A Proposal for Yield Prediction Using Multiple Models Based on the Average Number of Days to Harvest Tomato Fruit in Greenhouse Horticulture

Agricultural corporations that operate farms engage in contract transactions with supermarkets on a two-week to one-month basis. Therefore, it is critical for farmers to predict yield accurately two weeks ahead. In the experimental field of this study, facility-based cultivation enables some control over the growing environment, such as temperature and irrigation. However, uncontrollable factors, such as weather conditions, indi-vidual differences of plants, and variability in growing locations, make it difficult to predict yields with high accuracy. To predict yield with high accuracy under these circumstances, we propose a predictive method using multiple different machine learning models based on the average number of harvest days for different environments. We verified the accuracy of this proposed method in actual greenhouse tomato farming.