Modeling Pinot Noir Aroma Profiles Based on Weather and Water Management Information Using Machine Learning Algorithms: A Vertical Vintage Analysis Using Artificial Intelligence

Abstract

Wine aroma profiles are determinant for the specific style and quality characteristics from final wines. These are dependent on the seasonality, mainly weather conditions, such as solar exposure and temperatures and water management strategies from veraison to harvest. This paper presents machine learning modeling strategies using weather and water management information from a Pinot Noir vineyard from 2008 to 2016 vintages as inputs and aroma profiles from wines from the same vintages assessed using gas chromatography and chemometric analyses of wines as targets. Results showed that artificial neural network models (ANN) rendered the high accuracy in the prediction of aroma profiles (Model 1; R = 0.99) and chemometric wine parameters (Model 2; R = 0.94) with no indication of overfitting. These models could offer powerful tools to winemakers to assess aroma profiles of wines before winemaking, which could help to adjust some techniques to maintain/increase quality of wines or wine styles characteristic of specific vineyards or regions. These models can be modified for different cultivars and regions by including more data from vertical vintages to implement artificial intelligence in winemaking.

Full text



Foods2020,9,33;doi:10.3390/foods9010033www.mdpi.com/journal/foods
Article
ModelingPinotNoirAromaProfilesBasedon
WeatherandWaterManagementInformationUsing
MachineLearningAlgorithms:AVerticalVintage
AnalysisUsingArtificialIntelligence
SigfredoFuentes
1,
*,EdenTongson
1
,DamirD.Torrico
1,2
andClaudiaGonzalezViejo
1

1
SchoolofAgricultureandFood,FacultyofVeterinaryandAgriculturalSciences,UniversityofMelbourne,
Melbourne,VIC3010,Australia;eden.tongson@unimelb.edu.au(E.T.);damir.torrico@lincoln.ac.nz(D.D.T.);
cgonzalez2@unimelb.edu.au(C.G.V.)
2
DepartmentofWine,FoodandMolecularBiosciences,FacultyofAgricultureandLifeSciences,
LincolnUniversity,Lincoln7647,NewZealand
*Correspondence:sfuentes@unimelb.edu.au;Tel.:+61‐4245‐04434
Received:12November2019;Accepted:27December2019;Published:30December2019
Abstract:Winearomaprofilesaredeterminantforthespecificstyleandqualitycharacteristicsof
finalwines.Thesearedependentontheseasonality,mainlyweatherconditions,suchassolar
exposureandtemperaturesandwatermanagementstrategiesfromveraisontoharvest.Thispaper
presentsmachinelearningmodelingstrategiesusingweatherandwatermanagementinformation
fromaPinotnoirvineyardfrom2008to2016vintagesasinputsandaromaprofilesfromwinesfrom
thesamevintagesassessedusinggaschromatographyandchemometricanalysesofwinesas
targets.Theresultsshowedthatartificialneuralnetwork(ANN)modelsrenderedthehighaccuracy
inthepredictionofaromaprofiles(Model1;R=0.99)andchemometricwineparameters(Model2;
R=0.94)withnoindicationofoverfitting.Thesemodelscouldofferpowerfultoolstowinemakers
toassessthearomaprofilesofwinesbeforewinemaking,whichcouldhelpadjustsometechniques
tomaintain/increasethequalityofwinesorwinestylesthatarecharacteristicofspecificvineyards
orregions.Thesemodelscanbemodifiedfordifferentcultivarsandregionsbyincludingmoredata
fromverticalvintagestoimplementartificialintelligenceinwinemaking.
Keywords:winequality;machinelearningmodeling;weather

1.Introduction
Winequalitytraitsaredifficulttoassessinarapidandobjectivewayinvineyards,especially
beforewinemaking.Usually,qualityassessmentsthatareperformedinthewineindustryarerelated
totheacidityandsugarcontentinberries(BrixorBaume)toassessmaturity[1,2].However,this
assessmentonlygivesinformationabouttheamountofalcoholandacidityinthefinalwinethrough
fermentation.Hence,berrysugars/aciditydonotprovideusefulinformationonanyotherimportant
qualitytrait,suchasthepotentialaromaprofilesthatcouldbeobtainedinthefinalwine.
Alcoholpresentinbeverageshasbeenfoundtohaveaneffectontheperceptionofflavorand
aromas,asitaidsinthereleaseofvolatilearomaticcompounds[3].Furthermore,higheralcohol
wineshavebeensometimesregardedasbeneficialforthephysicochemicalexpressionofcolorand
otherqualitytraitsthatimpacttheirsensoryevaluation[4].However,increasingthealcoholcontent
inwinesisaproblemnowadaysduetoclimatechange,specificallyglobalwarming.Specifically,
highertemperaturesarecompressingphenologicalstages,resultinginearlierharvestduringhotter
monthsaroundtheglobe[5–8].Thisphenomenonproducesadoubleglobalwarmingeffectin
grapevines,whichcanresultinberryshrivelwiththeassociatedconcentrationofsugarinberries,

Foods2020,9,332of15
andthedegradationofcolorandaromacompounds,whichimpactthesensoryaromaandflavor
profileoffinalwines[7,9].Recently,theassessmentofmesocarplivingtissuehasbeenassociated
withqualitytraitsfordifferentgrapevinecultivarsforwinemaking[10].Berrycelldeathstarts
around90daysafterfullbloom;itisaprogrammedcelldeath,whichcanbeuncoupledfromsugar
accumulationandberryshrivel(bothexacerbatedbyhighertemperatures)andcandeterminethe
finalqualityofwines,aromaprofile,andsensoryappreciation[11,12].Hence,thereisadirectlink
betweentheseasonalweathercharacteristics,whicharemainlytemperatureexpressedinthermal
time(degreedays)accumulatedover10°Candphenologicalstagesoccurrenceandduration[13],
berrycelldeath,winequality,andaromaprofiles[11,12].Furthermore,theseberryqualitytraitscan
bemanipulatedusingdifferentirrigationtechniques,suchasregulateddeficitirrigation(RDI)[14–
20]andpartialrootzonedrying(PRD)[21–25].
Somemethodsusingproximalremotesensingwithinthenear‐infrared(NIR)lightspectrum
reflectivityhavebeendevelopedtoassessqualitytraitsfromberriesinanon‐destructiveway.Some
applicationshavebeenimplementedtoassessthesugarcontentinberries[26,27],berrypigments
[28,29],phenoliccompounds[30,31],andgrapematurityingeneral[28,32–34].However,sincethese
techniquesarestillmanual,theycannotaccountforthenaturalintra‐bunchandvineyardspatial
variability,requiringahugenumberofmeasurementsandmodelingstrategiestoobtainmeaningful
results.
Othertechniqueshavebeendevelopedthankstorecentadvancesinunmannedaerialvehicles
andremotesensingtechniquestoassessgrapematurity,whichcantakeintoaccountwithin‐vineyard
variabilityusinghigh‐resolutionmultispectralimageryanalysis[35–37].However,studieshavebeen
limitedtoafewflightsperseason,andtheindirectassessmentofberryqualityandmaturitymay
hamperresults.Furthermore,associatedcostsfordataacquisition,post‐processingtoobtain
orthomosaics,dataanalysisforclassification,andthematicmapproductionarestillcostly,requiring
inmanycountrieslicensedpilotsandhighdataanalysispowertoobtainmeaningfulmodels.
Thispaperpresentsmachinelearningmodelingstrategiesapplyingintegratedvineyardweather
andirrigationmanagementparametersasinputsandthearomaprofilesastargetsobtainedfroma
verticalwinelibraryfromaboutiquevineyard.Theresultsfromthismodelingstrategycouldoffer
animportanttooltowinemakerstoassessthearomaprofilesforfuturevintagesbeforewinemaking.
Theknowledgeofpotentialaromaprofilesofthefinalwinemayallowmakingadjustmentswithin
thewinemakingtomaintainorincreasequalitytraitsinthefinalwinetomaintainaspecificwine
stylethatischaracteristicofthewineregionorparticularvineyard.
2.MaterialsandMethods
2.1.StudyAreaandWeather/IrrigationManagementDataAcquisition
Thestudywasconductedusingweatherandmanagementdataandwinesamplesfroma
verticalwinelibrarybelongingtoacommercialvineyardlocatedatanelevationof540m.a.s.linthe
SouthoftheGreatDividingRangeoftheMacedonRangesinthesub‐regionofRomsey/Lancefield,
VictoriainAustralia.Thevineyardissituatedatadistancefromthemitigatinginfluenceoftheocean
(Figure1),andthecultivarsplantedconsistof69%Pinotnoir,26%Chardonnay,and5%Pinotgris,
andusemostlythelyretrainingsystem.Thestudywasconductedforverticalvintagesfrom2008to
2016ofPinotnoircultivars,andweather/irrigationmanagementdatawereobtainedfromthesame
siteforeachseason.Informationsuchas(i)solarexposurefromveraisontoharvest(V‐H),(ii)solar
exposurefromSeptembertoharvest(S‐H),(iii)maximumJanuarysolarexposure(MJSE),(iv)degree
daysfromS‐H(DD‐S‐H),(v)maximumJanuarytemperature(MJT),(vi)meanmaximumtemperature
fromV‐H(MeanMaxTV‐H),and(vii)meanminimumtemperaturefromV‐H(MeanMinTV‐H)was
extractedfromtheBureauofMeteorology(BoM).Furthermore,thewaterbalance(WB)was
calculatedusingtheirrigation(I),rainfall(RF),andevapotranspiration(ETc)datausingthefollowing
Equation(1):
𝑊𝐵  𝐼  𝑅𝐹0.85𝐸𝑇
(1)

Foods2020,9,333of15
whereWB=waterbalance;I=irrigationappliedinmegaliter(ML);RF=effectiverainfall,considering
85%ofthewaterisavailabletotheplant,andETc=cropevapotranspirationcalculatedusingthe
correspondingcropcoefficient(Kc)fordifferentphenologicalstages[14].


Figure1.Aerialimageofthestudyareaobtainedusinganunmannedaerialvehicle(UAV)inthe
2015–2016growingseasonfromatotalareaplantedof42hectares.
2.2.PhysicochemicalAnalysis
Winesfromeachvintagewereanalyzedintriplicatesforthedifferentphysicochemicaldata
measuredinthisstudy.Avolumeof20mLofeachwinesamplewaspouredina60×15mmGreiner
Bio‐OnePolystyrenePetridish(itemnumber628102;GreinerBio‐One,Kremsmünster,Austria)and
placedonawhiteuniformsurface.ColorinCIELabandRGBscaleswasmeasuredusingaNIXPro
colorsensor(NIXSensorLtd.Hamilton,Ontario,Canada).TheUV‐Visspectrafrom380to780nm
wereacquiredwithaLightingPassportProportablespectrometer(AsensetekIncorporation,New
TaipeiCity,Taiwan).Tocalculatecolorintensity,theabsorbanceof420,520,and620nmwere
summed,whileforcolorhue,theabsorbancefrom420nmwasdividedbythevaluefrom520nm.
FiftymLofeachwinesamplewereusedtodetermineliquiddensity(weightdividedbyvolume),pH
wasdeterminedusingapH‐meter(QM‐1670,DigiTech,Sandy,UT,USA),totaldissolvedsolids
(TDS)andelectricconductivity(EC)weremeasuredwithaYuelongYL‐TDS2‐Adigitalwaterquality
tester(ZhengzhouYuelongElectronicTechnologyCo.,Ltd,ZhengzhouCity,HenanProvince,
China),saltconcentrationwasmeasuredusingadigitalsalt‐meter(PAL‐SALTMohr,AtagoCo.,Ltd.
Saitama,Japan),andalcoholcontentusinganAlcolyzerWineMalcoholmeter(AntonPaarGmbH,
Graz,Austria).
2.3.GasChromatography–MassSpectroscopy
A5mLsampleofeachwinereplicatewaspouredintoa20mLscrewcapvialandsealedwith
an18mmmagneticscrewcapwithapolytetrafluoroethyleneandsiliconeliner.Thesesampleswere

Foods2020,9,334of15
analyzedwiththemethodproposedbyGonzalezViejoetal.[38]usingahigh‐efficiencygas
chromatographwithamassselectivedetector5977B(GC‐MSD;AgilentTechnologies,Inc.,Santa
Clara,CA,USA),coupledwithaPAL3autosamplersystem(CTCAnalyticsAG,Zwingen,
Switzerland).TheGC‐MSDhasadetectionlimitof1.5fg,andanHP‐5MScolumnwasattached
(length:30m,innerdiameter:0.25mm,film:0.25μ;AgilentTechnologies,Inc.,SantaClara,CA,
USA),whiletheflowratewassetto1mLmin−1ofthecarriergas(Helium).Headspacewithsolid‐
phasemicroextraction(SPME)andadivinylbenzene–carboxen–polydimethylsiloxanegreyfiber(1.1
mm;AgilentTechnologies,Inc.,SantaClara,CA,USA)wasused.Incubationtimewassetto20at45
°Cwitha5mincycleand1minforfiberconditioning(170°C).Furthermore,theextractiontimewas
setto40minwithagitation.Twoblanksampleswereused,oneatthestartandoneattheendto
avoidanycarryovereffect.Toidentifythevolatilecompounds,theNationalInstituteofStandards
andTechnologylibrary(NIST;NationalInstituteofStandardsandTechnology,Gaithersburg,MD,
UnitedStates)wasused.Onlythecompoundswith≥80%certaintywerereported.
2.4.StatisticalAnalysisandMachineLearningModeling
Datafromweather,physicochemical,andaromaprofilemeasurementswereanalyzedusinga
customizedcodewritteninMatlab®R2019a(Mathworks,Inc.Natick,MA.USA)toassesssignificant
correlations(p<0.05)betweenparameterswerereportedinamatrix.Thesedatawerealsousedto
developmachinelearningmodelsbasedonartificialneuralnetworks(ANN)usinganautomated
codeinMatlab®thattests17differenttrainingalgorithmsinaloop.Theweatherdatarelatedto(i)
solarexposureV‐H,(ii)solarexposurefromS‐H,(iii)MJSE,(iv)DD‐S‐H,(v)MJT,(vi)MeanMaxTV‐
H,(vii)MeanMinTV‐H,and(viii)waterbalancewereusedasinputsformachinelearningpurposes.
Twomodelsweredevelopedusingtheseinputstopredict(i)thepeakareaofninevolatilearomatic
compoundsmeasuredusingtheGC‐MSD(Model1)and(ii)14physicochemicalmeasurements
(Model2).Bothmodelsweredevelopedusingnormalizeddata(inputsandtargets)from−1to1,and
witharandomdatadivisionwith60%ofthesamplesusedfortrainingwithaLevenberg–Marquardt
algorithm,20%forvalidationwithameansquarederrorperformancealgorithm,and20%fortesting
withadefaultderivativefunction.Thenumberofneuronswasdefinedbyperformingatrimming
exercisewiththree,five,seven,and10neurons,with10neuronsgivingthebestmodelsthat
contributetotheabsenceofoverfitting.Themodelsconsistedofatwo‐layerfeedforwardnetwork
withatan‐sigmoidfunctioninthehiddenlayerandalineartransferfunctionintheoutputlayer
(Figure2).

(a)

Foods2020,9,335of15

(b)
Figure2.Artificialneuralnetworkmodeldiagramsshowingtheinputsandtarget/outputsof(a)
Model1topredictthearomaprofilebasedonthepeakareaofvolatilearomaticcompounds,and(b)
thephysicochemicaldataofPinotnoirwines.
3.Results
Table1showsthemeanvaluesoftheweatherdataforthevintageswithcontrastingwater
balancedata(2011–2014).Itcanbeobservedthat2011wasthewettestseasonwiththelowestsolar
exposureandmeantemperatures(MeanMaxTV‐HandMeanMinTV‐H),while2013wasthedriest
withthehighestMJSEandsolarexposure.Vintages2012and2014presentedvaluesinthemid‐range.

Foods2020,9,336of15
Table1.Meanvaluesofweatherdataonlyforthecontrastingvintagesbasedonwaterbalance.
YearSolarExposure
(V‐H;MJm2−1)
SolarExposure
(S‐H;MJm2−1)
MJSE
(MJm2−1)
DD‐S‐H
(days)
MJT
(°C)
MeanMaxTV‐H
(°C)
MeanMinTV‐H
(°C)
WaterBalance
(mm)
201115.619.124.61066.818.619.79.44673.7
201217.920.226.31147.319.422.610.75255.9
201321.821.828.91234.219.826.112.05−117.5
201419.020.027.61223.720.325.811.31−61.9
Abbreviations:V‐H=veraisontoharvest,S‐H=Septembertoharvest,MJSE=maximumJanuarysolarexposure,DD=degreedays,MJT=maximumJanuarytemperature,MaxTV‐H=maximum
temperatureveraisontoharvest,MinTV‐Hminimumtemperatureveraisontoharvest.

Foods2020,9,337of15
Table2showstheninevolatilecompoundsidentifiedinallthewinesamplestestedandthe
aromasassociatedwiththem.Itcanbeobservedfromthistablethatmostofthearomasarerelated
tofruityscents,especiallyapple,withtwospecificcompounds(phenylethylalcoholandethyl
laurate)withfloralandone(ethylpalmitate)withmilkyorcreamynotes.
Table2.Volatilecompoundsidentifiedusinggaschromatography–massspectroscopyandtheir
associatedaromas.
VolatileCompoundAroma*
EthylhexanoateApple/Greenbanana/Pineapple
PhenylethylalcoholRose/Bread/Honey
DiethylsuccinateCookedapple
EthyloctanoateApple/Banana/Pineapple
EthylnonanoateCognac/Apple/Winey/Nutty
Ethyl‐9‐decenoateFruity/Fatty/Roses
EthyldecanoateWaxy/Apple/Grape
EthyllaurateFloral/Soapy/Sweet
EthylpalmitateWaxy/Fruity/Creamy/Milky
*TheassociationbetweenthevolatilecompoundsandaromaswereobtainedfromTheGoodScents
Company[39],Genoveseetal.[40],Arcarietal.[41],andGonzalezViejoetal.[38].
Figure3showsthesignificant(p<0.05)correlationsbetweentheweatherinformation,the
aromas,andphysicochemicaldata.ItcanbeobservedthatthesolarexposurefromSeptemberto
harvestwaspositivelycorrelatedwithdiethylsuccinate(r=0.90),whilethedegreedaysfrom
Septembertoharvestwasnegativelycorrelatedwithethyl‐9‐decenoate(r=0.88).TheMJThada
positivecorrelationwithphenylethylalcohol(r=0.82)and“b”(r=0.88),andanegativecorrelation
with“B”.TheMeanMaxTV‐Hwasnegativelycorrelatedwithethyl‐9‐decenoate(r=−0.93)andcolor
intensity(r=−0.90),aswellaspositivelycorrelatedwithcolorhue(r=0.92)and“a”(r=0.84).Onthe
otherhand,theMeanMinTV‐Hhadanegativecorrelationwithethylhexanoate(r=−0.93),TDS(r=
−0.90),andEC(r=−0.90).Waterbalancewaspositivelycorrelatedwithethyl‐9‐decenoate(r=0.93)
andcolorintensity(r=0.90),andnegativelycorrelatedwithcolorhue(r=−0.95)and“a”(r=−0.86).
Meanvaluesofthearomaticvolatilecompoundsandphysicochemicaldataareshownas
supplementarymaterialinTableS1.

Foods2020,9,338of15

Figure3.Matrixshowingonlythesignificantcorrelations(p<0.05)betweentheweatherand
physicochemicaldataandvolatilearomaticcompoundsofPinotnoirwinesofvintagesfrom2008to
2016.Abbreviations:TDS=totaldissolvedsolids,EC=electricconductivity,V‐H=veraisonto
harvest,S‐H=Septembertoharvest,MJSE=maximumJanuarysolarexposure,DD=degreedays,
MJT=maximumJanuarytemperature,MaxTV‐H=maximumtemperatureveraisontoharvest,
MinTV‐Hminimumtemperatureveraisontoharvest.
InTable3,thestatisticalresultsfromtheANNmodelsareshown.Model1hadanoverallhigh
correlationcoefficient(r=0.99)withsimilarresultsforallstages(training,validation,andtesting;r>
0.97)topredictthepeakareaofninevolatilearomaticcompounds(Table2).Fromtheperformance,
itcanbeobservedthatbothvalidationandtestingmeansquareerror(MSE)valueswerethesame
(MSE=0.03),andthetraininghadalowerresult(MSE=0.003),whichcontributestotheabsenceof
overfittingofthemodel.Furthermore,theslope(b)forallstagesandtheoverallmodelwascloseto
theunity(b=0.97).Ontheotherhand,Model2hadanoverallcorrelationr=0.94topredict14
physicochemicalparameters(Figure2b).Theslopesfromthemodelsofthethreestageswerehigh
enough(b>0.83)withanoverallmodelb=0.90.SimilartoModel1,theperformanceofthetraining
stagefromModel2waslower(MSE=0.02)thanthevalidationandtestingstages,withthelasttwo
presentingsimilarresults(MSE=0.05andMSE=0.06;respectively).
Table3.Statisticsfromtheartificialneuralnetworkmodelstopredictthearomaprofilebasedonthe
peakareaofvolatilearomaticcompounds(Model1)andthephysicochemicaldata(Model2)from
Pinotnoirwines.
StageSamplesObservationsRSlope(b)Performance(MSE)
Model1
Training403600.990.980.003
Validation131170.970.980.03
Testing131170.970.920.03
Overall665940.990.97/

Foods2020,9,339of15
Model2
Training405600.960.910.02
Validation131820.930.830.05
Testing131820.900.940.06
Overall669240.940.90/
Abbreviations:R=correlationcoefficientandMSE=meansquareerror.
Figure4ashowstheoverallModel1topredictthearomaprofilebasedonthepeakareaof
volatilearomaticcompoundsofPinotnoirwines.Fromthe95%confidencebounds,only1.01%of
outliers(sixoutof594)werefound.Ontheotherhand,Figure4bdepictstheoverallModel2to
predictthephysicochemicaldataofthewines.Regardingthe95%predictionbounds,themodel
presented3.25%(30outof924)ofoutliers.Forbothmodels,severalretrainingattemptswere
performed,obtainingsimilarresultstothosepresentedinTable3andFigure4.Whenfeedingthese
modelswithnewdata,theoutputsvaluesaregivennormalizedfrom−1to1;however,thereverse
functionfornormalizationinMatlab®R2019a(MathworksInc.,Natick,MA,USA)providestheactual
valuesinthecorrespondingunits.

(a)

Foods2020,9,3310of15

(b)
Figure4.Overallartificialneuralnetworkmodelstopredict(a)thearomaprofile(Model1)and(b)
thephysicochemicalparametersofPinotnoirwines(Model2),bothusingtheweatherdataasinputs
(Figure2).Themodelsshowtheobserved(x‐axis)andpredicted(y‐axis)dataaswellasthe95%
confidencebounds.
4.Discussion
Thephysicochemicalparametersassessedinthisstudyhavebeenassociatedwithwinequality
byotherauthors.Aromasandcolor‐relatedparametersaresomeofthefactorsthathavebeenthe
mostassociatedwithwinequality[42,43].Sáenz‐Navajasetal.[44]foundthatthereisarelationship
betweenredwinecolorandthequalityperceptionfromconsumersandconcludedthatdarkerwines
withhigherredandloweryellowvalueswereratedashigherquality.Jacksonetal.[42]reporteda
significantandpositivecorrelationbetweenbothpHandcolorandoverallwinequality.The
importanceofTDS,EC,andsaltmeasurementsrelyonthefactthattheseareanapproachtominerals
content[45],whichareimportantinwinequality,asthemineralspresentinwinehavebeenrelated
tothosepresentinthesoil,andthesehavebeenassociatedwiththewine’snutritionalcomposition
andsafety[46].
TherewasasignificantvariabilitywithinthevintagesandtheparticularregioninVictoria
analyzedinthisstudy.Theextremescanbeconsideredforlow‐qualitywinesproducedinthe2010–
2011vintageduetoheavyrainsbeforeharvest,whichnegativelyaffectsthequalitytraitsinberries
andwine[47,48];thislow‐qualityassessmentwasobtainedfromanecdotalinformationfrompoints
receivedinthoseparticularyearsandthesensoryanalysisconductedbythevineyardstudied.On
thecontrary,dryseasonswerefoundforexamplein2013–2014and2014–2015,withincreasedberry
qualitytraitsthatwerepassedtotherespectivewines.Thelatterweremainlyduetosomecontrolof
thewaterreceivedbyplantsfromirrigationandwaterdeficits.Thesedifferencescontributetothe
robustnessofthemachinelearningmodelsfound,whichpresentednoindicationofoverfittingwith
highprecisioninthepredictionofthepeakareaofvolatilearomaticcompounds(Model1)and
physicochemicalwinecharacteristics(Model2).
Theeffectsofsolarexposureandcanopyarchitecture(whichisdependentonwaterbalance)on
thearomaprofilesofwineshavebeenpreviouslyreported,andtheyareconsistentwiththedata

Foods2020,9,3311of15
presentedinFigure3.Specifically,theseeffectsmanifestthroughtheinfluenceofthemicroclimate
withinbunches[49],phenoliccompounds[50,51],andtheflavonolprofile[52].Duetothedirecteffect
ofbunchexposuretoradiationinthearomaprofilesobtainedinwines,researchershaveinvestigated
theeffectofdefoliationasamanagementstrategytoincreaseberryqualityandaromatraits,which
dependsonthecultivar,timingofdefoliation,andclimaticregion[53–60].Theseresearches
demonstratetheimportanceoffruitexposuretosolarradiationandmicroclimateconditionsthatare
favorabletothedevelopmentofberryqualitytraits.
Aspreviouslymentioned,seasonaltemperaturesnotonlyinfluencetheoccurrenceandlength
ofdifferentphenologicalstagesingrapevines,suchasbudbreak,flowering,berryset,peasize,
veraison,andharvest,butalsothechemicalandaromacompositionofberries.Ofcriticalimportance
istheinfluenceofweatherparameters,suchastemperature[61–64],andwateravailabilityfrom
veraisononwardsinredcultivars,whichisdeterminanttothefinalwinequalityandaromaprofiles.
Severalstudieshavefocusedonthepreandpostveraisonphenologicalstagesforirrigation
treatmentstoincreaseberryandwinequalitytraits,especiallyinredcultivars[65–69].
Formachinelearningmodeling,ithasbeendemonstratedthattheimplementationofimportant
parametersasinputsthatdirectlyinfluencethetargetsproposedrendermorerobustmodelsin
contrasttotheusageofrawdata.Basedoncalculatedparametersratherthanrawdatainputs,there
arerecentstudiesimplementingmachinelearningtoassessbeerquality[70–72],interpretremote
sensingdataforplantwaterstatusassessmentinvineyards[73],chocolatequalityassessmentby
consumersusingNIR[74],andaromaprofilesincocoatreesbasedoncanopyarchitectureparameters
[75].Inthisstudy,relevantparametersfromweatherconditions,managementstrategies,and
physicochemicalparametersofwineswereobtainedandconsideredasinputsinthemachine
learningmodeling,whichcanexplainthehighaccuracyobtainedforthepredictionsofModels1and
2withoutsignsofoverfitting.
TheuseofANNformodelinghastheadvantageofbeingabletousemultipletargets,which
makesthemodelsmoreefficient.Thisisduetotheeasinessoffeedingonlyonemodeltoobtainall
theoutputdatainsteadofhavingtoaddthenewinputstoseveralsingle‐targetmodels.Several
studiesrelatedtofoodandagriculturehaveusedthistypeofmachinelearningalgorithmswithhigh
performanceandaccuracy[38,71,72,75–77].
Thetechniqueproposedconsidersthereadilyavailableweatherinformationfromvintagesclose
tothevineyardsandaverticalvintagelibrary,whichmostwineriescanobtaineasily.Themodels
developedassumethatthevineyardmanagementisconsistentthroughouttheseasons,includingthe
winemakingtechniquesandyeastused.Theimplementationofthesemodelstoothercultivars,
environments,andregionswillneedtheincorporationoffurthersite‐specificdataasinputsandwine
chemicalandaromaprofileanalysisfromavailableandcontrastingvintages.Thelatterbenefitfrom
thelearningaspectofthemodelsproposed,whichdoesnotrequireafulldevelopmentofnew
analysesfordifferentregions.
5.Conclusions
Artificialintelligencetechniquescanbeimplementedinthewineindustryfromreadilyavailable
weatherandmanagementpracticesdatatoassessqualitytraitsinfinalwines.Modelingstrategies
usingartificialneuralnetworksdevelopedforparticularregionscanbeimplementedforother
cultivars,environments,andregionsbyincludingextremevaluesfromtheirrespectivevintages.
Highaccuracymodelstodeterminethearomaprofileofwinesbeforethewinemakingprocesscan
offerapowerfultooltogrowersandwinemakersforthedecisionmakinginthevinificationprocess
tomaintainorincreasewinequalityandstyles.Furtherresearchisrequiredtoadaptthesetechniques
tocanopymanagementstrategiesandwithin‐seasonmodelingthatcanbeimplementedinreal‐time
withintheseasontomanipulatethefinalwineandaromaprofilestospecifictargetsusing
managementstrategies,suchascanopy,fertilization,andirrigationmanagement.
SupplementaryMaterials:Thefollowingareavailableonlineatwww.mdpi.com/xxx/s1,TableS1:Meansand
standarderror(SE)ofthevolatilearomaticcompoundsandphysicochemicalparametersofthewinedfrom
eachvintage.

Foods2020,9,3312of15
AuthorContributions:Conceptualization,S.F.;Datacuration,S.F.,E.T.andC.G.V.;Formalanalysis,S.F.and
C.G.V.;Investigation,S.F.andC.G.V.;Methodology,D.D.T.andC.G.V.;Projectadministration,S.F.;Validation,
S.F.andC.G.V.;Visualization,S.F.andC.G.V.;Writing—originaldraft,S.F.andC.G.V.;Writing—review&
editing,E.T.andD.D.T.Allauthorshavereadandagreedtothepublishedversionofthemanuscript.
Funding:Thisresearchreceivednoexternalfunding
Acknowledgments:TheauthorsacknowledgecontributionsofXiaoyiWangandPangzhenZhangfor
preliminarydatahandling.
ConflictsofInterest:Theauthorsdeclarenoconflictofinterest.
References
1. Coombe,B.G.;Dundon,R.J.;Short,A.W.Indicesofsugar—Acidityasripenesscriteriaforwinegrapes.J.
Sci.FoodAgric.1980,31,495–502.
2. Suklje,K.;Blackman,J.;Deloire,A.;Schmidtke,L.;Antalick,G.;Meeks,C.Grapestowine:Thenexus
betweenberryripening,compositionandwinestyle.InProceedingsoftheXInternationalSymposiumon
GrapevinePhysiologyandBiotechnology1188,Verona,Italy,13–18June2016;pp.43–50.
3. Ickes,C.M.;Cadwallader,K.R.Effectsofethanolonflavorperceptioninalcoholicbeverages.Chemosens.
Percept.2017,10,119–134.
4. Sherman,E.;Greenwood,D.R.;Villas‐Boâs,S.G.;Heymann,H.;Harbertson,J.F.Impactofgrapematurity
andethanolconcentrationonsensorypropertiesofWashingtonStateMerlotwines.Am.J.Enol.Vitic.2017,
68,344–356.
5. Webb,L.;Whetton,P.;Bhend,J.;Darbyshire,R.;Briggs,P.;Barlow,E.Earlierwine‐graperipeningdriven
byclimaticwarminganddryingandmanagementpractices.Nat.Clim.Chang.2012,2,259.
6. Anderson,K.;Findlay,C.;Fuentes,S.;Tyerman,S.Viticulture,wineandclimatechange.InGarnautClimate
ChangeReview;CambridgeUniversityPress:Cambridge,UK,2008,PP.1–16.
7. Webb,L.;Whetton,P.;Barlow,E.Modelledimpactoffutureclimatechangeonthephenologyof
winegrapesinAustralia.Aust.J.GrapeWineRes.2007,13,165–175.
8. Webb,L.;Whetton,P.;Barlow,E.ObservedtrendsinwinegrapematurityinAustralia.Glob.Chang.Biol.
2011,17,2707–2719.
9. Krasnow,M.;Weis,N.;Smith,R.J.;Benz,M.J.;Matthews,M.;Shackel,K.Inception,progression,and
compositionalconsequencesofaberryshriveldisorder.Am.J.Enol.Vitic.2009,60,24–34.
10. Fuentes,S.;Sullivan,W.;Tilbrook,J.;Tyerman,S.Anovelanalysisofgrapevineberrytissuedemonstrates
avariety‐Dependentcorrelationbetweentissuevitalityandberryshrivel.Aust.J.GrapeWineRes.2010,16,
327–336,doi:10.1111/j.1755‐0238.2010.00095.x.
11. Bonada,M.;Sadras,V.;Moran,M.;Fuentes,S.Elevatedtemperatureandwaterstressacceleratemesocarp
celldeathandshrivelling,anddecouplesensorytraitsinShirazberries.Irrig.Sci.2013,31,1317–1331,
doi:10.1007/s00271‐013‐0407‐z.
12. Bonada,M.;Sadras,V.O.;Fuentes,S.Effectofelevatedtemperatureontheonsetandrateofmesocarpcell
deathinberriesofShirazandChardonnayanditsrelationshipwithberryshrivel.Aust.J.GrapeWineRes.
2013,19,87–94,doi:10.1111/ajgw.12010.
13. Molitor,D.;Junk,J.;Evers,D.;Hoffmann,L.;Beyer,M.Ahigh‐resolutioncumulativedegreeday‐based
modeltosimulatephenologicaldevelopmentofgrapevine.Am.J.Enol.Vitic.2014,65,72–80.
14. Acevedo‐Opazo,C.;Ortega‐Farias,S.;Fuentes,S.Effectsofgrapevine(VitisviniferaL.)waterstatuson
waterconsumption,vegetativegrowthandgrapequality:Anirrigationschedulingapplicationtoachieve
regulateddeficitirrigation.Agric.WaterManag.2010,97,956–964.
15. Santesteban,L.;Miranda,C.;Royo,J.Regulateddeficitirrigationeffectsongrowth,yield,grapequality
andindividualanthocyanincompositioninVitisviniferaL.cv.‘Tempranillo’.Agric.WaterManag.2011,98,
1171–1179.
16. Greven,M.;Green,S.;Neal,S.;Clothier,B.;Neal,M.;Dryden,G.;Davidson,P.RegulatedDeficitIrrigation
(RDI)tosavewaterandimproveSauvignonBlancquality?WaterSci.Technol.2005,51,9–17.
17. Casassa,L.;Keller,M.;Harbertson,J.Regulateddeficitirrigationaltersanthocyanins,tanninsandsensory
propertiesofCabernetSauvignongrapesandwines.Molecules2015,20,7820–7844.

Foods2020,9,3313of15
18. Permanhani,M.;Costa,J.M.;Conceição,M.;DeSouza,R.;Vasconcellos,M.;Chaves,M.Deficitirrigation
intablegrape:Eco‐physiologicalbasisandpotentialusetosavewaterandimprovequality.Theor.Exp.
PlantPhysiol.2016,28,85–108.
19. Romero,P.;Gil‐Muñoz,R.;delAmor,F.M.;Valdés,E.;Fernández,J.I.;Martinez‐Cutillas,A.Regulated
deficitirrigationbaseduponoptimumwaterstatusimprovesphenoliccompositioninMonastrellgrapes
andwines.Agric.WaterManag.2013,121,85–101.
20. Ju,Y.‐l.;Liu,M.;Tu,T.‐y.;Zhao,X.‐f.;Yue,X.‐f.;Zhang,J.‐x.;Fang,Y.‐l.;Meng,J.‐f.Effectofregulated
deficitirrigationonfattyacidsandtheirderivedvolatilesin‘CabernetSauvignon’grapesandwinesof
Ningxia,ChinaFoodChem.2018,245,667–675.
21. Dry,P.;Loveys,B.Factorsinfluencinggrapevinevigourandthepotentialforcontrolwithpartialrootzone
drying.Aust.J.GrapeWineRes.1998,4,140–148.
22. dosSantos,T.P.;Lopes,C.M.;Rodrigues,M.L.;deSouza,C.R.;Maroco,J.P.;Pereira,J.S.;Silva,J.R.;Chaves,
M.M.Partialrootzonedrying:Effectsongrowthandfruitqualityoffield‐growngrapevines(Vitisvinifera).
Funct.PlantBiol.2003,30,663–671.
23. Bindon,K.;Dry,P.;Loveys,B.Influenceofpartialrootzonedryingonthecompositionandaccumulation
ofanthocyaninsingrapeberries(Vitisviniferacv.CabernetSauvignon).Aust.J.GrapeWineRes.2008,14,
91–103.
24. Gil,P.;Lobos,P.;Duran,K.;Olguin,J.;Cea,D.;Schaffer,B.Partialroot‐zonedryingirrigation,shading,or
mulchingeffectsonwatersavings,productivityandqualityof‘Syrah’grapevines.Sci.Hortic.2018,240,
478–483.
25. Gotur,M.;Sharma,D.;Joshi,C.;Rajan,R.Partialroot‐zonedryingtechniqueinfruitcrops:Areviewpaper.
IJCS2018,6,900–903.
26. Urraca,R.;Sanz‐Garcia,A.;Tardaguila,J.;Diago,M.P.Estimationoftotalsolublesolidsingrapeberries
usingahand‐heldNIRspectrometerunderfieldconditions.J.Sci.FoodAgric.2016,96,3007–3016.
27. Wu,G.;Huang,L.;He,Y.Researchonthesugarcontentmeasurementofgrapeandberriesbyusing
Vis/NIRspectroscopytechnique.GuangPuXueYuGuangPuFenXi=GuangPu2008,28,2090–2093.
28. Ribera‐Fonseca,A.;Noferini,M.;Jorquera‐Fontena,E.;Rombolà,A.D.Assessmentoftechnological
maturityparametersandanthocyaninsinberriesofcv.Sangiovese(VitisviniferaL.)byaportablevis/NIR
device.Sci.Hortic2016,209,229–235.
29. Chen,S.;Zhang,F.;Ning,J.;Liu,X.;Zhang,Z.;Yang,S.Predictingtheanthocyanincontentofwinegrapes
byNIRhyperspectralimaging.FoodChem.2015,172,788‐793.
30. Ferrer‐Gallego,R.;Hernández‐Hierro,J.M.;Rivas‐Gonzalo,J.C.;Escribano‐Bailón,M.T.Determinationof
phenoliccompoundsofgrapeskinsduringripeningbyNIRspectroscopy.LWTFoodSci.Technol.2011,44,
847–853.
31. GajdošKljusurić,,J.;Mihalev,K.;Bečić,,I.;Polović,,I.;Georgieva,M.;Djaković,,S.;Kurtanjek,Ž.Near‐
infraredspectroscopicanalysisoftotalphenoliccontentandantioxidantactivityofberryfruits.Food
Technol.Biotechnol.2016,54,236–242.
32. Herrera,J.;Guesalaga,A.;Agosin,E.Shortwave–nearinfraredspectroscopyfornon‐destructive
determinationofmaturityofwinegrapes.Meas.Sci.Technol.2003,14,689.
33. Larraín,M.;Guesalaga,A.R.;Agosín,E.Amultipurposeportableinstrumentfordeterminingripenessin
winegrapesusingNIRspectroscopy.IEEETrans.Instrum.Meas.2008,57,294–302.
34. Segade,S.R.;Giacosa,S.;Gerbi,V.;Rolle,L.GrapeMaturityandSelection:AutomaticGrapeSelection.In
RedWineTechnology;Elsevier:Amsterdam,TheNetherlands,2019;pp.1‐16.
35. Soubry,I.;Patias,P.;Tsioukas,V.MonitoringVineyardswithUAVandMulti‐sensorsfortheassessment
ofWaterStressandGrapeMaturity.J.UnmannedVeh.Syst.2017,5,37–50.
36. Matese,A.;DiGennaro,S.;Miranda,C.;Berton,A.;Santesteban,L.Evaluationofspectral‐basedand
canopy‐basedvegetationindicesfromUAVandSentinel2imagestoassessspatialvariabilityandground
vineparameters.Adv.Anim.Biosci.2017,8,817–822.
37. Matese,A.;DiGennaro,S.Practicalapplicationsofamultisensoruavplatformbasedonmultispectral,
thermalandrgbhighresolutionimagesinprecisionviticulture.Agriculture2018,8,116.
38. GonzalezViejo,C.;Fuentes,S.;Torrico,D.D.;Godbole,A.;Dunshea,F.R.Chemicalcharacterizationof
aromasinbeerandtheireffectonconsumersliking.FoodChem.2019,293,479‐485.
39. TheGoodScentsCompany.TheGoodScentsCompanyInformationSystem.Availabeonline:
http://www.thegoodscentscompany.com/data/rw1038291.html(accessedon3September2019).

Foods2020,9,3314of15
40. Genovese,A.;Piombino,P.;Gambuti,A.;Moio,L.Simulationofretronasalaromaofwhiteandredwine
inamodelmouthsystem.Investigatingtheinfluenceofsalivaonvolatilecompoundconcentrations.Food
Chem.2009,114,100–107.
41. Arcari,S.G.;Caliari,V.;Sganzerla,M.;Godoy,H.T.VolatilecompositionofMerlotredwineandits
contributiontothearoma:Optimizationandvalidationofanalyticalmethod.Talanta2017,174,752–766.
42. Jackson,M.G.;Timberlake,C.F.;Bridle,P.;Vallis,L.Redwinequality:Correlationsbetweencolour,aroma
andflavourandpigmentandotherparametersofyoungBeaujolais.J.Sci.FoodAgric.1978,29,715–727.
43. Rankine,B.;Fornachon,J.;Boehm,E.;Cellier,K.lnfluenceofgrapevariety,climateandsoilongrape
compositionandonthecompositionandqualityoftablewines.VITISJ.GrapevineRes.2017,10,33.
44. Sáenz‐Navajas,M.‐P.;Echavarri,F.;Ferreira,V.;Fernández‐Zurbano,P.Pigmentcompositionandcolor
parametersofcommercialSpanishredwinesamples:Linkagetoqualityperception.Eur.FoodRes.Technol.
2011,232,877–887.
45. Tariq,M.;Ali,M.;Shah,Z.Characteristicsofindustrialeffluentsandtheirpossibleimpactsonqualityof
undergroundwater.SoilEnviron.2006,25,64–69.
46. Karataş,,D.;Aydin,I.;Karataş,H.ElementalcompositionofredwinesinSoutheastTurkey.CzechJ.Food
Sci.2015,33,228–236.
47. Balint,G.;Reynolds,A.G.Irrigationlevelandtimeofimpositionimpactvinephysiology,yield
components,fruitcompositionandwinequalityofOntarioChardonnay.Sci.Hortic.2017,214,252–272.
48. Romero,P.;García,J.G.;Fernández‐Fernández,J.I.;Muñoz,R.G.;delAmorSaavedra,F.;Martínez‐Cutillas,
A.Improvingberryandwinequalityattributesandvineyardeconomicefficiencybylong‐termdeficit
irrigationpracticesundersemiaridconditions.Sci.Hortic.2016,203,69–85.
49. Martin,D.;Grose,C.;Fedrizzi,B.;Stuart,L.;Albright,A.;McLachlan,A.Grapeclustermicroclimate
influencesthearomacompositionofSauvignonblancwine.FoodChem.2016,210,640–647,
doi:10.1016/j.foodchem.2016.05.010.
50. Brillante,L.;Martínez‐Lüscher,J.;Kurtural,S.K.Appliedwaterandmechanicalcanopymanagementaffect
berryandwinephenolicandaromacompositionofgrapevine(VitisviniferaL.,cv.Syrah)inCentral
California.Sci.Hortic.2018,227,261–271,doi:10.1016/j.scienta.2017.09.048.
51. Song,J.;Smart,R.;Wang,H.;Dambergs,B.;Sparrow,A.;Qian,M.C.Effectofgrapebunchsunlight
exposureandUVradiationonphenolicsandvolatilecompositionofVitisviniferaL.cv.Pinotnoirwine.
FoodChem.2015,173,424–431,doi:10.1016/j.foodchem.2014.09.150.
52. Martínez‐Lüscher,J.;Brillante,L.;Kurtural,S.K.FlavonolProfileIsaReliableIndicatortoAssessCanopy
ArchitectureandtheExposureofRedWineGrapestoSolarRadiation.Front.PlantSci.2019,10,10,
doi:10.3389/fpls.2019.00010.
53. Radovanović,,V.;Stefanović,,D.;Radovanović,,B.Influenceofselectiveremovalofgrapevineleaveson
qualityofredwine.J.Process.EnergyAgric.2015,19,215–218.
54. Verdenal,T.;Zufferey,V.;Dienes‐Nagy,A.;Bourdin,G.;Gindro,K.;Viret,O.;Spring,J.‐L.Timingand
IntensityofGrapevineDefoliation:AnExtensiveOverviewonFiveCultivarsinSwitzerland.Am.J.Enol.
Vitic.2019,70,427–434.
55. Vargas,S.;Cazorla,M.;Bordeu,E.;Casaubon,G.;González,Á.Evaluationofleafremovalstrategiesand
clusterradiationprotectionongrapeandwinequalityofVitisviniferaL’CabernetSauvignonʹ.In
ProceedingsoftheXInternationalSymposiumonGrapevinePhysiologyandBiotechnology1188,Verona,
Italy,13–18June2016;pp.97–104.
56. Pessenti,I.L.;Ayub,R.A.;Botelho,R.V.Defoliation,applicationofS‐ABAandvegetalextractsonthe
qualityofgrapeandwineMalbeccultivar.Rev.Bras.deFrutic.2019,41,doi:10.1590/0100–29452019018.
57. Wang,Y.;He,L.;Pan,Q.;Duan,C.;Wang,J.EffectsofBasalDefoliationonWineAromas:AMeta‐Analysis.
Molecules2018,23,779.
58. Scafidi,P.;Barbagallo,M.;Pisciotta,A.;Mazza,M.;Downey,M.Defoliationoftwo‐wireverticaltrellis:
Effectongrapequality.N.Z.J.CropHortic.Sci.2018,46,18–38.
59. Peña‐Olmos,J.E.;Casierra‐Posada,F.FruitqualityandproductionofVitisviniferaL.Chardonnayaffected
bypartialdefoliationintropicalhighlands.Rev.Fac.Nac.DeAgron.Medellín2015,68,7581–7588.
60. Baiano,A.;DeGianni,A.;Previtali,M.A.;DelNobile,M.A.;Novello,V.;dePalma,L.Effectsofdefoliation
onqualityattributesofNerodiTroia(VitisviniferaL.)grapeandwine.FoodRes.Int.2015,75,260–269.
61. Jackson,D.;Lombard,P.Environmentalandmanagementpracticesaffectinggrapecompositionandwine
quality‐Areview.Am.J.Enol.Vitic.1993,44,409–430.

Foods2020,9,3315of15
62. Drappier,J.;Thibon,C.;Rabot,A.;Geny‐Denis,L.Relationshipbetweenwinecompositionand
temperature:ImpactonBordeauxwinetypicityinthecontextofglobalwarming.Crit.Rev.FoodSci.Nutr.
2019,59,14–30.
63. Kliewer,W.M.;Torres,R.E.Effectofcontrolleddayandnighttemperaturesongrapecoloration.Am.J.
Enol.Vitic.1972,23,71–77.
64. Teixeira,A.;Eiras‐Dias,J.;Castellarin,S.D.;Gerós,H.Berryphenolicsofgrapevineunderchallenging
environments.Int.J.Mol.Sci.2013,14,18711–18739.
65. Romero,P.;Fernández,J.;Botía,P.Interannualclimaticvariabilityeffectsonyield,berryandwinequality
indicesinlong‐termdeficitirrigatedgrapevines,determinedbymultivariateanalysis.Int.J.WineRes.2016,
8,3–17.
66. Intrigliolo,D.;Lizama,V.;García‐Esparza,M.;Abrisqueta,I.;Álvarez,I.Effectsofpost‐veraisonirrigation
regimeonCabernetSauvignongrapevinesinValencia,Spain:Yieldandgrapecomposition.Agric.Water
Manag.2016,170,110–119.
67. Jara,J.;Holzapfel,E.A.;Billib,M.;Arumi,J.L.;Lagos,O.;Rivera,D.Effectofwaterapplicationonwine
qualityandyieldin‘Carménère’underthepresenceofashallowwatertableinCentralChile.Chil.J.Agric.
Res.2017,77,171–179.
68. Romero,P.;Fernández‐Fernández,J.;Bravo‐Cantero,A.;Ayala,M.;Botía,P.ClimateInfluencesonYield,
BerryandWineQualityinMonastrellWineGrapesinaWarmWinegrowingRegion(JumillaArea,SE
Spain).GeoinforGeostatOverv.2016,4,3.8:2.
69. Conesa,M.R.;Falagán,N.;José,,M.;Aguayo,E.;Domingo,R.;Pastor,A.P.Post‐veraisondeficitirrigation
regimesenhanceberrycolorationandhealth‐promotingbioactivecompoundsin‘CrimsonSeedless’table
grapes.Agric.WaterManag.2016,163,9–18.
70. Viejo,C.G.;Fuentes,S.;Howell,K.;Torrico,D.;Dunshea,F.R.Roboticsandcomputervisiontechniques
combinedwithnon‐invasiveconsumerbiometricstoassessqualitytraitsfrombeerfoamabilityusing
machinelearning:Apotentialforartificialintelligenceapplications.FoodControl.2018,92,72–79.
71. GonzalezViejo,C.;Fuentes,S.;Torrico,D.;Howell,K.;Dunshea,F.R.Assessmentofbeerqualitybasedon
foamabilityandchemicalcompositionusingcomputervisionalgorithms,nearinfraredspectroscopyand
machinelearningalgorithms.J.Sci.FoodAgric.2018,98,618–627.
72. Viejo,C.G.;Torrico,D.D.;Dunshea,F.R.;Fuentes,S.DevelopmentofArtificialNeuralNetworkModelsto
AssessBeerAcceptabilityBasedonSensoryPropertiesUsingaRoboticPourer:AComparativeModel
ApproachtoAchieveanArtificialIntelligenceSystem.Beverages2019,5,33.
73. Romero,M.;Luo,Y.;Su,B.;Fuentes,S.Vineyardwaterstatusestimationusingmultispectralimageryfrom
anUAVplatformandmachinelearningalgorithmsforirrigationschedulingmanagement.Comput.
Electron.Agric.2018,147,109–117.
74. Gunaratne,T.M.;GonzalezViejo,C.;Gunaratne,N.M.;Torrico,D.D.;Dunshea,F.R.;Fuentes,S.Chocolate
QualityAssessmentBasedonChemicalFingerprintingUsingNearInfra‐redandMachineLearning
Modeling.Foods2019,8,426.
75. Fuentes,S.;Chacon,G.;Torrico,D.D.;Zarate,A.;GonzalezViejo,C.SpatialVariabilityofAromaProfiles
ofCocoaTreesObtainedthroughComputerVisionandMachineLearningModelling:ACover
PhotographyandHighSpatialRemoteSensingApplication.Sensors2019,19,3054.
76. Fuentes,S.;Tongson,E.J.;DeBei,R.;GonzalezViejo,C.;Ristic,R.;Tyerman,S.;Wilkinson,K.Non‐Invasive
ToolstoDetectSmokeContaminationinGrapevineCanopies,BerriesandWine:ARemoteSensingand
MachineLearningModelingApproach.Sensors2019,19,3335.
77. GonzalezViejo,C.;Fuentes,S.;Torrico,D.;Dunshea,F.Non‐ContactHeartRateandBloodPressure
EstimationsfromVideoAnalysisandMachineLearningModellingAppliedtoFoodSensoryResponses:
ACaseStudyforChocolate.Sensors2018,18,1802.



©2019bytheauthors.LicenseeMDPI,Basel,Switzerland.Thisarticleisanopenaccess
articledistributedunderthetermsandconditionsoftheCreativeCommonsAttribution
(CCBY)license(http://creativecommons.org/licenses/by/4.0/).

