Hydrological and pesticide transfer modeling in a tropical volcanic watershed with the WATPPASS model

Journal of Hydrology 529 (2015) 909–927

Contents lists available at ScienceDirect

Journal of Hydrology journal homepage: www.elsevier.com/locate/jhydrol

Hydrological and pesticide transfer modeling in a tropical volcanic watershed with the WATPPASS model Charles Mottes a,b,⇑, Magalie Lesueur-Jannoyer a,b, Jean-Baptiste Charlier c, Céline Carles a, Mathilde Guéné a, Marianne Le Bail d, Eric Malézieux b a

Cirad, UPR HortSys, Campus agro-environnemental Caraïbe, BP 214, 97285 Le Lamentin Cedex 2, Martinique, France Cirad, UPR HortSys, TA B-103/C Campus international de Baillarguet, 34398 Montpellier, France BRGM, 1039 rue de Pinville, 34000 Montpellier, France d AgroParisTech, UMR SADAPT, 16 Rue Claude Bernard, 75231 Paris Cedex 5, France b c

a r t i c l e

i n f o

Article history: Received 8 December 2014 Received in revised form 30 June 2015 Accepted 2 September 2015 Available online 10 September 2015 This manuscript was handled by Laurent Charlet, Editor-in-Chief, with the assistance of Hongtao Wang, Associate Editor Keywords: Volcanic catchment Conceptual modeling Hydrology Pesticide Chlordecone Water pollution

s u m m a r y Simulation of flows and pollutant transfers in heterogeneous media is widely recognized to be a remaining frontier in hydrology research. We present a new modeling approach to simulate agricultural pollutions in watersheds: WATPPASS, a model for Watershed Agricultural Techniques and Pesticide Practices ASSessment. It is designed to assess mean pesticide concentrations and loads that result from the use of pesticides in horticultural watersheds located on heterogeneous subsoil. WATPPASS is suited for small watershed with significant groundwater flows and complex aquifer systems. The model segments the watershed into fields with independent hydrological and pesticide transfers at the ground surface. Infiltrated water and pesticides are routed toward outlet using a conceptual reservoir model. We applied WATPPASS on a heterogeneous tropical volcanic watershed of Martinique in the French West Indies. We carried out and hydrological analysis that defined modeling constraints: (i) a spatial variability of runoff/ infiltration partitioning according to land use, and (ii) a predominance of groundwater flow paths in two overlapping aquifers under permeable soils (50–60% of annual flows). We carried out simulations on a 550 days period at a daily time step for hydrology (Nashsqrt > 0.75). Weekly concentrations and loads of a persistent organic pesticide (chlordecone) were simulated for 67 weeks to evaluate the modeling approach. Pesticide simulations without specific calibration detected the mean long-term measured concentration, leading to a good quantification of the cumulative loads (5% error), but failed to represent the concentration peaks at the correct timing. Nevertheless, we succeed in adjusting the model structure to better represent the temporal dynamic of pesticide concentrations. This modification requires a proper evaluation on an independent dataset. Finally, WATPPASS is a compromise between complexity and easiness of use that makes it suited for cropping system assessment in complex pedological and geological environment. Ó 2015 Elsevier B.V. All rights reserved.

1. Introduction Protecting surface and groundwater water resources is one of the major issues of this XXIiest century. This is particularly true in small volcanic islands where water resources is limited and facing up increasing anthropic demands and agricultural pressures (Rawlins et al., 1998). Volcanic ash soils are among the most productive soils in the world and are intensively cropped with horticultural cropping systems (Shoji and Takahashi, 2002), ⇑ Corresponding author at: Cirad, UPR HortSys, Campus agro-environnemental Caraïbe, BP 214, 97285 Le Lamentin Cedex 2, Martinique, France. Tel.: +596 596423073. E-mail address: [email protected] (C. Mottes). http://dx.doi.org/10.1016/j.jhydrol.2015.09.007 0022-1694/Ó 2015 Elsevier B.V. All rights reserved.

leading to high pesticides inputs to preserve production yields and qualities (Houdart et al., 2009). In the Caribbean and Central America, coastal resources, agricultural soils and aquatic fauna are highly contaminated by a suite of pesticides (Castillo et al., 2006; Coat et al., 2011; Henriques et al., 1997; Kammerbauer and Moncada, 1998; McDonald et al., 1999; Taylor et al., 2003). The long-term contamination of the French West Indies by chlordecone, an organochlorine insecticide with concentrations higher than 9 mg kg
1 in soils, and up to 10 lg L
1 in waters, dramatically illustrates pollutions in such context (Cabidoche et al., 2009; Gourcy et al., 2009). Studying pesticide fate in such environments remains a great challenge because of the hydrological specificities of tropical

910

C. Mottes et al. / Journal of Hydrology 529 (2015) 909–927

volcanic areas. First, humid tropical climate is characterized by high amount of annual rainfalls (from 2000 to 4500 mm in agricultural areas of the Lesser Antilles) and by high rainfall intensities. Both result in accelerated pesticide transfers from soil to stream water compared to temperate climate. Second, volcanic ash soils have high infiltration properties (Ks > 50 mm/h according to Poulenard et al. (2001) and Cattan et al. (2006)). Consequently, at the plot scale, pesticide infiltration toward the aquifer are predominant compared to the transport by surface runoff (Saison et al., 2008). Third, volcanic geological formations are complex and show important lithological variety having heterogeneous properties even in the same geological formation (Bernard et al., 2007). It gives rise to a very high spatial and vertical variability of the structure and of the hydrodynamic properties of the aquifers (Charlier et al., 2011; Foster et al., 1985; Lachassagne, 2006). This geological framework promotes underground flow paths that involve overlapping aquifers (Charlier et al., 2008). On a small volcanic watershed located in Guadeloupe (French West Indies), the mechanisms of pesticide transport in soils and surface waters were studied by Charlier et al. (2009a). Authors showed that two successive phases of stream water contamination occurred after application: a first event-dominated contamination phase when transport is linked to overland flow, and a second stabilized contamination phase originating mainly from the drainage of the aquifer. In this setting, to assess the impacts of cropping systems on the contamination levels of stream water at the watershed scale, it is first necessary to model hydrological and pesticide transfers from soil to stream accounting for the spatial variability of land uses and of the specificities of the tropical volcanic environment. Generally, models used are physically based distributed models such as MHYDAS (Moussa et al., 2002) or MIKE-SHE (Refsgaard and Storm, 1995), for reviews see Mottes et al. (2014) and Quilbé et al. (2006). These models usually require many input parameters which are either laborious or extremely costly to gather. For that reason, they are used in a degraded mode, with substantial simplifications of the physical and hydrological structures that these models aim at representing (Le Moine et al., 2008). Lumped conceptual models of solute and pollutant transport are an alternative approach. Examples are reservoir models simulating nitrate flows in the vadose zone (Philippe et al., 2011) or in groundwater as BICHE (Thiéry and Seguin, 1985) or TEMPO (Pinault et al., 2001), or simulating transfers of conservative (Hartmann et al., 2013) and non-conservative solutes in karst systems (Charlier et al., 2012). Recently, a lumped approach by Farlin et al. (2013) has shown the potential of models based on the groundwater residence time distribution to simulate atrazine evolution in fractured media. These models consider the watershed as an undivided entity, and use lumped values of input variables and parameters. Mathematical functions represent the different processes and fluxes between storage elements. For volcanic media, Charlier et al. (2008) proposed a lumped hydrological conceptual model adapted to tropical volcanic watershed. This model simulates stream discharge and groundwater fluctuations, but was not designed to model pesticide transport. Thus, there is a need to develop a new modeling approach of pesticide transport adapted to heterogeneous watersheds such as those in volcanic media. The aim of this paper is to present a new conceptual hydrological and pesticide transport model to assess water pollution at the watershed scale: the WATPPASS model (Watershed Agricultural Techniques and Pesticide Practices ASSessment). A relevant representation of water and pollutant transfers within the watershed is needed to operate a change in scale between cropping systems and water pollution at outlet. Such a representation is a prerequisite for more complex models aiming to assess the effects of cropping systems on water pollution in heterogeneous media, such as tropical volcanic watershed. At this stage, WATPPASS structure has two

nested modules. The field scale module generates water and pesticide runoff and infiltration inputs for the watershed module. The watershed module simulates water flows using a lumped approach at the watershed scale. Pesticide transfers are governed by water flows within watersheds using a conceptual spatial discretization of aquifers using compartments. WATPPASS takes into account the position of the infiltration flows on pesticide transfers toward groundwater ways. Regarding the intended application of WATPPASS which is cropping system assessment, we expect the simplified approach used in WATPPASS to correctly simulate long-term pesticide concentrations and loads in the stream within the range of the observed. This would make it possible to detect the potential long-term impact of pesticide use. In this article, we first present the model structure, second we applied WATPPASS on the Ravine watershed in Martinique (French West Indies). To do so, we performed a calibration of the model on hydrology only and checked results for pesticides without pesticide related calibration. As a case study, we used the persistent organic pesticide chlordecone that was mapped precisely on the Ravine watershed. Finally, we proposed and ran simulations with two model adjustments. Both adjustments still require an independent evaluation because they were designed to better represent the observed pesticide dynamics than with our initial model. 2. Calculation 2.1. Model structure WATPPASS uses a lumped representation of the watersheds to account for groundwater and pesticide fluxes. The overall structure of the model is described in Fig. 1. QO (L day
1), the discharge at outlet, is the sum of four contributions: surface runoff (Qsr), shallow aquifers discharge (Qsa), deep aquifer discharge (Qda) and flows between the two aquifers (Qsub). These four contributions also carry pesticides (PestQsr, PestQsa, PestQsub, PestQda) to the pesticide discharge at outlet (PestQO) (g) (Fig. 1b). 2.2. Field scale production functions 2.2.1. Surface runoff The surface runoff term of the watershed (Qsr) is generated using the curve number approach (USDA, 1986). Curve number is a method which makes it possible to estimate runoff while taking into account the effects of agricultural practices (Mottes et al., 2014). This method relies on more than 20 years of data collection and analysis (Ponce and Hawkins, 1996). The curve number method is currently the only method for which parameter estimations are provided for a large range of land uses and cropping systems in various locations including tropical ones (Cooley and Lane, 1982; Sartori et al., 2011). In this paper, we attributed to each field a curve number depending on its cropping system. Surface runoff (Qsr) is calculated using Eq. (1). The curve numbers are estimated at a daily time step using the retention capacity before runoff adjustment method presented in Eq. (4) which was developed especially for WATPPASS because of the inadequacy of existing adjustment methods (Kannan et al., 2008). Runoff is calculated with the SCS curve number equation (USDA, 1986)

( Qsr ¼

ðP
IaÞ2 ; P
IaþS

if ½P > Ia

0;

if ½P 6 Ia

ð1Þ

with Qsr the runoff amount in mm day
1; P the rainfall in mm day
1; S is the retention capacity before runoff (mm) and Ia

C. Mottes et al. / Journal of Hydrology 529 (2015) 909–927

911

Fig. 1. (a) Structure of the WATPPASS-lumped hydrological model, and (b) structure of the WATPPASS model for pesticide transfers in the watershed. ET: Evapotranspiration (mm), P: rainfall (mm), Qsr: runoff discharge, I: soil infiltration, Osa: outflow of shallow aquifer, Qsa: discharge of the shallow aquifer, Psa: percolation from the shallow aquifer, UH: unit hydrograph, Qsub: discharge resulting from the unit hydrograph, Lsa: leaching from the shallow aquifer, Oda: outflow from the deep aquifer, Lda: leaching from the deep aquifer, Qda: discharge from the deep aquifer. PestX: Pesticide transferred with water flows X.

the initial abstraction (mm). The initial abstraction is estimated using Ia = 0.2  S as suggested by USDA (1986) which results in

(

Qsr ¼

ðP
0:2SÞ2 P
0:8S

; if ½P > 0:2S if ½P 6 0:2S

0;

S ¼ 25:4 

  1000
10 CN

ð2Þ

ð3Þ

To account for the effect of antecedent moisture conditions on runoff, the curve numbers are updated daily by calculating the retention capacity before runoff S using constraints on the evolution of S between two values Smin and Smax:

St ¼ St
1 þ a1  PET t  

Smin
St
1 Smin

Smax
St
1 þ a2  ðP t
1
Qsrt
1 Þ Smax ð4Þ

where Pt
1 is the rainfall of the day before (mm), PETt is the potential evapotranspiration of the day (mm) and Qsrt
1 the runoff of the

day before (mm), St
1 is the value of S on the day before, a1 and a2 (unitless) are two empirical coefficients, Smax (mm) corresponds to maximum water retention before runoff in dry conditions (CN1) while Smin (mm) corresponds to maximum water retention in wet conditions (CN3). Both Smin and Smax are calculated with Eq. (3) and the values of CN1 and CN3 that are calculated using equations reported by Neitsch et al. (2011):

CN1 ¼ CN2


20  ð100
CN2 Þ 100
CN 2 þ expð2:533
0:0636  ð100
CN2 ÞÞ

CN3 ¼ CN2  exp ð0:00673  ð100
CN2 ÞÞ

ð5aÞ ð5bÞ

where CN2 is the curve number of the field for average moisture conditions. Finally, the curve number taking into account the antecedent moisture condition is calculated with the S value adjusted by Eq. (4):

CN ¼

25; 400 S þ 254

ð6Þ

912

C. Mottes et al. / Journal of Hydrology 529 (2015) 909–927

2.2.2. Water in soil WATPPASS conceptualizes the soil compartment and transfers in soil using a soil module which has been adapted from GLEAMS (Leonard et al., 1987) and SWAT (Neitsch et al., 2011) models. WATPPASS represents soil with 11 horizontal layers which makes it possible to represent pesticide temporal infiltration profiles. The first layer is 10 mm height. The heights of the other layers are calculated taking into account the total soil height heightsoil (mm). The height of each layers from 3 to 11 is heightsoil/10. The second layer height is heightsoil/10
10. A storage routing technique simulates percolation from upper soil layers to lowers (Leonard et al., 1987). For each soil layer, percolated water from the previous layer or from rainfall is added to the next layer. The variation of the water stock in each layer DSWlayer (mm) is expressed as



DSW layer ¼

Waterinput ;

if ½layer ¼ 1

ð7Þ

Percolayer
1 ; if ½layer > 1

where Waterinput = P
Qsr is the amount of water entering the first soil layer and Percolayer
1 (mm) is the amount of water percolated from the above layer. The water percolating from the layer Percolayer (mm) is calculated according to the following equation:

Percolayer


24 ¼ SW excess  1
exp TTperclayer

!!

ð8Þ

where SWexcess (mm) is the water content exceeding the field capacity of the layer and TTperclayer is the water percolation speed (mm day
1). SWexcess is calculated using



SW excess ¼

SW layer
FC layer ; if ½SW layer
FC layer > 0 0; if ½SW layer
FC layer 6 0

ð9Þ

and percolation speed is calculated using

TTperclayer

SAT layer
FC layer ¼ Ksat layer

ð10Þ

ð11Þ

Finally, I is the percolation resulting from the soil deepest layer

I ¼ Percolastlayer

ð12Þ

Soil percolation I (mm) recharges the shallow aquifer. Extraction of water from the soil due to evapotranspiration (ET in mm) is depth dependent. According to Neitsch et al. (2011), WATPPASS is parameterized so that 50% of the evapotranspiration is extracted by the 6% upper soil. If surface horizons cannot provide the requested water, water from deeper horizon is extracted. The initial water extraction at a specific depth z (mm) is calculated using

wup;z ¼

   ET z  1
exp
10  ½1
expð
10Þ height soil

ð14Þ With Qpest, the amount of pesticide in the layer (g); SAT, the amount of water in the layer at saturation (mm); Koc: the soil organic carbon–water partitioning coefficient ((mg kg
1)/ (mg L
1)); foc, the organic carbon fraction of the layer (g g
1); BulkDensity, the bulk density of the layer (g cm
3) and Height, the height of the layer (mm). For demonstration of the analytical differential equation solving resulting in Eq. (14), see Neitsch et al. (2011). Pesticide percolations from the deeper (11th) soil layer enter the shallow aquifer reservoir. According to Neitsch et al. (2011), pesticides extracted into runoff are weighted by a parameter which varies between 0 and 1 (PestCo) and the maximum amount of pesticide transported by water is constrained by the solubility of the pesticide in water (Sw in mg L
1). 2.4. Watershed scale production functions 2.4.1. Surface runoff Surface runoff at the watershed scale is the sum of the daily runoff calculated for all the fields of the watershed weighed by their relative area on the watershed. As a result, this version of WATPPASS assumes that the place of a field on the watershed has no impact on its runoff contribution at the watershed outlet. This hypothesis, which may only be valid for small watersheds having a dense hydrographic network, doesn’t require a specific routing function in the stream channel. 2.5. Groundwater equations

The water content of the soil layer varies according to

DSW layer ¼
Percolayer

Q pest;flow ¼ Q pest   
W flow  1
exp SAT þ K oc  f oc  BulkDensity  Height

ð13Þ

WATPPASS assumes that evapotranspiration demand can be extracted from the soil until a maximum depth of 1000 mm or less if the total depth of soil is less than 1 m. For each soil layer, the amount of water extracted is calculated as the difference between lower and upper bounds of the layer: wup,lay = wup,zl
wup,zu. 2.3. Pesticide in soil and in runoff The soil pesticide transfers component is adapted from SWAT (Arnold et al., 1998; Neitsch et al., 2011). Only the dissolved pesticide fraction it subjected to runoff and percolation. In each soil layer, dissolved pesticide transported by water flow Wflow (mm) is

The groundwater processes in the WATPPASS model are structured by two overlapping reservoirs which respectively represent a shallow aquifer and a deep aquifer (Fig. 1). Reservoir discharge is calculated linearly so that the outflow of the shallow reservoir (Osa) is

 Osa ¼

 1  W sa Dsa

ð15Þ

with Dsa in days characterizing the recession curve of the shallow aquifer and Wsa the amount of water in the shallow aquifer (L). The water level in the shallow aquifer is updated daily using

DW sa ¼ I
Osa

ð16Þ

A fraction of the shallow aquifer outflow is directly discharged into the stream. This fraction is

Q sa ¼ p1  Osa

ð17Þ

with p1 the fraction of outflow directly discharged at the outlet (unitless). The remaining fraction percolates from the shallow aquifer and is calculated using:

Psa ¼ ð1
p1Þ  Osa

ð18Þ

This percolated fraction is split into two fractions. The first fraction is added to the deep aquifer and is calculated using:

Lsa ¼ p2  Psa

ð19Þ

The remaining fraction (1
p2)  Psa is submitted to a unit hydrograph determined by the number of day of influence of a unitary rainfall on the watershed discharge at outlet. This unit hydrograph is used to simulate transfer between the two aquifers.

C. Mottes et al. / Journal of Hydrology 529 (2015) 909–927

913

Fig. 2. The Ravine watershed, split into 70 concentric rings of same width generates a 70 days response of the unit hydrograph (LUH = 70 days). The LUH value for that example has been arbitrary selected for illustration purpose.

The unit hydrograph (UH) is generated by splitting the watershed into LUH concentric rings centered at the outlet (Fig. 2). Finally, the unit hydrograph is defined by

UH ¼ ½UH1 . . . UHi . . . UHLUH  ¼

½S1 . . . Si . . . SLUH  PLUH i¼1 Si

ð20Þ

where Si is the surface of the iest ring starting from the outlet. Then, the variations of Qsub at different time steps ranging from t to t + LUH are calculated using

½DQ sub ðtÞ . . . DQ sub ðt þ iÞ . . . DQ sub ðt þ LUHÞ ¼ UH  ðð1
p2Þ  Psa ðtÞÞ

ð21Þ

Fig. 2 illustrates the discretization of the Ravine watershed with LUH = 70, corresponding to a response unit hydrograph for water discharge at outlet of 70 days. Reservoir outflow from the deep aquifer (Oda) is calculated linearly by

 Oda ¼

 1  W da Dda

ð22Þ

with Dda in days characterizing the recession curve of the deep aquifer and Wda the amount of water in the deep aquifer (L). At each time step: Wda is updated using:

DW da ¼ Lsa
Oda

ð23Þ

As for the shallow aquifer, a fraction of the deep aquifer outflow is discharged into the stream

Q da ¼ p3  Oda

ð24Þ

and the remaining fraction

Lda ¼ ð1
p3Þ  Oda

ð25Þ

is considered as definitely lost. Lda accounts for the water losses from the watershed. Finally, the total discharge at outlet is calculated by summing the flows calculated for the four contributing reservoirs (Fig. 1)

Q o ðtÞ ¼ Q sr ðtÞ þ Q sa ðtÞ þ Q sub ðtÞ þ Q da ðtÞ

ð26Þ

where Qo(t) is the discharge at outlet calculated by the model (L day
1).

2.6. The transfer function Water discharge from aquifers, calculated with the production functions (Eqs. (17), (21) and (24)), are considered as constraints for the ground water pesticides component of WATPPASS. Once pesticides enter to the groundwater system, they are assumed to have no interaction with the surrounding environment. Thus, no retardation factors are taken into account by the transfer function and no further parameters are needed for groundwater simulations. In the groundwater system, any transfer of water is associated with pesticide transfers using the equation:

Mpestexported ¼ C pest  V exported C pest ¼

M pest QWater

ð27Þ ð28Þ

where Mpestexported is the amount of pesticide exported from a compartment (g), Vexported is the volume of water flowing out of a compartment (L), Cpest is the concentration of pesticide in a compartment (g L
1), Mpest is the amount of pesticide in a compartment (g), and Qwater is the amount of water in a compartment (L). In WATPPASS, groundwater reservoirs are horizontally discretized by ‘‘LUH” units corresponding to circular concentric reservoirs. This kind of reservoir and their number facilitate computation and, we assumed this is suited to the response dynamic of the watershed according to its size. Pesticides percolated from each field are split between the circular concentric reservoirs using a transition matrix generated by the General Polygon Clipper Library. Percolated amounts of pesticides are instantaneously added to the corresponding circular concentric reservoirs constituting the shallow aquifer. Pesticides are assumed to be perfectly mixed in each concentric reservoir. Outflow water volumes and associated pesticides amount are removed from reservoirs using the following steps: (1) percolated water volumes (Psa and Lda) and associated pesticides amounts are removed on all sub-compartments according to their relative contribution to the total amount of water in the whole reservoir. Percolated fluxes (Lsa) and associated pesticides are added to their homologous underlying reservoirs. (2) Discharges at the outlet (Qsa and Qda) and associated pesticides are taken from the downstream reservoir (the one with the lower indices) until the total discharge demand generated by WATPPASS is achieved. Once

914

C. Mottes et al. / Journal of Hydrology 529 (2015) 909–927

reservoirs have been updated by removing outflow and percolated volumes associated with pesticides, equilibration of water in compartments is simulated. Equilibration is performed on the basis of water depth in reservoirs, the upper reservoir discharges into the lower one to achieve equilibrium of water depth.

(

Wtransfers ¼

Qw1 S1
Qw2 S2
S1 S2 ðhc1
hc2 Þ ; S1þS2

if ½hc1 > hc2

0;

if ½hc1 6 hc2

ð29Þ

In Eq. (29), the reservoir 1 stands for the upper reservoir (the one with a higher index), and reservoir 2 for the lower one (the one with the lower index). Wtransfers (mm) is the amount of water transferred from the upper reservoir to the lower reservoir. The water depth hci in a reservoir i is calculated as follows:

hci ¼

QW i Si

ð30Þ

with Si the surface of reservoir i (m2), and QWi the amount of water in reservoir i (L). Only transfers from upper zones to lower zones are allowed by WATPPASS. If more than one compartment is emptied by the discharge demand, equilibrium is performed by performing ‘‘n” passes of equilibration with ‘‘n” being the number of reservoirs emptied by the discharge to outlet (Qda or Qsa depending on the reservoir). During the equilibration of compartments, water transfers are associated with pesticide transfers between compartments according to Eq. (27). 2.7. Model parameters and calibration procedure 2.7.1. Parameters Surface water runoff function. The runoff production function requires rainfall inputs (P in mm) on the day and the day before, the potential evapotranspiration of the day (ETP in mm) and the runoff estimate from each field Qsr on the day before. The runoff production function requires two parameters which are evaluated at the watershed scale a1 and a2 and one parameter, curve number for average moisture conditions CN2, for each field of the considered watershed. The output of the function is the daily runoff of each field. Soil water function. For each layer, the soil water production function output is the water percolating from the soil layer (Percolayer in mm) and the input is the amount of water entering the soil layer by the upper border. It is either the rainfall (P in mm) for the first layer or Percolayer
1. Parameters of the soil water production function must be provided for the 11 layers and are the amount of water in the soil layer at field capacity (FClayer in mm), the water in the soil at saturation (SATlayer in mm), the water in the soil at wilting point (WPlayer in mm) and the saturated hydraulic conductivity (Ksatlayer in mm h
1). The soil water at saturation is estimated by

SAT layer ¼ porositylayer  Height layer

ð31Þ

where porositylayer is the porosity of the layer and Heightlayer is the height of the layer (mm). If only the soil bulk density is available, the porosity of each layer is calculated according to

porositylayer

  BulkDensitylayer ¼1
2:65

ð32Þ

where 2.65 (g cm
3) stands for the density of the material constituting soil (Ahuja et al., 2000). Pesticide extraction function. The inputs of the pesticide extraction function are the outputs from the surface water runoff Qsr and from the soil water function Percolayer. The pesticide extraction function requires a parameter related to the pesticide characteristics: the soil organic carbon–water partitioning coefficient of the

molecule (Koc in L kg
1). The pesticide extraction function requires the organic carbon fraction content for each soil layer (foc in g g
1). Groundwater function. The inputs of the ground water model are the percolations from each fields located on the watershed. The spatial coordinates of the outlet, contour maps of the watershed and of the fields of the watershed are needed in a shapefile format for WATPPASS to generate the watershed geometry. The groundwater function outputs are the discharges from each groundwater compartments (Qsa, Qsub and Qda). The groundwater function requires 6 empirical parameters: Dsa (days), Dda (days), LUH (days) and 1, p2, p3 (unitless). Transfers function. The WATPPASS transfer functions do not require any further parameters than those used for the watershed scale production functions. The conceptual compartmentalization of the catchment is used to control pesticide transfers. The transfers function outputs are the pesticide masses and concentrations at the watershed outlet. 2.7.2. Calibration and evaluation strategy Hydrology. We used three criteria related to the shape of the whole hydrograph to analyze model performances. The first criteria selected for hydrology is the Nash and Sutcliffe (Nashsqrt) criterion taken for the square root of values (Nash and Sutcliffe, 1970). Using the square root of values ensures equilibrium between low flows and peak flows. We also took the root mean square error (RMSE), and the water balance. WATPPASS uses a two steps calibration procedure for hydrology. The first step is to calibrate runoff because runoff related parameters are sensitive as for most hydrological models. That also ensures that surface runoff is not simulated by the shallow aquifer discharge (Section 4.3.1). The second step is a global calibration procedure that mobilizes a particle swarm optimization algorithm to maximize the Nashsqrt criteria. A correlation analysis of the data on the calibration period is used to define parameter’s calibration intervals (Section 3.5). Pesticides. Pesticides are not calibrated in this article because cropping systems assessment and design studies usually do not gather pesticides analyses required for pesticide calibration. We used three evaluation criteria that are the Nashsqrt, the RMSE and the pesticide load balance to assess the pesticide performances of the model. We also decomposed the mean square deviation (RMSE2) according to the methodology from Kobayashi and Salam (2000) to better understand pesticide modeling results. They detail the decomposition the mean square deviation into three terms: RMSE2 = SB + SDSD + LCS. SB assesses bias, SDSD assesses whether the model correctly reproduces the magnitude of fluctuation of the measurements and LCS assesses whether it simulates the temporal pattern of the fluctuation existing in the measurements. 3. Study site 3.1. Description Presentation. The Ravine watershed (14°490 200 N, 61°70 1400 W) is located on the North-East side of the Martinique Island in the French West Indies (Fig. 3). The Ravine watershed covers 131 ha. It is located on the eastern flank of the Martinique volcano La Pelée. Elevation of the watershed ranges from 315 to 630 m. The watershed divides into 2 areas according to local slopes, the downstream and the upstream areas with mean slopes ranging between 0% and 15%, and between 15% and 30%, respectively. The drainage network is represented by three main gullies which drain flows during rainy events. A perennial stream appears between 150 and 200 m upstream the outlet, indicating a permanent contribution of

C. Mottes et al. / Journal of Hydrology 529 (2015) 909–927

915

Fig. 3. Location of the Ravine watershed in the Caribbean (a) and in La Martinique (b); Contour map of the Ravine watershed and surrounding environment with localization of the measurement equipment (c).

groundwater to stream discharge in the downstream zone of the watershed. Climate. The West Indies are submitted to a maritime tropical humid climate. Climatic years are split into two seasons: a ‘‘dry” season lasting from February to April and a ‘‘rainy” season lasting from July to October. These two seasons are separated by transition period of variable rainfall intensities. Nevertheless, the dry season rarely includes high water deficits. A mean annual rainfall of 3607 mm is observed between 1978 and 2012 at the Eden station, which is the nearest (600 m) meteorological station from the outlet (Météo France, 2013).

Land use. A land use survey was performed from April 2011 to November 2011. For each field, land uses were recorded and classified. For this study, the land uses are assumed to be constant throughout the simulations period which is true for more than 80% of the recorded land uses (except for bare soil and fallow). This simplification is necessary to understand the hydrological functioning of the Ravine watershed. The watershed cuts across 20 farms with different production orientations: 18% of land uses are chayote (Sechium edule), 13% banana (Musa spp.), 6% pineapple (Ananas comosus), 6.5% fallow (Multiple species), 17% are covered by other horticultural species, and less than 2% are covered by

916

C. Mottes et al. / Journal of Hydrology 529 (2015) 909–927

roads and tracks roads. The remaining surface is mainly covered by forests, meadows and pastures. We assumed that these land uses generate different effects on hydrology and pesticides transfers. The diversity of land uses found on the Ravine watershed matches the diversity of the one found in comparable tropical volcanic context. Geology. Volcanic formations are pyroclastic deposits of andesitic composition (BRGM, 1983). Drilling sites and field surveys on the watershed showed two main geological formations which are (from the deepest to the most superficial): nuees ardentes exceeding 70 m height, and a pumice layer of 2–12 m depth (BRGM, 1983). Depending on the location, the nuees ardentes formation is composed by a different number of layers which can be identified by their alteration degree. They have been set on a paleovalley oriented ESE–WNW (Lalubie, 2010; Mouret, 1979). Soils. Soils of the watershed are Andosol (Colmet-Daage, 1969; WRB, 2006) which are young volcanic ash soils derived from pumices and ashes deposits. Andosol in the Lesser Antilles have high infiltration rates with a saturated hydraulic conductivity over 60 mm h
1 (Cattan et al., 2006). Charlier et al. (2008) have showed in same pedoclimatic environment that percolation through the water table is predominant, thus limiting lateral subsurface flows. 3.2. Hydrological data Daily rainfall and daily global radiation were acquired from Météo France. Rainfall comes from 2 stations surrounding the watershed downstream (Eden station) and upstream (Aileron station) (see Fig. 3 for locations). The daily rainfall on the watershed is estimated by the arithmetic mean of the two stations. The global radiation was measured using data from the closest meteorological station (Morne des Cadets) located at 9.5 km South–South West from the watershed. Due to difference in location (station located on the wind protected slope of Martinique) a 0.83 coefficient has been applied on the radiations measured at this station (Vittecoq et al., 2007). Potential evapotranspiration (PET) was calculated with the formula used for the French Caribbean archipelago (Khamsouk, 2001; Tixier, 2004):

PET ¼ 0:2392  ð0:83RgÞ þ 0:00037

ð33Þ

PET is the potential evapotranspiration in mm day
1 and Rg the global radiation in MJ m
2 day
1. In a first approximation, we considered evapotranspiration as equal to potential evapotranspiration due to very rare limiting water supply during the year. Discharge at the watershed outlet was measured at a 2 min time step using a PCDR1830-CS420-L sensor (Campbell ScientificÒ). Gauging of the stream has been performed using a SalinoMADD (MADDÒ) for low water level (less than 30 cm water depth). Manning–Strickler relationship was used for higher levels using measured sections of the stream since measurements in the river could not be performed due to security issues. The location of the station has been selected because of the existence of a natural weir in the streambed which controls ideally water flows. Pesticide samples were collected proportionally to measured discharge. The discharge was summed by the data logger (Campbell Scientific CR1000) until it reached a threshold value ranging from 300 m3 (during the dry season) to 1800 m3 (during the rainy season). When this value was reached, 200 mL of water were sampled. Weekly composite sample were obtained by mixing all the samples collected during the week. Time series of rainfall, evapotranspiration, and discharge for the period from 01 July 2011 to 18 April 2013 were synchronized at a time step of 1 day. These same data were also synchronized from 11 October 2011 to 01 February 2013 at a time step of 1 week to compare the pesticide time series.

3.3. Storm and baseflow partitioning Determining surface runoff is necessary to understand the functioning of the watershed and to calibrate the runoff module of WATPPASS. To do so, we performed stormflow/baseflow partitioning using the high frequency filter proposed by Arnold and Allen (1999):

QRðtÞ ¼ b  QRðt
1Þ þ

ð1 þ bÞ  ðQTðtÞ
QTðt
1ÞÞ 2

ð34Þ

QR(t) is storm discharge calculated at time step t, b is the parameter of the filter and QT(t) is total discharge at the outlet at time step t. First, this filter needs to be parameterized on a short period partitioned using manual separation. A 20-days period, lasting from the 21/07/2011 to the 11/08/2011, was chosen for parameterization. This period has daily rainfall ranging from 2.5 mm to 75.55 mm and mean daily rainfall of 22 mm which correspond to a high rainfall range compared to the rainfall events from the 01/01/2007 to the 18/04/2013. During this period 36 runoff events of different types have been identified. Stormflow and baseflow were separated according to the straight line method between the beginning point of the stormflow and the beginning of the recession curve (Chow et al., 1988). This method considers that the recession curve takes the form of an exponential decay that is linearized on a logarithmic scale (Fig. 4). The filtering has been performed from the 01/07/2011 to the 18/04/2013 on a 2 min time step with the b parameter equals to 0.99975. This parameter has been fitted on the 20-day period with the results of the manual separation given a Nash criterion of 0.99. A linear regression of daily filtered vs. daily manually calculated stormflow gave a coefficient slope coefficient of 0.99 with a R2 of 0.98. 3.4. Annual water balance The storm and baseflow separation makes it possible to estimate annual flows for the different components of the water balance:

Lann þ dW ann ¼ Pann
ET ann
Rann with Rann ¼ Bann þ Sann

ð35Þ

where Lann are the annual losses of the aquifer (mm), dWann is the annual variation in the aquifer stock (mm), Pann is the annual rainfall (mm), ETann is the annual evapotranspiration (mm), Rann is the annual streamflow (mm), Bann is the annual baseflow (mm), and Sann is the annual stormflow (mm). In West-indies, hydrological year range starts in February which is statistically the driest months of the year (Fig. 5). The hydrological year analyzed (2012–2013) is characterized by a very rainy month of May while having wet season drier than the mean year (Fig. 5). The annual water balance analysis shows that groundwater flows are predominant and that 77% of the streamflow comes from groundwater flows (Table 1). These values are close from the one reported by Charlier et al. (2008) on a watershed in a similar pedoclimatic context (Guadeloupe island – French West Indies). These results confirm that the attention given to groundwater flow in WATPPASS makes it suited to simulate hydrological functioning and solute transport on the Ravine watershed. 3.5. Hydrological response We performed cross-correlation analysis between rainfall and discharge on the calibration period (Section 4.1) to understand the response of the watershed to rainfall and better understand the hydrological behavior of the Ravine watershed. Crosscorrelation makes it possible to determine the influence of rainfall on discharge with no further treatment if rainfall is not auto-correlated (Mangin, 1984). Fig. 6a shows that daily rainfall

917

C. Mottes et al. / Journal of Hydrology 529 (2015) 909–927

10

Rainfall (mm)

a 5

0

100

200

300

400

500

600

300

b 200

100

0

scale (10−3.m3.s−1)

Discharge at log

Discharge (10−3.m3.s−1)

0

0

100

200

300

400

500

600

c 2

10

0

100

200

300

400

500

600

Time (minutes) Fig. 4. Example of flood event at the hourly time step and filtering used to separate storm and baseflow: the recession curve separation ‘‘dotted line”; the filter separation ‘‘dashed line”.

700

Average 1978−2012 2011 2012

600

Rainfall (mm)

500

400

300

200

100

0

F

M

A

M

J

J

A

S

O

N

D

J

Months Fig. 5. Annual distribution of rainfall at the Eden station, from February to January (Data from Météo France (2013)).

is almost not auto-correlated while the cross-correlation analysis shows that rain can influence discharge for 50 days (Fig. 6b). The watershed response shows two separate effects (Fig. 6b): (1) a first correlation peak (r = 0.6) lasting between 2 and 5 days; (2) a buffered peak lasting a bit less than 50 days. As expected in our initial hypothesis, two groundwater components can be char-

acterized on the watershed: one between 0 and 5 days standing for fast transfers including flows from the shallow aquifers, and one over 50 days which may stand for more buffered flows from a deeper aquifer. We thus selected interval for calibration within the range of the values observed by cross-correlation (Section 4.3.3).

918

C. Mottes et al. / Journal of Hydrology 529 (2015) 909–927

Table 1 Annual water balance of the Ravine watershed in year 2012.

4. Model application

01/02/2012 31/01/2013 Hydrological year

Auto−correlation

Rainfall (mm) PET (mm) Streamflow (mm) Stormflow (mm) Baseflow (mm) Losses and stocks (mm)

3710 (100%) 1438 (39%) 874 (23%) 203 (5%) 671 (18%) 1398 (38%)

a

0.8 0.6 0.4 0.2 0 −0.2

Cross−correlation

0

20

40

60

80

100

0.6 0.4 0.2 0 −0.2 0

20

40

60

80

The collected data were separated into two data sets. First, a calibration period extends from the 30/06/2011 to the 24/02/2012. The calibration period lasts 240 days including high flows as well as aquifers recessions. The validation period extends right after the calibration period from the 25/02/2012 to the 31/12/2012 and last 310 days and includes high flows and aquifers recessions. All the parameterization and calibration were performed on the calibration period. WATPPASS is evaluated by comparing the simulated data to the measurements of the validation period. WATPPASS estimates most of intermediate variables on the basis of the state of the watershed the day before. As a result, the model is initialized prior to the calibration period using a blank simulation during one year. The state of the model at the end of the initialization is used as a starting point for the calibration period and the state of the model at the end of the calibration period is used as a starting point for the validation period. 4.2. Sensitivity analysis

b

0.8

4.1. Calibration and validation periods

100

Lag [d] Fig. 6. (a) Rainfall auto-correlation function on the Ravine watershed, and (b) rainfall–discharge cross-correlation on calibration period at the Ravine watershed outlet.

We performed a sensitivity analysis of WATPPASS to determine the impact of the parameters on the simulated discharge. For this, we used the Nashsqrt calibration criteria. For sensitivity analysis, we ran the models on the calibration period and analyzed the effect of variations in the parameters values on the Nashsqrt coefficient in a 0–25% range with 5% steps (Fig. 7). All other parameters are set to their estimated or calibrated values. Fig. 7 shows that the most sensitive parameter is p2 which stand for the separation of percolation into the unit hydrograph and the deep aquifer. This is an empirical parameter, variation of
25% and +25% lead to degradation of Nashsqrt of more than 60%. WATPPASS is sensitive to the

80

b

a

60

40

20

0

-20

-10

0

10

Relative error on parameter (%)

p2 Dsa

20

Relative error on Nashsqrt (%)

Relative error on Nashsqrt (%)

15

10

5

p1 Dda p3 LUH CN

0 -20

-10

0

10

20

Relative error on parameter (%)

Fig. 7. Sensitivity of WATPPASS to major parameters: (a) percentage error of the Nashsqrt efficiency criteria comparatively to its calibration value. (b) Zoom on low percentage changes. p2: percolation from the shallow aquifer separation parameter between the unit hydrograph and the deep aquifer; Dsa: recession time of the shallow aquifer (days); p1: percolation separation parameter of shallow aquifer outflow between the discharge and percolation; Dda: recession time of the deep aquifer (days), p3: deep aquifer outflow separation between losses and discharge; LUH: length of the unit hydrograph (days), CN: curve number.

919

C. Mottes et al. / Journal of Hydrology 529 (2015) 909–927

4.3. Parameterization 4.3.1. Surface runoff We used Eq. (7) proposed by Hawkins (1993) and Eq. (6) to calculate the curve number value corresponding to each rainfall (P)– runoff (Qsr) couple determined by the runoff separation filter applied on discharge (Fig. 8):

  1=2 S ¼ 5 P þ 2Qsr
ð4Qsr2 þ 5PQsrÞ

ð36Þ

The asymptotic curve number of the watershed can be estimated using an adjustment in the form of

CNðPÞ ¼ ð100
CN 1 Þ  expðb  PÞ þ CN1

ð37Þ

with CN1 the asymptotic curve number of the watershed and b the recession parameter (mm
1). The solid line in Fig. 8 is the threshold of runoff on a rainfall 100 equation. depth (Hawkins, 1993). It is defined by the CN ¼ ð1þP=2Þ The observed pattern of curve number is close to the complacent response defined by Hawkins (1993) because the observed points tend to stay close to the solid line. Patterns close to complacent behavior can be associated with runoff generated by a partial area on a watershed. Nevertheless, the adjustment reaches an asymptotic value of 48 which is the curve number of the Ravine watershed (Fig. 8). The situation of runoff generated by a partial area is compatible with results found by Charlier et al. (2008) who, on a similar tropical volcanic watershed found that hortonian flow is generated mainly by impervious areas and plots cultivated with banana due to a rainfall concentration at the base of the banana trees increasing runoff at the plot scale (Cattan et al., 2009; Charlier et al., 2009b). This effect may be the result of banana and pineapple cropping systems in the Ravine watershed. The latter are cultivated on plastic mulch and the funnel-shaped organization of pineapple leaves could also concentrate flow and favor runoff. Runoff generated by pineapple on a soil with a high infiltration rate support this hypothesis because the pineapple (straight ridge) field was found to be the field which generates the most

100 90

Curve Number (unitless)

curve number. An over estimation of 25% of this parameter leads to 60% error on the Nashsqrt while an underestimation of 25% leads only to a 10% error on the Nashsqrt. The optimal curve number to achieve the best Nashsqrt is 2.5% lower than the initial value, but in this case, the improvement on Nashsqrt is low (less than 1%). WATPPASS is sensitive to the p3 parameter which is used to calculate the discharged fraction of the deep aquifer (7% modification of the Nashsqrt) and to underestimation of LUH. The other parameters induce a Nashsqrt deviance of less than 5% in their analyzed range. The low sensitivity of WATPPASS to parameters Dsa, Dna and LUH make it possible to estimate their values or to set a narrow range of value for their calibration. This makes it possible to reduce the number of calibration parameters. The other parameters are purely empirical and need to be optimized. Initialization of the watershed state has been performed by running the model during one year before the 01/07/2011 to equilibrate the compartments. To assess WATPPASS sensitivity to the length of the initialization period, and to ensure that there are no long term tendencies of the groundwater reservoir, the effects of different lengths of initialization (from no initialization to 3 years) have been tested on the calibration period. The Nashsqrt obtained for initialization of 2 and 3 years were equal to the one obtained with the one year initialization (0.76). This result indicates that WATPPASS has no long term tendency. Initialization of 6 month decreased the Nashsqrt to 0.74, 0.71 with 3 month and 0.67 without initialization. We conclude that a period longer than twice Dda is necessary to bring the groundwater system to equilibrium for further simulations.

80 70 60 50 40 30 20

0

20

40

60

80

100

120

Rainfall (mm) Fig. 8. Regression of CN as a function of rainfall on the calibration period; the fitted equation (dotted line) is CN(P) = (100
48)  exp(
0.024  P) + 48 with R2 = 0.97; points: calibration days; circles: validation days; solid line: complacent runoff behavior threshold (see explanations in the text).

runoff compared to bare soil, banana and sugar cane fields (Khamsouk, 2001). To account for the effect of antecedent moisture conditions on runoff, the parameters a1 and a2 of Eq. (4) were calibrated using the following procedure. First, the best asymptotic curve number (plotted in Fig. 8) for each rainfall-runoff couple was calculated using the following equation:

CN1 ¼

CN
100  expðb  PÞ with b ¼
0:024 1
expðb  PÞ

ð38Þ

Then, the parameters a1 and a2 were calibrated to achieve the best root mean square error between the CN1 estimated for rainfall-runoff couples and the curve number calculated by the runoff production function, the asymptotic curve number of the watershed is used as an initial condition. On the calibration period, a1 has been found to be equal to 2 and a2 equal to 0.21 for the Ravine watershed (Fig. 9). Then, we selected curve number values for the different land uses of the Ravine watershed and attributed them to fields of the watershed depending on their cropping systems (Table 2). These values rely on a literature survey only and have not been validated by experimentation in the Ravine watershed conditions. Nevertheless, they respect the hypothesis that specific land uses (impervious, banana and pineapple) generate significantly higher amount of runoff than the other land uses. We selected curve numbers corresponding to row crops with good and poor hydrological conditions for respectively banana and pineapple (USDA, 1986). The values are compatible with the results from Matamoros Camposano (2004) on banana and the processing of the data of Khamsouk (2001) on pineapple. Finally, the weighted curve number of the watershed was estimated by the linear combination of the CN of the different land uses weighted by their respective relative area. It differs by two points (46) from the asymptotic calibrated curve number (48). We were satisfied by the difference because of the lower sensitivity of WATPPASS to curve number underprediction (Fig. 7). 4.3.2. Field scale production function The soils of the Ravine watershed are andosols. To determine water allowed to percolate and to evapotranspirate in the soil water production function, field capacity (FC in mm) and wilting

C. Mottes et al. / Journal of Hydrology 529 (2015) 909–927

∞

Asymptotic Curve Number (CN )

920

70

Simulated CN

∞

Calculated CN

∞

60

50

40

30

0

50

100

150

200

250

300

350

400

450

500

550

Time (day) Fig. 9. Simulated curve numbers (solid line) and calculated asymptotic curve numbers (squares) for days with rainfall over 10 mm (calibration RMSE = 6.2, validation RMSE = 5.7).

Table 2 Curve numbers from the literature for the land uses and cropping systems of the Ravine watershed. Land use

Curve number selected

Surface covered in watershed (ha) (% of total)

CN value source (type of soil: A)

Banana (no-till)

67

17.12 (13.1%)

Pineapple (plastic mulched)

72

8.02 (6.1%)

Bare soil Forest Truck crops (dasheen, sweet potato, cabbage, tomato, cucumber) Chayote Pasture Impervious Meadow Orchard Roads (dirt) Road (impervious) Other Total

77 30 45

3.24 (2.5%) 44.27 (33.8%) 21.19 (16.2%)

Row crop with good hydrologic condition Matamoros Camposano (2004) Row crop poor hydrologic condition (CN higher than 66 found by Khamsouk (2001) due to plastic mulching) (USDA, 1986) Puerto Rico (fallow type) National Handbook (USDA, 1986) (USDA, 1986) Puerto Rico (truck crops reference)

48 39 98 30 43 72 98 48 46

23.88 (18.2%) 1.34 (1.0%) 0.60 (0.45%) 8.6 (6.6%) 0.94 (0.7%) 0.96 (0.7%) 0.8 (0.6%) 0.04 (0.03%) 131 (100%)

(USDA, 1986) Puerto Rico (coffee reference) National Handbook (USDA, 1986) National Handbook (USDA, 1986) National Handbook (USDA, 1986) National Handbook (USDA, 1986) National Handbook (USDA, 1986) National Handbook (USDA, 1986) Asymptotic mean Weighted CN

point (WP in mm) of layers have been calculated at respectively pF = 2.7 and pF = 4.2 by adapting the formula reported by Tixier (2004) for andosols: 1

Water content ðpFÞ ¼ 44:72 

0:4 þ ð4:36
pFÞ2  SAT 100

ð39Þ

where SAT is calculated using Eq. (2). All soil parameters have been set to fixed values from literature references (Table 3). Soil bulk density was assumed to be equal in all soil layers because concordant values among several authors (Table 3). The high Ksat of andosols have no impact on modeling results at a daily time step unless the daily infiltration is higher than 600 mm which never occurred in our data. It would require both a finer rainfall resolution and a finer calculation time step to impact the simulation results.

4.3.3. Groundwater production function Table 4 summarizes the calibration range and the fitted values for the 6 parameters on the Ravine watershed. The calibration range for Dsa and LUH were constrained by the correlation analysis we performed on the calibration period (Section 3.5). The Dda parameter was constrained to avoid very long term inertia of the deep aquifer that cannot be assessed with our data. The p1, p2 and p3 were allowed to vary in their full range: 0–1. 4.3.4. Pesticide extraction function Andosols contain tubular pores associated with micropores (Cabidoche et al., 2009; Prado et al., 2009; Woignier et al., 2012). As a result, the whole porosity is not accessed by water during percolation (Prado et al., 2009). This specificity requires adaptations of the equations of the model. We propose a simple representation of macropore flows on pesticide transfers: when tubular macropores are enabled, WATPPASS assumes that, each day, only the pesticide

Table 3 Values of soil parameters used to run WATPPASS on the Ravine watershed. Parameter

Value

Sources

Curve numbers Soil bulk density (g cm
3)

See Table 2 0.73

This article Colmet-Daage and Lagache (1965), Dorel et al. (2000), and Venkatapen (2012) Venkatapen (2012) Cattan et al. (2006) Sansoulet (2007) Neitsch et al. (2011)

Soil depth (mm) Ksat(0–30cm) (mm h
1) Ksat(30–100cm) (mm h
1) PestCo

1000 60 25 0.5

Table 4 Calibration range and fitted values of the WATPPASS parameters. Parameters (unit)

Calibration range

Fitted values

Dsa (days) p1 (unitless) p2 (unitless) Dda (days) p3 (unitless) LUH (days)

2–5 0–1 0–1 50–200 0–1 40–60

3.3 0.1 0.79 165 0.29 55

C. Mottes et al. / Journal of Hydrology 529 (2015) 909–927

fraction accessed by the first macroporous flows (SAT
FC) is loaded with pesticides. To do so, Wflow of Eq. (14) is adapted:

 Wflow ¼

SAT
FC; if ½Perco > SAT
FC Perco;

ð40Þ

Otherwise

Pesticide equilibration from the micropores to macropores is carried out at the end of each day. This modification leads to a dilution of pesticide concentration in the leachate when rainfall amount is high. 4.3.5. Pesticide and soil parameters Polluted fields by chlordecone on the watershed were parameterized using results of soil analysis (0–30 cm) required for the cultivation of several crops (Clostre et al., 2013; Lesueur-Jannoyer et al., 2012) (contamination map not shown due to restrictive conditions on data sharing). The intensive deep tillage performed with excavator during the last 30 years was assumed to have homogeneously mixed chlordecone throughout 60 cm. For deeper layers (60–100 cm), results after 38 years (1973–2013) of simulations (repetitions of climatic years 2011 and 2012) indicated concentrations representing 25% of the 0–30 cm layer. Plet (2012) performed vertical profiles of chlordecone concentrations and showed that the deep layer (60–90 cm) has concentrations ranging from 25% to 75% of the one of the 0–30 cm layer. The observed variability in concentration profiles might mainly depend on the frequency of very deep tillage practice during this period (Clostre et al., 2014). Finally, the 60–100 cm layers were set to half the concentration of the 0–30 cm layer. Organic matter content of each field of the watershed was attributed according to the cropping system (Table 5). Andosols have been reported to have a mean surface horizon at 4.3% of organic carbon with linear decrease until 60 cm depth to reach 1% in the 60–100 cm layers (Venkatapen, 2012). The distribution of organic carbon with depth was then calculated by an equation matching that pattern:

OC calc ðdepthÞ ¼ f oc;0
0:0055  depth  f oc ¼

1;

ð41Þ

if ½OC calc < 1

ð42Þ

OC calc ; Otherwise

The molecular characteristics of chlordecone have been selected from the literature with preference for local in situ studies. There is no significant environmental degradation of chlordecone which makes it possible to use the molecule as a conservative tracer (Baran and Arnaud, 2013; Cabidoche et al., 2009; Dolfing et al., 2012; Fernandez-Bayo et al., 2013b). We selected a solubility in water (Sw) of 3 mg L
1 (FOOTPRINT, 2013; Kenaga, 1980). Two values for the Koc coefficient were used in our simulations. First, we took a Koc value of 17,500 L kg
1 that is usually taken as a starting point for chlordecone transfers on andosol (Cabidoche et al., 2009). Second, we used the value (14,400 L kg
1) which was specifically reported by the same authors on the andosols of Martinique that correspond to the one of the watershed.

Table 5 Organic carbon fraction (foc,0) attributed to the first soil horizon for different cropping systems. Cropping system

Top layer organic carbon 0–10 cm, (foc,0) (%)

Source

Banana/orchard Pineapple Truck crops Forest–wood Pasture Other cropping systems

3.4 4.1 3.1 6.7 2.7 4.3

Albrecht et al. (1992) Albrecht et al. (1992) Albrecht et al. (1992) Albrecht et al. (1992) Albrecht et al. (1992) Venkatapen (2012)

921

4.4. Model adjustments In this section we present two adjustments of the model equations and structure that are intended to better represent the temporal dynamic of pesticides at the catchment outlet. The initial model structure is called ‘‘Initial model”. The model that integrates only the modification of pesticide transfer delays in aquifers is called ‘‘Model delay adjusted” (Section 4.4.1). The model that integrates both, the modification of pesticide transfer delays in aquifers and the modification of macropore flow (Section 4.4.2) is called ‘‘Model delay + macropores adjusted”. 4.4.1. Aquifer delay adjustment In this adjustment of WATPPASS, we slightly modified the transfer function described in Section 2.6 to simulate a retardation factor on pesticide transfers in aquifers. This modification does not require supplemental parameter. The adjusted model delays pesticides that leach out of soil before they enter into the shallow aquifer reservoirs according to the following rule: the pesticide mass initially percolating into the iest concentric reservoir on the day ‘‘d” enters the reservoir on the day ‘‘d + i”. 4.4.2. Macropore flow adjustment Equations representing macropore flows from Section 4.3.4 were adjusted. In our adjusted model, macropore flows also lead to a bypass of pesticide transfers. The global principle is the same than for our initial macropore hypothesis, except that we consider that, each day, only flows comprised between SAT
FC and twice SAT
FC are loaded with pesticides. In other words, we assumed that pesticide does not leach with the first macroporus flow (SAT
FC), and that only the fraction comprised between SAT
FC and twice SAT
FC are loaded with pesticide. As in the initial model, the equilibrium of pesticide between macropores and micropores is assumed to occur once a day. 5. Modeling results: calibration and validation 5.1. Hydrology Hydrological simulation results compared to observed discharge are shown in Fig. 10 for calibration and validation periods. The results for the efficiency criteria on both periods are given in Table 6. WATPPASS achieves a good adjustment on stream discharge of the Ravine watershed at a daily timestep (Table 6). Main flood peaks are well simulated (Fig. 10) as well as the water balance (Table 6). The good results for hydrologic modeling (Nashsqrt > 0.75) confirm the conceptual representation of a tropical volcanic watershed with two overlapping aquifers contributing to discharge at the watershed outlet. 5.2. Pesticide Pesticide simulations at the outlet are shown in Fig. 11 for the initial model as well as for one adjusted model: Model delay + macropore adjusted. 5.2.1. Initial model The shape of the simulated concentrations at a weekly time step with the initial model does not match exactly with the measurements as indicated by negative Nashsqrt criteria (Table 6). WATPPASS simulates a chlordecone concentration at outlet which is almost constant (Fig. 11c). Nevertheless, using a Koc of 17,500 L kg
1, the mean concentration simulated (0.21 ± 0.05 lg L
1) is close to the measured one (0.28 ± 0.13 lg L
1 with 30% error). Results obtained for variations of pesticide parameters have been found to be in the range

C. Mottes et al. / Journal of Hydrology 529 (2015) 909–927

Rainfall (mm)

922

a

100

50

0

0

50

100

150

200

250

300

350

400

450

500

550

Discharge (m3.s−1)

0.4 Calibration period

Simulated values Observed values

Validation period

0.3

b

0.2 0.1 0

0

50

100

150

200

250

300

350

400

450

500

550

days Fig. 10. (a) Daily rainfall; (b) daily discharge: simulation results for the Ravine watershed on calibration and validation periods.

Table 6 Calibration and validation values for the performance criteria on hydrology (calibrated) and pesticide (uncalibrated). Type

Indicator

Calibration

Validation

Nashsqrt RMSE (m3 s
1) Water balance (%)

0.76 0.012 101

0.78 0.013 102

Nashsqrt RMSE (lg L
1) Mass balance (%)

– – –


0.60 0.167 79

Nashsqrt RMSE (lg L
1) Mass balance (%)

– – –


0.42 0.158 95

Hydrology

Pesticide Koc = 17,500 L kg
1

Koc = 14,400 L kg
1

of measured pollution at outlet. By setting the chlordecone concentration in the 60–100 cm layers to respectively 25% and 75% of the values measured in the top layers, simulations gave concentrations of respectively 0.12 ± 0.03 lg L
1 and 0.32 ± 0.08 lg L
1. Variation of Koc around 17,500 L kg
1 and organic matter content (foc) resulted in variation of the same range: 0.29 ± 0.08 lg L
1 at
25% variation and 0.17 ± 0.05 lg L
1 at +25% variations. The local pesticide parameter (Koc = 14,400 L kg
1) leads to a good estimation of the mean pesticide concentration (0.25 ± 0.06 lg L
1). We conclude that the initial WATPPASS model is not suited to assess the dynamics of pesticides at outlet but is suited to identify potential mean long term concentrations at outlet that can result from pesticide applications on small tropical volcanic watersheds. Simulated cumulative loads stay within the analytical cumulative error threshold, but the two values of Koc show different behaviors (Fig. 11d). At the beginning of the simulation, results obtained with the Koc of 17,500 L kg
1 better fit the observed cumulated curve. After a short period (6  105 m3), there is a clear deviation of the cumulative load from observed values. With a Koc of 14,400 L kg
1, cumulative loads are close to the cumulative analytical error threshold at the beginning of the simulation while it better fits the observed cumulative load at the end of the simulation (95% mass balance). Deviance is mainly due to the misrepresentation by the model of the concentration peak that occurs around week 47.

5.2.2. Adjusted models The model adjusted for delay better simulates the mean chlordecone concentration at outlet and the spread of chlordecone concentrations around this mean but fails to represent the dynamics of concentrations (simulation curves not shown). Indeed, with a Koc = 14,400 L kg
1, the mean simulated concentration is now 0.29 ± 0.11 lg L
1 and is a better predictor of the observed one (0.28 ± 0.13 lg L
1). The decomposition of the mean square deviation of the concentrations detailed in Table 7 confirms this result (Kobayashi and Salam, 2000) by showing that both mean pesticide concentrations and the distribution of concentrations around this mean are better represented (SB and SDSD closer from 0) than with the model without delay. The weekly dynamic of concentrations was however degraded (higher LCS term). The model adjusted for delay + macropore better represents pesticide peaks and the temporal dynamics of pesticide concentration and load than the initial model (Fig. 11b and c). In particular, the MSE of the model is the lowest of the three models with a Koc of 17,500 L kg
1 (Table 7). The model adjusted for delay + macropore has low SB and SDSD meaning that the mean pesticide concentration and the spread of chlordecone around this mean are well represented by the model. A visual comparison of the concentrations and loads makes it possible to determine improvements in the simulation of pesticide concentration and load dynamics compared to the initial model (Fig. 11b and c). For instance, the pattern of the chlordecone concentration peak occurring around week 47 can be identified in simulations as delayed by one week (Fig. 11c). A variable delay ranging from 1 to 3 weeks can be observed between observed and simulated concentration values. This variable delay keeps the LCS term of the MSE at a high value. The model delay + macropore adjusted shows promising results for identifying periods where pesticide concentrations increase or decrease around the mean concentration observed in the stream.

6. Discussion The watershed model of WATPPASS aims to be simple and to do not require any calibration procedure on pesticides. Because of this simplicity, it was expected that the pesticide simulations would not perfectly match measurements. In the literature, field scale models have extensively been evaluated against measurements

923

C. Mottes et al. / Journal of Hydrology 529 (2015) 909–927

Fig. 11. Simulation results for the hydrologic and pesticide modules of WATPPASS at weekly timestep. (a) Discharge (m3 weeks
1), (b) chlordecone loads, (c) chlordecone concentrations, and (d) cumulated chlordecone load. Upper limit of pesticide simulation area: Koc = 14,400 L kg
1, lower limit: Koc = 17,500 L kg
1.

Table 7 Effect of modifying model structure on the decomposition of the mean square error (see Section 2.7.2). Methodology from Kobayashi and Salam (2000). Criteria

Koc = 14,400 L kg
1

Koc = 17,500 L kg
1

Initial

Delay adjusted

Delay + macropore adjusted

Initial

Delay adjusted

Delay + macropore adjusted

MSE (lg2 L
2) SB (lg2 L
2) SDSD (lg2 L
2) LCS (lg2 L
2)

0.0249 0.001 0.0048 0.0191

0.0292 0.000036 0.00037 0.0288

0.0275 0.0014 0.00033 0.0257

0.028 0.0057 0.0063 0.016

0.027 0.0019 0.0016 0.0235

0.0216 0.00035 0.00007 0.0212

(Branger et al., 2009; Köhne et al., 2009; Vanclooster et al., 2000a, 2000b). On the contrary, few studies present the assessment of pesticide models against measurements at the watershed scale (Borah and Bera, 2004; Holvoet et al., 2007; Payraudeau and

Gregoire, 2012; Quilbé et al., 2006). When they do, they focus on direct pesticide transfers (Holvoet et al., 2008), on surface or subsurface hydrology (Du et al., 2006; Gascuel-Odoux et al., 2009; Laroche et al., 1996) at the event scale (Kannan et al., 2006) or

924

C. Mottes et al. / Journal of Hydrology 529 (2015) 909–927

yearly timestep (Luo et al., 2008). Otherwise, authors mobilize very complex models that are not suited for management studies (Christiansen et al., 2004). We showed that WATPPASS correctly simulates the hydrological response of the catchment and that it identifies long-term mean pesticide concentrations and loads at watershed outlet. Thus, WATPPASS is able to assess the mid- to long-term pesticide concentrations in water that result from the combination of pesticide uses in cropping systems at the watersheds scale. This result is of high interest because there are no simple mitigation strategy available once pesticide have leached (Reichenberger et al., 2007). WATPPASS could be used by decision makers to assess the potential effect of modifying pesticide uses in cropping systems at the watershed scale. Also, it could be used to identify potential existing harmful situations of pesticide uses in watershed in order to prioritize modifications of cropping systems. We have seen that the initial model WATPPASS failed to represent the dynamics of pesticide concentration at the outlet. This pesticide dynamics may be the fact of specific transfer processes occurring in soil, aquifers or both. Several processes and compartments can be involved in this dynamic. Nevertheless, we are in the opinion that the representations of these compartments and processes would only increase the complexity of the model and the requirement of parameters. For instance, a vadose zone, solute retention in the aquifer porosity and on aquifer sediments (Baran and Arnaud, 2013; Legout et al., 2007), or the mixing zone between percolated pesticides and the saturated zone resulting from fluctuating levels of water in aquifers (Legout et al., 2007; Molenat et al., 2008) were shown to modify solute concentrations in aquifers. Some authors mobilize more complex groundwater models to represent the temporal dynamics of pesticides (Clement et al., 1998; Jorgensen et al., 2004; Refsgaard and Storm, 1995). In our case, such model complexifications would not necessarily produce better temporal pattern of weekly concentrations at a time scale of several years because of the accumulation of errors but would harm the use of WATPPASS for cropping system assessment studies (Affholder et al., 2012; Konikow, 2011). Thus, in order to keep its simplicity, WATPPASS was initially built assuming no interaction of pesticides with the aquifer environment system. We made this choice deliberately while knowing that aquifers produce multiple timescale mass transfers (Haggerty et al., 2004), and that pesticide reaching deep aquifers can be delayed by several years in West Indies (Gourcy et al., 2009). In the light of this discussion and of the results obtained with the initial model, we adjusted our model to integrate a retardation factor in aquifers. We also modified the representation of pesticides transfers with macropore flows so that it better modulates concentration in the leachate and at outlet. These two simple modifications combined improved the representation of the temporal pattern of pesticide concentrations and loads at the watershed outlet (Fig. 11b and c). However, the timing of pesticide peaks does not match perfectly with the measurements because, as stated above, real pesticide transfers in aquifers is more complex than the representation we made in WATPPASS with a delay based on the distance from the infiltration flows to outlet. The improvement in the representation of pesticide transfers with WATPPASS modifications has to be considered in comparison with the parameters requirement of the adjusted model that was kept unchanged. The results obtained with the adjusted models still require evaluations on independent datasets to confirm their improvements in pesticide simulations. The fact that a modification of the representation of macropore flows makes it possible to better simulates pesticides dynamics at the watershed outlet ask for more characterization of pesticide transfers on andosol with macroporus flows. In spite of the improved temporal dynamics of pesticides obtained by delaying pesticide transfers in aquifers and modifying pesticide transfers in macropores, we should considers that the

temporal variations of pesticide concentrations at outlet result from other factors that substantially modify the concentration of chlordecone in the water leaching into the shallow aquifer. For instance, tillage on polluted fields or the spatial distribution of several model inputs (e.g. rainfall and pesticide retention parameters) may have that effect. Tillage on polluted fields or dry periods may increase leachate concentrations. The fact that a significant contaminated surface of the catchment (>25% of the area polluted by chlordecone) was tilled between weeks 36 and 38 supports that hypothesis. On all soils, tillage destroys soil macro-structure, which in turn modifies water paths in soil and increases water flows access to pesticide (Alletto et al., 2010). On andosols, dry periods are known to alter physical and chemical properties of andosols (Quantin, 1972). In fact, the degradation of the soil structure may release molecules that are initially trapped in the fractal microporosity of the andosol clay and make them accessible to water flows (Woignier et al., 2012). In this setting, percolated water would have higher concentrations when reaching the shallow aquifers. The comparison between Koc values on andosol for chlordecone obtained using different protocols support this information. Indeed, Koc obtained with a protocol involving dried andosol and a destroyed soil structure led to an order of magnitude lower Koc values (Baran and Arnaud, 2013; Fernandez-Bayo et al., 2013a; OECD, 2000) from the ones reported using in situ measurements and modeling that we used in this study (Cabidoche et al., 2009). In the light of this discussion, we would recommend to better study the effect of the degradation of the andosol microstructure on pesticide transfers. In our analysis, we did not consider the spatial distribution of several parameters and inputs unless they could be attributed to a cropping system (foc, CN). On tropical volcanic island, it is known that rainfall is unevenly locally spatially distributed (Guiscafré et al., 1976; Mouret, 1979). It is likely that the observed variations of pesticide concentrations could result from rainfalls concentrated on polluted lands or on lands that are not. Spatial statistics and kriging could help to better estimates the relative repartition of rainfall on the watershed by taking into account orography for instance (Ly et al., 2011). Also, selecting a single fixed value of Koc on the watershed is not in accordance with the Koc variability that was observed by other authors for other pesticides (Coquet and Barriuso, 2002; Dubus et al., 2003; Ghafoor et al., 2013) and, as discussed previously, with the potential variability that may result from cropping systems. These two points are interesting for cropping systems assessment studies because they would modify the contributions of the fields of the watershed relatively to the amount of rainfall they statistically receive, to their cropping system, or both. Finally, WATPPASS use storage elements to represent aquifer systems that are highly adaptable to other complex hydrogeological contexts as illustrated by their use to simulate nitrate in karst springs (Pinault et al., 2001; Thiéry et al., 2009). The storage elements of WATPPASS can easily be adapted to reflects the diversity of existing aquifer system (Hartmann et al., 2013; Rimmer and Hartmann, 2012). WATPPASS simplifies transfers in the channel because of a case study that respects this condition. The use of WATPPASS on watersheds of more than several square kilometers requires adaptations such as the integration of a stream module or its use in a semi-distributed approach (Arnold et al., 1998; Polyakov et al., 2007). WATPPASS is adapted to pedoclimatic environment favoring water and pesticides percolation to aquifers while limiting subsurface flows. Thus, WATPPASS conceptual structure is simple, adaptable but is only applicable on small watersheds with fast aquifer response where leached pesticide quickly reaches the saturated zone and dispersion is limited.

C. Mottes et al. / Journal of Hydrology 529 (2015) 909–927

7. Conclusions The aim of our research was to design a simple modeling approach that can be mobilized for assessing the potential impact of cropping systems on water resource in complex hydrogeological environments. We conclude that WATPPASS is suited to simulate the daily hydrology of small volcanic watershed with two overlapping aquifer located under soils with high infiltration rates but also that the conceptual reservoir structure of WATPPASS can easily be adapted to other aquifer conditions. We also conclude that WATPPASS correctly predicts mean concentrations and loads of pesticide at outlet and proposed an adjusted model that better represent the pattern of temporal dynamic of pesticide at outlet. The adjusted modeling structure still requires an independent evaluation. Therefore, we recommend its use to assess the potential mid- to long-term impacts of pesticide uses on watershed where aquifers are major contributors to the discharge of the stream, and transfers are fast. To go further, our good results in a complex environment give insights to integrate an agronomical model in WATPPASS in order to assess the potential impacts of cropping systems on water resources at the watershed scale (Mottes et al., 2014). We recommend characterizing the role of macropores in pesticide transfers in andosol and studying tillage and drought impacts on pesticide leaching in andosol because they may significantly enhance that process. Acknowledgements This study was funded by Cirad, the European Regional Development Fund of Martinique, the Martinique French Water Office (O.D.E.) and the French Ministry of Overseas (M.O.M.). We are particularly grateful to the farmers of the Ravine watershed that help us to build the chlordecone contamination map and the cropping systems database. We thank the anonymous reviewer who helped improve the quality of the manuscript. References Affholder, F., Tittonell, P., Corbeels, M., Roux, S., Motisi, N., Tixier, P., Wery, J., 2012. Ad hoc modeling in agronomy: what have we learned in the last 15 years? Agron. J. 104, 735. http://dx.doi.org/10.2134/agronj2011.0376. Ahuja, L., Rojas, K.W., Hanson, J.D., Shaffer, M.J., Ma, L., 2000. Root Zone Water Quality Model: Modelling Management Effects on Water Quality and Crop Production. Water Resources Publication, Colorado. Albrecht, A., Brossard, M., Chotte, J.L., Feller, C., 1992. Les stocks organiques des principaux sols cultivés de la Martinique (Petites Antilles). Cah. ORSTOM, série pédologique 27, 23–36. Alletto, L., Coquet, Y., Benoit, P., Heddadj, D., Barriuso, E., 2010. Tillage management effects on pesticide fate in soils. A review. Agron. Sustain. Dev. 30, 367–400. http://dx.doi.org/10.1051/agro/2009018. Arnold, J.G., Allen, P.M., 1999. Automated methods for estimating baseflow and ground water recharge from streamflow records. JAWRA J. Am. Water Resour. Assoc. 35, 411–424. http://dx.doi.org/10.1111/j.1752-1688.1999.tb03599.x. Arnold, J.G., Srinivasan, R., Muttiah, R.S., Williams, J.R., 1998. Large area hydrologic modeling and assessment Part I: model development. JAWRA J. Am. Water Resour. Assoc. 34, 73–89. http://dx.doi.org/10.1111/j.1752-1688.1998.tb05961. x. Baran, N., Arnaud, L., 2013. Cartographie des risques de contamination des eaux souterrainses par les produits phytosanitaires en Martinique (Rapport final No. BRGM/RP-61976-FR). BRGM. Bernard, M.-L., Zamora, M., Géraud, Y., Boudon, G., 2007. Transport properties of pyroclastic rocks from Montagne Pelée volcano (Martinique, Lesser Antilles). J. Geophys. Res. Solid Earth 112, 1–16. http://dx.doi.org/10.1029/2006JB004385. Borah, D.K., Bera, M., 2004. Watershed-scale hydrologic and nonpoint-source pollution models: review of applications. Trans. ASAE 47, 789–803. Branger, F., Tournebize, J., Carluer, N., Kao, C., Braud, I., Vauclin, M., 2009. A simplified modelling approach for pesticide transport in a tile-drained field: the PESTDRAIN model. Agric. Water Manag. 96, 415–428. http://dx.doi.org/ 10.1016/j.agwat.2008.09.005. BRGM, 1983. Carte géologique de la Martinique à 1/50000: la montagne pelée. Cabidoche, Y.-M., Achard, R., Cattan, P., Clermont-Dauphin, C., Massat, F., Sansoulet, J., 2009. Long-term pollution by chlordecone of tropical volcanic soils in the French West Indies: a simple leaching model accounts for current residue.

925

Environ. Pollut. 157, 1697–1705. http://dx.doi.org/10.1016/j. envpol.2008.12.015. Castillo, L.E., Martínez, E., Ruepert, C., Savage, C., Gilek, M., Pinnock, M., Solis, E., 2006. Water quality and macroinvertebrate community response following pesticide applications in a banana plantation, Limon, Costa Rica. Sci. Total Environ. 367, 418–432. http://dx.doi.org/10.1016/j.scitotenv.2006.02.052. Cattan, P., Cabidoche, Y., Lacas, J., Voltz, M., 2006. Effects of tillage and mulching on runoff under banana (Musa spp.) on a tropical Andosol. Soil Tillage Res. 86, 38– 51. http://dx.doi.org/10.1016/j.still.2005.02.002. Cattan, P., Ruy, S.M., Cabidoche, Y.-M., Findeling, A., Desbois, P., Charlier, J.B., 2009. Effect on runoff of rainfall redistribution by the impluvium-shaped canopy of banana cultivated on an Andosol with a high infiltration rate. J. Hydrol. 368, 251–261. http://dx.doi.org/10.1016/j.jhydrol.2009.02.020. Charlier, J.-B., Bertrand, C., Mudry, J., 2012. Conceptual hydrogeological model of flow and transport of dissolved organic carbon in a small Jura karst system. J. Hydrol. 460, 52–64. http://dx.doi.org/10.1016/j.jhydrol.2012.06.043. Charlier, J.-B., Cattan, P., Moussa, R., Voltz, M., 2008. Hydrological behaviour and modelling of a volcanic tropical cultivated catchment. Hydrol. Process. 22, 4355–4370. http://dx.doi.org/10.1002/hyp.7040. Charlier, J.-B., Cattan, P., Voltz, M., Moussa, R., 2009a. Transport of a nematicide in surface and groundwaters in a tropical volcanic catchment. J. Environ. Qual. 38, 1031–1041. http://dx.doi.org/10.2134/jeq2008.0355. Charlier, J.-B., Lachassagne, P., Ladouche, B., Cattan, P., Moussa, R., Voltz, M., 2011. Structure and hydrogeological functioning of an insular tropical humid andesitic volcanic watershed: a multi-disciplinary experimental approach. J. Hydrol. 398, 155–170. http://dx.doi.org/10.1016/j.jhydrol.2010.10.006. Charlier, J.-B., Moussa, R., Cattan, P., Cabidoche, Y.-M., Voltz, M., 2009b. Modelling runoff at the plot scale taking into account rainfall partitioning by vegetation: application to stemflow of banana (Musa spp.) plant. Hydrol. Earth Syst. Sci. 13, 2151–2168. http://dx.doi.org/10.5194/hess-13-2151-2009. Chow, V.T., Maidment, D.R., Mays, L.W., 1988. Applied Hydrology. McGraw-Hill Education, New York. Christiansen, S.J., Thorsen, M., Clausen, T., Hansen, S., Refsgaard, C.J., 2004. Modelling of macropore flow and transport processes at catchment scale. J. Hydrol. 299, 136–158. http://dx.doi.org/10.1016/j.jhydrol.2004.04.029. Clement, T.p., Sun, Y., Hooker, B.s., Petersen, J.n., 1998. Modeling multispecies reactive transport in ground water. Ground Water Monit. Remediat. 18, 79–92. http://dx.doi.org/10.1111/j.1745-6592.1998.tb00618.x. Clostre, F., Lesueur-Jannoyer, M., Achard, R., Letourmy, P., Cabidoche, Y.-M., Cattan, P., 2014. Decision support tool for soil sampling of heterogeneous pesticide (chlordecone) pollution. Environ. Sci. Pollut. Res. Int. 21, 1980–1992. http://dx. doi.org/10.1007/s11356-013-2095-x. Clostre, F., Lesueur-Jannoyer, M., Achard, R., Letourmy, P., Cabidoche, Y.-M., Cattan, P., 2013. Decision support tool for soil sampling of heterogeneous pesticide (chlordecone) pollution. Environ. Sci. Pollut. Res. http://dx.doi.org/10.1007/ s11356-013-2095-x. Coat, S., Monti, D., Legendre, P., Bouchon, C., Massat, F., Lepoint, G., 2011. Organochlorine pollution in tropical rivers (Guadeloupe): role of ecological factors in food web bioaccumulation. Environ. Pollut. Barking Essex 1987 (159), 1692–1701. http://dx.doi.org/10.1016/j.envpol.2011.02.036. Colmet-Daage, F., 1969. Carte pédologique de la Martinique 1/20000. Colmet-Daage, F., Lagache, P., 1965. Caractéristiques de quelques groupes de sols dérivées de roches volcaniques aux Antilles française. Cah. ORSTOM, Série pédologique 8, 91–121. Cooley, K.R., Lane, L.J., 1982. Modified runoff curve numbers for sugarcane and pineapple fields in Hawaii. J. Soil Water Conserv. 37, 295–298. Coquet, Y., Barriuso, E., 2002. Spatial variability of pesticide adsorption within the topsoil of a small agricultural catchment. Agronomie 22, 389–398. http://dx.doi. org/10.1051/agro:2002017. Dolfing, J., Novak, I., Archelas, A., Macarie, H., 2012. Gibbs free energy of formation of chlordecone and potential degradation products: implications for remediation strategies and environmental fate. Environ. Sci. Technol. 46, 8131–8139. http://dx.doi.org/10.1021/es301165p. Dorel, M., Roger-Estrade, J., Manichon, H., Delvaux, B., 2000. Porosity and soil water properties of Caribbean volcanic ash soils. Soil Use Manag. 16, 133–140. http:// dx.doi.org/10.1111/j.1475-2743.2000.tb00188.x. Du, B., Saleh, A., Jaynes, D.B., Arnold, J.G., 2006. Evaluation of SWAT in simulating nitrate nitrogen and atrazine fates in a watershed with tiles and potholes. Trans. Asabe 49, 949–959. Dubus, I.G., Brown, C.D., Beulke, S., 2003. Sources of uncertainty in pesticide fate modelling. Sci. Total Environ. 317, 53–72. http://dx.doi.org/10.1016/S00489697(03)00362-0. Farlin, J., Gallé, T., Bayerle, M., Pittois, D., Braun, C., El Khabbaz, H., Elsner, M., Maloszewski, P., 2013. Predicting pesticide attenuation in a fractured aquifer using lumped-parameter models. Groundwater 51, 276–285. http://dx.doi.org/ 10.1111/j.1745-6584.2012.00964.x. Fernandez-Bayo, J.D., Saison, C., Geniez, C., Voltz, M., Vereecken, H., Berns, A.E., 2013a. Sorption characteristics of chlordecone and cadusafos in tropical agricultural soils. Curr. Org. Chem. 17, 2976–2984. Fernandez-Bayo, J.D., Saison, C., Voltz, M., Disko, U., Hofmann, D., Berns, A.E., 2013b. Chlordecone fate and mineralisation in a tropical soil (andosol) microcosm under aerobic conditions. Sci. Total Environ. 463–464. http://dx.doi.org/ 10.1016/j.scitotenv.2013.06.044. FOOTPRINT, 2013. The Pesticide Properties Database (PPDB) developed by the Agriculture & Environment Research Unit (AERU), University of Hertfordshire, funded by UK national sources and the EU-funded FOOTPRINT project (FP6-SSP-

926

C. Mottes et al. / Journal of Hydrology 529 (2015) 909–927

022704). [WWW Document]. (accessed 23.05.11). Foster, S.S.D., Ellis, A.T., Losilla-Penon, M., Rodriguez-Estrada, H.V., 1985. Role of volcanic tuffs in ground-water regime of Valle Central, Costa Rica. Ground Water 23, 795–801. http://dx.doi.org/10.1111/j.1745-6584.1985.tb01959.x. Gascuel-Odoux, C., Aurousseau, P., Cordier, M.-O., Durand, P., Garcia, F., Masson, V., Salmon-Monviola, J., Tortrat, F., Trepos, R., 2009. A decision-oriented model to evaluate the effect of land use and agricultural management on herbicide contamination in stream water. Environ. Model. Softw. 24, 1433–1446. http:// dx.doi.org/10.1016/j.envsoft.2009.06.002. Ghafoor, A., Jarvis, N.J., Stenström, J., 2013. Modelling pesticide sorption in the surface and subsurface soils of an agricultural catchment. Pest Manag. Sci. 69, 919–929. http://dx.doi.org/10.1002/ps.3453. Gourcy, L., Baran, N., Vittecoq, B., 2009. Improving the knowledge of pesticide and nitrate transfer processes using age-dating tools (CFC, SF6, 3H) in a volcanic island (Martinique, French West Indies). J. Contam. Hydrol. 108, 107–117. http://dx.doi.org/10.1016/j.jconhyd.2009.06.004. Guiscafré, J., Klein, J.C., Moniod, F., 1976. Les ressources en eau de surface de la Martinique. Monographie hydrologique ORSTOM. Office de la recherche scientifique et technique outre-mer, Paris. Haggerty, R., Harvey, C.F., Freiherr von Schwerin, C., Meigs, L.C., 2004. What controls the apparent timescale of solute mass transfer in aquifers and soils? A comparison of experimental results. Water Resour. Res. 40, W01510. http:// dx.doi.org/10.1029/2002WR001716. Hartmann, A., Wagener, T., Rimmer, A., Lange, J., Brielmann, H., Weiler, M., 2013. Testing the realism of model structures to identify karst system processes using water quality and quantity signatures. Water Resour. Res. 49, 3345–3358. http://dx.doi.org/10.1002/wrcr.20229. Hawkins, R.H., 1993. Asymptotic determination of runoff curve numbers from data. J. Irrig. Drain. Eng. 119, 334–345. http://dx.doi.org/10.1061/(ASCE)0733-9437 (1993) 119:2(334). Henriques, W., Jeffers, R.D., Lacher, T.E., Kendall, R.J., 1997. Agrochemical use on banana plantations in Latin America: perspectives on ecological risk. Environ. Toxicol. Chem. 16, 91–99. http://dx.doi.org/10.1002/etc.5620160110. Holvoet, K.M.A., Seuntjens, P., Vanrolleghem, P.A., 2007. Monitoring and modeling pesticide fate in surface waters at the catchment scale. Ecol. Model. 209, 53–64. http://dx.doi.org/10.1016/j.ecolmodel.2007.07.030. Holvoet, K., van Griensven, A., Gevaert, V., Seuntjens, P., Vanrolleghem, P.A., 2008. Modifications to the SWAT code for modelling direct pesticide losses. Environ. Model. Softw. 23, 72–81. http://dx.doi.org/10.1016/j.envsoft.2007.05.002. Houdart, M., Tixier, P., Lassoudière, A., Saudubray, F., 2009. Assessing pesticide pollution risk: from field to watershed. Agron. Sustain. Dev. 29, 7. http://dx.doi. org/10.1051/agro:2008042. Jorgensen, P.R., McKay, L.D., Kistrup, J.P., 2004. Aquifer vulnerability to pesticide migration through till aquitards. Ground Water 42, 841–855. Kammerbauer, J., Moncada, J., 1998. Pesticide residue assessment in three selected agricultural production systems in the Choluteca River Basin of Honduras. Environ. Pollut. 103, 171–181. http://dx.doi.org/10.1016/S0269-7491(98) 00125-0. Kannan, N., Santhi, C., Williams, J.R., Arnold, J.G., 2008. Development of a continuous soil moisture accounting procedure for curve number methodology and its behaviour with different evapotranspiration methods. Hydrol. Process. 22, 2114–2121. http://dx.doi.org/10.1002/hyp.6811. Kannan, N., White, S.M., Worrall, F., Whelan, M.J., 2006. Pesticide modelling for a small catchment using SWAT-2000. J. Environ. Sci. Health B 41, 1049–1070. http://dx.doi.org/10.1080/03601230600850804. Kenaga, E.E., 1980. Predicted bioconcentration factors and soil sorption coefficients of pesticides and other chemicals. Ecotoxicol. Environ. Saf. 4, 26–38. http://dx. doi.org/10.1016/0147-6513(80)90005-6. Khamsouk, B., 2001. Impact de la culture bananière sur l’environnement. Influence des systèmes de cultures bananières sur l’érosion, le bilan hydrique et les pertes en nutriments sur un sol volcanique en Martinique. (Cas du sol brun rouille à halloysite) (Ph.D. Thesis). ENSA Montpellier, Montpellier. Kobayashi, K., Salam, M.U., 2000. Comparing simulated and measured values using mean squared deviation and its components. Agron. J. 92, 345. http://dx.doi.org/ 10.2134/agronj2000.922345x. Köhne, J.M., Köhne, S., Šimu˚nek, J., 2009. A review of model applications for structured soils: (b) Pesticide transport. J. Contam. Hydrol. 104, 36–60. http:// dx.doi.org/10.1016/j.jconhyd.2008.10.003. Konikow, L.F., 2011. The secret to successful solute-transport modeling. Ground Water 49, 144–159. http://dx.doi.org/10.1111/j.1745-6584.2010.00764.x. Lachassagne, P., 2006. Chapitre XIII DOM-TOM. 1. Martinique. In: Roux, J.C. (Ed.), Aquifères et Eaux Souterraines En France. AIH, Paris, pp. 769–781. Lalubie, G., 2010. Les cours d’eau du massif de la montagne pelée: une approche multiscalaire pour appréhender les risques hydro-volcanogéomorphologiques, vol. I, II, Atlas (Ph.D. Thesis). Université des Antilles et de la Guyane, Martinique. Laroche, A., Gallichand, J., Lagacé, R., Pesant, A., 1996. Simulating atrazine transport with HSPF in an agricultural watershed. J. Environ. Eng. 122, 622–630. http://dx. doi.org/10.1061/(ASCE)0733-9372(1996)122:7(622). Legout, C., Molenat, J., Aquilina, L., Gascuel-Odoux, C., Faucheux, M., Fauvel, Y., Bariac, T., 2007. Solute transfer in the unsaturated zone-groundwater continuum of a headwater catchment. J. Hydrol. 332, 427–441. http://dx.doi. org/10.1016/j.jhydrol.2006.07.017. Le Moine, N., Andreassian, V., Mathevet, T., 2008. Confronting surface- and groundwater balances on the La Rochefoucauld-Touvre karstic system

(Charente, France). Water Resour. Res. 44. http://dx.doi.org/10.1029/ 2007WR005984. Leonard, R.A., Knisel, W.G., Still, D.A., 1987. GLEAMS: groundwater loading effects of agricultural management systems. Trans. ASAE 30, 1403–1418. Lesueur-Jannoyer, M., Cattan, P., Monti, D., Saison, C., Voltz, M., Woignier, T., Cabidoche, Y.M., 2012. Chlordecone aux Antilles: evolution des systemes de culture et leur incidence sur la dispersion de la pollution. Agron. Environ. Soc. 2, 45–58. Luo, Y., Zhang, X., Liu, X., Ficklin, D., Zhang, M., 2008. Dynamic modeling of organophosphate pesticide load in surface water in the northern San Joaquin Valley watershed of California. Environ. Pollut. 156, 1171–1181. http://dx.doi. org/10.1016/j.envpol.2008.04.005. Ly, S., Charles, C., Degré, A., 2011. Geostatistical interpolation of daily rainfall at catchment scale: the use of several variogram models in the Ourthe and Ambleve catchments, Belgium. Hydrol. Earth Syst. Sci. 15, 2259–2274. http:// dx.doi.org/10.5194/hess-15-2259-2011. Mangin, A., 1984. Pour une meilleure connaissance des systèmes hydrologiques à partir des analyses corrélatoire et spectrale. J. Hydrol. 67, 25–43. http://dx.doi. org/10.1016/0022-1694(84)90230-0. Matamoros Camposano, D.E., 2004. Predicting river concentration of pesticides from banana plantation under data-poor conditions (Ph.D. Thesis). Faculteit LandBouwkundige en Toegepaste Biologische Wetenschappen, Belgium. McDonald, L., Jebellie, S.J., Madramootoo, C.A., Dodds, G.T., 1999. Pesticide mobility on a hillside soil in St. Lucia. Agric. Ecosyst. Environ. 72, 181–188, doi:16/S01678809(98)00178-9. Météo France, 2013. METEO-FRANCE: Publithèque [WWW Document]. (accessed 10.08.13). Molenat, J., Gascuel-Odoux, C., Ruiz, L., Gruau, G., 2008. Role of water table dynamics on stream nitrate export and concentration in agricultural headwater catchment (France). J. Hydrol. 348, 363–378. http://dx.doi.org/10.1016/j. jhydrol.2007.10.005. Mottes, C., Lesueur-Jannoyer, M., Le Bail, M., Malézieux, E., 2014. Pesticide transfer models in crop and watershed systems: a review. Agron. Sustain. Dev. 34, 229– 250. http://dx.doi.org/10.1007/s13593-013-0176-3. Mouret, C., 1979. Contribution à l’étude hydrogéologique d’un bassin versant en milieu volcanique tropical – Rivière Capot (Martinique) (Ph.D. Thesis). Université des Sciences et Techniques du Languedoc, Montpellier. Moussa, R., Voltz, M., Andrieux, P., 2002. Effects of the spatial organization of agricultural management on the hydrological behaviour of a farmed catchment during flood events. Hydrol. Process. 16, 393–412. http://dx.doi.org/10.1002/ hyp.333. Nash, J.E., Sutcliffe, J.V., 1970. River flow forecasting through conceptual models Part I—a discussion of principles. J. Hydrol. 10, 282–290. http://dx.doi.org/10.1016/ 0022-1694(70)90255-6. Neitsch, S.L., Arnold, J.G., Kiniry, J.R., Williams, J.R., 2011. Soil and water assessment tool theoretical documentation – version 2009 (No. TR-406), Texas Water Resource Institute Technical Report. Texas Water Resources Institute, Texas. OECD, 2000. Test No. 106: Adsorption–desorption using a batch equilibrium method. Organisation for Economic Co-operation and Development, Paris. Payraudeau, S., Gregoire, C., 2012. Modelling pesticides transfer to surface water at the catchment scale: a multi-criteria analysis. Agron. Sustain. Dev. 32, 479–500. http://dx.doi.org/10.1007/s13593-011-0023-3. Philippe, E., Habets, F., Ledoux, E., Goblet, P., Viennot, P., Mary, B., 2011. Improvement of the solute transfer in a conceptual unsaturated zone scheme: a case study of the Seine River basin. Hydrol. Process. 25, 752–765. http://dx. doi.org/10.1002/hyp.7865. Pinault, J.-L., Pauwels, H., Cann, C., 2001. Inverse modeling of the hydrological and the hydrochemical behavior of hydrosystems: application to nitrate transport and denitrification. Water Resour. Res. 37, 2179–2190. http://dx.doi.org/ 10.1029/2001WR900017. Plet, J., 2012. Comment sélectionner les sites d’un bassin versant à instrumenter pour un suivi des processus de transfert de la chlordécone dans les sols et vers les eaux ? Cas du bassin versant de la rivière du Galion, en Martinique (M.Sc thesis). ISTOM, Cergy-Pontoise. Polyakov, V., Fares, A., Kubo, D., Jacobi, J., Smith, C., 2007. Evaluation of a non-point source pollution model, AnnAGNPS, in a tropical watershed. Environ. Model. Softw. 22, 1617–1627. http://dx.doi.org/10.1016/j.envsoft.2006.12.001. Ponce, V.M., Hawkins, R.H., 1996. Runoff curve number: has it reached maturity? J. Hydrol. Eng. 1, 11–19. http://dx.doi.org/10.1061/(ASCE)1084-0699(1996)1:1 (11). Poulenard, J., Podwojewski, P., Janeau, J.L., Collinet, J., 2001. Runoff and soil erosion under rainfall simulation of Andisols from the Ecuadorian Paramo: effect of tillage and burning. Catena 45, 185–207. http://dx.doi.org/10.1016/S0341-8162 (01)00148-5. Prado, B., Duwig, C., Márquez, J., Delmas, P., Morales, P., James, J., Etchevers, J., 2009. Image processing-based study of soil porosity and its effect on water movement through Andosol intact columns. Agric. Water Manag. 96, 1377–1386. http://dx. doi.org/10.1016/j.agwat.2009.04.012. Quantin, P., 1972. Les Andosols – Revue bibliographique des connaissances actuelles. Cah. ORSTOM, série pédologique 10, 273–302. Quilbé, R., Rousseau, A.N., Lafrance, P., Leclerc, J., Amrani, M., 2006. Selecting a pesticide fate model at the watershed scale using a multi-criteria analysis. Water Qual. Res. J. Can. 41, 283–295. Rawlins, B.G., Ferguson, A.J., Chilton, P.J., Arthurton, R.S., Rees, J.G., Baldock, J.W., 1998. Review of agricultural pollution in the Caribbean with particular

C. Mottes et al. / Journal of Hydrology 529 (2015) 909–927 emphasis on small island developing states. Mar. Pollut. Bull. 36, 658–668. http://dx.doi.org/10.1016/S0025-326X(98)00054-X. Refsgaard, J.C., Storm, B., 1995. MIKE SHE. In: Singh, V.P. (Ed.), Computer Models of Watershed Hydrology. Water Resources Publication, Colorado, pp. 809–846. Reichenberger, S., Bach, M., Skitschak, A., Frede, H.-G., 2007. Mitigation strategies to reduce pesticide inputs into ground and surface water and their effectiveness; A review. Sci. Total Environ. 384, 1–35. http://dx.doi.org/10.1016/j. scitotenv.2007.04.046. Rimmer, A., Hartmann, A., 2012. Simplified conceptual structures and analytical solutions for groundwater discharge using reservoir equations. In: Nayak, P. (Ed.), Water Resources Management and Modeling. InTech, pp. 217–238. Saison, C., Cattan, P., Louchart, X., Voltz, M., 2008. Effect of spatial heterogeneities of water fluxes and application pattern on cadusafos fate on banana-cultivated andosols. J. Agric. Food Chem. 56, 11947–11955. http://dx.doi.org/10.1021/ jf802435c. Sansoulet, J., 2007. Transferts d’eau et d’ions potassium et nitrate dans un sol à capacité d’échange anionique sous un couvert redistributeur de la pluie (Ph.D. Thesis). Institut National Agronomique de Paris-Grignon, Paris. Sartori, A., Hawkins, R., Genovez, A., 2011. Reference curve numbers and behavior for sugarcane on highly weathered tropical soils. J. Irrig. Drain. Eng. 137, 705– 711. http://dx.doi.org/10.1061/(ASCE)IR.1943-4774.0000354. Shoji, S., Takahashi, T., 2002. Environmental and agricultural significance of volcanic ash soils. Glob. Environ. Res. 6, 113–135. Taylor, M.D., Klaine, S.J., Carvalho, F.P., Barcelo, D., Everaarts, J., 2003. Pesticide Residues in Coastal Tropical Ecosystems: Distribution, Fate and Effects. Taylor & Francis, New York. Thiéry, D., Lopez, B., Bourguignon, A., Gutierrez, A., Baran, N., 2009. Modélisation des transferts de nitrate vers les eaux souterraines sur un site karstique soumis à des pressions agricoles (Trois Fontaines, Loiret): une contribution à la protection des captages vis-à-vis des pollutions diffuses. Géologues, 82–84.

927

Thiéry, D., Seguin, J.J., 1985. First model of nitrate transfer over a river basin with a lumped model. Application to Rembercourt basin (France), hydrogeology in the service of man. Presented at the Cambridge AIH congress, IAHS Publication, Cambridge, pp. 188–198. Tixier, P., 2004. Prototyping of banana-based cropping systems using models: An application to cropping systems of Guadeloupe F.W.I. (Ph.D. Thesis). Ecole Nationale Supérieure Agronomique de Montpellier, Montpellier, CIRAD. USDA, 1986. Urban hydrology for small watersheds (Technical Release No. 55). USDA, Soil Conservation Service, Engineering Division, Washington DC. Vanclooster, M., Boesten, J.J.T.I., Trevisan, M., Brown, C.D., Capri, E., Eklo, O.M., Gottesbüren, B., Gouy, V., van der Linden, A.M.A., 2000a. A European test of pesticide-leaching models: methodology and major recommendations. Agric. Water Manag. 44, 1–19. http://dx.doi.org/10.1016/S0378-3774(99)00081-5. Vanclooster, M., Ducheyne, S., Dust, M., Vereecken, H., 2000b. Evaluation of pesticide dynamics of the WAVE-model. Agric. Water Manag. 44, 371–388. http://dx.doi.org/10.1016/S0378-3774(99)00101-8. Venkatapen, C., 2012. Etude des determinants geographiques et spatialisation des stocks de carbone des sols de la Martinique (Ph.D. Thesis). Université des Antilles et de la Guyane, Fort de France. Vittecoq, B., Lachassagne, P., Lanini, S., Ladouche, B., Marechal, J., Petit, V., 2007. Elaboration d’un système d’information sur les eaux souterraines de la Martinique: identification et caractérisations quantitatives (Rapport final No. BRGM/RP-55099-FR). BRGM, Martinique. Woignier, T., Clostre, F., Macarie, H., Jannoyer, M., 2012. Chlordecone retention in the fractal structure of volcanic clay. J. Hazard. Mater. 241–242, 224–230. http://dx.doi.org/10.1016/j.jhazmat.2012.09.034. WRB (Ed.), 2006. World Reference Base for Soil Resources. A Framework for International Classification, Correlation and Communication, World Soil Resource Reports. FAO, Rome.