With avocado exports on the rise, the Colombian Agricultural Research Corporation (Agrosavia) and the National University of Colombia have joined forces to fight the most common pests in avocado exports.

With the U.S. being one of the main destination markets for Colombian avocados, the two organizations are using an Artificial Intelligence (AI) tool that seeks to predict risk areas and establish integrated pest management programs for plagues such as weevils and moths.

The predictive model identifies the smallest areas in the lots where quarantine pests can appear. This will not only allow more accurate control and monitoring, but will also make it possible to use fewer insecticides.

The tool uses machine learning techniques in addition to spatial and pattern analysis, spatial statistics and geostatistics.

The study was carried out in four experimental plots located in commercial Hass orchards in Timbío and Sotará, in Cauca, Colombia.

Related articles: AgroFresh expands ethylene management technology to Brazil and Mexico

Agronomist Juan Camilo Zapata Calero, Master in Agricultural Sciences, along with researcher Ph. D. Arturo Carabalí Muñoz, from Agrosavia, and professor John Josephraj Selvaraj from the Faculty of Engineering and Administration of the UNAL Palmira Campus were in charge of developing the tool.

To determine the presence or absence of potential damage, artificial neural networks were used to "process data in a manner inspired by the way the human brain does; they are capable of learning complex patterns and performing prediction and classification tasks".

Predictions are based on phenological variables such as planting date, day length, temperature, moisture supply, genetic component, plant management and fruit size.

The latter is a very important aspect because it defines whether or not the avocado will be affected by pests, since a smaller size usually means little to no pest damage.

With a portable meter, the researchers also recorded 15 climatic variables, including temperature, relative humidity, wind speed, altitude, wind direction and wind chill, which are the most influential in the prediction model.