More about predicting harvests
For small harvests, predicting grape maturity can be done in a spreadsheet.
Samples can be collected and predictions issued using linear regressions.
This method requires at least three samples reasonably spaced in time.
It is also possible to profile blocks using statistical and computational brute force against a very large number of historical
samples. This is the method at the core of this service.
Blocks receive maturation profiles in the form of one or more maturation equations taking into account
start, speed and limit of sugars contents.
These equations are adjusted for the current vintage and readjusted if samples do not track.
Overall, predictions change less often than when using a linear model; this results in more efficient planning.
The economic rationale
Grape maturation is often considered linear, this is the result of a traditional approximation, it is actually a sigmoid function.
As the early flat stage of the sigmoid is rarely sampled therefore little known, we use a fast to compute differential equation, which almost exactly
overlaps the sigmoid in its useful parts (a fast maturation phase, a slow maturation phase and eventually an asymptotic phase).
Where this is relevant to mass harvest is that the flattening of the curve starts relatively late in the maturation process
and happens relatively abruptly. Maturation is also subject to weather conditions and blocks ripen at different speeds
based on their location and variety. As a result, a linear model will work on average (looks good on a graph) but only
really provides an accurate result very late in the ripening process. Furthermore, a linear model will almost never predict
the same date with consecutive samples, this is causes much unnecessary re-planning with each wave of samples.
While wine makers can rectify sugar and acidity levels, this is additional work. The situation remains unfortunate though,
as it is in the latest stages of maturation that the grape flavour and characteristics develop the most.
Again wine makers know how to rectify this, with more work...
Conversely, too-ripe grapes may lead to fermentation starting before arrival to the winery and an overall loss of
control of the production process.
In some countries, the Law requires the cessation of chemical spraying for a certain time before harvesting,
at a time when the flattening may not be detectable yet.
In Australia for instance, this period is three weeks at the time of writing this document.
As a result, as one does not want to risk leaving a valuable harvest prey to pests and diseases, blocks predicted
using a linear model are often harvested too early.
Small producers can handle this because they can "see" the harvest. With experience, they can make educated decisions
and usually pick the right harvest time. Larger producers do not have the luxury of "feeling" their harvest;
it is too large, too varied, spread over too many hundreds of kilometers.
Profiling can provide a level of analysis that compensates for the loss of relation with the terroir in large operations.
Modeling the terroir by variety and location and historical maturation allows calculating what cannot be "felt";
as a result it is possible to pick harvest dates closer to the desired maturity level.
In addition, this method allows determining a harvest window during which the grape is acceptable,
which provides some flexibility to organise harvest.
Thoughtpool and harvest-plan.com are registered Trademarks of their respective owners.