You are on page 1of 14

INTERNATIONAL JOURNAL OF GEOMATICS AND GEOSCIENCES Volume 1, No 3, 2010

Copyright 2010 All rights reserved Integrated Publishing services

Research article

ISSN 0976 4380

Geospatial Modeling of Wine Grape Quality Indicators (Anthocyanin) for Development of Differential Wine Grape Harvesting Technology
Balaji Sethuramasamyraja1, Sivakumar Sachidhanantham1, Robert Wample2 1- Department of Industrial Technology, California State University, Fresno 2255 E Barstow Ave, M/S IT 09, Fresno, CA 93740-8002, USA 2- Viticulture and Enology Research Center, California State University, Fresno 2360 E Barstow Ave, Fresno, CA 93740-8003, USA balajis@csufresno.edu ABSTRACT Segregation of wine grapes based on quality during harvest is a growing need for producers and wineries as spatial variability of vineyard quality is well established. While wine grape quality indicators like anthocyanin (mg/g) are measurable, there is no commercial technology to differentially harvest using such parameters. Geo-referenced field samples of wine grapes were measured for anthocyanin and brix using a portable near-infrared (NIR) spectrometer. Data was collected from 437 sampling vines in a 45 acre block and 1330 in a 160 acre block of vineyards in the San Joaquin Valley of California (2006-2007). Geo-spatial modeling of anthocyanin yielded quality zones of high and low quality while the brix dataset was utilized to determine the timing of the harvest. The anthocyanin concentration used to differentiate between high and low quality was based on cut off values of 0.87 and 1.05 mg anthocyanin/g fruit for the two vineyards specified by winemakers. A differential harvest attachment was developed for a commercial mechanical grape harvester that utilized the geospatial quality map for segregation of wine grapes on-the-go. Three 40 tons lots of wine grapes representing the standard (average) field blend, high anthocyanin and low anthocyanin were differentially harvested from each vineyard. These wine grapes were fermented separately and subjected to analytical and taste panel analysis resulting in significant (99.4% confidence) difference in wines produced. Key words: Precision agriculture, precision viticulture, differential harvest, anthocyanin, wine grapes, geo-statistics, kriging. 1. Introduction Uniform harvest of vineyards has been the tradition where wine grapes from vineyard blocks were delivered to winery for wine making. Segregation of wine grapes based on quality was not feasible in the past due to inadequate sensing and mechanized harvest technology. Spatial variability of fruit yield and quality within wine grape vineyards could benefit the producer given the advances in sensing quality parameters and differentially harvesting technology to make different wines from the same vineyard block (Kaye et al., 2005).

372

INTERNATIONAL JOURNAL OF GEOMATICS AND GEOSCIENCES Volume 1, No 3, 2010


Copyright 2010 All rights reserved Integrated Publishing services

Research article

ISSN 0976 4380

The application of precision farming practices (Pierce and Nowak, 1999) to viticulture is relatively recent, but has taken advantage of technologies applied in other crops such as yield monitoring (Wample et al., 1999, Bramley and Hamilton, 2004). Besides yield, wine grape quality is a profitable aspect of the vineyard that could benefit from sitespecific crop management and precision farming (Cook and Bramley, 1998) [5]. Bramley and Hamilton (2004), Bramley (2005a) and Bramley et al. (2005b) reported spatial variability in both wine grape yield and quality over three years. The variability in yield showed spatial consistency within a vineyard, but the range in yield was significantly different temporally, that could be associated with high temperatures during the bloom period (Bramley and Hamilton, 2004). The tendency towards spatial consistency is encouraging with respect to understanding the source of this variability and hence to develop more precise management practices. The spatial variability in quality was neither consistent as yield nor well correlated with spatial yield variability suggesting factors controlling quality are more complex than yield and hence the latter cannot be used as criteria to determine wine quality management (Bramley, 2005a). Vineyard management practices like variable fertilizer applications, foliar nutrient programs and drip irrigation could help to minimize variability in vine growth as well as fruit quality. However, there are other factors such as slope, aspect, pests and disease, quality at harvest is still difficult to predict. To help overcome this problem, it has been desirable to practice differential (both temporal and spatial) harvests which are cost prohibitive in production settings (Dunn and Field, 2003) . Bramley (2005a) suggests the need for development of on-the-go wine grape quality sensing technology that could enable quality zone delineation for effective harvest and management of vineyards. In the absence of on-the-go quality sensing in wine grapes, and with the relatively poor correlation between yield and quality, temporally and spatially, within vineyards it seems desirable to look for alternative ways to improve the quality of fruit delivered to the winery. In such an attempt, global positioning systems (GPS), geographic information systems (GIS), portable near-infrared (NIR) spectroscopy could be utilized in combination, to identify quality zones in the vineyard that could be used to control mechanical harvesters and thereby segregate high and lower quality fruit for delivery to the winery. This segregation is expected to enable the winery to produce different quality wines as against an average wine. 2. Materials and Methods 2.1 Experimental Materials 2.1 1 Site Location: Vineyard Two different vineyards were utilized, in two different seasons, for the experiments. The vineyards were Twin Creeks Vineyard (2006) located in Lodi, California and Merjan Vineyard (2007) located in Madera, California (Figure 1). Cabernet sauvignon was the

373

INTERNATIONAL JOURNAL OF GEOMATICS AND GEOSCIENCES Volume 1, No 3, 2010


Copyright 2010 All rights reserved Integrated Publishing services

Research article

ISSN 0976 4380

wine grape varietal under study in both vineyards. Table 1 lists site specification of vineyards blocks. Table 1: Vineyard site information Vineyards Area, Vines/ac. Rows, ~Vines/ ac. Row Heading, Row & Vine spacing E-W, 8 ft. & 8 ft. N-S 10 ft. & 7.5 ft. Sample Vines Quality Paramtr.

Twin Creeks (Lodi, CA) Merjan (Madera, CA)

45, 681

157, 196 135, 335

437

Anthocyn . Anthocyn

160, 581

1330

2.2 Data Collection and Calibration Table 2 lists the sensors, global positioning system and software utilized in this research. The Geo XT and Geo XH are both hand held GeoExplorer series (Trimble Navigation Limited, Sunnyvale, CA) GPS systems, each of which is an integrated GPS receiver with Microsoft Compact Edition (CE) based mobile device. Trimble's Zephyr antenna (Trimble Navigation Limited, Sunnyvale, CA) was used with Geo XH handheld GPS system to achieve 8 inch (20 cm) accuracy for precision mapping and GIS data collection. The Trimble TerraSync (Trimble Navigation Limited, Sunnyvale, CA) is mobile computing software for spatial data collection and data maintenance. Trimble GPS Pathfinder Office Software Ver. 4.0 (Trimble Navigation Limited, Sunnyvale, CA) was utilized for post processing including data validation and conversion to Environmental Systems Research Institute (ESRI) ArcGIS native shape file format. ArcGIS 9 desktop software and geo-statistical extension (ESRI, Redlands, CA) was used for geo-statistical analysis. Luminar 5030 Miniature "Hand held" Near Infrared (NIR) Analyzer (Brimrose Corporation of America, Baltimore, Maryland) was utilized for the quantification of wine grape anthocyanin in Twin Creeks vineyard (2006). The spectral range used for the data collection was 1100 to 2300 nm in transmission mode, with a wavelength increment of 2 nm. Calibration for the NIR spectrometer was based on reference wet chemistry analysis using standard laboratory techniques. Zeiss Corona 45 VISNIR 1.7 spectrophotometer (Carl Zeiss Inc., Peabody, Massachuseets) was utilized to acquire anthocyanin values in the VIS/NIR range (400 nm to 1680 nm) at Merjan in 2007. Winegrape fruit samples were homogenized 45 s prior to spectral analysis resulting in 167 scans averaged as anthocyanin indicator per sample. The system was calibrated using reference method for grape color using Iland method (Iland, 1993) and twenty percent of the samples were correlated with a standard reference for winegrape color. 374

INTERNATIONAL JOURNAL OF GEOMATICS AND GEOSCIENCES Volume 1, No 3, 2010


Copyright 2010 All rights reserved Integrated Publishing services

Research article

ISSN 0976 4380

Figure 1: a) Twin Creeks vineyard, Lodi, CA b) Merjan vineyard, Madera, CA

375

INTERNATIONAL JOURNAL OF GEOMATICS AND GEOSCIENCES Volume 1, No 3, 2010


Copyright 2010 All rights reserved Integrated Publishing services

Research article

ISSN 0976 4380

Table 2: Sensors, Instrumentation and Software utilized in experiments Sensors & GPS Hardware Trimble GeoXTTM Twin Luminar 5030 Creeks Miniature Handhandheld GPS 1 Receiver4 held NIR Analyzer 5 Insight Monitor Raven 100S GPS Merjan Direct Command receiver6 Trimble GeoXHTM Module5 Zeiss CORONA hand held GPS VIS/NIR 1.7 Receiver4 2 Spectrometer 1 Brimrose Corporation, Baltimore, MD 2 Zeiss Microimaging, Thronwood, NY 3 Environmental Systems Research Institute, ESRI, Redlands, CA 4 Trimble Navigation Limited, Sunnyvale, CA 5 AgLeader Technologies, Ames, IA 6 Austin Technology Center, Austin, TX 2.3 Differential Harvest Attachment Spatial Management System (SMS) Basic software (Agleader Technologies, Ames, IA) was used for fruit quality coding. INSIGHT monitor, Direct Command liquid application control module (Agleader Technologies, Ames, IA) was installed on a twin boom Korvan 3016XL (Oxbo International Corporation, Byron, NY) for the differential harvest. Raven 100S GPS receiver (Austin Technology Center, Austin, TX) was used during the differential harvest. Figure 2 illustrates the block diagram of differential harvest attachment. The circuit consists of a AgLeader Insight Monitor (loaded with the final differential quality map shape file) that controls the master conveyor belts of the wine grape harvester through a liquid product control module hardware that sends a control signal to activate or deactivate the relay switch for directing the master conveyor to control the flow of wine grapes into the assigned gondolas A or B. The relay switch has a delay timer (15 s delay) to ensure residual flow of wine grapes in the conveyor to the appropriate gondola during the switch from one quality zone to the other. Vineyard Software ESRI ArcGIS 9.1, 9.2 Extension: GeoSpatial Analyst3 Trimble GPS Pathfinder Office Ver. 4.04 SMS Advanced5

376

INTERNATIONAL JOURNAL OF GEOMATICS AND GEOSCIENCES Volume 1, No 3, 2010


Copyright 2010 All rights reserved Integrated Publishing services

Research article

ISSN 0976 4380

Figure 2: Block diagram of the differential harvest attachment for a wine grape harvester. 3. Experimental Methods ArcGIS 9.0 software and geo-statistical extension (Environmental System and Research Institute, Redlands, CA) was utilized for stochastic interpolations associated with geospatial modeling of the data layers, which builds a bridge between statistics and GIS. Geo-statistical analyst extension has a comprehensive set of tools that can methodically perform geo-statistical analysis in the GIS environment starting from data representation, exploration of statistical and spatial properties of data, model fitting, perform diagnostics using validation tools and cross-validation statistics. The semivariogram estimation would follow according to (e.g. Isaaks & Srivastava, 1989) [8]. Semi-variance of the parameter under consideration is,

( h) =

N (h) i =1

[Z ( xi ) Z ( xi +h )]

/ 2 N (h)

(1) where, N(h) = the number of paired values Z(xi) and Z(xi+h), displaced by effective distance h.

377

INTERNATIONAL JOURNAL OF GEOMATICS AND GEOSCIENCES Volume 1, No 3, 2010


Copyright 2010 All rights reserved Integrated Publishing services

Research article

ISSN 0976 4380

Figure 3: Model of semivariogram parameters: nugget, range, and sill Cambardella Index (Cambardella et al., 1994) is defined as, Cambardella Index = 100(C 0 /(C 0 + C )) (2)

indicates the degree of randomness in the spatial variability of the variable (where, C0 = nugget, and C = partial sill, C1,=sill = C + C0). Cambardella Index less than 25 indicates strong spatially dependence, a value in the range of 25 and 75 indicates moderate spatial dependence, whereas a value greater than 75 indicates weak spatial dependence of the measured variable.The numerical parameters of semi-variogram such as: nugget effect, sill and range are as demonstrated in Figure 3. The semi-variance often flattens out at a particular sill value at a certain range. Spatial variability of wine grape quality parameters was investigated through stochastic geo-statistical ordinary kriging method. Table 3 shows statistics used to evaluate the performance of the various interpolation methods derived from the cross validation results. Kriging predicts use neighborhood weighting based on location for prediction and error analysis. Kriging is the best-unbiased predictor irrespective of the normality of data and best linear unbiased estimator as the mean residual error nears zero minimizes the variance of the errors (Pannatier, 1996).

378

INTERNATIONAL JOURNAL OF GEOMATICS AND GEOSCIENCES Volume 1, No 3, 2010


Copyright 2010 All rights reserved Integrated Publishing services

Research article

ISSN 0976 4380

Table 3: Cross validation statistics to evaluate stochastic interpolations (ordinary kriging) Error Analysis Prediction Accuracy Condition Mean Prediction Error ~ 0 Mean Standardized Error << Average Standard Error (ASE) = Root Mean Square Prediction Error (RMS PE) RMS Standardized Error ~ 1

Uncertainty of Prediction Standard Error

Case I: Overestimation If ASE > RMS PE Case II: Underestimation If ASE < RMS PE Case I: Overestimation If RMS Standardized Error <1 Case II: Underestimation If RMS Standardized Error >1

3. Results and Discussion Table 4 lists the exploratory data statistics of wine grape quality parameters. Anthocyanin ranged from 0.64 to 1.06 with a mean of 0.88 in twin creeks and 0.41 to 1.5 with a mean of 1.05 in Merjan vineyard. Trend analysis showed the anthocyanin dataset had global trend in north-south direction indicating long term management defects or natural variability with respect to topography. However, no transformation was utilized in kriging for both anythocyanin as global trend was difficult to model. Table 4: Exploratory Data Statistics for Anthocyanin (mg/g fruit) Vineyard Twin Creeks Merjan Sample Size 437 698 Mean 0.88 1.046 Standard Coefficient Deviation Variation 0.08 8.61 0.163 0.155 of

Semi-variograms individual attributes were governed by numerical parameters: range (distance beyond which there is no autocorrelation), sill (semi-variance associated with range) and nugget (micro scale variation or measurement error) effect. Range for anthocyanin was 79 m and 134.7 m for Twin creeks and Merjan, respectively. Table 5 summarizes the semi-variogram parameters used for ordinary kriging. Table 5 : Semivariogram parameters for Ordinary Kriging - Anthocyanin (mg/g) Vineyard Twin Creeks Merjan Nugget 0.0047 0.0192 Sill 0.0061 0.0249 Range (m) 79 134.7 Cambaradella Index 47.09 43.88

Nugget and sill for anthocyanin was 0.0047, 0.0061 and 0.0192, 0.0249 (mg/g)2 for Twin creeks and Merjan, respectively. Cambardella Index (nugget to sill ratio) index showed moderate spatial dependence for both the vineyards. Figure 4 and 5 represents the 379

INTERNATIONAL JOURNAL OF GEOMATICS AND GEOSCIENCES Volume 1, No 3, 2010


Copyright 2010 All rights reserved Integrated Publishing services

Research article

ISSN 0976 4380

anthocyanin interpolated maps using ordinary kriging with semi-variogram parameters (Table 5) for Twin creeks and Merjan vineyard, respectively. Figure 4 a represents the classified interpolated map with five zones while Figure 4 b represents the two quality zone used for differential harvest for twin creeks vineyard. Similarly, Figure 5 a represents the classified interpolated map with five zones while Figure 5 b represents the two quality zone used for differential harvest for Merjan vineyard. As evident in Figure 5, close to half of Merjan vineyard has high anthocyanin in the southern side, thereby eliminating the need for differential harvest in northern Merjan. So, Figure 6 a represents the classified interpolated map with five zones while Figure 6 b represents the two quality zone used for differential harvest for Merjan South vineyard.

Figure 4 : Twin Creeks a. Interpolated Quality (Anthocyanin) Map and b. Quality map (Anthocyanin) utilized for differential harvest (Q A - Quality A & Q B Quality B with 0.87 mg/g cut off value)

380

INTERNATIONAL JOURNAL OF GEOMATICS AND GEOSCIENCES Volume 1, No 3, 2010


Copyright 2010 All rights reserved Integrated Publishing services

Research article

ISSN 0976 4380

Figure 5: Merjan a. Interpolated Quality (Anthocyanin) Map and b. Quality map (Anthocyanin) utilized for differential harvest (Q A - Quality A & Q B Quality B

381

INTERNATIONAL JOURNAL OF GEOMATICS AND GEOSCIENCES Volume 1, No 3, 2010


Copyright 2010 All rights reserved Integrated Publishing services

Research article

ISSN 0976 4380

with 1.05 mg/g cut off value)

Figure 6: Merjan South a. Interpolated Quality (Anthocyanin) Map and b. Quality map (Anthocyanin) utilized for differential harvest (Q A - Quality A & Q B Quality B with 1.05 mg/g cut off value) Table 6 summarizes the cross validation results of ordinary kriging. For both vineyards, mean error & mean standardized error are both near zero and RMS prediction error is at a minimum, both of which indicate the accuracy of prediction. In the case of anthocyanin, the average standard error is very close to the RMS error with a possibility of minor underestimation while Merjan is overestimating. To substantiate earlier findings, RMS standardized error, is slightly higher than 1 for Twin creeks that indicates underestimation and is slightly lower than 1 for Merjan, indicating overestimation. Overall, the prediction errors were nominal with slight under/over estimations indicating applicability of ordinary kriging models in differential harvesting technology.

382

INTERNATIONAL JOURNAL OF GEOMATICS AND GEOSCIENCES Volume 1, No 3, 2010


Copyright 2010 All rights reserved Integrated Publishing services

Research article

ISSN 0976 4380

Table 6 : Ordinary Kriging Cross Validation Prediction Results for Anthocyanin (mg/g fruit) Average Std. Mean Stzd. RMS Stzd. Error Error Error Twin Creeks 0.0012 0.0726 0.0724 0.0154 1.0030 Merjan 0.0192 0.1477 0.1477 0.0069 0.9992 Final quality map (Figure 4 b and 6 b) were prepared with two quality zones (A and B) based on the anthocyanin cut off value of 0.87 for Twin creeks and 1.05 for Merjan (mg/g of fruit) specified by wine makers. The generated quality map was utilized for the wine grapes harvest using the newly developed differential harvest attachment. The final quality map was uploaded into Ag leaders INSIGHTTM monitor as a shape file to control the harvester using the differential harvest attachment in a standard commercialized wine grape harvester. Figure 7 illustrates the NIR sensor data collection, two zone interpolated quality maps loaded in differential harvester, and subsequent wine grape harvesting with Q A in one gondola and Q B in other. Brix (sugar content) was used to determine the timing of the harvest. Vineyard Mean Error RMS Error

Figure 7: Differential harvesting in the field Three 40 ton lots, a. 40 tons of Quality A (differentially harvested), b. 40 tons of Quality B (differentially harvested) and c. 40 tons of Control (harvested on a regular mode for an average or normal load of wine grape) of fruit were harvested for taste panel analysis. All the three categories were fermented separately and subjected to analytical and taste panel analysis. Taste panel analysis were performed to determine aroma, flavor, and /or mouth feel differences between wines made from low versus high and control anthocyanin of fruit samples collected from Cabernet Sauvignon harvest at Twin Creeks (2006) and

383

INTERNATIONAL JOURNAL OF GEOMATICS AND GEOSCIENCES Volume 1, No 3, 2010


Copyright 2010 All rights reserved Integrated Publishing services

Research article

ISSN 0976 4380

Merjan (2007) vineyard blocks. Taste panel analysis included 20 taste panelist and duo trio tests (replicated 5 times) performed at two different test groups with significant difference (99.4% confidence) in the wine taste and mouth feel. 4. Conclusions Global positioning systems (GPS), geographic information systems (GIS) and nearinfrared (NIR) spectroscopy were utilized in combination, to quantify wine grape quality variability, anthocyanin (mg/g fruit) in Twin Creeks vineyard (2006) and Merjan vineyard (2007). Geospatial interpolation techniques were utilized for interpolation of the wine grape quality indicators over the entire vineyard block with identification of two potential quality zones. The identified quality zones (a geo referenced map) was used to control the wine grape harvester for the separation of fruits located in quality zones i.e. quality A (Q A) and quality B (Q B) within vineyards. A differential harvesting attachment was developed for a commercial wine grape harvester to accomplish on-thego mechanized harvest of two different wine grapes from the same vineyard. This segregation enabled the winery to produce two different quality wines compared to what would have been a wine that was a blend of the different grape qualities in the vineyard. Acknowledgements Partial support for this research was provided by California Agricultural Technology Initiative Agricultural Research Initiative, American Vineyard Foundation, OXBO International Corporation and Constellation Wines US. Jim Orvis and Oren Kaye (Constellation Wines US - Mission Bell Winery, Madera, CA), John Gonsalves (Constellation Wines US Mission Bell Twin Creek Vineyard, Lodi, CA), Greg T. Berg (OXBO International Corporation, Kingsburg, CA), Antonio Odair Santos (IAC Instituto Agronomico, Sao Paulo, Brazil), Robert Cochran (Viticulture and Enology Research Center, California State University, Fresno, CA), and Jorge Rodriguez (Department of Industrial Technology, California State University, Fresno, CA), and Jeff Bentley (AgLeader Technologies, Ames, IA) are acknowledged for their contributions. References 1. Bramley, R.G.V. and R.P. Hamilton. 2004. Understanding variability in Winegrape production systems. 1. Within vineyard variation in yield over several vintages. Aust. J. Grape and Wine Research 10:pp 33-45. 2. Bramley, R.G.V. 2005a. Understanding variability in Winegrape production systems. 2. Within vineyard variation in quality over several vintages. Aust. J. Grape and Wine Research 11: pp 33-42. 3. Bramley, R.G.V., D.M. Lanyon and K. Panten. 2005b. Whole-of-vineyard experimentation An improved basis for knowledge generation and decision making. pp 883-890 In: Stafford, J.V. (Ed) Proceedings of the 5th European Conference on Precision Agriculture. Wageningen Academic Publishers, The Netherlands.

384

INTERNATIONAL JOURNAL OF GEOMATICS AND GEOSCIENCES Volume 1, No 3, 2010


Copyright 2010 All rights reserved Integrated Publishing services

Research article

ISSN 0976 4380

4. Cambardella, C.A., T.B. Moorman, J.M. Novak, T.B. Parkin, D.L. Karlen, R.F. Turco, and A.E. Konopka. 1994. Field-scale variability of soil properties in central Iowa soils. Soil Sci. Soc. Am. J. 58: pp 15011511. 5. Cook, S.E. and R.G.V. Bramley. 1998. Precision agriculture - Opportunities, benefits and pitfalls. Australian Journal of Experimental Agriculture 38, pp 753763. 6. Dunn, G. and S. Field. 2003. Crop Forecasting for better vineyard efficiency and wine quality. The Australian and New Zealand Grapegrower and Winemaker, 473, June, 2003, pp17-18. 7. Iland, P, Ewart, A, Sitters, J. 1993. Techniques For Chemical Analysis and Stability Tests Of Grape Juice and Wine. Patrick Iland Wine Promotions, Campbelltown, Australia. 8. Isaaks, E. H., and R.M. Srivastava. 1989. An Introduction to applied geostatistics. 1st Edition, Oxford, Oxford University press. pp 140-182. 9. Kaye, O., R. L. Wample, and A. O. Santos. 2005. American Society of Enology and Viticulture Annual International Meeting, Sacramento, CA. 10. Pannatier, Y. 1996. VarioWin: Software for Spatial Data Analysis in 2D. New York, Springer-Verlag, NY. pp 20-21. 11. Pierce, F.J., and P. Nowak. 1999. Aspects of precision agriculture. Adv. in Agron. 67: pp 1-85. 12. Wample, R.L., L. Mills, and J.R. Davenport. 1999. Use of precision farming practices in grape production. Pp. 897-905. In: P. Robert, R.H.Rust and W.E.Larson (eds) Proc. 4th International Conference on Precision Agricultrue. Minneapolis-St.Paul 19-22 July 1998. ASA/CSAA/SSSA Press, Madison, WI.

385

You might also like