Page 101 - Modelisation du devenir des pesticides...

3.3 Results and discussion

Table 3.8 presents the comparison between the normalised root mean square
error and the efficiency scores obtained by STICS and the minimum, maximum
and median values obtained at Vredepeel from 10 models, 69 for the sampling
dates (103, 278 and 474 DAA) , and at Thiverval-Grignon from 3 models. 58
The results at Vredepeel are given for both calibrated and uncalibrated runs.
Compared to other capacity models tested at Vredepeel, STICS predicts bentazon
residues that are better reproduced (i.e. less overestimated). 64,65 The statistical
scores obtained by STICS are in the same range as for Richards type models. An
exception is found for the efficiency on the bentazon profiles, which is within the
range of uncalibrated values but not in the range of calibrated values. This is also
consistent with the fact that the reference run in STICS was not calibrated. A
detailed comparison is first made with 2 capacity models, PELMO and PLM, and
then with the Richard’s type model PESTLA. The bromide concentration profiles
simulated with the PELMO model are close to the STICS simulation at day 103
and 278, whereas there is greater discrepancy between the two models for the soil
water content and bentazon concentration profile. 26 The discrepancies between the
STICS and PELMO predictions for Bentazone profiles are mainly attributed to
DT50 values and sorption parameters. Moreover, the Bentazone profiles estimated
by STICS are relatively close to simulations performed with PLM, 28 which is
linked to the use of the same value for field capacity and wilting point. However,
the deviations between the two models are attributed to the different value of
DT50 , i.e. 62 day for STICSa and 44 or 100 days for PLM. 28 In accordance
with the results obtained with PELMO and PLM, the poor predictions of the soil
moisture profile and bromides have a weak effect on the accuracy of the bentazon
prediction regarding pesticide properties. The STICS parametrization for DT50 is
similar to the value chosen by JB for the PESTLA application at the Vredepeel
site. 62 Despite the use of the same value for DT50, and a weak sorption parameter,
a large discrepancy is found between the two models in the Bentazone predictions
at day 103 and 278 DAA. The maximum concentration peaks predicted by STICS
are close to PESTLA results, but the Bentazone distributions along the profile
are differents. The difference can mainly be attributed to the formalism used to
simulate water transfer, since the PESTLA model is able to simulate the variation
of the shallow ground water. Nevertheless a good match between observed and
simulated concentration profiles do not necessary imply an accurate simulation of
pesticide losses via leaching. 62 At Kerlavic, only the Agriflux model was used for
year 2002/2003 , 56 which restricted the comparison to one year. The drainage flow
and the loss of isoproturon by leaching is poorly predicted by STICS compared to
Agriflux. The bias on drainage flows predicted by Agriflux ranges from -15 to 9 %,
whereas the bias obtained with STICS is close to 25%. Moreover the isoproturon
flow simulated with Agriflux is less underestimated with bias from -42 to 12%56as

89