Helping farmers to reduce herbicide environmental impacts

(1)

HAL Id: hal-02630233

https://hal.inrae.fr/hal-02630233

Submitted on 27 May 2020

HAL is a multi-disciplinary open access

archive for the deposit and dissemination of

sci-entific research documents, whether they are

pub-lished or not. The documents may come from

teaching and research institutions in France or

abroad, or from public or private research centers.

L’archive ouverte pluridisciplinaire HAL, est

destinée au dépôt et à la diffusion de documents

scientifiques de niveau recherche, publiés ou non,

émanant des établissements d’enseignement et de

recherche français ou étrangers, des laboratoires

publics ou privés.

Helping farmers to reduce herbicide environmental

impacts

F. Le Bellec, A. Velu, P. Fournier, Sandrine Le Squin, S. Michels, D. Tendero,

Christian Bockstaller

To cite this version:

F. Le Bellec, A. Velu, P. Fournier, Sandrine Le Squin, S. Michels, et al..

Helping farmers to

reduce herbicide environmental impacts.

Ecological Indicators, Elsevier, 2015, 54, pp.207-216.

(2)

EcologicalIndicators54(2015)207–216

ContentslistsavailableatScienceDirect

Ecological

Indicators

jou rn al h om ep a g e :w w w . e l s e v i e r . c o m / l o c a t e / e c o l i n d

Helping

farmers

to

reduce

herbicide

environmental

impacts

F. Le

Bellec

a,∗

_,

_A.

_Vélu

a

_,

_P.

_Fournier

b

_,

_S.

_Le

_Squin

c

_,

_T.

_Michels

a

_,

_A.

_Tendero

d

_,

_C.

_Bockstaller

e a_CIRAD,_UPR_HortSys,_F-97455_{Saint-Pierre,}_France

b_CIRAD,_UPR_26,_F-97455_{Saint-Pierre,}_France c_CIRAD,_UMR_PVBMT,_F-97410_{Saint-Pierre,}_France d_CIRAD,_UPR_AIDA,_F-97410_{Saint-Pierre,}_France

e_INRA,_UMR1121_{INPL/ENSAIA/INRA,}_F-68021_Colmar,_France

a

r

t

i

c

l

e

i

n

f

o

Articlehistory: Received21March2014

Receivedinrevisedform10February2015 Accepted12February2015 Keywords: Herbicides Environmentalindicator Decision–supportTool Field

Bestmanagementpractice Tropicalclimate CART

a

b

s

t

r

a

c

t

Whilepesticideshelptoeffectivelycontrolcroppests,theircollateraleffectsoftenharmtheenvironment. OntheFrenchislandofReunionintheIndianOcean,over75%ofthepesticidesusedareherbicides andtheyareregularlydetectedinwater.Agri-environmentalmodelsandpesticideriskindicatorscan beusedtopredictandtohelppesticideuserstoreduceenvironmentalimpacts.However,whilethe complexityofmodelsoftenlimitstheirusetothefieldofresearch,pesticideriskindicators,whichare easiertoimplement,donotexplicitlyidentifythetechnicalleversthatfarmerscanactupontolimit suchtransfersontheirscaleofaction(thefield).Theaimofthisarticleistocontributetodeveloping adecisionsupporttooltoguidefarmersinimplementingrelevantpracticesregardingthereductionof pesticidetransfers.Inthisarticle,weproposeamethodologybasedonclassificationandregressiontrees. Weappliedourmethodologytoapesticideriskindicator(I-PHYindicator)foridentifyingtheimportance ofthevariables,theirinteractionsandrelativeweightincontributingtothescoreoftheindicator.We appliedourmethodologytotheassessmentoftransferriskslinkedtotheuseof20herbicidesapplied toallsoilsinReunionandaccordingtodifferentclimate,plotmanagementandproductapplication scenarios(4096scenariostested).Weconstructedregressiontreeswhichidentified,foreachherbicide oneachsoiltype,thecontributionmadebyeachinputvariabletotheconstructionoftheindicatorscore. Thetreeisrepresentedgraphically,andthisaidsexplorationandunderstanding.The20herbicideswere dividedinto3groupsthatdifferedthroughthemaincontributingvariabletotheindicatorscore.These variableswerealltechnicalleversavailabletofarmerstolimittransferrisks.Thesetreesthenbecome decisionsupporttoolsspecifictoeachpesticideuser,enablingthemtotakeappropriatedecisionswith aviewtoreducingpesticideenvironmentalimpacts.

1. Introduction

Whilepesticideshelpineffectivelycontrollingthemaincrop pests(includingweeds),theircollateraleffectsareoftenharmful tohumanhealth(Pedlowskietal.,2012)andmoregenerallyto biodiversity(Pedlowskietal.,2012;Reichenbergeretal.,2007). AccordingtotheFAO (FoodandAgricultureOrganizationofthe United Nations), based on statistics from 58 countries, almost 43% of thepesticidevolumes appliedworldwide in 2010were herbicides (FAOSTAT, 2013). On the French island of Reunion in the western Indian Ocean (located at 21◦06S, 55◦36E),the predominanceofherbicidesisevenmoremarked.In2011,75%of thepesticidevolumessoldwereherbicides(Maillary,2012).The

∗ Correspondingauthor.Tel.:+262262969587. E-mailaddress:lebellec@cirad.fr(F.LeBellec).

sustainabledevelopmentobjectivesoftheislandarespearheaded bytheParcNationaldelaRéunion(70%oftheisland’sareabelongs totheUNESCOWorldHeritageList)andencompassthereduction ofsurfacewaterandgroundwatercontamination.Indeed,in2010, outof21 activeingredientsfoundin thewatersof Reunion,17 wereherbicidesortheirdegradationproducts(Badat,2011).Such contaminationcanbelinkedtopooragriculturalpractices(wrong choiceofherbicide,equipmentorapplicationconditions,etc.)but alsotolocalpedoclimaticandtopographicalconditions(Oliverand Kookana,2006;Davisetal.,2011;Mottesetal.,2013),whichcan varysubstantiallyintheTropics.Forexample,inReunion,rainfall variesfrom600to7000mmyear−1andslopesfrom0to45%.Inthe circumstances,replacingoneherbicidebyanotherand/or appro-priateapplicationconditionswouldhelptolimitenvironmental impacts(Reichenbergeret al.,2007).Decision–supporttoolsare neededtohelpfarmersidentifyrelevanttechnicalleverstodeal withthisissueofreducingwatercontaminationbyherbicides.

http://dx.doi.org/10.1016/j.ecolind.2015.02.020

(3)

Thedegreeofherbicidetransportintheenvironmentdepends onseveralfactors,suchastheapplicationrate,herbicide persis-tenceand mobility,rainfall,topography, andclimate(Lin etal., 1999).Pesticidesinsoilaresubjecttosorptionaswellasseveral biologicalandchemical degradationmechanisms.Theseinvolve chemical, microbial, and photodecomposition, which lead to a decreaseinpesticideconcentrationsinthesoil.Pesticidesmaybe transportedtodifferentpartsoftheenvironmentbyvolatilization, runofferosion,andleaching.Transportbyrunoffandleachingmay causethecontaminationofsurfacewaterandgroundwater.Many modelscanbeusedtosimulatethesedifferentpollutionprocesses (Mottesetal.,2013)buttheircomplexityoftenlimitstheiruseto thefieldofresearchandtospecificassessmentsituations(Voltz etal.,2005).Manypesticideenvironmentalriskindicatorsbasedon operationalmodelsthatareeasiertoimplementhavebeen devel-opedforpredictingorassessingtheenvironmentalriskslinkedto pesticideuse(Bockstalleretal.,2009;Payraudeauand Gregoire, 2012).Thesepesticideriskindicatorscanbeusedtoassessthese risks,whichareoftendifficulttoperceiveasawhole(Payraudeau andGregoire,2012;Reusetal.,2002).However,astheseindicators arelessaccuratethanmodels,Voltzetal.(2005)recommended usingthemasadecision–supporttooltoprioritizerisksituations ratherthanasatooltopredictpesticideflows.

Thesepesticiderisk indicatorscombinea moreor lesslarge numberofvariablesandconsideragriculturalpracticesand appli-cationconditionstovaryingdegrees(Devilliersetal.,2005).All indicatorsproduceascoretoreﬂectenvironmentalriskor perfor-mance;thisoutcomeisgenerallytheonlyresult.Indeed,indicators donotexplicitlyprovideinformationontheleversforactiontobe acteduponinordertoimprovepractices.Thelargenumberof vari-ablesusedandtheirinteractions,butalsotheaggregationmethods andtheoftenincompletesensitivityanalysesoftheseindicators (Devillersetal.,2005),makethesearchforleversacomplex busi-ness.Thisﬁndingcontributestothe“blackbox”imagethatsuch toolsareoftencriticizedfor;thisisalsowhytheseindicatorsare rarelydecision–supporttoolsforfarmers(Reusetal.,2002).

Inthisarticle,weproposesomesolutionsforanalysingthese leverswhich,toourknowledge,haveyettobeexplored.Weapplied ourmethodology,basedonclassiﬁcationandregressiontrees,to theI-PHYindicatoroftheassessmentmethodINDIGO(Bockstaller etal.,2009).Wetestedthisindicatorandourmethodtoassessthe environmentalrisksassociatedwiththeapplicationof20herbicide activeingredientsusedin Reunion.We thendiscussourresults andourmethod toidentifyandprioritize importancevariables forthisindicatorinordertomakeitadecision–supporttoolfor farmers.

2. Materialsandmethods

2.1. I-PHYindicator

ThepesticideriskindicatorI-PHYwasdevelopedinparallelto otherenvironmentalindicatorsfortheassessmentmethodINDIGO (Bockstalleretal.,2009).Thescoreoftheindicatorwaspublishedby

VanderWerfandZimmer(1998)andenhanced,adaptedandtested forarablefarming(Bockstalleretal.,2008).Sincethen,I-PHYhas beenadaptedtootherfarmingsystemssuchaswinegrowing,fruit production,ﬁeldvegetableproduction,palmtrees,etc.Inthelast 5years,theI-PHYindicatorhasbeenusedinmorethan100cases inFrancebyadvisersmainlyworkingontheassessmentofrisks atﬁeld/farmlevelorworkingonthedevelopmentofinnovative croppingsystems(Bockstalleretal.,2008).

Calculationoftheindicatorisbasedonfourcomponents respec-tivelyassessingtherisklinkedtotheamountofactiveingredient (a.i.)appliedandtheriskforgroundwater,surfacewaterandair.

Inasecondstep,anoverallindicatoriscalculated.Threetypesof inputvariablesareused(Table1):

1.Pesticidepropertiesrelatedtoenvironmentalfateortothe eco-toxicologyeffect,

2.Site-speciﬁcconditions(e.g.runoffsensitivity),

3.Characteristicsofpesticideapplication(e.g.rateofapplication). Afuzzyexpertsystemis usedtoaggregateallthese hetero-geneous variables into indicator modules and to subsequently aggregate these modules into a synthetic indicator. By using fuzzy subsets the effect of a knife-edge limit of a given class can be avoided. Output values for each module, as well as for the overall indicator, are expressed on a qualitative scale used in the INDIGO method: between 0 (maximum risk) and 10 (no risk)with a referencevalue of 7 (maximumacceptable risk).

Fig.1showsanexampleforgroundwaterriskwherethemain weightis given toa pesticide propertyvariable (ground water ubiquityscore–GUS)(Gustafson,1989),withlessweightgiven toposition(cropinterceptionhere)andsoilsensitivityto leach-ing.Aggregationrulesaredeﬁnedaccordingtoknowledgeabout theprocessesforeachmodule.Itshouldbenotedthatforsurface water(Fig.2),thesensitivityoftheﬁeldtorunoffanddriftplaysa majorroleincomparisonwiththepesticideproperty(half-lifeof a.i.(DT50)variable).Thisaggregationmethodenablestocopewith casesofcompensationbetweeninputvariablesaswellascasesof non-compensation(Sadoketal.,2008).Forthegroundwater sur-facewaterandaircomponentofI-PHY,aggregationrulesintegrate knowledgeontheprocesses.Compensationbetweenvariablesmay occuronlywhenavariablebelongingtothe‘favourable’class lim-itspesticide transfer.For instance,this is the case of ‘position’ variable that indicatesthat pesticidemay be in a position out ofreachforleachingorrunoff byinterceptionbycropcoveror incorporationinto soil(seeFigs. 1and 2).For theenvironment componentofI-PHY(Fig.3),weassumedthatlowsprayingrate compensatesa hightransfer risk.Thisis based onsome litera-turedata(e.g.Battaglinet al.,2000;De Lafontaineet al.,2014) andpesticideregistrationdata(Tomlin,2006).Conversely,wedo notacceptahighlevelofcompensationbetweenthethreerisk components,groundwatersurfacewaterandair.Weconsidered thatwhenoneisunfavourableandtherateishigh,thesituation isunacceptableforstakeholders.Thisisgivenbyalowscoreof 2out10fortheenvironmentcomponentofI-PHY(Fig.3), show-ingahighrisk.Finally,inallcomponentsofI-PHY,thetoxicityor ecotoxicityvariablecanincreasebutnotdecreasetherisk.In con-clusion,innocase cantotal compensationoccurbetweeninput variables.

In order to carry out our study, we constructed some use scenarios for20 activeingredientsfor thewholearableareaof Reunion.Thesescenarioswereconstructedfromtheknown val-uesrelatedtothepesticidepropertiesandthecharacteristicsof thesoil(Table1,variables1–12)andfromvaluesrelatedtothe useofthepesticide,consideredbythescenarios(Table1,variables 13–27).

2.2. Choiceofactiveingredientsstudiedandtheircharacteristics We studied 20 herbicide active ingredients (Table 2). Six-teen active ingredients were herbicides marketed in Reunion between 2009 and 2011, three of which (glyphosate, 2,4-D and S-metolachlor) alone amounted to 80% of herbicide sales (Maillary,2012).Threeotheractiveingredients(isoxaﬂutole, thien-carbazone and nicosulfuron), earmarked to eventually replace some active ingredients, were also studied. Lastly, atrazine

(4)

F.LeBellecetal./EcologicalIndicators54(2015)207–216 209

Table1

VariablesneededforI-PHYcalculation,theirshortnameusedforthispaper.

Nr. Variables Valuestested

Shortname Fixed Favourable Unfavourable 1 Henry’slawconstantat25◦C(Pam3_mol−1₎ _KH _Pesticide _– _–

2 Half-lifeofactiveingredient(day−1₎ _DT50 _Pesticide _– _–

3 Acceptabledailyintake(mgkg−1_j−1₎ _ADI _Pesticide _– _–

4 Aquatox(EC50,mgL−1₎ _Aquatox _Pesticide _– _–

5 Leachingpotentialindex GUS Pesticide – –

6 Organiccarbonsorption(mL−1₎ _Koc _Pesticide _– _–

7 Soiltype(sandy,loamy,clayey) Soiltype Soil – –

8 Soildepth(<30,30–60,60–90,>90)(cm) Soildepth Soil – –

9 pH(<5,5–5.5,5.5–6,6–6.5,>6.5) pH Soil – –

10 Organiccarboncontent(<2,2–5,>5)(%) OC Soil – –

11 Absence/existenceofswellingclay Clay Soil – –

12 Slop(0,<5,5–10,>10)(%) Slop Soil – –

13 Drift-reducingnozzle? Reducingnozzle Yes – –

14 Maintenanceofnozzle? Maintenance Yes – –

15 Isthesoilisdraining? Draining No – –

16 Isthesoilﬁltering? Filtering – No Yes

17 Distancebetweenthewaterwayandtheplot Water-dist – >15m <5m

18 Stateofthewaterway Water-state – Dry Fullofwater

19 Dryorhumidweather Weather – Humid Dry

20 Numberofdayswithoutrainfall(>5mm) Rainfall – >DT50a.i. 0

21 Plotcovercropstate Cover-state – Total Withoutplants

22 Covercropoccupation(%) Cover-occup. – 100 0

23 Numberof10-dayperiodsbetweenthetreatmentand thelastmower

Mower – >4 <1

24 Non-cultivatedplantstripsdownslope(m) Plantstrip – <12m >6m 25 Applicationpositionoftheproduct Position – Intothesoil Onbaresoil

26 Typeofspray Spray – Hose Pressureandairblast

withaxialfanspray

27 Typeofspraynozzle Nozzle – Othertypesofnozzle Classicslit

Valuestested:variables1–12ﬁxedbypesticideandsoilproperties;13–15ﬁxedbyReunionIslandconditions;16–27testedattheirfavourableandunfavourablesubsetsin thescenarios.ThesubsetsweredeterminedduringI-PHYconstruction.DT50:half-lifeofactiveingredient(a.i.).GUS:groundwaterubiquityscore(Gustafson,1989)where GUS=log10(DT50)×[4−log10(Koc)],thisindexindicatestheriskofpesticidetransferstogroundwaters.TheKocvalueisameasureofhowtightlypesticideisretainedby theorganicfractioninsoils.

Fig.1.DecisiontreeofthegroundwatercomponentofI-PHY,GUS:groundwaterubiquityscore(Gustafson,1989)whereGUS=log10(DT50)×[4−log10(Koc)]whereDT50 isthehalf-lifeoftheactiveingredient,Koc:soilsorptioncoefﬁcient,position;functionofcropcover,leachingsensitivityfunctionofsoilproperties.ADI:acceptabledaily intakeoftheactiveingredient.

was added to our study as a control for past phytosanitary uses.

Pesticidepropertiesrelatedtoenvironmentalfateortothe eco-toxicologyeffectforI-PHYscorecalculationwere(Table1): -Henry’slawconstant(KH)which characterizestheabilityofa

substanceinasolutiontovolatilize.

-Thehalf-life(DT50)ofanactiveingredientprovidesinformation onthetimetakentobreakdownhalfthequantityoftheactive ingredientused.

-Acceptabledailyintake(ADI)indicatestoxicityforhumans. -Aquatox,thisvariableresultsfromthetoxicityoftheactive

ingre-dientforaninvertebrateaquaticorganism(Daphniamagna)after 48h.

(5)

Fig.2.DecisiontreeofthesurfacewatercomponentofI-PHY,runoffsensitivityasafunctionofslope,soilproperties,driftsensitivityasafunctionofdistancetosurface water,position;asafunctionofcropcover,DT50:half-lifeoftheactiveingredient,aquatictoxicityofactiveingredientforDaphniamagna.

-Groundwater ubiquity score (GUS). GUS is an index used to estimatethepotentialofpesticidestocontaminate groundwa-ter(Gustafson,1989).GUS=log10(DT50)×[4−log10(Koc)];the higher theGUS value,the higher the potential for pesticides tomovetowardsgroundwater.GUSis aprettysimple indica-torbutothervariablesmayexplainthistransferofpesticidesto groundwater(e.g.soiltype,pH,mattercontent...);theyarealso consideredbyI-PHYwhentheGUSvalueisunfavourable(Fig.1).

Thecharacteristicsoftheactiveingredientscomefromthe Pes-ticidePropertiesDatabase(UniversityofHertfordshire,2013)and aregiven inTable 2.Theactive ingredientswerestudied sepa-ratelywithoutconsideringanypossiblecombinations(e.g.such as in commercial specialities) at the approved rate for weed-ing(activeingredientrateha−1)(Ministèredel’agricultureetde l’agroalimentaire,2013).Finally,wedidnotstudythemetabolites oftheactiveingredientsbecauseI-PHYdoesnotallowit.

Fig.3. DecisiontreeoftheenvironmentcomponentofI-PHY,Rgw,Rsw,Rair:respectivelygroundwater,surfacewaterandaircomponent,rate:activeingredientrateper ha.

(6)

Table2

Characteristicsofthe20activeingredientsstudied.ThecharacteristicscomefromthePesticidePropertiesDatabase(UniversityofHertfordshire,2013).

Activeingredient Herbicides

KH(25◦C)

(Pam3_mol−1₎ DT50(days) ADI(mgkg −1

day−1)

Aquatox (mgl−1)

GUS Registeredrate (gha−1)

Quantitiessold 2009–2011(kg)

2,4-D 1.30E−05 10 5.00E−02 1.00E+02 1.62 960 50,531

Aclonifen 3.03E−03 117 7.00E−02 1.20E+00 0.30 2700 447

Asulam 1.31E−10 3.2 3.60E−01 5.79E+01 1.36 3600 5539

Atrazine 1.20E−07 75 2.00E−02 8.50E+01 3.30 1500 Removed

Bentazone 7.20E−05 13 1.00E−01 6.40E+01 2.30 1218 92

Diquat 5.00E−12 1000 2.00E−03 2.20E+00 −2.68 800 1900

Fluazifop-P 4.90E−02 1 1.00E−02 6.20E-01 0.00 250 126

ﬂuroxypyr 1.69E−10 13.5 8.00E−01 1.00E+02 2.42 200 659

Glufosinate 8.96E−09 7.4 2.10E−02 6.68E+02 1.06 750 2105

Glyphosate 4.20E−07 12 3.00E−01 4.00E+01 0.91 2160 52,361

Isoxaﬂutole 1.87E−05 2 2.00E−02 1.50E+00 0.59 99.75 New

Linuron 2.50E−06 48 3.00E−03 3.10E-01 2.03 750 169

Mesotrione 5.10E−07 32 1.00E−02 6.22E+02 3.43 150 2141

Metribuzin 2.00E−05 11.5 1.30E−02 4.90E+01 2.57 875 7678

Nicosulfuron 1.48E−11 26 2.00E+00 9.00E+01 3.79 22,5 New

Oxadiazon 3.80E−02 502 3.60E−03 2.40E+00 2.40 2400 98

Pendimethalin 2.73E−03 90 1.25E−01 2.80E-01 -0.39 1200 4245

S-metolachlor 2.40E−03 45 1.00E−01 2.35E+01 2.72 1920 21,228

Thiencarbazone 7.88E−13 11.6 1.20E+00 9.86E+00 2.13 40 New

Triclopyr 2.90E−03 39 3.00E−02 1.31E+02 3.69 3600 5131

Favourablesubset* _>2.63E−6 _<1 _>1 _>100 _<1.8

Unfavourablesubset* _<2.63E−4 _>30 _<0 _<0.01 _>2.8

Fuzzylogic* _]2.63E−6;

2.63E−6[

]0;30[ ]0;1[ ]0.01;100[ ]1.8;2.8[

KH:Henry’slawconstantat25◦_C_(Pa_m3_mol−1₎_which_{characterizes}_the_ability_of_a_substance_in_a_solution_to_volatilize,_DT50:_half-life_of_active_ingredient_(day−1_),_provides

informationonthetimetakentobreakdownhalfthequantityoftheactiveingredientused,ADI:acceptabledailyintake(mgkg−1_j−1₎_indicates_toxicity_for_humans,_Aquatox:

thisvariableresultsfromthehighesttoxicity(EC50,mgL−1₎_of_the_active_ingredient_for_an_invertebrate_aquatic_organism_(Daphnia_magna)_after₄₈_h,_GUS:_groundwater

ubiquityscore(Gustafson,1989)whereGUS=log10(DT50)×[4−log10(Koc)],thisindexindicatestheriskofpesticidetransferstogroundwaters.TheKocvalueisameasure ofhowtightlypesticideisretainedbytheorganicfractioninsoilsandtheregisteredrate:theapproveddoseforweeding(gha−1₎₍_Maillary,₂₀₁₂_).

*_Subset_value_for_each_variable_in_I-PHY₍_Van_der_Werf_and_Zimmer,₁₉₉₈_).

2.3. Variablescharacterizingthesoilsofthestudysite

Pesticidemobilityandfateinasoilareinfluencedbysorption anddegradationprocesses.Theadsorptionofapesticideincreases withincreasedsoilorganiccarbonandclaycontent,andcanslow itsmovementinsoil.Degradationprocessesdependonthesoiltype andenvironmentalconditions.Inaddition,theextentofleaching andrunoffisinfluencedbythesoilproperties,slopeandsoildepth. Toaccountfortheseprocesses,I-PHYconsiders7variables char-acterizingsoil(Table1):soiltype,soildepth,pH,organiccarbon content,absenceofswellingclay,slopeandfilteringnatureofthe soil.Fig.4showsthemappingofallthearablesoilsontheisland ofReunion,carriedoutwithArcGISsoftware(ESRI,ArcMap10.0, 2013).Thismapwasobtainedbysuperposing6thematiclayers cor-respondingtothevariablescharacterizingsoil,exceptthefiltering natureofthesoil;itcouldnotbeincorporatedintothismapthrough lackofcartographicaldata.Wethereforeconsidereditasavariable ofthescenariotested(seeSection2.4.Wethendividedthetotal arableareaoftheislandintogeographicalunitsof5haeach char-acterizedbythecombinationofthe6soilvariablesderivedfromthe centrepointofeachunit.Themapcomprises34,645geographical unitsfromwhich118distinctsoilcombinationswereidentified. 2.4. Creationofactiveingredientusescenarios

Twenty-sevenvariablesareneededtocalculatetheI-PHYscore (Table1).Inordertocarryoutourstudy,wecreatedactive ingre-dientusescenariosfromthevariablesneededtotakeintoaccount thediversityofpossibleapplicationconditions.Therewere15of thesevariables(Table1,variables16–27)giventhattheother12 arelinkedtopesticideandsoilproperties(Table1,variables1–12). Eachvariabletakes2thresholdvalues(unfavourableorfavourable) whichwereusedtocreatethescenarios.Inordertolimitthe num-berofscenarios,2variablesrelated tothecharacteristicsofthe equipmentused(Nos.13and14,Table1)wereblockedattheir

favourablethresholdvalues,whichamountedtoregulatory con-ditionsforgoodagriculturalpractices.Likewise,variableno.15, “draining”wastestedwithitsfavourablethresholdvalueasitwas a practicenotfoundinReuniongiven themajortopographyof theisland enabling suchdrainagenaturally. We therefore con-structedouractiveingredientusescenarioswiththeremaining12 variables,consideringtheirtwothresholdvalues.Thisgave4096 usescenariosforeachpesticidetested(212_)._The_scenarios_were

recordedandprocessedinadatabasedevelopedwithAccess®

soft-ware(MicrosoftCorporation2010). 2.5. Statisticalanalyses

Inthestatisticalanalysis,wesoughttoprioritizetheimportance ofthevaluesrelatedtotheuseofthepesticideintheconstruction oftheI-PHYscoreinordertodeterminewhichofthemshouldbe acteduponasaprioritytoincreasethescoreoftheindicator.The initialdatasetincludedthe4096usescenariosappliedtoeachofthe 118soils,foratotalof483,328I-PHYscoresperactiveingredient. AllthestatisticalanalyseswerecarriedoutwithRsoftware(Version 2.15.2,RDevelopmentCoreTeam2013,Vienna,Austria).

Firstly,wesoughttoreducethedatasetforeasierinterpretation oftheresults.Wedidthisusinghierarchicalclusterclassiﬁcation accordingtotheWardcriterionovertheEuclidiandistance.The resultsofthisstageenabledustoidentifysomedifferentsoiltypes leadingtothesameI-PHYscoreforthesamescenario.Wetherefore reducedthedatasetaccordingly,excludingallthedoublets.

Secondly,theI-PHYscorevariation(calledgain)wascalculated foreachsoiltype,eachactiveingredient,andeachvariable.The gainwascalculatedfromthedifferencebetween2equal scenar-ios,exceptthevalueofthevariableanalyzed.Ifavariablehadno gain,thevariableandthe2scenariosweredeletedfromthedataset. Then,basedonthisreduceddataset,weusedCART(Breimanetal., 1984),astatisticalmethodthatsuccessivelydividesup observa-tions,bybinarydivisions,intosubsets(callednodes),whichare

(7)

Fig.4. Maprepresentationofthe118soilscharacterizedby6variables:typeofsoil,depth,pH,organicmatter,swellingclayandslope.Noneofthemisprevailinginthis territory.Thewhitezonesarenon-arableareas(volcano,riverbed,etc.)createdwithArcMAPsoftware(ESRI,2013).

increasinglyuniforminrelationtoatargetvariable;inthiscase theI-PHYscore.Theresultsweredisplayedintheformofa regres-siontree(Fig.5).Thequalityofthedivisionswasmeasuredbythe decreaseinthemeansquarederrorinducedbythedivision.The selectedvariablewastheonethatinducedthelargestdecrease inmeansquarederror.Theprocedurewasrepeateduntilsubsets wereobtainedthathadveryfewobservations.Iftheconditionof thisselectedvariablewastrue(unfavourablesubsetofvariable,

Table1),theinitialobservationsweresenttotheleftoftheparent node.Iftheconditionofthisselectedvariablewasfalse(favourable subsetofvariable,Table1),theobservationsweresenttotheright oftheparentnode.Finally,theI-PHYminscorecorrespondedto thescenariowhereallthevariables wereintheirunfavourable conditions,i.e.the poorestapplication conditionsfor theactive ingredient.TheI-PHYminimumscoreisgivenatthebottomleftof theregressiontree(Fig.5).Conversely,theI-PHYmaximumscore correspondedtothescenariowhereallthevariableswereintheir favourableconditions,i.e.thebestapplicationconditionsforthe activeingredient.TheI-PHYmaxscoreisgivenatthebottomright oftheregressiontree(Fig.5).

Lastly,wecomparedtheimportanceofthevariablesinthe con-structionoftheI-PHYscorebetweenactiveingredientsandallsoils combined.Wedidthisusingtherandomforestsstatisticalmethod (Breiman,2001).Thismethodisbasedontheaggregationof sev-eraltreesdisruptedbydoublerandomization,ofboththesample (bootstrapsample,i.e.withthedrawingandreturningofn observa-tionsoutoftheinitialnobservations),butalsosomeexplanatory

variables(sub-setsofexplanatoryvariablesselectedfromallthe initialvariables).Aregressiontreewasconstructedfromallthese samples(500inourcase),witharandomdrawofthesub-setsof explanatoryvariables (ofthepinitialvariables,p/3weredrawn randomly)tosearchforeachoptimumdivision.Theimportance ofthevariablesintheconstructionoftheI-PHYwasthendeﬁned bythemeandecreaseinheterogeneity.Asthismeasurementwas notcalculatedonthesamesamplefortheactiveingredients,we roundeditofftoanindexbetween0and100.Theseindexvalues thenenabledustoconstructuniformgroupsofactiveingredients accordingtotheimportanceofthevariablesfortheI-PHYscore usingaprincipalcomponentsanalysis(PCA)followedby hierar-chical ascendantclustering.Lastly, thedifferences inthemean, minimumandmaximumI-PHYscoresforthedifferentgroupswere testedusingKruskal–WallisandWilcoxoncomparisonofmultiple meanstests.

3. Resultsanddiscussion

3.1. ThevalidityofI-PHYscores

Themean,minimum(min)andmaximum(max)I-PHYscores forthe4096scenariostestedonthe118soilsforeachactive ingre-dientarepresentedinTable3.Themeanscoresvariedfrom3.72 (triclopyr)to8.72(thiencarbazone).TheI-PHYminscoresofthe differentactive ingredients variedfrom 2.36(triclopyr)to 6.04

(8)

Table3

Theminimum,maximum,roomforprogressbetweenminimumandmaximum,mean(±standarddeviation)oftheI-PHYscoresfor20activeingredients.

Activeingredient I-PHYminimum I-PHYmaximum Roomforprogress(max–min) I-PHYmean

2,4-D 4.78 8.37 3.59 7.25(±0.88) Aclonifen 3.51 6.02 2.51 4.97(±0.73) Asulam 4.60 9.03 4.43 8.30(±0.85) Atrazine 3.63 8.93 5.30 6.04(±1.52) Bentazon 3.98 7.90 3.92 6.28(±0.87) Diquat 4.48 9.21 4.73 7.59(±1.24) Fluazifop-P 4.63 7.54 2.91 6.41(±0.84) Fluroxypyr 5.00 9.50 4.50 6.90(±0.63) Glufosinate 4.92 9.21 4.3 8.29(±0.90) Glyphosate 4.86 9.07 4.21 8.28(±0.85) Isoxaﬂutole 5.02 8.94 3.92 7.90(±0.87) Linuron 4.37 9.21 4.84 7.55(±1.16) Mesotrione 3.85 9.26 5.41 6.57(±1.51) Metribuzin 3.87 8.11 4.24 6.10(±1.05) Nicosulfuron 4.90 9.51 4.61 7.52(±1.22) Oxadiazon 2.81 4.48 1.67 3.80(±0.39) Pendimethalin 4.24 6.67 2.43 5.78(±0.67) S-metolachlor 3.16 6.19 3.03 4.59(±0.75) Thiencarbazone 5.73 9.60 3.87 8.72(±0.78) Triclopyr 2.36 5.28 2.92 3.72(±0.72)

(ﬂuoxypyr).TheI-PHYmaxscoresvariedfrom4.48(oxadiazon)to 9.60(thiencarbazone).Theseresultsmadeitpossibleatthisstage toidentifysomeactiveingredientsforwhichthescorewasunder theacceptableleveloftheI-PHYindicator(score<7).Tenoutof20 pesticideswereinthiscase.

The question of the validity of the score still remained. In thatrespect,BockstallerandGirardin(2003)andBockstalleretal. (2008)proposedcomparingtheresultsoftheindicatorwith mea-surementsofpesticideconcentrationsinwater.Suchacomparison meansacceptingachangeinstudyscalebutalsoacceptingits lim-itations.Indeed,watercontamination,especiallysurfacewater,is alwaysfoundafteramajorclimateevent(Louchartetal.,2001) while samplingtomonitor water qualityis usually carried out withoutconsideringtheseriskperiods(Voltzetal.,2005).When comparingtheseI-PHYscoreswithwaterqualitymeasurementsin Reunion(Badat,2011),wefoundthat7activeingredients(andtheir associatedmetabolites)studiedinthisarticleaccountedfor72% oftotaldetections(insurfacewatersandgroundwaters):atrazine (56%),metolachlor(8%),bentazon(2%),triclopyr(2%),glyphosate (2%),2,4-D(1.5%)andmetribuzine(0.5%).Apartfrom2,4-Dand glyphosate,theotheractiveingredientshadthelowestmeanand minimumI-PHYvalues.Inaddition,thesameherbicidesare regu-larlyincriminatedinwaterpollutioninmanytropicalortemperate countries(Danieletal.,2002;HaarstadandLudvigsen,2007;Ghosh andSingh,2013).Indeed,2,4-Dandglyphosatealoneaccountfor over2/3ofpesticidesalesinReunion(Maillary,2012).Itislikely thatdetectionofthoseherbicidesinthewatersinReunionisalso theconsequenceofstrongpressurelinkedtothevolumesapplied and/ortoinappropriateculturalpractices.I-PHYcannotconsider thatriskonaterritoryscalebecause,byitsveryconstruction,it assessesrisksonsmallquantitiesofpesticidesappliedonaﬁeld scale.However,theI-PHYminvaluesforthesetwoactive ingredi-entswere4.78for2,4-Dand4.86forglyphosate(Table3);these resultsindicateahighriskoftransfertotheenvironmentunder thepoorestapplicationconditions.Thisbriefcomparisonofthe resultsofourstudywithmeasuredcontaminationenablesusto suggestthattheI-PHYindicatorseemsrelevantforidentifyingmost substancespresentingenvironmentalrisks.

3.2. Regressiontreesasdecision–supporttools

The hierarchical cluster classiﬁcation enabled us to identify somedifferentsoiltypesleadingtothesameI-PHYscoreforthe samescenario.Thisclassiﬁcationledto9differentsoiltypesfor

11activeingredients(2,4-D,aclonifen,asulam,diquat,fluazifop-P, glufosinate, glyphosate,isoxaflutole,linuron,pendimethalinand thiencarbazone), 37 soil types for triclopyr and 45 soil types for the last 8 active ingredients studied (atrazine, bentazon, fluroxypyr,mesotrione,metribuzine,nicosulfuron,oxadiazonand S-metolachlor). Then, 496 regression trees wereconstructed (1 regression tree for each type of soil and per activeingredient) usingtheCARTmethod.Noneoftheregressiontreesconstructed wasidentical.Thenumberofnodesmakingupthesetreesnotably determinedtheircomplexity.Fig.5showsthe2,4-Dregressiontree forthemostcommonsoilinReunion;9otherregressiontreesfor9 activeingredientsaregiveninthesupplementarydata(Figs.7–15). Theseregressiontreesenabledustovisualizetheimportanceofthe variablesintheconstructionoftheI-PHYscore.Thescoreatthe topofthetreecorrespondstothemeanofthescenarios,whichis reducedorincreaseddependingontheunfavourableorfavourable thresholdvaluesofthevariablesinvolvedintheconstructionof thetree.The leavesofthetrees (red,orange orgreen squares) indicatethefinalI-PHYscores,withtheminimumscorefurthest leftandthemaximumscorefurthestrightfortheactive ingredi-entonthesoiltypeconsidered.Thedifferencebetweenthesetwo scoreswasusedtoestimatethepossibleroomforimprovement. Thesetreesalsomadeitpossibletovisualizethevariables accord-ingtothegainstheycouldprovide.Themorethevariableislocated nearthetopofthetreethegreaterwasthegainitprovidedinthe constructionoftheI-PHYscore.Fromonesoiltoanother,orfrom oneactiveingredienttoanother,theprioritizationofthese vari-ablesandtheirinteractionsdiffered(seesupplementarydata,Figs. 7–15).

TheCARTmethod(Breimanetal.,1984)isoneofthefew non-parametricmethodsthatofferawayofcomputingtheimportance of variables (Belet al., 2009).Thismethod buildsclassification andregressiontreesforpredictingcontinuousdependentvariables (regression) and categorical predictor variables (classification). CARTprocedureshaveproventobeveryusefulinecologicaland environmentalcontextsbecausetheoutputsareeasilyunderstood (De’athandFabricius,2000).CARTmodelshaveseveraladditional advantagesoverothertechniques:inputdatadonotneedtobe nor-mallydistributed;itisnotnecessaryforpredictorvariablestobe independent;andnon-linearrelationshipsbetweenpredictor vari-ablesandobserveddatacanbemodelled(Beletal.,2009).Thus, classificationandregressiontreesbyCARTrepresentan alterna-tivetomanytraditionalstatisticaltechniques,includingmultiple regression, analysis of variance, logistic regression, log-linear

(9)

Fig.5. Regressiontreeof2,4-DforthemostcommonsoilinReunion(soiltype:clayey;soildepth:>90cm;pH:5–5.5;organicmatterrate:2–5%;absenceofswellingclay; slope:>10%;ﬁlteringsoil).TheexplanatoryvariablesrepresentedarethoseselectedbytheCARTmethod(seematerialsandmethods).Theytakeontheunfavourablevalue (reddottedbranch)orthefavourablevalue(greencontinuousbranch).Thesedifferentconditionsledtosomemeanscores(I-PHYscores)shownintheleavesofthetree. Foranexplanationofthevariables(spray...),seeTable2.Ascoreover7indicatesasituationwherethepesticidetransferriskisacceptable.

models and linear discriminant analysis (De’ath and Fabricius, 2000).Under theseconditions,regression treescan beused for interactiveexplorationandforthedescriptionandpredictionof patternsandprocesses.Inourcase,regressiontreesenabledusto identifyandprioritizeimportancevariablesfortheI-PHYindicator, foreachherbicideunderourstudyconditions.Thetreeis repre-sentedgraphically,andthisaidsexplorationandunderstanding. Suchananalysisprovesdecisiveforusingthistypeofindicator(Hak etal.,2012)asitmakesitpossibletodeterminetheexactweightof eachofthevariablesmakinguptheindicator(Ferraraetal.,2012). 3.3. Roomformanoeuvretoreduceherbicideenvironmental impacts

Therandomforestsstatisticalmethodwasusedtodetermine theimportanceofthevariables intheI-PHYscoreconstruction. Even though all the variables contributed to this construction exceptthe‘cover-occup’variable(Table4),onlyfourvariableswere verysignificant.Thesewerethe‘spray’,‘water-dist’,‘water-state’ and‘position’variables.However,theimportanceofthesevariables variedwiththepesticide.Thisfirstanalysiscompletedbya princi-palcomponentsanalysisbroughtout3groupsofactiveingredients accordingtotheirvariablesofimportanceintheconstructionofthe I-PHYscores.Thesegroupswererepresentedinthefirstfactorial planeoftheanalysis(Fig.6).Axes1and2explainedover83%ofthis graphicrepresentation.Eachofthe3groupsincludedoneofthe3 activeingredientsmostusedinReunion, sowethereforecalled thesegroupsbytheingredientnames:‘2,4-D’group,‘glyphosate’ groupand‘S-metolachlor’group.Fortheingredientsinthe ‘2,4-D’group,the‘spray’variablewasthemainvariableofimportance intheI-PHYscorethen, toa lesserdegree,the‘water-dist’and

‘water-state’ variables. Conversely, for the ingredients in the ‘glyphosate’group,the2variablesofimportancewere‘water-dist’ and‘water-state’thenthe‘spray’variable.Fortheactive ingredi-entsinthe‘S-metolachlor’group,thevariableofimportancewas

Fig.6. IndividualfactormapofthePCAofimportancemeasurementsforthe vari-ablesofalltheactiveingredients.Thetwoaxesrepresent83%.Threegroupscan bedistinguished,eachrepresentedbyoneofthemostwidelyusedactive ingredi-entsontheislandofReunion:group1:2,4-D,group2:glyphosateandgroup3: S-metolachlor.

(10)

Table4

ImportanceofthevariablesintheconstructionoftheI-PHYscore.Thisimportanceiscalculatedwiththerandomforestmethodroundedofftoanindexbetween0and 100.Thestatisticalanalysesbroughtout3groups(PCAgroup)ofactiveingredientsaccordingtotheirvariablesofimportanceintheconstructionoftheI-PHYscorescalled: ‘2,4-D’group,‘glyphosate’groupand‘S-metolachlor’group(S-metointhetable).VariablesneededforI-PHYcalculation:seeTable1.

ActiveIngredient Spray Water-dist Water-state Posi. Cover-state Rain. Mower Filte. Plantstrip Nozzle Weather Cover-occup PCAgroup

2,4-D 37.2 27.1 27.7 1.3 2.7 3.0 0.2 0.0 0.4 0.3 0.0 0.0 2,4-D Aclonifen 52.2 20.7 19.7 2.1 4.2 0.0 0.2 0.0 0.6 0.3 0.0 0.0 2,4-D Asulam 6.8 39.8 41.0 1.8 4.5 5.0 0.2 0.0 0.7 0.2 0.0 0.0 glyphosate Atrazine 12.2 6.2 6.3 68.1 0.5 0.2 4.5 1.9 0.1 0.0 0.1 0.0 S-meto Bentazon 21.8 20.1 20.1 29.2 2.4 2.5 2.6 0.9 0.2 0.2 0.0 0.0 S-meto Diquat 56.0 17.8 17.7 2.5 4.8 0.0 0.2 0.0 0.7 0.3 0.0 0.0 2,4-D Fluazifop-P 67.7 14.2 14.1 0.8 1.3 1.3 0.1 0.0 0.2 0.3 0.0 0.0 2,4-D Fluroxypyr 19.5 18.7 19.1 33.9 2.4 2.5 2.6 0.9 0.2 0.2 0.0 0.0 S-meto Glufosinate 18.1 36.5 35.2 1.8 3.6 4.0 0.2 0.0 0.5 0.2 0.0 0.0 glyphosate Glyphosate 10.3 39.1 39.6 1.6 3.9 4.4 0.2 0.0 0.6 0.2 0.0 0.0 glyphosate Isoxaﬂutole 35.0 28.5 27.6 1.6 3.1 3.4 0.2 0.0 0.4 0.2 0.0 0.0 2,4-D Linuron 54.9 18.6 18.6 2.7 2.2 2.0 0.5 0.0 0.3 0.3 0.0 0.0 2,4-D Mesotrione 12.0 5.1 5.1 71.5 0.3 0.3 3.4 2.1 0.0 0.0 0.1 0.0 S-meto Metribuzin 13.7 11.3 11.3 57.1 1.0 1.1 3.2 1.0 0.1 0.1 0.1 0.0 S-meto Nicosulfuron 15.8 12.1 11.8 53.4 0.9 1.0 2.0 2.8 0.1 0.1 0.1 0.0 S-meto Oxadiazon 15.6 13.8 14.1 50.5 4.3 0.0 0.4 0.6 0.6 0.1 0.0 0.0 S-meto Pendimethalin 40.3 26.5 26.3 1.5 3.0 1.4 0.2 0.0 0.5 0.3 0.0 0.0 2,4-D S-metolachlor 13.8 11.2 11.0 55.5 1.0 0.3 5.4 1.5 0.1 0.1 0.0 0.0 S-meto Thiencarbazone 14.0 33.7 33.4 9.2 3.8 4.0 1.3 0.1 0.5 0.1 0.0 0.0 glyphosate Triclopyr 15.3 9.9 10.0 56.2 0.5 0.5 6.7 0.7 0.1 0.1 0.0 0.0 S-meto

the‘position’variable then, toa lesser degree,the ‘water-dist’, ‘water-state’ and ‘spray’variables. Whenconsidering theI-PHY scores,themeanandminimumscoresfortheactiveingredients inthe‘glyphosate’groupweresigniﬁcantlygreaterthanthoseof the‘S-metolachlor’group.Themeanscoreswere5.57forthe ‘S-metolachlor’groupasopposedto8.54forthe‘glyphosate’group withap-valueof0.00062.Theminimumscoreswere3.57forthe ‘S-metolachlor’groupasopposedto5.38forthe‘glyphosate’group withap-valueof0.0018.However,theI-PHYmeanandminimum scoresforthe‘S-metolachlor’groupdidnotdifferfromthoseofthe ‘2,4-D’group.Lastly,themaximumI-PHYscoresforthe3groups werenotstatisticallydifferent:7.96forthe‘2,4-D’group,9.32for the‘glyphosate’groupand7.45forthe‘S-metolachlor’group.This lastpointshowsthattheconditionsunderwhichtheseactive ingre-dientswereusedhadastrongimpactonthescores.

Withaviewtolimitingtheenvironmentalimpactsofherbicides, ourresultsshowedthatreplacingtheactiveingredientsofthe ‘S-metolachlor’groupwiththoseofthe‘glyphosate’groupwouldbe worthwhile.Unfortunately,thisisnotalwayspossiblebecausethe activeingredientsdonotalwayshavethesametarget (monocotyle-donousordicotyledonousweeds)andthesamemodesofaction (preorpost-emergence).Ifthesubstitutionofactivesubstances isnotpossible,regressiontreescanhelpfarmerstoidentify vari-ablesofimportanceconstitutingtheirownroomformanoeuvre. Wefoundthatfarmershadpossibilitiestoactuponthese differ-entvariablesofimportance.Thosepossibilitieswereeitherpartial forthe‘water-dist’and‘water-state’variables,ortotalforallthe others.Ourresultsshowedthatfourvariablesmainlyaffectedthe I-PHYscoreandthisimportancevarieddependingonthepesticide group.However,thecomparisonoftheseresultswithliterature, suggeststhatI-PHYprobablyminimizesthespecificriskslinked tomajor rainfalleventsgiven themain variables we identified thatinfluenceitsscore.Infact,whilethetwovariableslinkedto theproximityofawaterpoint(‘dist-state’and‘water-state’) log-icallyinfluencedtheI-PHYscoreswhatevertheactiveingredient (HaarstadandLudvigsen,2007),the‘rainfall’variableonlyhada moderateimpactonthisscore.Yet,thatvariablewasconsidered veryimportantinsomeothercasestudies(Louchartetal.,2001; Hermosinetal.,2013)andparticularlyinatropicalclimate(Damm andVandenBrink,2010).Bernardetal.(2005)showedthat what-everthetypeofsoilconsidered,therisksofgroundwaterpollution duringheavyrainfall(>50mmday−1)byalltheactiveingredients

theystudiedwereproven.Includedintheseingredientsandthose weassessedﬁguredatrazine,triclopyrand2,4-D.Davisetal.(2011)

alsoshowedthattherisksofsurfacewaterpollutionlinkedto herbi-cides(atrazine)ona cropsuchassugarcanemainlyarosefrom irrigationpracticesjustafterherbicideapplication.Thispractice iswidelyusedinReunion,particularlytoincreasetheeffectiveness ofpre-emergenceherbicidessuchasS–metolachlor.I-PHYdoesnot considerirrigationpracticesandtheycouldbeincludedwiththe ‘rainfall’variable,whichwouldallthemorewarrantincreasingthe weightofthatvariable.

4. Conclusion

Ourmethodologyenabledustoidentifyandprioritize impor-tancevariablesfortheI-PHYindicator,foreachherbicideunder ourstudyconditions.Such ananalysisprovesdecisiveforusing this typeof indicatoras itmakesit possibletolimit the“black box”imageoftheseindicators.Assuch, thesevariables become leversforactionuponwhichusersoftheseherbicidescanactto limit therisk of transferstotheenvironment. We tested some scenariosbasedontheextremethresholdvaluesofthevariables (favourableandunfavourable)inordertodeterminethemaximum roomforimprovement.Inpractice,andforanexanterisk assess-ment,somevariablescanbeblockedattheirtruevaluesrelativeto theactualconditionsofuse.Thismakesitpossibletoreducethe numberofscenariosand,thereby,thecomplexityoftheregression trees.Thesetreesthenbecomedecision–supporttoolsspeciﬁcto eachuserandprovidecustomizedadviceenablinguserstotake appropriate decisionswithaviewtoachievingtheirobjectives. Inordertohelpfarmers,thisdecision–supporttoolisnow avail-ableasanapplication,calledPhyto’Aide,onourwebapplications platform(www.margouilla.net/phytoaide).Thistool,basedonan opensourcecode,iscompletelyfreeandtheuserfriendlyinterface allowsthefarmer,oranyuser,toselectanactiveingredientandsoil parameterstoacquirearegressiontreeandthenadvice.Itprovides userswithatutorialtounderstandthemethod,howtointerpret thetreeandtoknowwhichactionscanbetaken.

Acknowledgments

Weshouldliketoexpressourwarmthankstoallstakeholders oftheagriculturalsectorandtheFrenchadministrationfortheir

(11)

extensiveinvolvementinthiswork.WethankPeterBigginsfrom CIRADforhislinguisticsupport.Wethanktheanonymous review-erswhohelpedimprovethequalityofthemanuscript.Theproject wasco-ﬁnancedbytheEuropeanUnionand theReunionIsland Region.

AppendixA. Supplementarydata

Supplementary data associated with this article can be found,intheonlineversion,athttp://dx.doi.org/10.1016/j.ecolind. 2015.02.020.

References

Badat,2011.Etatdelaressourceetdesusagesdel’eauàlaRéunion–2010.In:Ofﬁce del’eauRéunion,Saint-Denis,France.

Battaglin,W.A.,Furlong,E.T.,Burkhardt,M.R.,Peter,C.J.,2000.Occurrenceof sul-fonylurea,sulfonamide,imidazolinone,andotherherbicidesinriver,reservoirs andgroundwaterintheMidwesternUnitedStates,1998.Sci.TotalEnviron.248, 123–133.

Bel,L.,Allard,D.,Laurent,J.M.,Cheddadi,R.,Bar-Hend,A.,2009.CARTalgorithm forspatialdata:applicationtoenvironmentalandecologicaldata.Comput.Stat. DataAnal.53,3082–3093.

Bernard,H.,Chabalier,P.F.,Chopart,J.L.,Legube,B.,Vauclin,M.,2005.Assessmentof herbicidesleachingriskintwotropicalsoilsofReunionIsland.J.Environ.Qual. 34,534–546.

Bockstaller,C.,Girardin,P.,2003.Howtovalidateenvironmentalindicators.Agric. Syst.76,639–653.

Bockstaller,C.,Wohlfahrt,J.,Hubert,A.,Hennebert,P.,Zahm,F.,Vernier,F.,Mazzela, N.,Keichinger,O.,Girardin,P.,2008.Lesindicateursderisquedetransfert deproduitsphytosanitairesetleurvalidation:exempledel’indicateurI-Phy. Ingénieries86,103–114.

Bockstaller,C.,Guichard,L.,Keichinger,O.,Girardin,P.,Galan,M.B.,Gaillard,G.,2009.

Comparisonofmethodstoassessthesustainabilityofagriculturalsystems,a review.Agron.Sustain.Dev.29,223–235.

Breiman,L.,2001.RandomForests.StatisticsDepartment,UniversityofCalifornia, Berkeley,USA.

Breiman,L.,Friedman,J.H.,Olshen,R.A.,Stone,C.J.,1984.Classiﬁcationand Regres-sionTrees.Chapman/HallWadsworth,USA.

Damm,M.A.,VandenBrink,P.J.,2010.Implicationsofdifferencesbetween tem-perateandtropicalfreshwaterecosystemsfortheecologicalriskassessmentof pesticides.Ecotoxicology19,24–37.

Daniel,P.E.,Bedmar,F.,Costa,J.L.,Aparicio,V.C.,2002.Atrazineandmetribuzin sorptioninsoilsoftheArgentineanhumidpampas.Environ.Toxicol.Chem.21, 2567–2572.

Davis,A.M.,Thorburn,P.J.,Lewis,S.E.,Bainbridge,Z.T.,Attard,S.J.,Milla,R.,Brodie, J.E.,2011.Environmentalimpactsofirrigatedsugarcaneproduction:herbicide run-offdynamicsfromfarmsandassociateddrainagesystems.Agric.Ecosyst. Environ.180,123–135.

DeLafontaine,Y.,Beauvais,C.,Cessna,A.J.,Gagnon,P.,Hudon, C.,Poissant,L., 2014. Sulfonylurea herbicides in an agricultural catchmentbasin and its adjacentwetlandin theSt.LawrenceRiverbasin.Sci. TotalEnviron.479, 1–10.

Devillers,J.,Farret,R.,Girardin,P.,Rivière,J.-L.,Soulas,G.,2005.Indicateurspour évaluerlesrisquesliésàl’utilisationdespesticides.Tec&Doc,Lavoisier,Paris.

De’ath, G., Fabricius, K.E., 2000. Classiﬁcation and regression trees: a pow-erful yet simple technique for ecological data analysis. Ecology 81, 3178–3198.

FAOSTAT. Pesticides (use). http://data.fao.org/dataset?entryId=5e70fee4-fb65-43b6-8da1-b6de4626b9bd&tab=metadata(accessed18.02.13).

Ferrara,A.,Salvati,L.,Sateriano,A.,Nolè,A.,2012.Performanceevaluationandcost assessmentofakeyindicatorsystemtomonitordesertiﬁcationvulnerability. Ecol.Indic.23,123–129.

Ghosh,R.K.,Singh,N.,2013.Adsorption–desorptionofmetolachlorandatrazine inIndian soils:effectofﬂyashamendment.Environ.Monit.Assess. 185, 1833–1845.

Gustafson,D.I.,1989.Groundwaterubiquityscore:asimplemethodforassessing pesticideleachability.Environ.Toxicol.Chem.8,339–357.

Haarstad,K.,Ludvigsen,G.H.,2007.TenyearsofpesticidemonitoringinNorwegian groundwater.GroundWaterMonit.R27,75–89.

Hak,T.,Kovanda,J.,Weinzettel,J.,2012.Amethodtoassesstherelevanceof sus-tainabilityindicators:applicationtotheindicatorsetoftheCzechRepublic’s SustainableDevelopmentStrategy.Ecol.Indic.17,46–57.

Hermosin,M.C.,Calderon,M.J.,Real,M.,Cornejo,J.,2013.Impactofherbicidesused inolivegrovesonwatersoftheGuadalquivirriverbasin(southernSpain).Agric. Ecosyst.Environ.164,229–243.

Lin,Y.J.,Makesh,K.,Anugrah,S.,Gian,G.,1999.Effectofsimulatedsunlighton atrazineandmetolachlortoxicityofsurfacewaters.Ecotoxicol.Environ.Saf. 43,35–37.

Louchart,X.,Voltz,M.,Andrieux,P.,Moussa,R.,2001.Herbicidesrunoffatﬁeldand watershedscalesinaMediterraneanvineyardarea.J.Environ.Qual.30,982–991.

Maillary,2012.,2012.Basenationaledesventesdesdistributeurs2009–2010et 2011.DAAF,Saint-Pierre,France.

Ministèredel’agricultureetdel’agroalimentaire,2013.e-phy,lecataloguedes pro-duitsphytopharmaceutiquesetdeleursusagesdesmatièresfertilisanteset dessupportsdeculturehomologuésenFrance,http://e-phy.agriculture.gouv.fr/

(accessed07.06.12).

Mottes,C.,Lesueur-Jannoyer,M.,LeBail,M.,Malézieux,E.,2013.Pesticide trans-fermodelsincropandwatershedsystems:areview.Agron.Sustain.Dev.,

http://dx.doi.org/10.10007/s13593-013-0176-3(onlineﬁrst).

Oliver,D.P.,Kookana,R.S.,2006.On-farmmanagementpracticestominimizeoff-site movementofpesticidesfromfurrowirrigation.PestManag.Sci.62,899–911.

Payraudeau,S.,Gregoire,C.,2012.Modellingpesticidestransfertosurfacewaterat thecatchmentscale:amulti-criteriaanalysis.Agron.Sustain.Dev.32,479–500.

Pedlowski,M.A.,Canela,M.C.,DaCostaTerra,M.A.,RamosdeFaria,R.M.,2012.

ModesofpesticidesutilizationbyBraziliansmallholdersandtheirimplications forhumanhealthandtheenvironment.CropProt.31,113–118.

Reichenberger,S.,Bach,M.,Skitschak,A.,Frede,H.G.,2007.Mitigationstrategiesto reducepesticideinputsintogroundandsurfacewaterandtheireffectiveness, areview.Sci.TotalEnviron.384,1–35.

Reus,J.,Leendertse,P.,Bockstaller,C.,Fomsgaard,I.,Gutsche,V.,Lewis,K.,etal., 2002.Comparisonandevaluationofeightpesticideenvironmentalrisk indica-torsdevelopedinEuropeandrecommendationsforfutureuse.Agric.Ecosyst. Environ.90,177–187.

Sadok,W.,Angevin,F.,Bergez,J.-E.,Bockstaller,C.,Colomb,B.,Guichard,L.,Reau,R., Doré,T.,2008.Exanteassessmentofthesustainabilityofalternativecropping systems:guidelinesforidentifyingrelevantmulti-criteriadecisionaidmethods. Agron.Sustain.Dev.28,163–174.

Tomlin,C.D.S.(Ed.),2006.ThePesticideManual.,14thed.BritishCropProtection Council,Farnham,UnitedKingdom,p.1349.

UniversityofHertfordshire,2013.ThePesticidePropertiesDataBase(PPDB) Devel-opedby theAgriculture&Environment ResearchUnit(AERU). University ofHertfordshire,2006–2013.http://sitem.herts.ac.uk/aeru/ppdb/en/atoz.htm

(accessed07.06.12).

VanderWerf,H.M.G.,Zimmer,C.,1998.Anindicatorofpesticideenvironmental impactbasedonafuzzyexpertsystem.Chemosphere36,2225–2249.

Voltz,M.,Alix,A.,Barriuso,E.,Bedos,C.,Bonicelli,B.,Caquet,T.,Dubus,I.,Gascuel, C.,Gril,J.N.,2005.Devenirettransfertdespesticidesdansl’environnementet impactsbiologiques.In:Aubertot,J.N.,Barbier,J.M.,Carpentier,A.,Gril,J.N., Guichard,L.,Lucas,P.,Savary,S.,Voltz,M.(Eds.),Pesticides,agricultureet environnement–Réduirel’utilisationdespesticidesetenlimiterlesimpacts environnementaux,ExpertisescientiﬁquecollectiveINRA-Cemagref,chapitre, vol.3,pp.1–219.