Online yield forecasting tool hungry for data

Two Sydney researchers can already make a pretty accurate prediction of what Shiraz yields will be like in the Barossa next year, but their vision is even greater.

Drs Steve Van Sluyter and Alex Ferry from Macquarie University are one year into a three-year, Wine Australia-funded project designed to develop an online tool for predicting yields at the block level that could eventually be adapted to any region and grape style.

They know about the Barossa because that’s where most of their initial data has come from, courtesy of Treasury Wine Estates. They now need more – and more varied – data for two reasons.

The first is that a reliable tool for a given region will need key data about that region and its various grape varieties. The second is that if all the data comes from vineyards in close proximity to each other, it is not sufficiently independent (in a statistical sense) to use for broader modelling.

‘Essentially we need to more accurately measure our accuracy’, Dr Van Sluyter  said. ‘With these small datasets we’re using to train and test the models. The numbers are looking really good, but we have no way of knowing how that would extend to larger models across more regions and cultivars.

‘We’re reaching out to wineries in other regions now to get the data we’re looking for, including historical yield data and dates for budburst and flowering.’

Three sector partners are already involved with the project, which is both ambitious and complicated.

Previous work in New Zealand has shown the potential of the concept, but it used a simple model and only two variables. Adding more variables, such as weather at key vine stages, adds a lot more complexity and requires use of a technique known as machine learning.

‘Modelling this level of complexity at the block level is not trivial and has not been undertaken in the past, nor has the development of a yield estimation tool from an understanding of these interactions been previously attempted’,  Dr Van Sluyter said.

‘A lot of people think it’s impossible and the reason they think that is because it’s human psychology that we overweight anomalies. When something expected happens we just say “of course” and ignore it, but when something unexpected happens we place huge importance on that.

‘The reality is that the vast majority of the time the grapevines are behaving on average and we can predict that.’

The final tool will incorporate weather data interactions with vine characteristics at key developmental stages. Crop forecasts will be available as early as winter pruning and will be continually updated with increasing precision as the season progresses.

The aim is to only include data that growers are likely to collect for other reasons, so that the labour required to use the tool will be minimised. A browser interface will be developed to make the yield predictions easily accessible.