evaluation of a commercial grape yield monitor for use mid...

This work is licensed under a Creative Commons Attribution 4.0 International License

Newcastle University ePrints - eprint.ncl.ac.uk

Taylor JA, Sanchez L, Sams B, Haggerty LL, Jakubowski R, Djafour S, Bates TR.

Evaluation of a commercial grape yield monitor for use mid-season and at-

harvest. Journal International des Sciences de la Vigne et du Vin 2016, 50(2).

Copyright:

This paper is published with open access by Institut des Sciences de la Vigne et du Vin

DOI link to article:

http://dx.doi.org/10.20870/oeno-one.2016.50.2.784

Date deposited:

24/08/2016

http://creativecommons.org/licenses/by/4.0/

http://eprint.ncl.ac.uk/

http://dx.doi.org/10.20870/oeno-one.2016.50.2.784

EVALUATION OF A COMMERCIAL GRAPE YIELD MONITOR FOR USE MID-SEASON AND AT-HARVEST

James A. Taylor1,4, Luis Sanchez2, Brent Sams2, Luke Haggerty3, Rhiann Jakubowski3, Sarah Djafour1 and Terence R. Bates1

1 : Cornell Lake Erie Research and Extension Laboratory, School of Integrative Plant Science, Cornell University, 6592 West Main St, Portland, NY, 14769, United States

2 : E&J Gallo Winery, PO Box 1130, Modesto, CA, United States3 : Cornell Cooperative Extension, Cornell Lake Erie Research and Extension Laboratory,

6592 West Main St, Portland, NY, 14769, United States4 : School of Agriculture, Food and Rural Development, The University of Newcastle upon Tyne,

Cockle Park Farm, Morpeth, NE61 3EB, United Kingdom

*Corresponding author : [email protected]

Aims : Yield monitors are becoming more common inNorth America. This research evaluates the precision andaccuracy of a retro-fitted, commercially available grapeyield monitor mid-season, for crop estimation and cropthinning applications, and at harvest for yield mapping.Methods and Results: Several grape yield monitors weremounted on the discharge conveyor belt of grape harvestersin both commercial and research vineyards in NorthAmerica. Sensor response was compared to manualmeasurements at multiple masses, ranging from 20 kg to28 Mg over the course of three seasons. Measurementswere taken during crop thinning and estimation (mid-season) and at harvest. Results showed that the grape yieldmonitor performance was sufficient to generate goodspatial maps of the relative variation in harvest yield andmid-season thinned yield. However, at harvest the sensorshowed a shift in response between days of up to ±15 %,such that the generation of absolute yield maps required adaily calibration against a known mass. Within a day(single harvest operation) the sensor response did notappear to drift. Mid-season applications required a differentcalibration to harvest applications.Conclusion: The yield sensor worked well for both mid-season and at harvest operations in North Americanvineyards but required a daily calibration to avoid driftissues. The mid-season yield calibrations were differentbetween seasons ; however, the harvest calibration factorwas stable between seasons.Significance and Impact of study: The study showed thata commercial yield monitor with correct calibration waseffective at even low fruit flow. This opens the possibilityof using a harvest sensor mid-season to mechanicallyestimate fruit load from small point samples and to map theamount of fruit removed during fruit thinning operations.This will improve the quality of information available toviticulturist to understand fruit and crop load. Thecommercial yield monitor is suitable for use in NorthAmerican vineyards.Keywords : on-the-go proximal sensing, yield mapping,crop estimation.

Objectifs: Les capteurs de rendement deviennent de plusen plus communs en Amérique du nord. Cette étude vise àévaluer la précision et l’exactitude d’un capteur derendement viticole disponible dans le commerce, étalonnéavec des valeurs réelles. Il est utilisable à mi-saison pourestimer la récolte et durant les vendanges pourcartographier le rendement.Méthodes et résultats : Plusieurs capteurs de rendementviticole ont été montés sur le convoyeur de décharge demachines à vendanger, à la fois au sein de vignoblescommerciaux et de recherche en Amérique du nord. Laréponse du capteur a été comparée à des mesures manuellespour plusieurs masses, allant de 20 kg à 28 Mg au cours detrois saisons. Les mesures ont été réalisées au cours del’éclaircissage et de la prédiction de récolte (à mi-saison) etdurant les vendanges. Les résultats ont montré que laperformance du capteur de rendement viticole étaitsuffisante pour cartographier de manière fidèle la variationrelative du rendement à la récolte et celui de l’éclaircissageà mi-saison. Cependant lors des vendanges, le capteur amontré un décalage dans sa réponse selon les jours, allantjusqu’à 15 %. Au sein d’une journée (pour une vendangeunique), la réponse du capteur ne semble pas dériver.L’utilisation à mi-saison exige un étalonnage différent decelui pour les vendanges.Conclusion : Le capteur de rendement fonctionne biendans les vignobles d’Amérique du nord mais demande unétalonnage quotidien afin d’éviter des problèmes de dérive.Les étalonnages de mi-saison et de récolte sont différents,bien que le facteur d’étalonnage durant les vendanges a étéstable au fil des saisons.Signification et impact de l’étude : L’étude a montréqu’un capteur de rendement disponible dans le commerceavec un étalonnage correct a été efficace, même pour unfaible flux de raisins. Cela ouvre les possibilités d’utiliserun capteur de rendement à mi-saison afin d’estimermécaniquement la charge en fruit et de cartographier lesfruits enlevés lors de l’éclaircissage. Cela permettrad’améliorer la qualité de l’information disponible pour lesviticulteurs afin de comprendre et interpréter la charge enfruit et la production potentielle.

Abstract Résumé

J. Int. Sci. Vigne Vin, 2016, 50, 2, 57-63©Vigne et Vin Publications Internationales (Bordeaux, France)- 57 -

manuscript received 28th July 2015 - revised manuscript received 19th April 2016

01_taylor_05b-tomazic 05/07/16 14:13 Page57

- 58 -J. Int. Sci. Vigne Vin, 2016, 50, 2, 57-63©Vigne et Vin Publications Internationales (Bordeaux, France)

James A. Taylor et al.

INTRODUCTION

On-harvester grape yield monitors have beencommercially available since the late 1990s. The firstcommercially available grape yield monitor was theHM-570 grape yield monitor released byHarvestMaster (Logan, UT, USA). The HM-570 wasa volumetric sensor that used ultra-sonic sensorsmounted above the discharge conveyor to measurethe height (volume) of grapes on the conveyor belt.Volume was then calibrated to mass using a densitycoefficient. Considerable issues with the operation ofthe system in Australian conditions between 1999and 2002 were noted by researchers at CSIRO andThe University of Sydney (Taylor, 2004). Inparticular, the density coefficient required constantadjustment, not only between varieties but alsothroughout the day as the ratio of berries to mustbeing offloaded changed.

In 2001, Farmscan Ltd (Western Australia, Australia)released a load cell-based grape yield sensor thatproved to be more reliable than the HarvestMastersystem (Taylor 2004) ; however problems withFarmscan’s parent company effectively resulted inthe Farmscan grape yield sensor being withdrawnfrom the market by 2007. A second load cell-basedsystem was commercially released in 2005 inAustralia by Advanced Technology Viticulture(ATV) (Adelaide, South Australia, Australia) Since2005 there has been a slow but steady increase inadoption of the ATV grape yield monitor,predominantly in Australia but increasingly in NorthAmerica with 6 yield monitor installations in theUSA in 2012, 9 in 2013 and >30 in 2014.

Examples of yield maps have been presented in theliterature (Bramley & Hamilton, 2004 ; Tisseyre etal., 2007 ; Arnó et al., 2009), as well as ageostatistical analysis of the spatial variationobserved in Australian and European viticulture(Taylor et al., 2005) ; however, there has been noindependent evaluation of this technology published.Given the increased interest in grape yield monitorsin the USA, and the expected continued growth inadoption, an evaluation of the accuracy and thepotential uses of this technology is consideredpertinent.

MATERIALS AND METHODS

1. System and Installation

The only currently available commercial grape yieldmonitor is manufactured, installed and supported byAdvanced Technology Viticulture (ATV). The ATVgrape yield monitor (GYM) is designed to be

retrofitted to a discharge conveyor belt and can bealtered to fit most makes of grape harvesters that usea discharge conveyor. The yield sensor consists of aweigh-frame on load cells under the dischargeconveyor belt of the harvester that weighs the grapesas they are being off-loaded. A belt speed sensor, ajunction box and a data logger/controller in the tractorcabin complete the system. For mapping capabilitiesa Global Navigation Satellite System (GNSS)receiver is also required. After each installation, astatic calibration of the load cells is performed byplacing known dead weights on the dischargeconveyor belt over the sensor. This is to ensure thatthe system is operating within acceptable limits andhas been correctly installed. A daily, dynamiccalibration (rezeroing) of the system should beperformed prior to operation to tare the force beingexerted on the load cells by the empty belt. In total, 7 GYM systems were used within this study : 5 systems were installed on commercial harvestersthat operated primarily in winegrape vineyards in theSan Joachin valley in California, 1 GYM system wasinstalled on a commercial harvester in the Lake Erieregion, NY State and 1 GYM system was installed ona harvester associated with the Cornell Lake ErieResearch and Extension Laboraty (CLEREL)juicegrape vineyards in the Lake Erie viticulture areain NY State.

2. Evaluation

The GYM was designed for use at harvest. Theaccuracy and precision of the GYM systems atharvest was assessed by comparing sensormeasurements against recorded masses of bin loads(0.1 – 0.7 Mg) and truck loads (5 – 30 Mg) in bothCalifornia and New York. In addition to harvestassessment, the GYM was evaluated for use mid-season during fruit load estimations (5 – 70 kg) inNew York. Mid-season use is a novel application ofthis technology in this study. It has two potentialapplications 1) as a means of automated cropestimation in production systems, such as juicegrapesystems, where destructive sampling for cropestimation is common, and 2) for mapping fruit loadremoval when crop-thinning is employed as a mid-season management tool.

Truck scale measurement at harvest : For the 2012,2013 and 2014 harvests, the daily mass of grapessensed was compared to the mass of grapes deliveredto the crush. In 2012, these data were only from theCLEREL GYM (n 11). In 2013, additional data fromsensors on 5 different commercial harvesters inCalifornia were available (n 31), as well as theCLEREL harvester (n 12). In 2014, data were only


obtained from the CLEREL harvester (n 13). Allthese data were collected at the truck-load scale, withthe CLEREL data varying from small flat-bed trucks(~5 Mg) to full semi-trailer loads (~25 Mg). Thecommercial data in each year were only available asfull semi-trailer load measurements (20-25 Mg). Datawere collected over a period of several weeks in eachyear, though not always daily. All the GYM sensorswere calibrated, but the data were not corrected toadjust the daily total GYM mass to the deliveredcrush mass. Each measurement was therefore affectedby the quality of GYM operation (re-zeroing andmaintenance) and the nuances of the harvesteroperation. The daily error (as a percentage) wascalculated for each truck load; however, to compareacross production systems and harvesters, only errordata associated with full semi-trailer loads ispresented (n 51).

Bin scale measurements at harvest: In both 2012 and2013, the CLEREL GYM was used to harvest half-row trials (16 rows - 32 trials with 45 vines/trial) atthe CLEREL vineyard, Portland, NY. Each trial wasindividually harvested and weighted on a Cardinal708 floor scale (2268 kg ±0.45 kg) (Cardinal ScaleManufacturing Co., Webb City, Missouri, USA) sothat a comparison between the scale mass and theGYM-sensed mass could be made. Fruit load per trialvaried but was typically about 400 kg (approximatelyhalf a harvest bin). In both years, all data werecollected on the same day and the total daily masssensed was adjusted to the total delivered (crush)mass. This trial was not run in 2014.

Bucket-scale measurements mid-season for cropestimation: In 2013 and 2014 the yield monitor wasevaluated for use mid-season (July) during yieldestimation. In July 2013, measurements (n 57) weretaken over two days on the CLEREL vineyards fromareas as small as a single panel (typically 7.32 mcontaining 3 vines) to multiple panel lengths using theCLEREL harvester and GYM. In July 2014, datawere obtained from two GYMs in the Lake Erieviticulture area; the CLEREL harvester (n 15) and aGYM on a commercially operated harvester (n 69).The CLEREL harvester obtained data from two-panelsections, whilst the commercial operator used a fixeddistance (14.6 m) to avoid issues with irregularlyspaced panels in the vineyards. In all cases, theGYMs were operated mid-season with the harvestcalibration factors. For each sampling site theharvested fruit was captured off the end of thedischarge conveyor into a bucket and weighed usingeither an Ohaus D-5-MO platform scale (20 kgcapacity) (Nänikon, Switzerland) or a hand-heldspring scale (25 kg capacity) (INS-T, Chatillon, NY,

USA). Where yield was > 20 kg, the captured berrieswere sub-divided and multiple weights taken. TheGYM readings were captured by the GNSS-enableddata logger. Post-processing was performed to extractthe GYM recorded mass at each point and relate it tothe manual scale measurements. For plotting andanalysis, these data were not corrected against thetotal mass of grapes harvested at this time.

The intent here is to provide information on theprecision and accuracy of the GYM against actualmass at different scales. Only the harvest truck load(crush) data had a temporal dimension, with datacollected on multiple days over several weeks in eachyear. The other data were collected on either onesingle day or over short time periods (2-4 days).

For all three approaches, plots of the GYM vs actualscale mass were generated. Linear regression wasperformed with the intercept constrained to (0,0).The coefficient of variation (R2) and root meansquare error (RMSE) were calculated. The histogramof the percentage errors (%E = [(Actual-Sensor)/Actual]*100) for all the truck measurementsfrom both the research (CLEREL) and commercialGYMs were also calculated. For the truck-scale data,linear regression fits were performed separately forthe CLEREL and commercial data sets due to thedifference in range of values (truck loads). Allanalysis was performed in JMP Pro v11.1 (SASInstitute, Cary, NC, USA).

RESULTS

Truck-scale measurements : Figure 1 shows therelationship between the GYM readings and the massof weighed grapes for truck-scale data collected atharvest in both California and NY State. TheCLEREL truck data had a large range of values(4.5–23.6 Mg) and a strong fit (R2 0.95) (Fig 1a). TheRMSE was 1.43 Mg or 7.87 % of the mean(18.18 Mg) response. The fit was stable over thethree years of the study at CLEREL. The commercialdata from California had a much lower range(21.6–28.1 Mg) as it was restricted to full semi-trailerloads, and had a lower fit for the regression (R2 0.4).This is probably due in part to the shorten range(relative to the CLEREL data) and the fact that thedata was derived from 5 different GYMs (harvesters)across multiple sites and states compared to a singleGYM at CLEREL. Even with the lower fit, thegradient is very similar between the two sets of data,and very close to a 1:1 fit. This is indicative of wellcalibrated sensors. The RMSE error from thecommercial data set was 1.75 mg, which is higherthan the CLEREL data set. The mean response was

- 59 -J. Int. Sci. Vigne Vin, 2016, 50, 2, 57-63

©Vigne et Vin Publications Internationales (Bordeaux, France)


also higher (24.63 Mg), such that the RMSE was7.11 %, or similar to the CLEREL data set. The %Error distributions are shown in Fig. 1b for all thedata combined (CLEREL and commercial). Therewas a slight bias, with the GYM reading lower thanthe crush (truck mass) on average (~2.5 %). The %Errors were in the range of ±15 % of truck weightand were independent of the truck load size or theyear or the type of harvester for this data set (data notshown).

Bin-scale measurements: The precision of the GYMon any given day at harvest was assessed from theplot trial data (Fig. 2). The yield data were correctedagainst the delivered crush mass in each year (tocorrect for the daily error observed in Fig. 1). Thecorrections were respectively -5.13 % and -3.35 %for 2012 and 2013. The yield in 2012 was lower thanin 2013. The mass of harvested fruit per trial rangedfrom 89 to 816 kg over the two years. As expectedwith the correction, the linear regression follows a1:1 fit across the two years, and the fit is very strongwith R2 0.95 and RMSE 35 kg (or 9.70 % of themean response). There was no indication that therewas any shift in the quality (error) of the sensorresponse over the course of the day when plots wereharvested (data not shown).

Mid-season measurements : Figure 3 showsuncorrected fits of GYM-sensed vs actual mass fromvery low masses of grapes picked mid-season duringcrop estimation practices in the Lake Erie juicegrapeviticulture region. There were three datasets – twofrom CLEREL (2013 and 2014) and one from acommercial grower (2014). The range of harvested

grapes was 1.3 to 97 kg across all data. In Fig. 3 thethree data sets are fitted separately and show verydifferent gradients ; however all three show areasonable to strong fit for field-collected data withoutliers removed (R2 0.70-0.92). The 2013 and 2014mid-season CLEREL regression fits had differentgradients between years, which contrasts with thestability of the inter-annual harvest regression fit(Fig. 1a). The difference in gradient between years isprobably associated with yield estimations atdifferent stages of berry development. The CLEREL2014 mid-season data had the fewest points (n 15)and the worst fit (R2 0.29). The CLEREL 2014 datahad three points that were outliers (shown as opencircles in Fig. 3). These points were probablyassociated with harvester problems, particularlyproblems associated with belt operations. Theseissues were identified at the time but were not loggedagainst individual data points. The linear fit shown inFig. 3 was generated without these outlying points(open circles) and the fit improved considerably (R20.73, n 12). This highlights the need for care in bothmanual measurements and machine operation duringcrop estimation. The commercial 2014 mid-seasondata were collected by a grower under commercialoperating conditions, i.e. not with the rigour of theCLEREL research-oriented measurements. Therewere some points omitted due to untidy recordkeeping, however these data still had a strong linearrelationship between the GYM and scalemeasurements (R2 0.70).



Figure 1. A) Plot of the actual vs. sensed mass of grapes at truck-load scale (squares 2012 CLEREL; circles 2013CLEREL; diamonds 2014 CLEREL; triangles 2013 commercial data).Dashed line is fit of a linear regression to the CLEREL data (2012-14).

Solid line is the fit of a linear regression to the 2013 commercial data. R2 and RMSE values for the linear regressions are provided on the plot. B) A histogram of the daily errors associated with 51 individual full semi-trailer measurements.


DISCUSSION

Assessment of the yield monitor at multiple scales ofmass measurement showed that the sensing systemprovides good information across all scales. The yieldmonitor is subject to some temporal shift that cancreate absolute errors of ±15 % in measurementbetween days at harvest. Consequently, for optimumresults at harvest, the daily sensor total should beadjusted against the total mass of grapes weighted atthe crush. In single harvester systems this is relativelystraight forward; however, when multiple harvesters(sensors) are used in a vineyard it can be moredifficult to track weights back to a single machine.Unfortunately, proper correction is also moreimportant in multi-sensor vineyards, as differentmachines are likely to have different errors on a givenday. The importance of this daily correction willdepend on how the information is used and whetherrelative information on high and low yielding areas orabsolute information is needed by the producer. Thisis a potential limitation to adoption of the GYM inlarger, multi-harvester vineyards.

Daily calibrations are even more important for anymid-season applications of the GYM. These dataindicate that for mid-season use of the GYM adifferent calibration coefficient is needed to that usedat harvest, and it appears that this coefficient isdynamic as the berries develop. It is hypothesised thatsensor performance is affected by differences in berryphysiology, with lighter, smaller, harder berries mid-season compared to large juicy grapes at harvest, andthe absence of the must (juice) weight. Further studiesare needed to test this hypothesis. However, the linear

fits observed in this evaluation indicate thatrecalibration should be relatively simple and willproduce sensible results. The CLEREL 2013 and2014 mid-season data indicate that the calibration islikely to be year-specific, despite the harvestcalibration being stable between years. Again thereason for this is not known, but if the calibration isaffected by berry physiology, then the rapid growthand change in berry development during this periodis likely to generate a dynamic calibration coefficientuntil berry size and mass stabilise. It may be possibleto have a generic calibration if crop estimation can betimed for the exact same stage of berry developmenteach year, but this is considered difficult to achieve incommercial enterprises.

This is the first reported use of the GYM mid-seasonand the quality of the results in Fig 3 show greatpromise for the application of the GYM for 1) cropload estimation from destructive harvesting mid-season and 2) mapping removed fruit during thinningoperations. The former has the potential torevolutionise yield estimation in viticultural systemswhere destructive sampling is common, while thelatter will allow precision viticulture to be betterapplied to crop load management. Future studies areneeded to further quantify these potentialapplications, especially in commercial systems.

The stability and linearity of the response for a givenday, either mid-season or at harvest, is indicative of asystem that is suitable for pattern identification



Figure 2. Plot of the actual vs. sensed mass of grapes at the partial bin-load scale from half-row experimentsin a Concord (Vitis labrusca) vineyard at CLEREL

vineyard (circles 2012 and triangles 2013).The linear regression was fit across all data.

The R2 and RMSE values are provided on the plot.

Figure 3. Plot of the actual vs. sensed mass of Concordjuicegrapes at a bucket-scale

(squares 2013 CLEREL; circles 2014 CLEREL;diamonds 2014 commercial vineyard data).

The sensor calibration coefficient was held constant, whichresulted in different dates having different gradients for thelinear fits. The open circles were data from CLEREL 2014that were assumed to be erroneous and not used for the

linear regression fits.


regardless of the quality of the sensor calibration.That is, even though the data may contain someabsolute error, the relative patterns generated by theseyield data are correct. An example of the observedspatial patterns in a mid-season crop thinning mapand a final (harvest) yield map are shown in Fig 4.The difference in mass removed is evident (medianvalues of 4 and 18 Mg/ha for the thinned and harvestyield respectively) even when considering that at thismid-season stage the berry weight is approximatelyhalf that of berry weight at harvest. Broad, coherentpatterns (patches) are evident in both maps, althoughthere are different patterns in the midseason andharvest maps. The reasons for this are not clear at themoment, but are of obvious interest for viticulturistslooking to manage fruit load.

It is important to understand that these spatial grapeyield data are different to spatial yield data fromother crops, particularly combinable arable crops.Principally this is because grape harvesters are veryefficient at removing and offloading the berries andthere is very little convolution or mixing of theberries within the harvester. This is in contrast to thelevel of convolution and resultant data smoothingthat occurs in combine harvesters (Whelan andMcBratney, 2002). The efficiency of the grapeharvester means that a lot of the short-range andstochastic variation in production is transferred to theyield sensor. Visually this makes a ‘noiser’ point rawyield and correct interpolation is needed to bettervisualise, analyse and interpret the spatial patterns inyield (Bramley & Williams, 2001 ; Bramley 2005 ;Bramley et al., 2008).

CONCLUSIONS

The evaluation of the sensor response against manualmeasurements showed a strong linear relationshipduring use mid-season and at-harvest. The sensorresponse was stable and linear within a harvestingoperation; however, between days (operations) therewas often a shift in the sensor response that couldlead to an absolute error of ±15 %. The grape yieldmonitor was determined to be effective forvisualizing relative spatial yield patterns, but the yielddata requires a daily adjustment (re-calibration)against a measured mass if these relative patterns areto reflect absolute yield variations. Evaluation of theGYM mid-season showed that it was suitable for useat this stage with possible applications in cropestimation and mapping of crop thinning.

Acknowledgements : The authors would like toacknowledge the assistance provided by Dr BerndKleinlagel, Advanced Technology Viticulture with sensorinstallation and initial calibration. This work is possiblebecause of the financial support provided by the NationalWine and Grape Initiative (NGWI). The project teamwould like to express our gratitude to NGWI and thevarious viticultural organizations that support NGWI.Additional financial support was also provided by the NewYork Grape and Wine Foundation.

REFERENCES

Arnó, J., Martínez-Casasnovas, J.A., Ribes-Dasi, M. andRosell, J.-R. 2009. Precision Viticulture. Researchtopics, challenges and opportunities in site-specificvineyard management. Span. J. Agric. Res., 7(4),779-790



Figure 4. Example of the spatial patterns associated with grape yield monitoring during crop thinning mid-season (left)and at harvest (right) at the CLEREL Concord juicegrape vineyard, Portland, NY.

All data collected with a GNSS-enable grape yield monitor. Thinned yield data (left) has been post-processed to correct for mid-season bias in operation. Note the difference in spatial patterns between the two maps.


Bramley R.G.V. 2005. A protocol for the construction ofyield maps from data collected using commerciallyavailable grape yield monitors.Supplement No. 1, February, 2005 (available atwww.cse.csiro.au/client_serv/resources/protocol_supp1.pdf - accessed 14/07/2015)

Bramley, RG.V. and Williams, S.K. 2001. A protocol forthe construction of yield maps from data collectedusing commercially available grape yield monitors.CRC for Viticulture, Adelaide (available atwww.cse.csiro.au/client_serv/resources/CRCVYield_Mapping_Protocol.pdf - accessed 14/07/2015)

Bramley, R.G.V., Kleinlagel, B. and Ouzman, J. 2008. Aprotocol for the construction of yield maps from datacollected using commercially available grape yieldmonitors. Supplement n° 2. April 2008 - Accountingfor ‘convolution’ in grape yield mapping (available atwww.cse . c s i ro . au / c l i en t_ se rv / r e sou rce s /protocol_supp2.pdf - accessed 14/07/2015).

Bramley, R.G.V. and Hamilton, R. 2004. Understandingvariability in winegrape production systems 1. Within

vineyard variation in yield over several vintages. AustJ. Grape Wine R., 10(1), 32-45

Tisseyre, B., Ojeda, H. and Taylor, J. 2007. Newtechnologies and methodologies for site-specificviticulture. J. Int. Sci. Vigne Vin, 41(2), 63-76

Taylor, J.A. 2004. Precision Viticulture and DigitalTerroir : Investigations into the application ofinformation technology in Australian vineyards.Ph.D. Thesis. The University of Sydney (available atwww.digitalterroirs.com – accessed 14/07/2015).

Taylor, J.A., Tisseyre, B., Bramley, R.G.V. andReid, A. 2005. A comparison of the spatial variabilityof vineyard yield in European and Australianproduction systems. In : Precision Agriculture ’05.Proceedings of 5th ECPA, Uppsala, Sweden. Ed. J.V.Stafford, Wageningen Academic Publishers.

Whelan, B.M. and McBratney, A.B. 2002. A parametrictransfer function for grain-flow within a conventionalcombine harvester. Precis. Agric. 3(2), 123-134.




evaluation of a commercial grape yield monitor for use mid...

Documents