Methodolgy

Background

Accurately predicting the quality of winegrapes is extremely difficult. We may believe that aspect and the diurnal temperature variation of the soil effect grape quality. But how much? There is no consensus on what the combination should be.

Take temperature for example. Everyvine’s analysis of growing-degree-days has found little direct correlation between indicators of quality and degree-days. It appears that being within a range of temperatures is necessary but there is no perfect temperature. Instead characteristics are interrelated. The optimal temperatures depend on many characteristics of the site as well as how the vineyard is managed.

A reasonable question might then be, “How is Everyvine’s Vineyard Rating possible?” The answer is modern statistical methods and principally machine-learning. In his recent article in Scientific American, Yaser S. Abu-Mostafa explained the technique: “At its simplest, machine-learning algorithms take an existing data set, comb through it for patterns, then use those patterns to generate predictions about the future…and it is everywhere. It makes Web searches more relevant, blood tests more accurate, and dating services more likely to find you a potential mate.”

Everyvine Vineyard Rating Chart

Everyvine Vineyard Rating Process

Everyvine’s Method

Everyvine’s analysis is built from the following assumptions:

  1. Grapes that sell for above average prices are likely to be high quality
  2. Grapes that that go into award winning wines are likely to be high quality
  3. Site characteristics like aspect, elevation, climate, and soil effect grape quality.
  4. Vineyard Management and planning effects grape quality.

From these assumptions Everyvine has built rating algorithms. Our algorithms leverage some of the latest developments in the field of statistical learning. They are complex and evolve as available data increases but a basic idea of how our ratings are achieved is quite understandable.

Vineyards are rated by individual planted block and grouped by varietal. The ratings algorithms comb through the available data for each wine grape varietal and identify common patterns of characteristics among vineyards producing high quality wine grapes. If enough data is available about a grape varietal to draw accurate conclusions, our algorithms then rate any vineyards growing that varietal. If a vineyard is has a strong track record of performance and fits patterns our algorithms have identified as important it is rated highly.

Further reading

Want to learn more about this topic?

  1. Wikipedia Machine Learning

    Overview of machine learning techniques.

  2. Abu-Mostafa, Yaser S. “Artifical Intellegence – Machines that think for themselves.” Scientific American. 7 2012: 78. Print.

    Great introduction for the layperson to machine learning.

  3. The Elements of Statistical Learning: Data Mining, Inference, and Prediction (Second Edition) by Trevor Hastie, Robert Tibshirani and Jerome Friedman (2009)

    Our favorite reference for modern statistical learning methods.

  4. Gladstones, J. (1992). Viticulture and Environment. Winetitles, Adelaide.

    A good overview of how site characteristics impact winegrapes.

Skip to toolbar