Development of novel surveillance tools for rapid detection of citrus psyllids

South Africa’s citrus industry is lucrative but its profitability is threatened by pest insects, either due to direct damage or via the transmission of pathogens causing disease that negatively impact citrus production. The citrus psyllids Diaphorina citri and Trioza erytreae are vectors for ‘Candidatus Liberibacter asiaticus’, the pathogen of one of the most devastating citrus diseases in the world known as Huanglongbing (HLB). Vector control is essential to minimise the disease’s spread. Huanglongbing and D. citri are not yet present in South Africa but have become established in parts of the African continent and are both spreading. My project attempted to improve the attractiveness and identification process of yellow sticky traps; a psyllid monitoring technique recommended in citrus orchards for T. erytreae. This was done to help prevent the introduction of HLB and D. citri, the invasive vector of its pathogen, to South Africa. To improve attractiveness, various plant semiochemical odour lures were tested using T. erytreae as model organism. For improvement of the identification process, an automated vision-based artificial intelligence (A.I.) driven system was developed and tested. The successful development and implementation of these tools has the potential for the speedy implementation of control measures to prevent the establishment or spread of the HLB pathogen and its insect vectors. Yellow sticky trap augmentation was done using eight plant semiochemicals, a commercially available D. citri lure, and hexane as a solvent control. All test attractants were dispensed from sealed polyethylene bulbs. Field cage trials were used to determine the most effective semiochemicals for further field tests. The field tests were done in a pesticide free lemon orchard using a randomised 5 × 6 grid. Temperature and humidity were recorded so that their effect on semiochemical release rate could be determined. Using gas chromatography-mass spectrometry, odorant composition and release rates were evaluated. None of the semiochemicals improved psyllid catch during the field cage or field trials, and weathering in the field did not affect the composition of odorants. However, temperature influenced odorant loss, and release rate from polyethylene bulbs decreased over time. Development and training of the automated psyllid identification system was done by setting out, then collecting 544 traps around South Africa, Mauritius, and Reunion. They then underwent manual processing where target psyllid groups were identified, and relevant traps photographed. Photographs were then annotated and uploaded onto Roboflow for data augmentation and training, validation, the testing of the A.I. models. Five models were developed using YOLOv8 with two models, YOLOv8s and YOLOv8m, showing promise as a workable means to speed up and improve the psyllid identification process. While semiochemicals did not improve psyllid captures in this study, they should not be ruled out to improve yellow sticky trap monitoring. Both YOLOv8 models, while promising, have limitations that need to be addressed. Further studies into yellow sticky trap augmentation should investigate blends of semiochemicals with overlapping attractiveness to citrus psyllids as well as the possibility of using pheromones as an alternative. For the A.I. models, increasing the number of images used during training could increase effectiveness and accuracy, though another option is to develop and test a tandem model system where the same input is fed into two separate models so that the outputs can be compared.