Early Prediction of Fruit Size Based on Flower Characteristics of Strawberry

Abstract

In Japan, wholesale prices of strawberries fluctuate markedly, making the timing of peak harvest with periods of high market value a critical issue. Consequently, there is a strong demand for technologies that enable harvest control. We aimed to achieve harvest control with fruit maturation management using environmental control. Harvest control requires accurate prediction of fruit yield, which predicts fruit size and harvest date. In this study, we focused on predicting fruit size. To construct an early prediction model of fruit size as the foundation for harvest control decisions, we attempted predictions using only floral and environmental parameters available at the time of flowering. From the collected parameters, variable selection based on the variance inflation factor was performed to mitigate the effects of multicollinearity. To avoid overfitting, a 4-fold cross-validation was applied, and a multiple regression model was constructed to predict fruit weight. As a result of variable selection, six parameters were selected as explanatory variables: flowering order within the inflorescence, receptacle area, pedicel diameter, flowering day, average air temperature, and direction of inflorescence. The multiple regression model achieved an R2 of 0.775. Additionally, the feasibility of substituting future temperature data with historical data or target set-point temperatures was demonstrated. Furthermore, a flower thinning experiment showed that the influence of post-flowering source–sink balance on fruit size was minimal, indicating that it does not need to be incorporated into the model. These results demonstrate that fruit size can be predicted with reasonable accuracy using data available at flowering, and will play an important role in early yield prediction systems that will be developed in the future.

Abstract from the paper’s OpenAlex record; open copy licensed cc-by-nc. Read at the source ↗

The text the model read, with the source record’s inline markup (italics, sub- and superscripts) rendered and its character entities decoded; quote offsets refer to the raw text.