Development of a Sap Flow Prediction Model for 'Shingo' Pear Trees Using Growth and Meteorological Information

Understanding sap flow is crucial for comprehending tree growth. However, it is impractical for all farms to directly measure sap flow. Although many studies have attempted to predict sap flow, most of them have relied on hard-to-obtain data in addition to weather information. This study aims to propose a model for predicting sap flow in ‘Shingo’ pear trees using only easily accessible weather variables, civil twilight time, and DAFB (Days After Full Bloom). Additionally, to account for the time-lag effect in the nonlinear relationship between sap flow and weather variables, we performed lagged Spearman correlation analysis. For solar radiation, the correlation coefficient was found to peak at a 4-hour lag. For the sap flow prediction model, we used machine learning models such as Random Forest and XGBoost, as well as deep learning models including LSTM, GRU, BiLSTM, and a hybrid CNN-GRU-BiLSTM model. All models achieved R² values above 0.94, demonstrating that sap flow prediction is feasible using only the proposed variables. In particular, the CNN-GRU-BiLSTM model performed the best with an MAE of 225.4, RMSE of 427.6, and R² of 0.9550. We concluded that the CNN-GRU-BiLSTM model is well-suited for predicting sap flow in environments with seasonal and localized climate phenomena, such as in South Korea. Through this, we expect to provide a data-driven irrigation standard to farms across the country.