Apple Fruitlet Abscission Prediction. I. Development and Evaluation of Reflectance Spectroscopy Models

Chemical thinning, the most common and cost-effective thinning method, is conducted during early apple fruit development over a 3- to 4-week period using multiple applications of plant growth regulators. It critical provide growers with tools assess efficacy chemical thinners quickly accurately because visible responses are not apparent for up 2 weeks after application. The objective this study was build model predict fruitlet abscission following thinner application in situ reflectance data obtained portable near infrared (Vis/NIR) spectrophotometer. Developed models were compared currently available (FGM). ‘Honeycrisp’ diameter measured on dates around across 2-year period. After June drop, fruitlets determined have either persisted or abscised. Random forest, partial least squares regression, XGBoost classification used from data. Each developed 2021, 2022, combined 2021 + 2022 For each dataset, 5-fold cross validation three performance metrics: 1) overall accuracy, 2) recall, 3) specificity. Datasets tested unbalanced, majority class down-sampled, minority up-sampled synthetic oversampling technique. In both years, FGM reliably estimated 9 days Before time point, had low prediction accuracy years—persisting abscising 2022. spectroscopy, random forest that balanced over-sampling technique found be best combination predicting efficacy. dataset ranged 84% day before 93% at These results show Vis/NIR promising tool This technology high accuracies range potential two growing seasons. Further testing cultivars, timings, regions would facilitate commercialization technology.