Optimization of the single - layer time series prediction model for reference crop evapotranspiration

Reference crop evapotranspiration is essential for agricultural irrigation, hydrological prediction, crop growth simulation, and climate disaster monitoring. This study develops a single-layer Long Short-Term Memory neural network model to forecast ET₀, focusing on the Yangling area. The optimal configuration includes five hidden layers, a lookback period of 10, and 500 training steps. Results show the model accurately predicts ET₀, with predictions closely matching actual values within ±0.75 mm. The study reveals seasonal error variations, with slightly higher inaccuracies in summer. The single-layer LSTM model offers a robust method for ET₀ prediction, maintaining stability and reliability across different months. This research provides a reliable theoretical basis for water-saving irrigation and enhanced crop productivity.