Key Words: American oak, sensory analysis, oak barrels
Introduction-need some literature and some time.
Goals of the experiment
To determine if multiple small scale barrel trials can yield results consistent to single large scale trials.
To determine is vintage, region, and varietal sensory effects on oak interactions.
To determine is large-scale industry trade tastings can yield similar results to a controlled laboratory sensory analysis
There is ample discussion in the literature regarding the number of barrels required to get a valid trial. The work determining the number of barrels is based on chemical analysis, not sensory analysis. This experiment is based on trials of between the four and 10 barrels per treatment. The underlying assumption is that the confidence in assigning any differences found to the oak source increases if that difference is found in multiple trials. Alternatively, evaluation of all the results, controlling for wine and oak source, may be performed. To test this hypothesis, six different wines, representing two vintages and three varieties, were evaluated by the trained panel.
Materials and Methods
Mills in Missouri, Minnesota, Pennsylvania, and Virginia provided oak wood sourced within a radius of approximately 100 miles from each mill. Although all American oak used in wine barrel production was traditionally believed to be Quercus alba, forests in these areas consist of mixed species; the actual species provided by each mill may reflect the local mix (Table 1). The wood was cut into staves and aged 18 to 24 months at their respective mill sites and shipped to Calistoga, California, for coopering. Wood moisture contents were comparable for each lot. Barrels were coopered using Tonnellerie Française Nadalie's standard "Bordeaux Export" 225 liter design. The barrels were bent over an oak fire and toasted with their "house" toast. Barrel heads were also toasted. All barrels were made over a five day period, each day dedicated to producing barrels from one mill's wood, with the same one cooper performing all the bending and another one cooper performing all the toasting for the entire five day production. Barrels produced on one of the five days were made from a blend of all four mill's woods to be used as a control. To code for the wood origin, barrels were then marked with a single identifying letter (Table 2). The letter codes used for the 1998 barrels were different than those used for the 1997 barrels.
Sensory Descriptive Analysis of the Six Wine Subset
Ten panelists were recruited from production and retail in the wine industry and from local college wine making program. Six of the ten tasters had been determined as discriminators in previous evaluations of these wines when determining if sensory differences existed. One taster was unable to complete the screening and training process. Therefore, a total of nine tasters completed all scoring tastings. Panelists were told that all wines were from similar experiments, but from different wineries. Panelists were told that they cannot know the wine, the winery, or the nature of the experiment. Panelists were paid for their participation.
Four Chardonnays, each made by the two different wineries over two vintages, one Merlot, and one Sauvignon blanc were chosen from the available fourteen wines for descriptive analysis. Thus, six different wines each made with five different oak barrel sources were evaluated for a total of 30 wines (Table 3).
Descriptor and Standard Generation
Panelists were presented with an oak source subset of an individual wine, and asked to generate a list of aroma and flavor descriptors that best differentiated these wines. A written list of these descriptors was kept for each day's tasting, and re-evaluated with a different subset of the same experiment on a following day for each of six days. Physical standards were then created to match these descriptors. Panelists smelled these physical standards and evaluated their appropriateness for each of the six experiments. A total of thirty-nine descriptors were generated for all six experiments. This list was refined to 27 by combining some similar descriptors. For example artificial strawberry was combined with cherry, and lemon, citrus, and orange combined together. Frequency of use for these 27 descriptors (or combinations of descriptors) with then evaluated for the six experiments. Eleven of these terms were specifically fruit and variety related. After discussion with the panelists, it was decided to develop variety-specific fruit descriptors. The remaining 16 terms were ranked by frequency of use and presented to the panelists with a subset of all experiments. The final list of descriptors contained nine terms: fruity (specific to varietal), smoky, toasted, cedar, spicy, coconut, vanilla, bitter, and astringent (Table 4). To confirm that the panelists were using the terms clearly, a wine was spiked with the physical descriptors used to make the smoky, toasted, bitter, and astringent standards. Panelists then practiced using a scorecard identical to the one used in the experiment comparing these four spiked wines to the unspiked control. Panelists were unaware that they were tasting spiked wines. Statistical analysis of these results showed that the panelists were using these terms consistently.
One wine type (that is, the five treatments of one wine) was evaluated each day over six different days within a two week period. Three bottles of each oak source barrel type of the day's wine were opened in the morning and examined, decanted, and blended together and transferred back into clean bottles. Three hundred seventy five milliliters were taken from this larger sample of each barrel type, and blended together in equal proportions to create a reference wine. Thus, the wines presented included: Missouri, Virginia, Pennsylvania, Minnesota, the Blend of all four wood types, and the laboratory produced blend of all five wines. Bottles were fitted with pour spouts and remained at room temperature (approximately 20 to 22 Celsius) throughout the testing day. Tasting occurred by appointment and all wines were tasted within 8 hours of opening. All aroma and taste standards were available to panelists in a nearby countertop, and they were encouraged to visit these standards at any time. Panelists were seated in individual booths and presented with all five wines in random order together with the laboratory produced reference wine. All sample sizes were 30 milliliters, poured into by and INAO-ISO glasses between 15 and 35 minutes before being tasted, and immediately covered with the plastic Petri dish. Sixty milliliters of reference wine was supplied; more was supplied if panelists requested it. Wines were coded with random three digit numbers. Panelists were instructed to first smell and taste the reference wine, then smell and taste the first coded sample. A 100 mm horizontal line marking scale with indented goal posts was used (Figure 1). The reference wine was used as a midpoint for all descriptors. Panelists were instructed to mark the score sheet with the vertical line corresponding to the relative intensity of each descriptor for the first wine. They were then instructed to rinse with the water provided, and repeat with the next wine. They were to proceed through the rest of the set using this protocol. Afterr they had finished scoring each of the five wines, they took a five to ten minute break, and then proceeded onto the next set. This was repeated one more time per day, resulting in a total of 15 scored wines per day per panelist. Panelists had not been informed of the total number of experiments and thus could not guess that they were doing three replications. Scores were converted to numerical values (with the midpoint at 50 mm) and entered into a Microsoft Excel spreadsheet for later importation into Minitab v 13 for statistical analysis.
Industry Trade Tastings of all Fourteen Wines
Between December of 1998 and February of 2000, eight tastings of subsets of the various wines were conducted. Attendance at each tasting varied between 26 and 105 winemakers or other wine industry professionals. The combinations of wines poured and attendance resulted in some trials being evaluated by as few as 26 tasters and others by as many as 222 tasters (Table 5: Descriptors and recipes for scoring terms). Wines were poured in flights of white wines followed by flights of red wines, using 30 mL self-measuring pouring spouts, in alphabetical order based on barrel codes. Tasters were asked to self-randomize their tasting order, with various locations in the room instructed to taste left to right or right to left, starting with different samples. Tasters were asked to score the intensity of 13 descriptive terms on a nine point scale using a modified CERA barrel committee scoresheet, with 1=lowest intensity, 5=average intensity, and 9=highest intensity. In addition, a hedonic rating (1=lowest pleasure, 5=average pleasure, and 9=highest pleasure) and a preference ranking (1=most preferred, 5=least preferred) were recorded (Figure 2). Physical aroma and flavor standards corresponding to the scoresheet were not provided. Written descriptions of the characteristics were provided and reviewed verbally prior to tasting (Table 6). Results were tabulated in a Microsoft Excel spreadsheet and data transferred to Minitab v13 for statistical analysis.
Results and Discussion
Trained Panel Descriptive Analysis of the Six Wine Subset
Analysis of Variance
Analysis of variance was performed on the sensory data obtained using the trained panel for the six wines and the nine terms. Significance differences were found for various terms throughout the wines. Terms which are significant when analyzed by individual wines were astringent, bitter, cedar, coconut, smoky, spicy, toasted, and vanilla (Table 7).
Feulliat has shown that the cis oak lactone changes qualitatively as the concentration increases. Lower concentrations are described as "vanilla-like" and "oak-like", while higher concentrations use the term of "coconut". The trained panel, unaware that they were tasting oak trials, used the two terms independently. Because they may have been describing different concentrations of the same compound, an extra manipulation of these data is included. The scores for coconut and toasted oak were combined and averaged to give the term "average coconut and toasted oak". The manipulated term "average coconut and toasted oak" was also significant within some individual wines.
Analysis of variance on the entire data set was also performed, and differences between the oak sources were again found (Table 8). In every case, the effect of the wine itself was significant, and in many cases a significant wine x oak source interaction existed. That is, the effect of the wine itself was universal, and in some cases the effect of the oak source was not consistent across wines. When the data for all six wines are pooled and analyzed together, controlling for wine and oak source, the significant descriptive terms are coconut, smoky, toasted, and vanilla.
Mean value differences of significant terms
Differences in the average scores were compared using Fishers LSD for the appropriate significance level. These differences are notated using the traditional letter abbreviations (Table 9).
The significance and level of these values for each term and each wood source were not consistent across the different wines. To summarize when these values appeared to show a trend and when they appeared random the number of times a term was scored significantly high and significantly low was summarized (Table 10). Based on this summary, each of the five oak sources showed these tendencies:
Blend: The blend of the woods appeared to be low in the term coconut, and high in the term toasted.
Minnesota: The Minnesota oak source appeared to be lower in smokiness, toasted, and the manipulated term of "coconut and toasted". It did not appear to be higher in any of the terms.
Missouri: The oak sourced from Missouri appeared to be lower in the terms astringent and vanilla, and higher in the terms coconut, toasted, and the manipulated term "coconut and toasted".
Pennsylvania: The Pennsylvania barrels appeared to be lower in coconut and fruity. It appeared to be higher in smokiness.
Virginia: The Virginia oak appeared to be lower in astringents, smokiness, toasted, vanilla, and the manipulated term "coconut and toasted". However, it seemed higher in the individual term coconut.
Principle component analysis Principle component analysis on the average data of the six wines was performed using four terms (fruity, astringent, smoky, and the manipulated term "average coconut and toast") (Figure 3). These terms were chosen for their significance, their ability to separate the data, and for their similarity in terms used to differentiate the trade tasting results. The first axis accounted for 55% of the variation and is related to the terms astringent, smoky, and "coconut and toast". The Missouri oak and the Blend of all four woods were high in these characters, while the Virginia oak was low in these characters. The second axis accounted for 29% of the variation and was primarily related to Fruitiness; the Minnesota wood was high in this character, while the Pennsylvania wood was low.
Industry Tastings of All Fourteen Wines
The use of untrained panels to do casual descriptive analysis of wines is increasingly common as the industry has become aware that descriptive results can be more valuable than simple preference responses, particularly when discussing new techniques of production. Many people in the trade are familiar with the concept of scoring wines based on descriptors. The purpose of these trade tastings was twofold: to gather as much sensory data as possible on the trials, and to generate interest in the experiment in general. From a classical sensory standpoint, the latter purpose, which requires that the tasters have some prior knowledge of the type of experiment, will bias the results and invalidate the former. Although the samples were coded, and tasted in a self randomized order, the mere knowledge that these were oak barrel trials coded with the simple letters A through E could induce complications and compromise the results. In addition, the terms for descriptive use were chosen for the panelists and no physical standards were provided. It was hoped that most of those participating had tasted enough barrel trials to be familiar with the common words used to describe them. It was also hoped that a written descriptions of the terms would be adequate to clarify the descriptors. Again, this situation is in conflict with classical sensory methods. Sixteen terms were scored during the trade tastings. Fourteen of these could be considered descriptive terms, one a "hedonic" score, and the final score was a preference ranking. This paper will only deal with the 14 descriptive terms. To determine if the trade tasting results were valid enough for further evaluation, the results from these tastings on the six wines which were also evaluated by the trained panel were first compared and contrasted.
Analysis of Variance
Analysis of variance was performed on the results obtained from the trade tastings. And as with the data on the trained panel, analysis was performed on individual wines and on the sum of all wines, controlling for the wine and the oak source (Table 11). Significant differences were found amongst the various terms. To avoid missing trends, significance was considered to be p<0.2 or better.
Unlike the laboratory tests, the trade tastings were not balanced, so combining all data skewed the results toward the tastings with more data. To control for this, mean scores from each of the fourteen wines for each of the fourteen terms were also analyzed; eight terms were significant of p<0.2 or better (Table12).
Mean value differences of significant terms
The eight significant terms from the mean scores for all fourteen wines were evaluated for significance using Fisher's LSD. (Table 13). The overall terms found to be significant in the trade tastings across all the wines include: fruit flavor, oak aroma, vanilla, spicy, raw, roasted, smoky, and bitter.
Validation of the trade tasting data
The same six wine subset evaluated by the trained panel were analyzed separately from the trade tasting data. Similarities in the description and the significance were found. Both panels used some description of fruit intensity, vanilla, spicy, the terms smoky, bitterness, and astringency. The trade tasting used a term to describe overall oak aroma and flavor, while the trained panel developed two terms which may be related to this overall oak impression: coconut aroma, and toasted oak aroma. To make the trained panel data more comparable to the trade tastings, the manipulated term "toasted oak and coconut" (an average of the scores for toasted oak and coconut) was used in a summary principle component analysis.
Principal component analysis of the six wine subset
Principal component analysis was performed in the average data from the trade tastings on the six wines using the terms of fruit aroma, astringents, oak aroma, and smoky (Figure 4). These terms were chosen for their significance, their ability to separate the data, and for their similarity in terms used to differentiate the trained panel tasting results.
Comparison of the PCAs of the six wines from both the trained panel data and the trade tasting data demonstrates two things: not only are there distinctive sensory characteristics in wines made from different American oak sources, but also that a trained sensory panel and untrained trade panel found similar differences amongst the wines and differentiated the trials using similar descriptors.
Principal Component Analysis of All Fourteen Wines
Further analysis of the trade tasting data using all fourteen wines and the same for descriptors showed some effect of the addition of the eight additional data sets (Figure 5).
Analysis of the average data from each of the fourteen wines was also subject to PCI using the eight terms found the significant when the average data was analyzed. This data is displayed in two forms: one identifying the sources of the oak, and the other identifying the wine itself (variety and vintage). It should be noted that despite significant differences found due to the origin of the oak, the wine itself, (that is the variety and the vintage), have more impact on the flavor than the oak wood source (Figures 6 and 7).
A trained sensory panel found differences in six wines due to the regional sources of the American oak wood, and industry tastings of the same wines plus an additional eight found similar differences. Barrels made from wood sourced in Minnesota, Virginia, Missouri, and Pennsylvania, along with a control of a blend of the wood, were significantly differentiated by a trained panel across six wines using the terms coconut, smoky, toasted, and vanilla. Terms which were significant when analyzed by the individual wines include, in addition, astringent, bitter, cedar, smoky, and spicy. Trade tastings using a standardized list of descriptors associated with barrel trials significantly differentiated across fourteen different wines using the terms fruit flavor, oak aroma, vanilla, spicy, raw, roasted, smoky, and bitter. Each experiment consisted of between four and 10 barrels per treatment, and within each of the many experiments there are wines which disagree with the findings the averaged results. Whether this is a reflection of the natural variation to be expected in a barrel trial, or whether this variation is due to a manifestation of a barrel-to-wine or a barrel-to-vintage interaction based on wine chemistry or other variations is unknown and impossible to determine from this data set.
Feuillat, F., Keller, R., Sauvageot, F., and J.-L. Puech. Characterization of French oak cooperage (Quercus robur L., Q. petraea Liebl.). Research of the study group on barrel-aging Burgundy wines. Am. J. Enol. Vitic. 50:513-518 (1999).
Lindblom, B. CERA barrel committee tasting protocol. Am. J. Enol. Vitic. 50:527-533 (1999).
Marco, J., Artajona, J., Larrechi, M. S., and F. X. Ruis. Relationship between geographical origin and chemical compostion of wood for oak barrels. Am. J. Enol. Vitic. 45:192-200 (1994).
Mosedale, J. R., Puech, J.-L., and F. Feuillat. The influence on wine flavor of the oak species and natural variation of heartwood components. Am. J. Enol. Vitic. 50: 503-512 (1999).
Sauvageot, F., and F. Feuillat. The influence of oak wood (Quercus robur L., Q. petraea Liebl.) on the flavor of Burgundy pinot noir. An examination of variation among indiviual trees. Am. J. Enol. Vitic. 50:447-455 (1999).
Towey, J. P., and A. L. Waterhouse. Barrel-to-barrel variation of volatile oak extractives in barrel-fermented chardonnnay. Am. J. Enol. Vitic. 47:17-20 (1996).
Towey, J. P., and A. L. Waterhouse. The extraction of volatile compounds from French and American oak barrels in Chardonnay during three successive vintages. Am. J. Enol. Vitic. 47:163-172 (1996).
Table 2: Letter codes for identifying oak sources
Table 3: Wines produced, number of barrels per lot
Table 4: Wines tasted, dates, locations, and number of tasters, and evaluation method
Table 5: Descriptors and recipes for scoring terms
Table 6: Attribute descriptions for industry tasting
Table 7: AOV of Individual Wine by Trained Panel
Table 8: AOV of Average of Six wines
Table 9: LSD on means of trained panel data
Table 10: Tabulation of times significant means of Six wines were scored High or Low by trained panel
Table 11: AOV of Individual Wines by Trade Tastings
Table 12: AOV of Average of Fourteen wines by Trade Tasting
Table 13: LSD on means of Trade Tasting data
Figure 1: Descriptive Analysis Scoresheet
Figure 2: Barrel Trial Evaluation Sheet
Figure 3: PCA of Average Trained Panel Data
Figure 4: PCA of Average Trade Tasting Data: 6 wines only
Figure 5: PCA of Average Trade Tasting Data: 14 wines
Figure 6: PCA of Average of 14 Individual Wines by Trade Tasting Data: Identified by Oak Source
Figure 7: PCA of Average of 14 Individual Wines by Trade Tasting Data: Identified by Variety and Vintage