Researchers Introduce AI Tool to Help Olive Farmers Predict Harvest Timing

Using machine learning to analyze a range of data points from model farms, researchers were able to predict the timing of the olive harvest with 90 percent accuracy.

Following more than three years of development, the results of the Predic 1 Operational Group’s work were presented last month at a conference in Mengíbar, Jaén.

The group’s remit was to deliver a platform capable of predicting olive harvests an entire season in advance, a goal they said they accomplished with an accuracy of up to 90 percent.

The work was carried out by a consortium comprising the University of Jaén, Cetemet, Citoliva, Cooperativas Agro-alimentarias de Andalucía, a farmers’ union, and Nutesca, using traditional Picual olive groves in Jaén, Córdoba and Granada as test cases.

According to María Isabel Ramos, a professor at the University of Jaén’s Department of Cartographic, Geodetic and Photogrammetric Engineering and corresponding author of a 2022 study about the technology, predictive systems are crucial to the future of the olive sector.

At the scientific level, crop harvest prediction is one of the most complex problems within precision agriculture,” she said. There are several studies that make these predictions based on the close relationship between the emission of pollen and fruit production, others from aerobiological, phenological and meteorological variables, all with efficient and acceptable accuracies from July onwards.”

We intend to advance this prediction and be able to make optimal predictions in the period before flowering… long before the farmer carries out their strategic planning and economic investment in the farm,” Ramos added.

The group used data mining methodologies previously used in predictive healthcare projects to create regression models from meteorological data and historical harvest data from across the initial target area.

This was combined with current data from drones equipped with thermographic sensors and multispectral cameras, satellite imagery, phenological assessments, foliar and soil analyses and data collected from model farms.

The model utilizes machine learning, the best-established field of artificial intelligence and one with a proven track record in agriculture, to predict crop yields as accurately as possible.

Using a support vector machine algorithm made it possible to use multiple kernels, namely the linear and Gaussian kernels. This makes it easier for the algorithm to adapt to the nature of the data, allowing infinite transformations to be carried out. More