Harvest-plan aims to ease the vintage planning process by producing non-linear harvest predictions of grape maturity. Non linear predictions can be hard going even for expert mathematicians with specialist statistical modelling software.
It calls on a body of mathematics and IT called “machine learning” (machine learning is a fairly vast topic, and this post is not really the place to discuss it, so for those who are happy to plunge down the rabbit hole here’s a few links, provided mostly to persuade you not to break your teeth trying this with spreadsheet software …
- Machine learning wikipedia article
- Machine learning Matlab article
- Machine Learning SAS article
- Machine Learning Mathematica case study).
So why do we bother with the maths? So viticulturists and winemakers don’t have to; we’d prefer to see them growing grapes and making wine rather than spending time on maths or synchronising meteorological data. We also don’t want them to sell the house or give up weekends to write software.
We’re also not trying to be prescriptive with predictions. While we can predict the date grapes will reach a desired ripeness, the point of the exercise is to give sufficient, and sufficiently accurate, notice of that event in advance of it happening. The final decision on when to pick will depend on a much broader range of factors, particularly sensory assessments. What we DO want to do is allow people to plan confidently five to six weeks ahead of harvest … so that all the activities related to spraying, transport, winery capacity and harvesters can be in place; and so disruptions like adverse weather events can be handled better.
That’s WHY we predict; the following posts will go into a bit more detail about HOW we predict.