It is possible to make more money when you make a good forecast of your yield, that has been known for years by growers and traders. However growers, growers associations and traders are struggling with this for years: the inaccuracy of the prognoses makes it not clear enough how much product is expected in the coming period. These prognosis regularly differ dozens of percent from reality. LetsGrow.com has developed, through the use of Machine Learning, for tomato a harvest forecast with an average more than 90% accurate!
The past as input
Many growers make crop prognoses for the sale of their products. With this information the sales process can be optimized. No shortage or surplus of product, ensures a better return on sales. It is also possible to optimize internal processes , such as work scheduling, at the nursery with this information.
Peter Hendriks, Business Unit Manager at LetsGrow.com: "I am regularly approached by growers with the question whether we can build an accurate and universal crop prognosis system. We have for some years now a plant physiological crop prognosis model for tomato which is doing reasonably well. However, it takes time to do the counting for the harvest forecast accurately, so that the model remains accurate. In addition, this way is error-prone, since the model only looks at the present. We also should use data from the past."
Artificial Intelligence vs. plant physiology
For over 17 years, LetsGrow.com has been collecting data from and for their customers. How nice would it be if these data can be separately used by each grower by making it an efficient yield forecasting model? Hendriks: "The technology is developing rapidly. In recent months, we specialized ourselves in Machine Learning. This form of Artificial Intelligence is really an incredibly powerful tool! By combining our plant physiological background with this technique, we can switch quickly to provide very accurate analyzes, such as this prognosis , for our customers."
Is it a lot of work for a grower to do this forecast? Hendriks: "We train the system for the grower. The system is trained per grower with its own data. The system recognizes patterns from the past and in combination with crop registrations that are already being done, the harvest forecast for the coming 4 weeks is predicted. If the historical data of a grower is already in our system, we can immediately start working on it. If not, then we have to find a way how we can retrieve the correct data. In fact, it costs the grower very little work. And it delivers a lot!"
Has the end been achieved considering accuracy with this forecast? Hendriks: "No, certainly not! The name Machine Learning says it all: the system continues to learn itself. It is getting better and better over time. In addition, we are also further refining the underneath model.
We are also working on integrating new data sources to be able to use them in the model, such as the measurements made by HortiKey. They measure, among other things, automatically the inventory of the tomatoes that are still in the greenhouse. If we combine this with our technique, we will reduce the manual input even more and can we further increase this forecast to 99.9%!”
Are you interested in this new and accurate crop forecast? Contact us or visit us at the Horticontact at stand 401. The analysis is now available for tomato growers. Other crops follow quickly.
For more information:
+31 - 657583579
Ton van Dijk
+31 - 653735670