Abstract
As part of a crop forecasting study, the variability in grape berry maturity and weight within individual clusters of Cabernet Sauvignon were determined. Each cluster was weighed as a whole and the number of berries within the cluster was counted. The difference between bottom berry weight and average berry weight was found to be statistically significant. Four berries were chosen systematically from the four sides of each cluster’s top, middle, and bottom region, yielding measurements of four variates: (1) top berry Brix, (2) middle berry Brix, (3) bottom berry Brix, and (4) Brix determined by mixing the juice extracted from all of each cluster’s remaining berries. Each of the replicates of variate (4) values were then subtracted from the corresponding measurements of variates (1), (2), and (3). The mean of the bottom berries’ difference in Brix from the entire cluster’s Brix was found to differ from zero significantly. This suggests that the Brix of a cluster may be represented poorly by berries near the bottom of the cluster.
In considering the factors influencing the size and soluble solids of individual berries within bunches of Pi-not noir and greenhouse-grown Cabernet Sauvignon vines in New Zealand, Trought et al. (1997) found that measurements did not reveal any consistent influence of berry position on the rachis for any fruit parameters examined. Here we describe a study of Cabernet Sauvignon (CS) clusters with different findings for two variates, berry weight and soluble solids (Brix). It is perhaps the case that the differences between the conclusions reached by these two studies are due to the study design and statistical analysis procedures chosen. The variates whose distributions we studied were all differences between two measurements both taken from the same cluster. According to Marriott (1990), such a design can yield better estimates of differences by “removing” the possible effects of other variables. Similarly, Dixon and Massey (1983) assert that in sampling from two populations extraneous factors may cause a significant difference in means, even though there may be no difference in the effects that are being measured. Conversely, extraneous factors can mask or obscure a real difference. An experimental design that sometimes overcomes part of this difficulty is the observation of pairs. By means of a paired measurement experimental design, it may be possible to increase statistical power substantially.
Relationships between important berry characteristics and the position of a berry along the rachis main axis are relevant to sampling procedures influencing crop harvest. Varying Brix levels within a cluster relate especially to sugar sampling, which attempts to estimate the Brix of an entire field by sampling a few berries from many clusters. The accuracy of these estimates would be significantly improved by understanding which berries on a given cluster yield Brix readings closest to the Brix of the entire cluster.
Materials and Methods
All clusters were collected from a single block, coded Dm-CS, of Cabernet Sauvignon vines located at the Wappo Hill Vineyard in Napa, California. The block contains cordon-pruned vines on a vertical trellis system planted in 1997, with 5 x 6 spacing and rows oriented northeast/southwest. The rootstock for the entire block is 101-14, with two clones: clone 4 (rows 1–121) and clone 5 (rows 122–253). The two clones were not sampled or analyzed separately. Thirty vines were analyzed, five selected randomly from each of six randomly selected rows within the block. On 3 September 2003, five clusters were gathered from each vine as described below and used for further analysis.
Cluster selection.
To select five clusters from each of the randomly sampled vines, five points were first marked on a short, 3.18-cm-wide ribbon of elastic material. When this material, in its nonstretched condition, extended from −1 to 1, the coordinates of five points, referred to in the mathematical literature as Gaussian nodes, were specified in Stroud and Secrest (1966) as A = −1, B = −0.6546537, C = 0, D = 0.6546537, and E = 1. Nodes A, E, and C were easily designated as the two ends and the middle of the elastic, respectively. Nodes B and D were determined by folding the elastic at its midpoint and marking each side at a distance from the fold equal to 0.6546537 times the length of the folded material. This method was used to mark three elastic ribbons of differing lengths, enabling selection at each vine of an elastic ribbon that, in its nonstretched state, was slightly shorter than the cordon arm from which the clusters were to be gathered. Cordon arms were chosen by alternating sides on the vine.
At each randomly selected vine, an elastic ribbon (shorter than the cordon arm) was stretched so that the ends of the elastic were located at the points where the outermost shoots joined the cordon. The shoot closest to the trunk was designated as Gaussian node A, while the shoot farthest from the trunk was Gaussian node E. One non-nubbin cluster was selected randomly from each of the five shoots closest to the five marks on the stretched elastic. (The term nubbin clusters refers to clusters comprised of 15 berries or less, not including shot/damaged berries.) For one vine, the cordon arm sampled had no non-nubbin cluster-bearing shoots near Gaussian node C that were also between Gaussian nodes B and D, resulting in a total sample size of 149 clusters.
Grape cluster analysis.
Each one of the 149 clusters was weighed as a whole, and its bottom berry removed and weighed. A two-stage sample was drawn using the method described by Guenther (1965). Fourteen vines, distributed across all rows, were selected and all five of their selected clusters reserved for Brix analysis as described below. After consideration of the time within which measurements could be taken prior to significant cluster changes, additional clusters were selected randomly for Brix analysis, one from each of the remaining 16 vines. This resulted in a Brix analysis subsample of 86 clusters. For all 149 clusters, berries were removed, counted, and weighed, and each rachis was weighed. Shot berries were excluded from all calculations.
Brix analysis.
Based on trisected length measurements along the rachis main axis, berries on each cluster within the size 86 subsample were divided among three regions: top, middle, and bottom. Of these 86 clusters, 11 were too small to allow for effective differentiation between their top, middle, and bottom regions, and were thus excluded from the sample. (As our Brix analyses were concerned primarily with regional differences in Brix within a given cluster, the exclusion of these smaller clusters is unlikely to have biased our results. Because of their small size, they are effectively all one region and, therefore, not especially relevant to our study of Brix. Because it was impossible to allocate their berries definitively into top, middle, and bottom regions, wings were also excluded. A separate study is underway to study characteristics that differentiate clusters with wings from other clusters.) For the remaining 75 clusters, four berries (shot berries excluded) were selected from each region, crushed, and their Brix determined. When a cluster had one or more large wings, these were removed from the cluster. Individual sets of four selected berries were chosen so that, within any given set, berry centers were close to a plane perpendicular to the rachis main axis and adjoining berries were separated maximally. For example, suppose the position and shape of a given cluster to be that of an inverted elliptical cone. Each lowest subset (quartet) of berries was selected so that its center was near a plane perpendicular to axis of the cone (the rachis main axis), and its position was near the ends of the major and minor axes of the ellipse formed by intersection of the plane with the cone. Following the three assessments of the Brix of the three sets of four selected berries, the remaining berries of each cluster were removed. These were counted, weighed, and crushed. Then the Brix of the berries that remained on the cluster was determined. Data-checking procedures revealed that one of the 75 sampled clusters had incorrect measurements of the Brix of the bottom and middle berries. Since this was determined too late for accurate reassessment, the calculations discussed below that involve bottom and middle berry Brix were based on measurements of the remaining 74 clusters.
Results and Discussion
SPSS for Windows (ver. 11.5; Chicago, IL) enabled us to compute a Student’s t test of the hypothesis that the average difference between a cluster’s average berry weight and the weight of this same cluster’s bottom berry equaled zero. The two-tailed significance level of this test was p = 0.002.
To determine whether this highly significant finding might be due to the presence of a few outliers (Barnett and Lewis 1978), Figure 1⇓ was constructed, which shows a histogram based on the same difference between the average berry and bottom berry (DIFAVBT) data that yielded the p = 0.002 finding, with the exception that data concerning the 11 clusters that had the lightest bottom berries were removed from the file. The histogram suggests that the distribution of this censored DIFAVBT variate will be best represented by a mixture model and that even though a minority of bottom berries may have weights whose distribution is similar to that of other berries within a given cluster, most bottom berries have a different distribution.
The variate top berry Brix minus average berry Brix (TOPDELTA), whose estimated distribution is described by the rightmost box plot in Figure 2⇓, equals the Brix of the four top berries minus the Brix measured from the juice extracted from all remaining berries of a given cluster. The mean of the 74 values of this variate equaled −0.037, while their standard deviation was 0.37, suggesting that it would make very little difference whether the Brix of the four top berries were calculated or, alternatively, whether Brix were determined from a random sample of size four from among all of a cluster’s berries.
The variate MIDDELTA is here defined to equal the Brix of the four middle berries minus the Brix measured from the juice extracted from all of a given cluster’s remaining berries. The mean of the 74 values calculated was −0.059, with a standard deviation of 0.41. Ordinarily, this would suggest that the Brix of the middle berries does not differ significantly from that of the entire cluster. However, because of the presence of an outlier with a value close to two, this finding is less reliable than the TOPDELTA finding.
The variate BOTDELTA, whose estimated distribution is described by the leftmost box plot shown in Figure 2⇑, equals the Brix of the four bottom berries minus the Brix measured from the juice extracted from all remaining berries of a given cluster. The mean of the 74 BOTDELTA measurements was determined by SPSS calculations to equal 0.25, with a two-tailed Student’s t test value of t = 3.5 (p = 0.001, 73 degrees of freedom) (Table 1⇓). This demonstrates a significant deviation from our null hypothesis (that the mean of the BOTDELTA variate’s distribution is zero) and suggests that a given berry taken from the cluster’s bottom region is likely to have significantly higher Brix than the overall cluster.
A detailed study of cluster’s characteristics has been conducted by Glynn (Glynn and Boulton 2001, Glynn 2002). The latter work refers to early research of Kasimatis et al. (1975) that, according to Glynn, supports the traditional viticultural concept of berries having the highest Brix in the shoulder berries, followed by midcluster berries, and the lowest Brix berries in the tip; the Brix values of the berries in the tip of the cluster were the most variable.
While the box plots in Figure 2⇑ do show the inter-quartile range of the bottom region-associated distribution to be roughly twice that of the top region-associated distribution, both estimated distributions are due largely to the paired measurement experimental design, more symmetric than comparable estimated distributions shown in Kasimatis et al. (1975). In addition, a single variate results from the subtraction of cluster Brix from bottom Brix. On the other hand, when a data vector consists of two separate variates, such as cluster Brix and bottom Brix, both measured on the same cluster, multivariate versions of procedures such as analysis of variance should be used in place of univariate procedures.
The choice of multivariate versus univariate procedure notwithstanding, the Kasimatis et al. (1975) paper describes a substantial scientific contribution. In particular, among the 14 vineyards studied, only in vineyard G was what Kasimatis calls Tip Brix greater than Shoulder Brix. Thus, it seems quite likely that, given other studies that corroborate the basic finding of our study, Thompson Seedless within-cluster Brix distribution will be found to differ substantially from Cabernet Sauvignon distribution.
Conclusions
The present study indicates that, with respect to berry weight and percent Brix, significant berry differences can exist regionally within a cluster. Further research would be valuable in assessing whether these cluster-specific regional Brix relationships are consistent across different varieties, cluster architectures, year, and growing regions. For vines similar to those sampled in this study, our findings suggest that sugar sampling is best confined to the top region of a cluster. While middle berries may also yield accurate readings, at least for the block studied, the bottom of the cluster should be avoided, since Brix measurements from that region could bias assessments of a crop’s readiness for harvest.
Footnotes
Acknowledgments: Funding for this project was provided by a UC Discovery Grant awarded by the Industry-University Cooperative Research Program (IUCRP). This grant was a three-way partnership between UC Berkeley, Robert Mondavi Winery, and the State of California. Special thanks to Louise Tolbert, who helped conceive the idea for this study. Thanks also to Daniel Bosch and Patrick Mahaney.
- Received April 2004.
- Revision received August 2004.
- Revision received October 2004.
- Copyright © 2005 by the American Society for Enology and Viticulture