LSTM-Based Prediction of Apple Fruit Growth Using Past Diameter and Climatic Condition Data

Reliable yield projections require both fruit counting and forecasting fruit size at harvest. The prediction of final fruit size primarily depends on established growth models, which rely mainly on isolated fruit growth measurements taken throughout the season. This study presents a novel predictive model for apple fruit growth based on past diameter measures and atmospheric pressure. The studied growth data was acquired within a season by continuously monitoring apple diameter using a connected fruit dendrometer. Preliminary analysis showed that atmospheric pressure also contributes to accurate diameter forecasting. Utilizing Long Short-Term Memory (LSTM) modeling, the proposed approach predicts fruit diameter over eight days based on past diameter and atmospheric pressure data collected over a week, achieving a Mean Absolute Error below 0.5 mm. Such a tool holds strong potential for optimizing harvest prediction and orchard management efficiency.