A new algorithm uses satellite images to distinguish olive grove types without field visits

A study by the universities of Cordoba and Seville develops a method that makes it possible to verify, easily and quickly, whether an olive plantation is traditional, intensive, or super-intensive

The traditional olive grove, with large and well-separated trees, is being replaced by plantations with greater density, such as intensive or super-intensive ones, where the space between them is reduced. Productivity increases with this type of intensive and super-intensive plantations, but so does their use of resources, such as water. 

Due to this transformation's agronomic and environmental consequences, but also its economic and socio-cultural ones, public administrations implement policy and issue aid to modulate it. To do this, they need to have updated information at all times to know how many hectares of each type of plantation there are and how fast they are changing.

This need is met by a method based on convolutional networks, a type of neuron network, developed by a team at the University of Cordoba in collaboration with the University of Seville, which automatically identifies these patterns in olive groves using free open satellite images from Sentinel-2. 

"The problem we faced is that, until now, we had this information, thanks to the images of the PNOA (National Aerial Orthophotography Plan), which feature great spatial resolution, but they are updated every 3 years, so we had very outdated information," explains Isabel Castillejo, a researcher in the Department of Graphic and Geomatic Engineering of the UCO. To overcome this problem, the team turned to the use of Sentinel – 2 images (an Earth observation mission developed by ESA) that are available for free every 5 days. These images, however, feature lower spatial resolution, which makes it more difficult to identify patterns, as the treetops cannot be directly observed in the images. This is where Convolutional Neural Networks (CNN) come in, which are a type of Deep

Learning analysis technique used for advanced pattern recognition tasks in data.

"We trained 3 different learning algorithms to detect planting systems using these satellite images and found that the best of the three (approach B) was 80% accurate, a very high percentage considering the difficulty of the problem and the resolution of the input images," explains Cristina Martínez, a researcher in the UCO's Department of Electronic and Computer Engineering. 

The researchers point to another crucial advantage of this method, which is that everything is automated: "just by entering a text file with the code of the plot or the cadastral reference, the plot is defined, with its boundaries, and with that information the satellite images for the period requested are identified, downloaded and entered into the network, which determines the type of olive grove, all in an automated way." 

This innovative method eliminates dependence on traditional methods, which usually involve field visits and random sampling, thus offering a more efficient and precise alternative for the management and monitoring of olive groves. The team is already conducting research to apply this type of processing using neural networks and satellite images to the study and prediction of water stress in olive groves. 

Reference: 
Martínez Ruedas, C., Yanes Luis, S., Linares Burgos, R., Gutiérrez Reina, D. y Castillejo González, I.L. (2025). Assessment of CNN-based methods for discrimination of olive planting systems with Sentinel-2 images. Computers and Electronics in Agriculture, 234, 110311. https://doi.org/10.1016/j.compag.2025.110311