A model to predict peach phenology and maturity using meteorological variables

Abstract

This paper outlines the development of regional-scale peach [Prunus persica (L.) Batsch] phenology models for the southeastern United States. We constructed regression-based models to predict full bloom and first harvest using phenological data for eight peach cultivars grown in Georgia, South Carolina, and Texas and meteorological data from nearby cooperative weather stations. The mean absolute error (MAE, absolute value of predicted date minus observed date) of the all-cultivar full bloom model was 3.48 days. Bloom model errors were roughly 6 to 7 days (or less) 90% of the time, while comparable natural bloom date variability at a single station is about 12 to 13 days. Only two cultivars ('Dixired' and 'Elberta') had first harvest models with 90% error thresholds at or less than the natural harvest date variability at a single station (9 to 10 days). Thus, first harvest appears to be less predictable than full bloom, at least in terms of the temperature variables used in this study. The bloom model could serve as a prognostic tool, while the two cultivar harvest models are appropriate only for diagnostic studies, such as evaluating the potential response of a peach cultivar in a new location. The appeal of the models is that they consider numerous cultivars over a wide region and they use readily available meteorological data.