GUIDE
forthe
Gradingoflucernehay
2015
Dr.GDJScholtz LTC♥Prof.HJvdMerwe
Inordertocreateanaccurateanduniformgradingsystem,itisimperativethatallprescribedproceduresregardingsampling,thehandlingofsamples,NIRanalyses,identificationofoutliers,qualitystandardsandcertificationbemeticulouslyapplied.1. SamplingThefollowingprocedureshouldbetakenduringsampling:1.1 Identify a single lot of lucerne bales, in other words, a homogeneous
amountoflucernebaleswith:
* samecut* samevariety* samefarmland* samephysiologicalstage* cutwithinaperiodof48hours
1.2 Samplingshouldbedonerandomly
This implies that nopredetermined reason should exist for a particularbaletobeselectedand/orbeselectedonaccountofcolour,leafiness,etc.
1.3 Sampler/Probespecifications
A standardized hollow core probe (Fig. 1) as prescribed by the NLT, isusedtoestablishquickandeasy,andrepresentativesampling.
Figure1. Prescribedlucernehayprobe
1.4 Thecorrect drilling techniqueshould be applied (Alsoseehayprobe
instructions)
* Insert the drill into the centre of the topside of the bale at 90ºdegrees or in the centre of the side of a bale at an angle of 45ºdegrees.
* The drill should be inserted to at least 325mm (as indicated ondrill).
* Thebitofthedrillshouldbesharpenedconstantly.1.5 Amplesub-samplesextractedproportionatelyspreadoverthelotshould
betaken.* aminimumof20isneededtolimitvariation.*numberofsub-samplesperbale *smallbales=one
*largeorroundbales=two1.6 Samplesize
* 180to250gisexpedient.* A sample less than 180g would be deemed to be non-
representative.
* If a sample should exceed 250g, handling during the grindingprocesswould be impaired, because the entire sample has to bemilled. If samples are found to be too large, it is normally theresultofthedrillbeingtoolarge.
1.7 Handlingofsamples
* Unmilled sub-samples should be immediately combined,
thoroughlymixedandsealedintoair-tightbagsfortransportingtotheclosestNLT-registeredNIR-instrument,whereapplicable.
* Allsamplesshouldbeprotectedagainstheatanddirectsunlight.
1.8 Divisionofasample
* unmilledsamplesshould,undernocircumstances,besub-divided.
2. Handlingofsamples2.1 PertenDA7200instrument
Equipment(collectionbowl,plasticbag,scraper)(Fig.2)
Figure2. Gradingequipment
Step1: Poursamplefromcollectionbottleintoplasticbag.Step2: Mixthoroughlyinplasticbag.(Fig3).
Figure3. Mixsampleinplasticbag.
Step3: Dividesampleequallybetweentwobowls.(Fig.4)
Figure4. DivisionofsampleStep4: To level, scrape bowls slightly with a ruler or other flat
object,withoutcompactingcontents(Fig.5)
Figure5. LevellingofsampleStep5: Takereadingsofbothbowls.(Fig.6)
Figure6. TakingaNIRreading
Step6: Pour contents of bowls and residue back into plastic bag
andrepeatsteps1to4(Fig.7)toobtainaveragereadingsfor4bowls.
Figure7. Repetitionofprocess
2.2 FossinstrumentStep1: Grind entire sample in collection bottle by way of a
prescribedlaboratorymillwitha1mmsieve(Fig.8).Thismillshouldbeservicedannuallytoensurecorrectfineness.
Figure8. PrescribedlaboratorymillStep2: Mix sample thoroughly in bottle before transferring to
applicableFosscontainer,placeinNIR-instrumentandtakeareading.
Step3: Clean mill thoroughly after every milling of a sample to
preventcontamination.
3. IdentifyingoutliersItisoftheutmostimportancethatoutliersshouldbeidentifiedandthoseparticularsamplessavedandpresentedtotheNLTforfurthercalibrationdevelopment.Thefollowingprocessshouldbetakenfortheidentificationofoutliers.
3.1 PertenOutlierIdentificationforLucernesamplesUsingaDA72XXDiodeArrayBasedNIRAnalysisSystemNIRtechnologyhastheabilitytoindicatewhetherornotsamplespresentedtoinstrumentswillmatchtheexistingcalibrationbeingused.ThePertenDA72XXmakesuseofMahalanobisdistancesto indicateacceptableorunacceptablespectralmatchesagainstexistingcalibrations.
Onceananalysishasbeencompleted,asamplewillbeflaggedunderthe‘match’columnaseither‘yes’,‘possible’or‘no’.‘Yes’indicates:1.Goodspectralagreementwiththecalibrationsamplespectraaccordingtothe”limits” in the “Mahalanobis Distance / Leverage” setting (done uponinstallation)2. The component/parameter results are within the limits expected for theproduct.3.Thecomponent/parameterresultsarewithinthelimitssetinthecalibrationmodel.‘Possible’indicates:Fair spectral agreement with the calibration sample spectra according to the”limits” in the Mahalanobis Distance / Leverage” setting (done uponinstallation)‘No’indicates:1.Poorspectralagreementwiththecalibrationsamplespectraand/or…2.thatthecomponent/parameterresultsareoutsidethelimitsexpectedfortheproductand/or…3.thatthecomponent/parameterresultsareoutsidethelimitsofthecalibrationmodel.‘Yes’resultscanbeacceptedascorrectandcanbeusedinconjunctionwiththeNLTguidelinesforgradingofLucerne.‘Possible’ resultsshouldbere-analysedbeforeofficialgrading takesplace. Ifasecond ‘Possible’ result is obtained, the sample should be retained for furthercalibrationpurposes.‘No’ results should be retained immediately for further calibration purposes.Thesesampleswillbeusedtoensurethatfutureversionsofthecalibrationaremademore robust and these sample’s spectrawill be added to the calibrationdatabase.‘No’samplesaretobereportedimmediatelytotheNLT.Thesesampleswillmostprobablycontainexcess foreignmaterialwhichmayberemovedandthesamplere-analysed.Ifafurther‘No’matchisobtained,theresultscannotbeusedandshouldbereportedtotheNLTassuch.
3.2 FossOutlieridentification
ForaquickgraphicalexplanationofHoutliers,pleaseseeOutlieridentification.
InfrasoftInternational(ISI)hasbeenusingtheconceptofHoutliersasatestfortheconfidencethatcanbeplacedinthepredictedvaluessince1980.AsNIRSisasecondarymeasurement(ameasurementbasedonaprimaryreferencesuchaswet chemistry), it is important that some measure of confidence in themeasurementisexpressed.TheunderlyingtheoryintheHtestisthatthecloser
theunknownistothesamplesthatwereusedcalibration,themoreconfidencetheusercanhavethattheunknownwillbecorrectlymodelled.
Houtliersaredeterminedsolelyon theunknownsample'sspectrum,andhaveno relation to the predicted values. They are calculated by comparing thespectrumoftheunknownsampletothatofthesamplesthatwereusedtocreatethecalibration.
HoutliersarebasedonMahalanobisdistancemeasurements.Thisisastatisticalcalculationbasedon thevariance in thepopulation's spectra.TheMahalanobisdistancecanbethoughtofasdescribinghowwelltheunknownsamplefitswiththesamplesthatwereusedtocreatethecalibration.ThelowertheHvalues,thecloserthefit.Iftheunknownsamplehasaspectrumthatisverysimilartothosein the calibration set, itwillhave lowHvaluesand it canbeassumed that thecalibration will do a good job of predicting that sample. Conversely, if theunknown isverydifferent from thecalibrationsamplesandhashighHvalues,thesampledoesnotfitinthecalibrationandlessconfidenceshouldbeplacedinthepredictedvalues.
There are two types of H outliers that ISIscan™ uses; G H (Global) and N H(Neighborhood).Ineachcase,thelowertheHvalue,thehighertheconfidenceinthe predicted values. G H describes the fit of the unknown in the calibrationpopulation.Generally,sampleswithaGH>3.0areconsideredoutliers,butthisissomewhatcalibrationandproductdependent.
NH indicates how close the nearest neighbouring calibration sample is to theunknown. The closer the sample readings, the higher the confidence that theunknownismodelledwellinthecalibration.AhighNH(>1.0),indicateseitherthat the unknown has found a "hole" in the calibration or that the sample isoutsidethecalibrationset. Inthefirstcase,theunknownwillhaveaGH<3.0,indicatingthatthesamplefits intheoverallpopulation. Ifpossible, thissampleshould be saved and addedback to the calibration set to "fill the hole." In thesecondexample,thesamplewillhaveaGH>3.0alongwiththeNH>1.0.Thisunknown does not belong and should either be repredicted with a differentpredictionmodelornotused.
BothGHandNHcalculationsstartwithaPCA(principalcomponentanalysis)that analyzes the samples in the calibration set and characterizes all samplesaccording toprincipalcomponentscores.Theunknownsare thencompared tothisdataset.
4. Qualitystandards
Qualitystandards–NewLucerneQualityIndex(NLQI)Grade/Class NLQI1 Foreignmaterial(%)Prime 104 &
higherAbsent
1 97-103 Absent
2 93-96 Present
3 92&less Present HJ van der Merwe & GDJ Scholtz, 2010
1 = New lucerne quality index (NLQI)
� Maximum moisture allowed to prevent mould and heating, as determined by a
NIR instrument = 16%.
� Loads containing harmful plants and proclaimed weeds are returned (Fig. 9).
� Loads issuing strange smells, mould and any signs of over-heating can be
downgraded or returned.
� Loads which can obviously be identified as of mixed quality will be classified
as the lowest grade within the load.
Figure9. Forbiddenandharmfulplants
5. CertificationonPerteninstruments PrintingofNLTCertificateusingPertenDA7200/DA7250:After the sample has been correctly prepared and analysed by the PertenDA7200orPertenDA7250,theresultspageisautomaticallydisplayed.
At therightbottomcornerof thepage,youwillhaveabutton“ExportGradingxml”.
When you click on this button, another pop upwill appearwhere youwill berequiredtocomplete:
“GenerateGradingCertificateXML”isthefirstpopupthatwillcomeup.InthereyouwillneedtoentertheBuyerandSeller.Whenyouclickonthebrowsebuttonnext to the Buyer/Seller fields, another pop up will come up where you canSelect,Insert,ChangeorDeleteaBuyer/Seller.WhenyouhavecompletedtheBuyerandSeller fields,clickon“GenerateXML”button.YouwillbetakenbacktotheResultsscreen.
Nextstep,youwillneedtogototheNLTCertificatePrintingSoftware.
The NLT Software will automatically bring the results through into the NLTprogram.On the right, there is a column that is labelled “Printed”. New certificates thathavebeenbrought inbutnotbeen completedandprintedwill nothave a tickmarknexttoit.Clickontheappropriatecertificateandclickon“PreviewCertificate”whichwilltakeyoutothedetailsofthecertificate.
Inthisscreen,alloptionsexcept“Batch/TruckNo”.arerequiredfields.Click on: “Save and Print” to update the certificate details and preview thecertificate.Youwillthenbetakentotheactualcertificatewhereyoucanprintitout.
5.1 CertificationCertificate
5.1.1 InterpretationofanalysesNLQIiscalculatedfromthefollowinganalyses:1. ADF
ADFresultscanvarybetween21and47%.LowerADFisconcomitantwithahigherNLQI.Inpracticethiscouldbeattainedbyearliercuttingandlimitingofleaf-lossduringthehaymakingprocess.
2. ASHTheashanalysesmayvarybetween7and30%. Lowerash contentwillleadtoahigherNLQIvalue.Avalueras14%isanindicationofextremelyhighsoilcontamination,whichshouldbelimitedduringthehaymakingprocess.Furthermore,ahighcalciumfertilizerwithahighK–andtherefore,ashcontent–isconcomitant. Extremelylowashvaluesshouldbe treatedand interpretedwith theutmostcare. Thiscanbeanindicationofanoutlierwhichcannormallyoccurduringanyseason.AvalueofthiskindshouldbereportedtotheNLTassoonaspossible,forinvestigationandpossiblerectification.
3. LIGNINThe lignin content may vary between 4 and 16%. Low lignin isconducive forahigherNLQIvalue.Onceagain, thismaybeachievedwithanearliercuttingstageandlimitationofleaf-loss.
The inter-relationshipbetweendifferentparametersmayvary fromseasontoseason,thusinfluencingthelucernehaymatrix.Suchadifferentiationcansubsequentlyhaveanimpactonthelucernehaycalibration.Participationintheannualvalidationprocessand identificationofoutliers isof theutmostimportancetoascertainconsistentpredictionsbyNIRinstruments.Values outside the abovementioned confines are to beconsidered as outliers and should be saved for continuinganalysesandcalibrationdevelopmentpurposes. Furthermore,die NLQI value of this particular lucerne hay sample is not acorrect indication of lucerne quality. The NLT should becontactedinsuchcasesforrecommendations.
LucerneTechConsult(LTC)hayprobeInstructions
The patented design of the LTC-Probe™ allows for quick and representativesamplingoflargestacksofhay.Theprobeisself-unloadingandissuppliedwithacollectingbottle(capacityfor20coresamples)andacleaningrodwithplasticprotecting sleeve, which protects the operator against personal injuries whenusingthecleaningrod,andalsoprotectingthecuttingedgeoftheprobeduringstorage.Twofollowingsamplingproceduresareprescribedtogetarepresentativecoresamplefrombaledhay:Removethecleaningrodandpushtheprobeintoeitherendofthebalenearthecenterata90°angle,oratthesideofthebaleata45°angle.Besuretopushtheentire shaft into the bale. By using the cleaning rod, drive each core into thecollectionbottleaftereachsamplingattempt.Repeatthisproceduretoobtainaminimumof20samplespertruckload.Wethaywillwedgeintheprobeifmorethenonesampleistakenasdescribedintheaboveprocedure.Thecuttingedgecanbesharpenedwithasmallroundfilefromtimetotime.Donot change the angle of the cutting edge. For protection of the cutting edgealwaysreplacethecleaningrodandprotectionsleevebackontotheprobe.This probe should be used according to the standardised NLT sampling protocol asdiscussed in theNLT lucernehaygradingguide. Forany informationregarding theLTCHay Probe or Sampling protocol please contact our office at 0613104100 or [email protected].