the soil parameter input. The influence of soil parameterization has a weaker
impact on the statistical comparison with the observed pesticide profile than on
the leaching flux, although it leads to a lower concentration. Such analysis high-
lights the difficulty to assess the leaching flux based on concentration profile, since
concentration is affected both by degradation and leaching.
As several intercomparison of pesticide fate models was done at Vredepeel, si-
milar statistical results on similar dates were computed with stics.
Table 3.11 presents for the root mean square error and the efficiency the scores
obtain by Stics and the minimal, maximal and median value of the 10 models

94