Artificial intelligence anticipates thrips outbreaks

Researchers at Texas A&M AgriLife Research have developed artificial intelligence models to predict population growth in thrips (Frankliniella occidentalis) in tomato and pepper production systems. Their study showed an accuracy of 87,7% in open fields and 84,9% in high tunnels. The tool can anticipate the risk of outbreaks and support management decisions before crop damage occurs.

The research evaluated machine learning models in two contrasting productive environments. Random Forest showed the best performance in open fields. XGBoost achieved the highest accuracy in high tunnels. The authors also tested Gradient Boosting Machine, or GBM. All three algorithms analyzed environmental and biological variables related to insect dynamics.

Data used

The study used data from 1.686 yellow sticky traps installed weekly in tomato and pepper fields at the Texas A&M AgriLife research station in Bushland, Texas. Of these, 903 traps came from tall tunnels and 783 from adjacent open fields. After standardization, the researchers worked with 2.254 modeling units.

The researchers combined thrips counts with meteorological variables. The list included average, maximum, and minimum temperature, relative humidity, precipitation, wind speed, and wind direction. The study also included the population recorded 14 days prior to collection. This interval corresponds to the approximate development time of the insect, from egg to adult, under the evaluated conditions.

The previous insect population, referred to in the study as the "parent population," emerged as the main predictor of severity in both environments. Temperature followed. Humidity and wind had secondary effects. In open fields, the combination of a high previous population and higher relative humidity contributed to high severity levels. In high tunnels, wind had a greater impact on predicting high severity.

Difference between environments

The difference between the environments was decisive. Models trained in one system failed to predict the population in the other. Accuracy was 44,13% when the high tunnel model was applied to the field. The field model reached 38,22% when applied to the high tunnels. The authors concluded that open field and high tunnel function as distinct microecosystems, even when they are side by side.

This finding reinforces the importance of microclimate in pest management. According to the authors, high tunnels and open fields differ in thermal stability, humidity, wind, and exposure. These conditions alter the development, dispersal, and transmission potential of Frankliniella occidentalis viruses. The study indicates that forecasting tools need to consider these differences in order to generate useful alerts for the producer.

Change in management

Anticipating risk can change the logic of management. According to Kiran Gadhave, an entomologist at AgriLife Research and assistant professor in the Department of Entomology at Texas A&M, identifying risk a week in advance shifts control from a response to damage to a preventative strategy.

The results also point to limitations. The authors report that the models do not yet incorporate biological regulators, such as natural enemies and interspecific competition. The study also used meteorological data from an open-field station. For tall tunnels, sensors installed inside the structures can improve spatial resolution and the ecological interpretation of the relationships between climate and thrips.

More information at doi.org/10.1016/j.ecoinf.2026.103690