Using machine learning to identify the geographical drivers of Ceratitis capitata trap catch in an agricultural landscape

Abstract

Abstract The spatial distribution of Ceratitis capitata Wiedemann (Diptera: Tephritidae) trap catch was classified and related to a set of geographic variables to identify its main geographical drivers. Trap catch data were sourced from an area-wide integrated pest management (AW-IPM) 1 programme and classified into statistically significant hot- and cold spots (HCSs) 2 . Trap data of four consecutive fruiting seasons were combined to identify monthly and seasonal long-term HCSs. The main geographic drivers of the HCSs were identified using variable importance lists produced by the random forest (RF) machine learning (ML) algorithm. Long-term climate, topography, landscape and fruit fly management variables were used as predictor variables in RF to classify HCSs. The resulting RF models produced classification accuracies of up to 80%. In most cases, the most important variable was long-term rainfall, suggesting that this was the most prominent driver of C. capitata HCSs in our study region. The result of this study highlights the value of long-term pest monitoring data and long-term environmental data in understanding the spatial distribution of C. capitata trap catch in complex agricultural systems. This study sets out a framework to spatially quantify C. capitata trap catch into HCSs using monitoring data from AW-IPM programmes, enabling the investigation of complex ecological relationships through the use of ML algorithms. The results of these analyses could improve area-wide integrated fruit fly management programmes through more precise spatial planning of management actions, such as the sterile insect technique (SIT) 3 , leading to better programme performance and reduced costs.