Webinar recap: AI poised to improve accuracy and reduce time in the vineyard



Team Terraview

Wineries, consultants, and AI experts discussed current and advanced models.

Vineyards are one of the few crops where sometimes less is more. As wineries transition towards better wines, they are becoming more selective about which berries they send to crush. To achieve this goal, winemakers will need improved tools to manage the variances in quality of pre-production berries and tooling that extends to growers who work for them. The receipt of tons of grape at the press can prove a daunting task for any operation. This need for insight is further compounded by climate change. Thus, accurate harvest estimation has become the holy grail for many producers. Embracing the challenge, Terraview organized a webinar where different perspectives came together.

Álvaro Maestro, Technical Director of the renowned winery Emilio Moro, is always keen on new technologies and high-quality wines. Together with Maestro, Manuel del Rincón brought his experience as an agronomic consultant to the table. Lluís Nache explored the potential of AI to improve current models. Finally, Romain Bonnaud, Terraview Country Manager for Spain, assessed the current situation of the company and the winemaking sector.

Why do wineries need harvest estimation?

In his intervention, Maestro pointed out some of the reasons to estimate yields. Besides avoiding a surplus or not reaching the desired production, meeting the quotas from appellations of origin was high on the agenda. In appellations like Rioja, producers cannot go over 7K per hectare for white grapes. Production levels, of course, also have an impact on prices paid for grapes. In Ribera del Duero, Emilio Moro’s stronghold, limits are similar. Besides regulatory concerns, predicting harvest quality is one of the main challenges for Maestro’s winery. ‍

The above considerations have a substantial impact on viticultural practices. This translates into thinning grape clusters to reduce yields or resorting to irrigation and fertilizers to increase them. Of course, financial strategies also need to adapt to the final output—prices per kilo of grapes, bottles produced, or storage costs depend on that.

Manual methods are still the primary source of intelligence

As Maestro remarked, Emilio Moro uses manual methods for their harvest estimation. This means selecting a certain number of vines per hectare and counting berries. The whole procedure requires only 25 minutes per plot, but accuracy can be relatively low, sometimes below 20%. ‍

Manuel del Rincón explained his own method, which generates a higher accuracy, but is more labor-intensive. Rincon’s method uses reference vines every year and sort grape clusters into large, medium, and small types. Then berries, and even seeds, are counted to arrive at the final estimation. Other factors also include measuring nutrient deficiencies, water pooling, vigor, and diseases like powdery mildew. With this method, del Rincón claimed an accuracy of within 10% of actual harvest.

The AI angle

Lluís Nache provided an overview of how different technologies come together to offer AI-driven insights to wineries. The starting point was artificial intelligence and its two main branches, machine learning (ML) , and deep learning. ML allows algorithms’ training over time, while deep learning uses neural networks to crunch through vast amounts of data.

The second element for a robust AI-driven estimation model would be a wide range of data sources. The most important would be aerial imaging, both from satellites and drones, open weather maps (OWM) and weather stations, and historical data, mainly harvest and weather data. Aerial imaging can give insights into vigor and health. In turn, weather stations measuring temperature, rain, wind, dew points, and other parameters, feed valuable data into the AI engine. Finally, historical data will be essential for machine learning, to achieve higher accuracy in earlier stages.

Nache stressed that many other parameters could also be fed into the systems—from grape varieties and soil types to plot orientation, altitude, and slopes. Coupled with 3D grape modeling on the ground, AI systems can achieve up to 98 % accuracy over time.    

The challenge for wineries is to start leveraging the data sources they already have, as well as introducing new ones, to obtain better insights. When labor is increasingly scarce in the fields, recommendation systems such as those provided through AI can allow winery managers to make the most of their time. This sentiment is shared by wineries such as La Rioja Alta and Roberto Frías, its technical director, who Terraview interviewed a few weeks ago.

As Bonnaud explained in his opening speech, improved decision-support systems are not only key for better harvest estimation, but also to ensure the sustainability of wineries in the long run. If you are interested, you can watch the whole webinar here.