Yes it does, says Benjamin Lewin MW—even if we cannot yet demonstrate it by objective means.
By
Before I get myself into trouble with the title of this column, let me say straightaway that I absolutely believe in terroir based on my experience of tasting wine, but I have never been convinced by any attempt to demonstrate a basis for it. (There might be a career to be made in debunking claims that the effects of terroir on wine have been “proved.”) What I want to consider here is the nature of the gap between subjective experience and the methods that might be used to demonstrate terroir on an objective basis.
No one who is interested in Burgundy can really doubt the existence of terroir. There are simply too many cases where two adjacent vineyards give consistently different results, even though, so far as we can tell, all variables other than the site are similar. I could fill this whole column with examples from the Côte d’Or, but one will suffice.
Cazetiers and Combe aux Moines are neighboring premier crus in Gevrey-Chambertin. Thanks to its more northerly exposure, Combe aux Moines boasts a cooler milieu. Conversely, Cazetiers sprawls further down the incline boasting a marginally lower mean altitude bundled with a richer soil. This could delineate why Combe Aux Moines harvests almost two days post Cazetiers. The Faiveley plots in both vineyards share boundaries, with their technical director Jérôme Flous humorously asserting that “The tractor doesn’t stop.” Both vineyards receive identical tending and vinification methods. A side by side taste test of the wines from different years reaffirms a consistent difference: Cazetiers invariably presents a more refined impression, whereas Combe Aux Moines exudes a more robust aura. This embodies the essence of terroir.
The case for terroir in Bordeaux is somewhat less convincing, largely owing to the fact that all wines here are blends. Manifesting persistent and nuanced terroir shades across different vintages becomes a challenging task, especially when there is a variability in proportion of varieties blended each year. With production scale exceeding that of Burgundy, wines here are an amalgam of different vineyards and varieties, occasionally originating from extensively scattered plots, thus representing a medley of terroirs.
As far as distinctiveness in Bordeaux’s terroir is considered, the most suitable candidate is the famed gravel mounds on the Left Bank. This term pertains to geology rather than gardening-associated gravel. A gravel mound is a construct of topsoil atop packed sand, that rests on a gravel bed, a few centimeters to 2-3 meters thick. Underneath this are alternate layers of compressed sand and clay lens.
The first growths occupy prime positions on the gravel mounds, these are located on deep gravel with a low water table. The favored locations are warmer than surrounding areas, which might have been advantageous historically but might become a mixed blessing in the face of climate change. Other leading châteaux also mainly lie along the gravel band, but their terroir is more varied. Excluding the first growths, there’s no real geological evidence supporting a hierarchy among them or distinguishing them from nearby châteaux. The claims that terroir has any connection with the 1855 Classification, for instance, are incredibly tenuous.
Considering this background, it’s intriguing that a claim to discern the uniqueness of specific producers should come from Bordeaux. This claim is founded on using gas chromatography to identify a range of volatile molecules in wines from seven châteaux—four on the Left Bank and three on the Right Bank. Chromatography is a method for separating compounds due to their differing movement rates when carried by a fluid. The publication was accompanied by considerable media hype—the New York Times proclaimed “Bordeaux Wine Snobs Have a Point, According to This Computer Model.”
The University of Geneva (which collaborated with the University of Bordeaux on this research) made fairly dramatic assertions in a press release. “Artificial intelligence has succeeded in identifying with 100 percent precision the chemical identity of red wines coming from seven grand domaines in Bordeaux […]. This paves the way for potential new tools to counter fraud.”
The primary assertion of the article is that the chemical composition of its wines can distinguish each estate. In principle, this isn’t exactly surprising. Unless one harbors supernatural beliefs, it’s evident that the chemical makeup of any two wines must differ. However, given the complexity of wine, all previous attempts to identify individual molecular markers for châteaux or regions have failed to persuade, although markers for different varieties certainly exist.
The compounds analyzed in the raw study fall into three groups: esters (the main class of volatile compounds in wine), oak (specifically those introduced through oak aging), and off-flavour (attributed to known defects). The machine learning method used condenses each château and vintage data to a single point in a two-dimensional chart (one point per tested vintage, effectively combining the data from the three categories for that vintage). Points for each château create a rather distinct cluster, indicating the potential to identify each château by a distinct profile.3
The findings are significantly undermined by the fact that the châteaux are not named. All we know are the appellations: one wine from Pomerol, two from St-Emilion (but it’s unknown if they’re from the plateau or the plain, which is critical), two from Pauillac, one from Margaux, and one from Pessac-Léognan. Consequently, the study doesn’t provide enough information for others to replicate it, breaching a fundamental scientific principle.
To arrive at the straightforward two-dimensional chart, a considerable amount of original data manipulation is necessary. This doesn’t imply anything negative; it has been an increasing trend in science over the past half-century for scientists to distance themselves from their data due to complex analytical methods. Ultimately, AI might be the method that most separates scientists from their data.
In my book, Inside Science, I explore the appropriate level of data convolution. The book aims to explain the reality of science, including the need for transparency and repeatability, which is challenged by artificial intelligence. By its nature, AI makes it difficult to maintain traditional criteria for accepting scientific outcomes. A potential solution to verifying AI-generated conclusions in experimental science is to examine downstream impacts and see if other methods can validate predictions. However, such testing isn’t possible here.
With some skepticism, I consider whether the significance of this type of analysis should be accepted. Essentially, artificial-intelligence methods reduced up to 30,000 individual data points from each wine’s original chromatogram into a single data point on the final plot. Accepting this requires a leap of faith for those traditionally trained in the scientific method, given the lack of any conventional error bar to indicate the reliability range.
Proceeding with the argument, let’s assume this could be a method identifying chemical differences between wines. What would likely cause this difference? Probably varietal constitution. The wines vary drastically. Excluding a varietal Merlot, all are blends, with Cabernet Sauvignon constituting from 0 to 90 percent, Cabernet Franc from 0 to 56 percent, and Merlot from 0 to 100 percent. Given the high aromatic profile of the Cabernets, this range suggests that varietal range alone would result in significant chromatogram variation.
If we are to seriously consider this study indicative of individual châteaux wines’ distinctness, we must set aside the possibility of varietal differences predominantly accounting for results. The study lacks conventional controls to address this or any other variable impacting a wine’s aromatic profile. However, there are data plots intended to show that varietal constitution doesn’t explain everything. These are based on computational analysis, which in some way is a type of double manipulation. If it’s true that each château can be identified by its profile, despite the blend’s annual variation, it poses that varietal constitution is not the whole explanation.
Understanding the identities of the château is quite tricky as the varietal compositions stated for several châteaux don’t match any property in the defined appellations. The wines from the château equivalent to Pessac-Léognan are said to have 75-90 percent Cabernet Sauvignon and around 5 percent Petit Verdot. Locating a classified growth in Pessac with this high level of Cabernet Sauvignon would be quite intriguing.
I made the decision to decipher the identity of the châteaux by referring to the yearly stated varietal compositions of their wines.5 To my surprise, the results suggested the wines in question were Lafite Rothschild, Latour, Margaux, and Haut-Brion from the Left Bank. The identity of the châteaux on the Right Bank was a bit harder to figure out, but the patterned coding hinted that the 100% Merlot might be Château Petrus (with an alternative being Le Pin). It also became clear that one of the St-Emilions was definitely a Château Cheval Blanc . However, the details for another potential St-Emilion did not seem to match up with the Ausone data. It seemed to me that the researchers had only paid attention to the most acclaimed wines of Bordeaux.
The only issue with this though, was that multiple châteaux seemed to have been wrongly assigned to appellations. For instance, the data for Château Haut-Brion was linked to an estate in St-Emilion; the information for Château Margaux matched a Pessac-Léognan estate; and the details for the château in Margaux seemed to be for Lafite Rothschild instead. This understandably raised doubts whether the extensive data reduction that was used to create the basic graphs of the study was trustworthy.
I then chose to revisit the core data, which seemed to suggest that the three châteaux of Médoc all fell closely together. This was separate from the clustering of the châteaux coming from the Right Bank and the single château from Pessac-Léognan. If only things were neatly such that the geographical arrangement of Bordeaux was mirrored in the plots that represented their aromatic profiles. The issue came when I chose to identify the châteaux based on their varietal composition rather than their stated appellations, at which point, the châteaux from the same area stop clustering together. For starters, the estate lodged in between Médoc and Libournais was, in fact, not in Pessac-Léognan but Château Margaux.6
Ignoring these muddles for a moment and assuming each château does have an individual profile, a new question arises. A question which spreads some uncertainty over whether the study’s conclusions can be extended further. Considering the analysis was based only on seven of the most unique châteaux in Bordeaux, is it truly justified to conclude the method can certainly identify every château? It’s no surprise the most renowned châteaux in Bordeaux have distinctive features. Would the same results have been found had the seven châteaux been at the level of cru bourgeois or were they petits châteaux?7
Other factors that might alter the outcome could include the types of yeasts utilized during fermentation, the extraction methodologies employed during fermentation, the maceration duration, the quantity of new wood utilized during aging, and the length of the aging process. The amount of new oak, for example, is speculated to be a significant influence on the aromatic portfolio. Possibly, exposure to new oak during aging could be discerned through chromatographic analysis. However, due to the lack of information concerning the oak-aging regimes utilized in these specific findings, it’s impossible to ascertain to what degree it might have affected the outcome. If my assumptions about the châteaux’s identities are correct, new oak would have comprised a significant component and might have contributed nearly 100 percent in all instances.
The amalgamation of numerous influences resulting in the aromatic variations in the wines could be distinctive for each château. However, it appears too early to deduce that individual châteaux can be evaluated based on the outcomes of seven out of 10,000 châteaux in Bordeaux, especially considering that these seven are among the most distinctive. The two samples originating from Pauillac are the sole instances where the châteaux are sufficiently proximate for individual variation, as opposed to generic geographical origin, to be a plausible explanation. The sparse data provided by only two châteaux doesn’t lend much impetus to conclude that this is a promising new approach.
Thus, the aspiration of detecting fraud with this method is still remote. It might, however, prove beneficial for broadly examining geographical origins, possibly even for investigating assertions of 100 percent new oak usage, or more ambitiously, to differentiate barrique usage from staves or oak dust. If we take the results at face value, the technique wouldn’t be useful in detecting cases where one vintage is misrepresented as another.
The scenario presents an intriguing conflict between producers and terroir. Traditional terroir theory argues that individual vineyards should be recognizable through common attributes in the wines’ aromatic profiles. Would a similar analysis of Burgundy identify groups of wines by geography (terroir) or by producer (winemaking)? Without considering this issue, generating a reliable interpretation of the results is impossible. While I don’t dismiss the prospect of uncovering something fascinating, the case is not definitively proven, and the assertions made for the findings are overstated, at the very least (a situation unfortunately quite common in modern science).
Imagine we want to utilize this kind of analysis to verify that terroir generates a unique chromatographic profile. What steps would be taken? Fundamentally, it’s crucial that the vineyards under examination grow the same single variety of vines of the same age and cultivar. Clearly, the vineyards must also be maintained in an equivalent manner. Moreover, the winemaking process for each should be identical. The ideal situation would be for the wines to be aged in conditions without oak, thus minimizing external influences on the aromatic profile. Attaining these conditions are challenging enough that future fulfilment seems unlikely.
There’s another vital component necessary to differentiate between the effects of terroir and the impacts of winemaking. Grapes ought to be analyzed right after they’re harvested and the completed wines, with the most suitable point of analysis being at bottling. Supposing the differences are due to terroir, it’s logical that these would be noticeable in the grapes. Although fermentation could either amplify or diminish the differences, those who use indigenous yeasts for fermentation (considered part of the terroir) would contend that adding a specific strain for fermentation lessens the terroir’s impact. Analyzing only grapes might be met with criticisms that any differences found may not significantly influence the wine’s organoleptic profile; however, only analyzing the wine may leave room for the possibility that the effects are a result of particular features of winemaking.
The biggest contributor to aromatic variability in wine is yeasts – the hefty majority of aromatic compounds come about from fermentation rather than originating from the grape. This is especially true in the case of Brett yeast.
The yeast strain Brettanomyces, or Brett, when it contaminates wine, leads to a uniquely potent aromatic profile, characterized by the compounds 4-ethylphenol and 4-ethylguaiacol. Between the 1980s and ’90s, numerous Bordeaux châteaux wines were characterized by Brett, resulting from winery-wide contamination and not isolated incidents of bottle contamination. Might this be seen as a terroir component? During that era, this emerging method probably would’ve been successful in differentiating between the wines from the contaminated châteaux and those from the uncontaminated ones. To some extent, the yeasts used in fermentation could potentially contribute to distinctive outcomes within individual châteaux.
The debate on whether yeasts are part of terroir is complex. Personally, I believe they are not because the strains of indigenous yeasts differ annually. Inclusion of yeasts would imply that Brett was part of some châteaux’s terroir, albeit briefly, during infection times. This oxymoron illuminates the difficulty of the situation, highlighting the necessity to exclude yeast from terroir.
Three years ago, I evaluated some inflated assertions regarding the “undeniable evidence” of terroir through chemical analysis, published in WFW. The article titled “How to Prove Terroir Exists (and Do We Really Want to?)”.8 This research focused on Malbec grapes harvested from various individual vineyards in Mendoza. My conclusion was that the results evidenced that the chemical distinctions are altitude-induced, predominantly because increasing ultraviolet exposure elevates tannins and total phenols due to grape skin thickening. This example shows that the laws of physics apply to viticulture, but, it does not demonstrate any influence of terroir apart from altitude.
Does terroir exist or would it express itself differently if a region changed the types of grapes it grows? Let’s imagine we are in 2050. It is much hotter everywhere, and Bordeaux has finally conceded to the heat force and permitted the growth of Syrah vines, even replacing Cabernet Sauvignon and Merlot. Would the first growths still remain superior? Will Pauillac and St-Julien still contrast as opulence versus delicacy? Could the disparities be different?
A few years back, I posed a question to Frédéric Engerer of Château Latour, asking him whether he believed that its character would retain its essence with a different grape variety. He responded that it might be possible but would be complicated to grow Syrah, although the existing effects would still be observable. According to Engerer, the terroir is the noble aspect. The Cabernet Sauvignon merely serves as the tool to express the differences amongst terroirs. He theorized that the same could be true for another grape variety.
Terroir finds its expression through variety. Each variety thrives in its favored terroir: Cabernet Sauvignon in gravel, Pinot Noir in limestone, and Syrah in granite. When we discuss the influence of terroir, we’re predominantly talking about how a variety adapts in variations of its preferable terroir. It is not a certainty that it would have the same sensitivity to other terroirs, or that differing terroirs would have identical impacts on all varieties. This does not negate the significance of terroir; on the contrary, but may necessitate some prudence in extrapolating results from certain situations that have probably undergone strict selection.
So does terroir exist? Absolutely. Can we objectively demonstrate it at this point? Not yet.
I must confess that terroir remains a matter of faith. While this is somewhat uncomfortable for a scientist, it is preferable to accepting substandard science.
Notes
M Schartner, JM Beck, J Laboyrie, L Riquier, S Marchand, and A Pouget, “Predicting Bordeaux Red Wine Origins and Vintages from Raw Gas Chromatograms,” Communications Chemistry 6, article 247 (2023); https://doi.org/10.1038/s42004-023-01051-9
Contrary to the title’s indication, predicting a wine’s origins from raw gas chromatograms requires intricate data calculations. The calculated data plots demonstrate these differences.
These wine samples range from 1990 to 2007 and are collected from the châteaux in 2019. Surprisingly, there isn’t much vintage variation in the results, even though the original objective was to study the aging in red Bordeaux. Justine Laboyrie, “Composition et origine du bouquet de vieillissement des vins rouges de Bordeaux. Influences du terroir dans l’expression aromatique des vins vieux,” Médecine Humaine et Pathologie, Université de Bordeaux, 2020.
The château identities are stated as A (Pomerol), B and C (St-Emilion), D and E (Pauillac), F (Margaux), and G (Pessac-Léognan). In the varietal constitutions table, however, they are termed A, B, F, G, S, T, and V. We can only speculate how S, T, and V correlate to C, D, and E. From the varietal compositions provided, it’s suggested that B is Haut-Brion, F is Lafite Rothschild, G is Château Margaux, and T is Latour. Possibly, S is Petrus, V is Cheval Blanc, and A is unidentifiable. I deduce it’s a St-Emilion château—with Pomerol stated—capable of consistently creating a wine from 50 percent Merlot and 50 percent Cabernet Franc each year.
I surmised that the data pertains to a specific château if every year there’s a perfect match between the varietal composition reported by the château and that provided for the sample, given that no more than two years contradict this.
If our analysis’s sample IDs are accurate, the findings are questionable. This is since the cluster groups would signify different appellations than those the authors identified. As far as the IDs can reveal, it seems that the supposed Right Bank group includes Haut-Brion; Margaux is the intermediate château, not one in Pessac-Léognan as mentioned earlier; and the purported Médoc group must contain at least one Right Bank wine.
By recognizing the château through the varietal composition, I conclude that the grand vins are the wines analyzed, not the second wines. However, it underscores the idea that comparing each grand vin with the second wine might be enlightening.
“Terroir and Vintage Discrimination of Malbec Wines Based on Phenolic Composition Across Multiple Sites in Mendoza, Argentina” by R Urvieta, G Jones, F Buscema, R Bottini, & A Fontana was published in Scientific Reports volume 11 in 2021, article number 2863.