The En Primeur 2021 campaign has come to an end. Defined by most – first of all by us “thanks to our satellite-derived data” – an heterogeneous vintage, it also raised some concerns, given the shy reaction of the market.
As you may have already heard, thanks to artificial intelligence, Saturnalia leverages weather and vegetation data – acquired by satellites – to derive a vintage score. It all started with vintage 2018 and, as usual, we are going to check how our model compared with the critics. Before anything else, let me clarify a point: the scores come clearly from two separate worlds. We do not taste the wines, but we assess instead the growing conditions of the grapes, deriving a potential quality score after the harvest.
Now let’s get to the point. How far is our score in comparison of the main critics? Is the Saturnalia Vintage Score – computed in advance – reliable? I’m Daniele De Vecchi, Saturnalia project manager. Let me prove our point with some comparison.
The chart is ordered by distance from the average. Click here to see a more detailed comparison with more wines. At a first glance, it seems like the vintage score assessed well most of the wines. Let’s dig a bit deeper.
The absolute deviation (absolute value of the difference between Saturnalia Vintage Score and the average of all the critics) is represented by this histogram. The average of absolute deviation is around 1.68 points, with 77% of the wines with a deviation of less than 2.5 points. On average, our score fits in +/- 1.7 window around critics’average. A quick comparison with 2020 is shown in the chart below. For 2020, the average absolute deviation is 1.52, with 83% of wines with a deviation of less than 2.5 points. The slight reduction in performance can be explained by the higher heterogeneity of the vintage, leading to a spread in the critics judgements.
You may notice from the chart above that there is a wide variation among all the critics. The minimum range is 3 points while the maximum reaches 16 points. To better understand the variability and heterogeneity of the vintage, we have to consider the distribution (chart below)
Most of the scores are concentrated between a 5 and 8 points window. A quick comparison with 2020 shows that the distribution of 2021 range is wider and centred between 6 and 7. On the other hand, 2020 which was a less heterogeneous vintage, is centred between 5 and 6.
To summarise the main findings:
© 2021 Ticinum Aerospace. All rights reserved