Catena 119 (2014) 97–103

Contents lists available at ScienceDirect

Catena
journal homepage: www.elsevier.com/locate/catena

Application date as a controlling factor of pesticide transfers to surface

water during runoff events
Laurie Boithias a,b,⁎,1, Sabine Sauvage a,b, Raghavan Srinivasan c, Odile Leccia d, José-Miguel Sánchez-Pérez a,b,⁎
a
University of Toulouse, INPT, UPS, Laboratoire Ecologie Fonctionnelle et Environnement (EcoLab), Avenue de l'Agrobiopole, 31326 Castanet Tolosan Cedex, France
b
CNRS, EcoLab, 31326 Castanet Tolosan Cedex, France
c
Spatial Sciences Laboratory, Texas A&M University, College Station, TX 76502, USA
d
Irstea, Adbx Research Unit, 50 avenue de Verdun, 33612 Gazinet-Cestas Cedex, France

a r t i c l e i n f o a b s t r a c t

Article history: In agricultural watersheds, pesticide contamination in surface water mostly occurs during stormflow events.
Received 2 September 2013 When modelling pesticide fate for risks assessment, the application timing input is one of the main uncertainty
Received in revised form 30 January 2014 sources among all the parameters involved in the river network contaminations process. We therefore aimed to
Accepted 18 March 2014
assess the sensitivity of the river network pesticide concentration patterns to application timing shifts within a
Available online 13 April 2014
plausible range of application dates, considering two pre-emergence herbicides (metolachlor and aclonifen)
Keywords:
characterised by two different octanol/water partition coefficients (Kow). The Soil and Water Assessment Tool
Application timing (SWAT) was applied in the 1110 km2 agricultural watershed of the river Save (south-western France), where
Sorption properties wheat, maize, sorghum and sunflower are intensively grown. The pesticide application date was changed within
Metolachlor a one-month interval and the pesticide concentration at catchment outlet was simulated from March to June
Aclonifen 2010. Total metolachlor concentration prediction could be improved by an application timing shift to 3 days
SWAT model later (Daily R2 = 0.22 and PBIAS = −57%). By testing the behaviour of the two molecules, it was shown that
Save river sorption processes were influencing the control of application timing on the transfer to surface water:
metolachlor concentration in the channel depended on both discharge and delay between application date and
first stormflow event whereas the transfer of aclonifen depended on rainfall intensity for exportation with
suspended sediments through surface runoff. At last, the study discusses the potential implications of the sensi-
tivity in terms of regional agricultural management practice design.
© 2014 Elsevier B.V. All rights reserved.

1. Introduction groundwater unfit for human consumption. Drinking water quality

European Maximum Permissible Level (MPL) is of 0.1 μg L−1 for an in-
The detrimental effect of intensive agriculture on surface water and dividual pesticide concentration and 0.5 μg L−1 for all pesticide concen-
groundwater quality has been shown by various authors (Burt, 2001; tration (EC, 1998). Recent studies showed the role of one-off and
Cullum, 2009; Ulrich et al., 2013; Zalidis et al., 2002; Zeiger and intense events, such as floods, on water quality degradation regarding
Fohrer, 2009). The transfer of excessive pesticide loading from cultivat- pesticides, including in the south-western France area (Boithias et al.,
ed land to surrounding surface water, either dissolved or sorbed onto 2011, 2014a; Taghavi et al., 2010, 2011). Intensity and timing of rain
particles, may be harmful to terrestrial and aquatic ecosystems and irrigation were shown to be the main inducers of pesticide transfers
(Martin et al., 2011; Niemi et al., 2009; Polard et al., 2011). The partition (Chiovarou and Siewicki, 2008; Vryzas et al., 2009). Short-term (5-day)
between both dissolved and particulate fractions controls the bioavail- precipitation and antecedent soil water deficit were identified as the
ability of the chemical for living organisms' contamination. Pesticide ex- two most important explanatory variables for maximum pesticide con-
portations, from either point losses (e.g. through leaking tools) or centrations in drainflow (Lewan et al., 2009). Reichenberger et al.
diffuse sources (i.e. mostly through runoff and droplet drift) (Holvoet (2007) listed the shift of the pesticide application to an earlier or later
et al., 2005; Müller et al., 2003), may make stream water and date as an efficient mitigation strategy. Modelling studies corroborated
observations for runoff incidence on pesticide exportation (Boithias
et al., 2011; Chu and Mariño, 2004; Zhang and Zhang, 2011) and for ap-
⁎ Corresponding authors at: EcoLab, Avenue de l'Agrobiopole, 31326 Castanet Tolosan plication timing role at seasonal scale (Luo et al., 2008) and at rainfall
Cedex, France. Tel.: +33 5 34 32 39 20; fax: +33 5 34 32 39 01. event scale (Fohrer et al., 2014; Holvoet et al., 2005; Neitsch et al.,
E-mail addresses: [email protected] (L. Boithias),
[email protected] (J.-M. Sánchez-Pérez).
2002; Vazquez-Amabile et al., 2006). Dubus et al. (2003) highlighted
1
Author's present address: Catalan Institute of Water Research, Emili Grahit 101, the uncertainties inherent in pesticide fate modelling, including applica-
Scientific and Technological Park of the University of Girona, 17003 Girona, Spain tion timing, which depends on the farmer and varies from year to year

http://dx.doi.org/10.1016/j.catena.2014.03.013
0341-8162/© 2014 Elsevier B.V. All rights reserved.
98 L. Boithias et al. / Catena 119 (2014) 97–103

(Beernaerts et al., 2002; Campbell et al., 2004). Indeed, large-scale sur- Tertiary period. Calcic soils stem from molasses and represent 61% of
veys with farmers often do not give precise enough information about the whole catchment area with a clay content ranging from 35% to
application sites, application dates and pesticide doses, i.e. pesticide ap- 50%. They are located on the top of the hills and on their slopes. Non-
plication rates, for catchment-scale daily time-step modelling purpose calcic silty soils represent 30% of the soil in this area (40–60% silt).
(Boithias et al., 2011). They are mainly located downstream, close to the Garonne alluvial
In south-western France, spring floods (i.e. spring flushes) were plain. Alluvial deposits are found along the streams and represent 9%
shown to be the main inducers of pre-emergence herbicide stream net- of the catchment area (Boithias et al., 2014b). Top soil organic matter
work contamination, as they are mostly applied on bare soils in the content is about 2% (Veyssy et al., 1999).
most rainy periods (Boithias, 2012; Macary et al., 2013, 2014). When The climate is temperate oceanic. The river Save hydrological regime
applied, pesticide doses are assumed to be at the most equal to manu- is mainly pluvial with a maximum discharge in May and low flows lasting
facturer recommendation. Thus, for contaminant fate modelling and from July to October (1998–2010). The annual precipitation is
possible catchment-scale risk assessment, uncertainty lies in temporal 600–900 mm and the annual evapotranspiration is 500–600 mm
and spatial patterns of pesticide application. Boithias et al. (2011) con- (1998–2010). Mean annual discharge is about 6.1 m3 s−1 (1998–2010).
cluded that the Soil and Water Assessment Tool (SWAT—Arnold et al., During low flows, river flow is sustained upstream by the Neste canal
1998; Gassman et al., 2007) was an appropriate catchment-scale (about 1 m3 s−1) (data from Compagnie d'Aménagement des Coteaux
model to simulate the fate of dissolved and sorbed phases of pesticides de Gascogne—CACG).
at a daily time-step. To our knowledge, no studies were yet published About 90% of the catchment surface is devoted to agriculture. The up-
that related the impact of the application timing to the hydrophobicity stream part of the catchment is a hilly agricultural area mainly covered
of applied chemicals. As a first step to assess the uncertainty of the pes- with pasture and forest with cereals and maize on small plateaus. The
ticide inputs (application site, timing, and dose) when modelling pesti- downstream part is devoted to intensive agriculture with mainly both
cide fate at catchment-scale with SWAT, the aims of this study were maize grown as monoculture and a 4-year crop rotation alternating win-
twofold: (1) to assess the sensitivity of the river network pesticide con- ter wheat with sunflower and maize, sorghum or soybean. Water supply
centration patterns to application timing shifts within a plausible range for irrigation is 210 mm for maize from July to September (Boithias et al.,
of application dates, considering two herbicides characterised by two 2014b). A Cemagref/Irstea-ADBX 3-year survey (2007–2009) was per-
different octanol/water partition coefficients, and (2) to discuss the po- formed anonymously with catchment farmers in order to avoid any
tential implications of the sensitivity in terms of agricultural manage- risk for them to be identified. The survey reports 3-year average spatial
ment practice design. and temporal information about site, timing and dose of pesticide appli-
cation. The most applied pesticides are metolachlor and aclonifen, both
2. Material and methods are pre-emergence herbicides. Each year, 28 tonnes of metolachlor, a
highly soluble and poorly hydrophobic chemical (solubility in water
2.1. Study area Sw = 480 mg L−1, and hydrophobicity expressed by log(Kow) = 2.9),
and 56 tonnes of aclonifen, a poorly soluble and highly hydrophobic
The river Save is located in south-western France and drains an area chemical (Sw = 1.4 mg L−1, log(Kow) = 4.37) (Tomlin, 2009), are ap-
of 1110 km2 (Fig. 1). Altitudes range from 663 m at its source in the Pyr- plied throughout the catchment. On average, metolachlor is applied
enees piedmont to 92 m at the confluence with the river Garonne after a each year to maize and sorghum around the 5th of April, whereas
140 km course at a 0.4% average slope. The catchment is monitored at aclonifen is applied each year to maize and sorghum around the 5th of
the Larra gauging station, whose elevation is 114 m (Fig. 1). The geolog- April and to sunflower around the 20th of April. In 2009, sunflower fields
ical substratum is built from impermeable molassic deposits stemming covered 9% of the catchment (100 km2), maize covered 10% of the catch-
from the erosion of the Pyrenees Mountains during the end of the ment (112 km2) and sorghum covered 6% (70 km2).

Fig. 1. Location of the Save catchment, the Larra gauging station and the 5 meteorological stations.
 L. Boithias et al. / Catena 119 (2014) 97–103 99

The present study focuses on the 4-month spring high flow period of (Table 1). Dissolved pesticides are transported with surface and
2010, lasting from March to June. subsurface runoff, while sorbed pesticides are transported with
surface runoff only;
2.2. Measured data (3) in-stream pesticide processes, including degradation, volatilisation
and settling. The processes depend on the pesticide's phase, based
The river Save discharge was monitored from July 2009 to June on the partition coefficient CHPST_KOC (Table 1).
2010 at the Larra hydrometric station. Hourly discharges (Q) were ob-
tained from CACG. The hourly discharge was plotted by the rating SWAT predicts both dissolved and sorbed concentrations of
curve Q = f(H) in which the water level (H) was measured continuously pesticides: total simulated concentration is obtained by adding up
and then averaged for each day. both dissolved and sorbed simulated concentrations. Authors refer to
Suspended sediments and pesticides were monitored from July 2009 Neitsch et al. (2009) for detailed description of the model's equations.
to June 2010, both manually and automatically, as described in previous
studies on Save catchment (Boithias et al., 2011, 2014a; Oeurng et al., 2.3.2. SWAT data inputs
2010): an automatic water sampler, connected to the probe, was pro- Inputs maps are (1) a digital elevation model with a 25 m × 25 m
grammed to activate pumping water for 30 cm water level variations resolution from Institut Géographique National, France (BD TOPO R);
during high flows, for the rising and falling stages (from 1 to 29 river (2) a land use map from the classification of three 2009 Landsat 5TM
water samples of 1 L were grabbed per stormflow event depending on images with associated management practices provided by
its intensity). Grab sampling was also undertaken near the probe position Cemagref/Irstea-ADBX: spatial and temporal 3-year averages of
at weekly intervals during low flow. In addition, continuous suspended planting and seedling dates, amounts, type and dates of fertilisation,
sediment data were collected by turbidity measurements. Samples of pesticide application, irrigation, grazing, tillage, and harvest opera-
1 L-river water were not aggregated before analysis. Pesticide laboratory tions, including crop rotations, stemming from a 3-year survey
analyses were performed as described by Taghavi et al. (2010, 2011) on (2007–2009) with catchment farmers; and (3) a soil map digitised
both filtered and unfiltered extracts of the same sample of water with a and aggregated by Cemagref/Irstea-ADBX, from paper maps pre-
limit of detection ranging between 0.001 and 0.003 μg L−1 depending pared by soil scientists of the CACG in the 1960s with associated
on the molecule. Dissolved and particulate concentrations of pesticides soil layer properties. Climate data from 5 stations (Fig. 1) were pro-
were then summed up to get the total concentration. In case of sub- vided by Météo-France. Two stations in the upstream section had a
daily samplings, daily concentrations were calculated as an average con- complete set of measurements of daily minimum and maximum air
centration balanced by the instantaneous discharge. Thus, 89 daily con- temperature, wind speed, solar radiation, and relative humidity
centration data were available for the June 2009–July 2010 period, that were used to simulate the reference evapotranspiration by the
including 43 daily concentration data for the March–July period. Penman–Monteith method. Daily discharge data for the Neste
canal, supplying water as an upstream point source to the Save
2.3. Modelling approach river network, was contributed by CACG. Version 2009.93.7a of
ArcSWAT and SWAT Editor were used to set up SWAT inputs and
2.3.1. The SWAT model run the model. The catchment was discretised into 73 sub-basins
The Soil and Water Assessment Tool (SWAT—Arnold et al., 1998; whose minimal area was 5 km2. 2985 HRUs were generated integrat-
Gassman et al., 2007) is a physically-based agro-hydrological model. It ing 23 land uses classes, 6 soil classes and 5 slope classes (%: 0–2,
was chosen because its daily operating time-step was appropriate for 2–5, 5–10, 10–15 and above 15) (Boithias et al., 2014b).
the modelling of floods on the Save catchment, and because it simulates
pesticide fate in both dissolved and sorbed phases in land and in river 2.3.3. Model calibration and validation
channel (Boithias et al., 2011). SWAT discretises catchments into sub- Pesticides can be transported in solution or attached to suspended
basins. Sub-basins are then further subdivided into Hydrological Re- matter. Therefore, calibration and validation of dissolved phase (e.g. ni-
sponse Units (HRUs). HRUs are areas of homogenous land use, soil trate) and particulate phase (e.g. suspended matter) were of major con-
type and slope. HRU outputs are inputs for the connected stream net- cern. This study used a SWAT project which was validated for discharge,
work. One sub-basin is drained by one reach. Pesticide processes in nitrate loads and concentrations, crop yields and biomasses from 2007
SWAT are divided into three components: to 2010 (Boithias et al., 2014b). For this study, suspended matter and
both dissolved and sorbed phases of pesticide concentrations
(1) pesticide processes in land areas, including biotic and abiotic (metolachlor and aclonifen) were calibrated at the outlet from July
degradation, volatilisation from plant and soil surface, infiltra- 2009 to June 2010 (12 months) and without recalibrating hydrological
tion, and leaching; parameters. The performance of the model was evaluated using the co-
(2) transport of pesticides from land areas to the stream network in efficient of determination (R2) and the percentage of bias (PBIAS) com-
both the dissolved and the sorbed phases, depending on the soil puted with daily suspended matter concentration and daily total
adsorption coefficient Koc normalised for soil organic content metolachlor and aclonifen concentrations, both simulated and ob-
served, for both the 4-month (March–June 2010) and the 12-month
(July 2009–June 2010) periods. We deemed daily R2 satisfactory when
Table 1 higher than 0.5 (Green et al., 2006) and daily PBIAS satisfactory if within
Physicochemical properties and mass-transfer calibrated coefficients for metolachlor and
aclonifen in the SWAT model: partition in soil and channel, half-life and degradation rate
± 55% for suspended matter and ± 70% for pesticides (Moriasi et al.,
in the channel water and in the sediment bed (respectively CHPST_REA and SEDPST_REA). 2007). We assumed that the average pesticide application date sur-
veyed in the previous years (2007–2009) would also match the farmers'
Parameters Name in SWAT File Metolachlor Aclonifen
decisions of 2010. Survey average values of pesticide application date
Partition in Koc L kg−1 pest.dat 200 8203 and applied amounts are reported in Fig. 2, shown as red bars. Parame-
soil
ters values are given in Table 1.
Partition in CHPST_KOC L kg−1 .swq 2.5 × 10−4 7.2 × 10−3
channel
Soil half-life HLIFE_S days pest.dat 90 90 2.4. Application date scenarios
Degradation CHPST/SEDPST_REA days−1 .swq 0.025 0.025
rate Based on the average values of application date given in Fig. 2 (the
Volatilisation CHPST_VOL m day−1 .swq 0.3 0.3
“0” scenario), 10 application date scenarios were run by shifting the
100 L. Boithias et al. / Catena 119 (2014) 97–103

(a) 1-Mar 15-Mar 29-Mar 12-Apr 26-Apr 10-May 24-May 7-Jun 21-Jun
0
10

Rainfall
(mm)
20
30 2010
40
50
(b) 50
Observed Simulated
40
Discharge
(m3s-1)
30
20
10
0
(c) 1500
Observed Simulated
Sediments
(mg L-1)

1000

500

0
Application "+15"
(d) 10 "+12" "+9" 3
"+6" "+3"
9 "0" "-3"
"-6" "-9"

Metolachlor application (kg ha-1)

8 "-12" "-15"
Obs.
Metolachlor (µg L )

7
-1

2
6
"0"
5

4
1
3

0 0

(e) 5 3

Aclonifen application (kg ha-1)

"0"
4
-1
Aclonifen (µg L )

2
3

2
1

1
"0"

0 0
1-Mar 15-Mar 29-Mar 12-Apr 26-Apr 10-May 24-May 7-Jun 21-Jun

Fig. 2. Simulations at the outlet of the Save catchment from March to June 2010: (a) catchment average SWAT interpolated rainfall; (b) observed and simulated discharge (m3 s−1);
(c) observed and simulated suspended sediment concentration (mg L−1); (d) simulated total metolachlor concentration (μg L−1) for 11 application date scenarios and observed concentrations;
and (e) simulated total aclonifen concentration (μg L−1) for 11 application date scenarios and observed concentrations. In (d) and (e), bars represent the amount of respective pesticide applied
for initial “0” scenario (kg ha−1).

application dates by steps of 3 days, down to 15 days before average “0” indices (Si), each of them calculated as shown in Eq. (1) (Melching
scenario and up to 15 days after average “0” scenario. Simulated pesti- and Yoon, 1996):
cide total (dissolved + sorbed) concentration patterns were analysed
at the outlet during the 2010 spring flood (March to June 2010). Each ∂P I
Si ¼  ð1Þ
simulation scenario's performance was evaluated using R2 and PBIAS ∂I P ðIÞ
computed with daily total metolachlor and aclonifen concentrations,
both simulated and observed. The latter performances were then com- With P the prediction (total load at outlet) and I the input value
pared to the performances of the “0” scenario. (application date). I was − 15, − 12, − 9, − 6, − 3, + 3, + 6, + 9,
+ 12 and + 15 days.
2.5. Sensitivity analysis
3. Simulation results
The sensitivity of the application date input was assessed for both
metolachlor and aclonifen total load at catchment outlet (2010 spring For the 4-month period, the R2 and the Nash-Sutcliffe Efficiency
flood). It was calculated as the average (S) of 10 relative sensitivity (NSE) of the daily discharge were 0.64 and 0.48 respectively (they
 L. Boithias et al. / Catena 119 (2014) 97–103 101

were of 0.74 and 0.62 respectively during the 12-month period and of Table 3
0.70 and 0.61 respectively during the 2007–2010 period). The R2 and For 11 application date scenarios: fluxes (kg) of total metolachlor and aclonifen at the
outlet of the river Save from March the 1st to June the 30th, 2010; fraction (%) of spring
the PBIAS of the daily suspended sediment concentration were 0.50 flood duration with a concentration of pesticide exceeding the EU water quality standard
and −35% respectively (they were of 0.45 and −43% respectively dur- of 0.1 μg L−1.
ing the 12-month period). Fig. 2 shows average SWAT catchment-scale
Metolachlor Aclonifen
interpolated rainfall, observed and simulated discharge, observed and
−1
simulated suspended sediment concentration, and the concentration Flux [Met] N 0.1 μg L Flux [Acl] N 0.1 μg L−1
(kg) (% of days) (kg) (% of days)
at outlet of total observed and simulated metolachlor and aclonifen
for the 11 pesticide application date scenarios, from March to June “−15” 38 68 46 40
2010. Suspended sediment concentration was correlated to discharge “−12” 42 65 48 40
“−9” 44 61 49 41
(R2 = 0.77, n = 122), itself responding to rainfall with runoff peaks. “−6” 52 61 50 41
The concentration patterns of both pesticides are different: metolachlor “−3” 57 57 51 42
concentration peaks are distributed depending on the application date, “0” 66 54 52 43
whereas aclonifen peaks are concentrated around the same date (the “+3” 66 44 54 42
“+6” 68 44 56 42
4th of May). For total metolachlor concentration, scenario “+3” fitted
“+9” 69 44 57 43
observations the best (Table 2), where R2 for dissolved and sorbed con- “+12” 71 44 59 44
centrations were 0.21 and 0.10 respectively (4-month period). For “+15” 82 43 36 39
aclonifen, the best scenario was “+ 15”, where R2 for dissolved and
sorbed concentration is 0.03 and 0.05 respectively (4-month period).
Goodness-of-fit indices for the 12-month calibration period were slight- month spring period were weaker than those of the 2007–2010 period.
ly higher than the ones for the 4-month period for both molecules So was the suspended sediment transport, which drives sorbed pesti-
(Table 2). cides transfers. Bias between observations and model predictions of
Shifts in application date had the highest impact on metolachlor pesticide concentrations may also come from inadequate values of the
total load exportation at outlet during the spring flood period: exported pesticide parameters (e.g. half-life and reaction coefficients, although
fluxes from the 1st of March to the 30th of June were ranging between their calibrated values were in the range of the pesticide properties in-
38 and 82 kg for metolachlor, but were ranging between 36 and 59 kg puts reported by Neitsch et al. (2002)), given that the uncertainty on
for aclonifen (Table 3). Metolachlor exportation rates (input/output their value is large (Dubus et al., 2003; Walker et al., 2002). Uncertainty
ratio) were ranging between 1.4‰ and 2.9‰. Aclonifen exportations also exists among observed data and laboratory analysis (Dubus et al.,
rates were ranging between 0.7‰ and 1.1‰. Metolachlor concentration 2003). The spatial and the temporal patterns of pesticide applications
exceeded the MPL of 0.1 μg L−1 during 43% to 68% of the time depending are uncertain because of the limitations of the large-scale anonymous
on the application date, whereas aclonifen concentration exceeded the survey that provided average spatial and temporal values of application
same MPL during 39% to 44% of the time (Table 3). In addition, the sen- site and doses, together with application timing. The farmers did not ac-
sitivity S of metolachlor was −0.20 whereas it was −0.06 for aclonifen. tually apply pesticides on all their crops (e.g. sunflower, maize or sor-
ghum) the same day: farmers staggered the application depending on
4. Discussion the weed pressure and on the amount of rainfall announced by weather
forecast, contrarily to simulation where only one application date per
Simulation of pesticide concentration was improved by shifting the land use was considered (for that purpose Gevaert et al. (2008) sug-
application date, although simulations of pesticides did not meet all sat- gested to describe application date input as a probability distribution).
isfactory standards. The quality of the simulation for both molecules The metolachlor modelling results in Fig. 2 show concentration peaks
was in the range of the previous daily time-step modelling pesticide immediately after application dates, which are not observed from the
fate study in the Save catchment of Boithias et al. (2011). The quality “−15” to “0” cases. As highlighted by the goodness-of-fit values report-
of the simulation was however lower than other previous modelling ed in Table 2, metolachlor was probably actually applied later, from the
pesticide fate studies at daily time step, e.g. recently Fohrer et al. 8th of April to the 18th of April. The aclonifen modelling results in Fig. 2
(2014), who benefited from detailed flufenacet and metazachlor input also show concentration peaks that are not observed. One explanation is
data in the 50 km2 Kielstau catchment. that aclonifen, that has a Kow higher than the one of metolachlor, is
Hydrological processes, that drive pesticide transfers, were satisfac- mostly sorbed to suspended matter and trapped with them when they
torily modelled although the goodness-of-fit indices during the 4- sediment along the river course. In addition, given that pre-emergence

Table 2
Goodness-of-fit indices calculated for both the 4-month (n = 43) and the 12-month (n = 89) periods with metolachlor and aclonifen total observed and simulated concentrations for 11
application date scenarios.

Metolachlor Aclonifen

4 months 12 months 4 months 12 months

(March–June 2010) (July 2009–June 2010) (March–June 2010) (July 2009–June 2010)

R2 PBIAS (%) R2 PBIAS (%) R2 PBIAS (%) R2 PBIAS (%)

“−15” 0.01 −91 0.01 −69 0.01 −788 0.01 −478

“−12” 0.00 −100 0.01 −76 0.01 −810 0.01 −491
“−9” 0.00 −120 0.01 −93 0.01 −831 0.01 −504
“−6” 0.00 −137 0.02 −107 0.01 −851 0.01 −516
“−3” 0.00 −142 0.03 −111 0.01 −867 0.01 −526
“0” 0.01 −150 0.04 −118 0.01 −885 0.02 −538
“+3” 0.22 −57 0.25 −41 0.01 −893 0.02 −542
“+6” 0.22 −60 0.25 −42 0.01 −916 0.02 −557
“+9” 0.22 −64 0.25 −45 0.01 −940 0.02 −572
“+12” 0.22 −67 0.25 −48 0.01 −978 0.02 −596
“+15” 0.19 −104 0.22 −79 0.09 −535 0.09 −321
102 L. Boithias et al. / Catena 119 (2014) 97–103

pesticide as metolachlor and aclonifen are usually applied before germi- of a wide range of parameters including application site, date and
nation of seeds, there is little chance that aclonifen had been applied be- amount and pesticide properties. Plausible scenarios stemming from
fore the 21st of March and after the 22nd of April. Given the best combinations may be later validated by spatially distributed exten-
precipitation pattern, the latter suggests that aclonifen might not have sive field surveys. The same approach could be later performed for
been applied by farmers in 2010, or maybe in much lower quantity point-source contaminations, including pharmaceuticals and hydrocar-
(dose and spread surface) than they did in average from 2007 to 2009. bon compounds. Applied to a regional scale, such a tool would later
Indeed, farmers use to change the molecules they use from one year help water managers to localise the main contamination sources
to the next in order to avoid weed resistance (e.g. Service, 2007). Ac- avoiding time- and money-consuming field campaigns and, in the case
cording to the 3-year survey, most of aclonifen was usually applied of pesticides, to assess the contamination risk and the environmental
around the 20th of April. In 2010, if farmers missed the weather window impacts of future agricultural practices changes. This will allow them
lasting from the 8th to the 18th of April, they had no time to spread pes- to suggest farmers appropriate mitigation practices, i.e. best manage-
ticide later as it had been raining every two days during the following ment practices (BMP) such as pesticide application date shifts or applied
weeks. dose reduction during rainy periods, or such as tillage limitations in
The latter suggests that if aclonifen had been applied in 2010, then most erosion prone agricultural areas. Such measures will help to
aclonifen observations would have had a similar pattern to the simulat- avoid excessive concentrations in surface water and achieve for example
ed one. Application date shift had more impact on metolachlor consid- the objectives of water policies such as the Water Framework Directive
ering the 2010 spring flood period. Because of its high solubility and in Europe.
low hydrophobicity, metolachlor was mostly found in the dissolved
phases, thus allowing transfers with surface and sub-surface runoffs
whatever the rainfall intensity (Müller et al., 2003; Ulrich et al., 2013). Acknowledgment
Its concentration in the channel therefore depended on both discharge
and delay between application date and first stormflow event. Con- This work was performed as part of the EU Interreg SUDOE IVB pro-
versely, aclonifen was mostly sorbed to soil particles. Its transfer not gram (SOE1/P2/F146 AguaFlash project, http://www.aguaflash-sudoe.
only depended on the delay between application date and first eu) and funded by ERDF and the Midi-Pyrénées Region. We sincerely
stormflow event, but also on rainfall intensity for exportation with thank the CACG for discharge data, Météo-France for meteorological
eroded suspended sediments through surface runoff (Jin et al., 2009; data, and Ecolab staff for field and laboratory support.
Otto et al., 2012). Therefore, the effect of application time on pesticide
transfers itself depends on pesticide specific controlling factors, as
their sorption ability, or hydrophobicity, that can be quantified by the References
partition coefficient Kd (Boithias et al., 2014a; Nakano et al., 2004), Abbaspour, K.C., 2008. SWAT-CUP2: SWAT Calibration and Uncertainty Programs - A User
known in SWAT as Koc and CHPST_KOC (Table 1). The transfer of pesti- Manual. 95.
cides in soil, and hence their bioavailability and transfer to other com- Arnold, J.G., Srinivasan, R., Muttiah, R.S., Williams, J.R., 1998. Large area hydrologic model-
ing and assessment. I. Model development. J. Am. Water Resour. Assoc. 34, 73–89.
partments, also depends on volatilisation, chemical and biological Beernaerts, S., Debongie, P., DeVleeschouwer, C., Pussemier, L., 2002. Het pilootproject
degradation processes, together with soil and suspended matter prop- voor het Nil bekken. Groenboek Belgaqua-Phytophar 2002, Belgium.
erties (Boulange et al., 2012). Considering the volatilisation rates similar Boithias, L., 2012. Modélisation des transferts de pesticides à l’échelle des bassins versants
en période de crue. PhD diss. Institut National Polytechnique de Toulouse, France
for both pesticides (Table 1), the combined effect of application timing (220 pp.).
and hydrophobicity thus depends on the degradation rate in both soil Boithias, L., Sauvage, S., Taghavi, L., Merlina, G., Probst, J.L., Sánchez Pérez, J.M., 2011. Oc-
and channel (Ghafoor et al., 2011): the role of application timing is currence of metolachlor and trifluralin losses in the Save river agricultural catchment
during floods. J. Hazard. Mater. 196, 210–219.
only relevant for pesticide whose half-lifes are significantly longer Boithias, L., Sauvage, S., Merlina, G., Jean, S., Probst, J.L., Sánchez Pérez, J.M., 2014a. New
than the duration of the period without rainfall. Therefore, the role of insight into pesticide partition coefficient Kd for modelling pesticide fluvial transport:
rainfall timing may either be reduced or increased in a climate change application to an agricultural catchment in south-western France. Chemosphere 99,
134–142.
context. For instance in the south-western part of Europe, where occur-
Boithias, L., Srinivasan, R., Sauvage, S., Macary, F., Sánchez-Pérez, J.M., 2014b. Daily nitrate
rence of rainfall is likely to decrease and the intensity of the rainfall and losses: implication on long term river quality in an intensive agricultural catchment
subsequent runoff is likely to increase (García-Ruiz et al., 2011; Lehner (south-western France). J. Environ. Qual. 43, 46–54.
Boulange, J., Kondo, K., Phong, T.K., Watanabe, H., 2012. Analysis of parameter uncertainty
et al., 2006), the effects of degradation on pesticide fate may exceed
and sensitivity in PCPF-1 modeling for predicting concentrations of rice herbicides. J.
the effects of their mobility to river networks in both the dissolved Pestic. Sci. 37, 323–332.
and the sorbed phases. Burt, T.P., 2001. Integrated management of sensitive catchment systems. Catena 42,
275–290.
Campbell, N., D’Arcy, B., Frost, A., Novotny, V., Sansom, A., 2004. Diffuse Pollution: An In-
5. Conclusion troduction to the Problems and Solutions, First ed. IWA Publishing, London, UK.
Chiovarou, E.D., Siewicki, T.C., 2008. Comparison of storm intensity and application timing
Results suggest that the delay between pesticide application and on modeled transport and fate of six contaminants. Sci. Total Environ. 389, 87–100.
Chu, X., Mariño, M.A., 2004. Semidiscrete pesticide transport modeling and application. J.
first rain is a more significant driving factor of the transfer to surface Hydrol. 285, 19–40.
water of molecules of low Kow than of high Kow. They imply that pesti- Cullum, R.F., 2009. Macropore flow estimations under no-till and till systems. Catena 78,
cide concentration signal at the outlet is a combination of spatial and 87–91.
Dubus, I.G., Brown, C.D., Beulke, S., 2003. Sources of uncertainty in pesticide fate model-
temporal patterns of application date and dose, in addition to pesticide ling. Sci. Total Environ. 317, 53–72.
specific properties such as half-life and hydrophobicity. This study leads EC, 1998. Directive 98/83/EC of 3 November 1998 on the quality of water intended for
us to recommend assessing the uncertainty on pesticide application site, human consumption. Off. J. Eur. Communities L330.
Fohrer, N., Dietrich, A., Kolychalow, O., Ulrich, U., 2014. Assessment of the environmental
date and amount, as an integral part of the pesticide calibration process fate of the herbicides flufenacet and metazachlor with the SWAT model. J. Environ.
when modelling pesticide fate at catchment scale. Qual. 43, 75–85.
Therefore, future work will consist in testing all combinations of spa- García-Ruiz, J.M., López-Moreno, J.I., Vicente-Serrano, S.M., Lasanta–Martínez, T., Beguería,
S., 2011. Mediterranean water resources in a global change scenario. Earth Sci. Rev.
tial and temporal application date patterns, together with a range of
105, 121–139.
plausible doses, in an automatic spatially explicit expert system to get Gassman, P.W., Reyes, M.R., Green, C.H., Arnold, J.G., 2007. The soil and water assessment
the combination fitting the best the observed signal at outlet. Tools tool: historical development, applications, and future research directions. Trans.
such as SWAT-CUP (Abbaspour, 2008) are appropriate for that inverse ASABE 50, 1211–1250.
Gevaert, V., Van Griensven, A., Holvoet, K., Seuntjens, P., Vanrolleghem, P.A., 2008. SWAT
modelling exercise: by comparing outputs with observed concentra- developments and recommendations for modelling agricultural pesticide mitigation
tions, they allow assessing the uncertainty and calculating the sensitivity measures in river basins. Hydrol. Sci. J. 53, 1075–1089.
 L. Boithias et al. / Catena 119 (2014) 97–103 103

Ghafoor, A., Jarvis, N.J., Thierfelder, T., Stenström, J., 2011. Measurements and modeling Niemi, R.M., Heiskanen, I., Ahtiainen, J.H., Rahkonen, A., Mäntykoski, K., Welling, L.,
of pesticide persistence in soil at the catchment scale. Sci. Total Environ. 409, Laitinen, P., Ruuttunen, P., 2009. Microbial toxicity and impacts on soil enzyme activ-
1900–1908. ities of pesticides used in potato cultivation. Appl. Soil Ecol. 41, 293–304.
Green, C.H., Tomer, M.D., Di Luzio, M., Arnold, J.G., 2006. Hydrologic evaluation of the soil Oeurng, C., Sauvage, S., Sánchez-Pérez, J.-M., 2010. Dynamics of suspended sediment
and water assessment tool for a large tile-drained watershed in Iowa. Trans. ASABE transport and yield in a large agricultural catchment, southwest France. Earth Surf.
49, 413–422. Process. Landf. 35, 1289–1301.
Holvoet, K., Van Griensven, A., Seuntjens, P., Vanrolleghem, P.A., 2005. Sensitivity analysis Otto, S., Cardinali, A., Marotta, E., Paradisi, C., Zanin, G., 2012. Effect of vegetative filter
for hydrology and pesticide supply towards the river in SWAT. Phys. Chem. Earth 30, strips on herbicide runoff under various types of rainfall. Chemosphere 88, 113–119.
518–526. Polard, T., Jean, S., Gauthier, L., Laplanche, C., Merlina, G., Sánchez-Pérez, J., Pinelli, E., 2011.
Jin, K., Cornelis, W.M., Gabriels, D., Baert, M., Wu, H.J., Schiettecatte, W., Cai, D.X., De Neve, Mutagenic impact on fish of runoff events in agricultural areas in south-west France.
S., Jin, J.Y., Hartmann, R., Hofman, G., 2009. Residue cover and rainfall intensity effects Aquat. Toxicol. 101, 126–134.
on runoff soil organic carbon losses. Catena 78, 81–86. Reichenberger, S., Bach, M., Skitschak, A., Frede, H.G., 2007. Mitigation strategies to reduce
Lehner, B., Döll, P., Alcamo, J., Henrichs, T., Kaspar, F., 2006. Estimating the impact of global pesticide inputs into ground- and surface water and their effectiveness; a review. Sci.
change on flood and drought risks in europe: a continental, integrated analysis. Clim. Total Environ. 384, 1–35.
Chang. 75, 273–299. Service, R.F., 2007. A growing threat down on the farm. Science 316, 1114–1117.
Lewan, E., Kreuger, J., Jarvis, N., 2009. Implications of precipitation patterns and anteced- Taghavi, L., Merlina, G., Probst, J.L., 2011. The role of storm flows in concentration of pes-
ent soil water content for leaching of pesticides from arable land. Agric. Water ticides associated with particulate and dissolved fractions as a threat to aquatic eco-
Manag. 96, 1633–1640. systems. Case study: the agricultural watershed of Save river (Southwest of France).
Luo, Y., Zhang, X., Liu, X., Ficklin, D., Zhang, M., 2008. Dynamic modeling of organophos- Knowl. Manag. Aquat. Ecosyst. 400, 11.
phate pesticide load in surface water in the northern San Joaquin Valley watershed Taghavi, L., Probst, J.L., Merlina, G., Marchand, A.L., Durbe, G., Probst, A., 2010. Flood event
of California. Environ. Pollut. 156, 1171–1181. impact on pesticide transfer in a small agricultural catchment (Montousse at Aurade,
Macary, F., Leccia, O., Almeida Dias, J., Morin, S., Sánchez-Pérez, J.-M., 2013. Agro- south west France). Int. J. Environ. Anal. Chem. 90, 390–405.
environmental risk evaluation by a spatialised multi-criteria modelling combined Tomlin, E., 2009. The Pesticide Manual, Fifteenth ed. C.D.S Tomlin.
with the PIXAL method. Int. J. Geomatics Spatial Anal. 23, 39–70. Ulrich, U., Dietrich, A., Fohrer, N., 2013. Herbicide transport via surface runoff during
Macary, F., Morin, S., Probst, J.-L., Saudubray, F., 2014. A multi-scale method to assess intermittent artificial rainfall: a laboratory plot scale study. Catena 101, 38–49.
pesticide contamination risks in agricultural watersheds. Ecol. Indic. 36, 624–639. Vazquez-Amabile, G., Engel, B.A., Flanagan, D.C., 2006. Modeling and risk analysis of
Martin, S., Bertaux, A., Ber, F., Maillard, E., Imfeld, G., 2011. Seasonal changes of nonpoint-source pollution caused by atrazine using SWAT. Trans. ASABE 49, 667–678.
macroinvertebrate communities in a stormwater wetland collecting pesticide Veyssy, E., Etcheber, H., Lin, R.G., Buat-Menard, P., Maneux, E., 1999. Seasonal variation
runoff from a vineyard catchment (Alsace, France). Arch. Environ. Contam. Toxicol. and origin of particulate organic carbon in the lower Garonne River at La Reole
62, 29–41. (southwestern France). Hydrobiologia 391, 113–126.
Melching, C.S., Yoon, C.G., 1996. Key sources of uncertainty in QUAL2E model of Passaic Vryzas, Z., Vassiliou, G., Alexoudis, C., Papadopoulou-Mourkidou, E., 2009. Spatial and
River. J. Water Resour. Plan. Manag. 122, 105–113. temporal distribution of pesticide residues in surface waters in northeastern
Moriasi, D.N., Arnold, J.G., Van Liew, M.W., Bingner, R.L., Harmel, R.D., Veith, T.L., 2007. Greece. Water Res. 43, 1–10.
Model evaluation guidelines for systematic quantification of accuracy in watershed Walker, A., Bromilow, R.H., Nicholls, P.H., Evans, A.A., Smith, V.J.R., 2002. Spatial variability in
simulations. Trans. ASABE 50, 885–900. the degradation rates of isoproturon and chlorotoluron in clay soil. Weed Res. 42, 39–44.
Müller, K., Deurer, M., Hartmann, H., Bach, M., Spiteller, M., Frede, H.G., 2003. Hydrological Zalidis, G., Stamatiadis, S., Takavakoglou, V., Eskridge, K., Misopolinos, N., 2002. Impacts
characterisation of pesticide loads using hydrograph separation at different scales in a of agricultural practices on soil and water quality in the Mediterranean region and
German catchment. J. Hydrol. 272, 1–17. proposed assessment methodology. Agric. Ecosyst. Environ. 88, 137–146.
Nakano, Y., Miyazaki, A., Yoshida, T., Ono, K., Inoue, T., 2004. A study on pesticide runoff Zeiger, M., Fohrer, N., 2009. Impact of organic farming systems on runoff formation
from paddy fields to a river in rural region—1: field survey of pesticide runoff in processes—a long-term sequential rainfall experiment. Soil Tillage Res. 102, 45–54.
the Kozakura River. Jpn. Water Res. 38, 3017–3022. Zhang, X., Zhang, M., 2011. Modeling effectiveness of agricultural BMPs to reduce sedi-
Neitsch, S.L., Arnold, J.G., Kiniry, J.R., Williams, J.R., 2009. Soil and Water Assessment Tool ment load and organophosphate pesticides in surface runoff. Sci. Total Environ.
theoretical documentation—version 2009 (TR406). 409, 1949–1958.
Neitsch, S.L., Arnold, J.G., Srinivasan, R., 2002. Pesticides fate and transport predicted by
the Soil and Water Assessment Tool (SWAT)—atrazine, metolachlor and trifluralin
in the sugar creek watershed. p. 96.