Improving onion thrips management: Developing machine learning-enhanced, spatially optimized sequential sampling plan

Onion thrips, Thrips tabaci Lindeman (Thysanoptera: Thripidae), is a significant economic pest of onion worldwide. Onion thrips reduce bulb yields through feeding injury on leaf tissue and the transmission of destructive plant pathogens. Current action thresholds have been used successfully to provide season-long thrips control, but with 1-5 fewer insecticide applications compared to a weekly spray program. While the adoption of this action-threshold-based program has reduced costs, control decisions still depend upon significant fixed-sampling efforts, which can be a barrier to adoption and lead to unnecessary pesticide applications, increasing the risk of insecticide resistance.

To address this issue, we developed a sequential sampling plan using spatio-temporal thrips data collected during the 2021 and 2022 growing seasons to reduce the overall sampling time required to make control decisions. The sampling plans was designed to determine the optimal decision by calculating the likelihood ratio between the null (no spray) and alternative hypotheses (spray), based on the observed data at each step of the sampling process.

We incorporated machine learning models to enhance the precision of our sampling plan with the same spatio-temporal data to predict thrips population dynamics and optimize sampling efforts. Our results demonstrate that the machine learning-enhanced sequential sampling plan effectively reduced the sampling effort required to estimate populations of onion thrips by more than 80% compared to the traditional fixed sampling approach. This approach integrating insect ecology and machine learning, resulted in significant reductions in sampling time and control decisions leading to substantial savings for onion growers.