AI Model Improves Grape Berry Temperature Forecasting for Vineyards

Researchers have developed an advanced AI model, the Feed-Forward Attention Long Short-Term Memory (FAM-LSTM) network, specifically designed for accurate, multi-step, and high-resolution forecasting of grape berry temperature. This capability is vital for vineyard managers to implement timely strategies against heat stress, which can severely impact grape quality and yield. The FAM-LSTM model was trained using environmental data from vineyards in Washington state and rigorously validated against other established models like standard LSTM, GRU, RNN, and Random Forest. It consistently demonstrated superior performance across various forecasting horizons, ranging from 15 minutes to 72 hours. A key finding was the significant improvement in forecasting accuracy when the model incorporated in-vineyard microclimate measurements compared to relying solely on nearest open-field weather station observations. While errors increased with longer forecast horizons and during peak daytime heat, the FAM-LSTM framework provides a robust tool for precision heat stress management in viticulture.