<![CDATA[Accurate and adaptive irrigation water management is essential for maintaining rice productivity in regions with high climatic variability, such as Pinrang Regency, Indonesia. Conventional irrigation practices based on fixed schedules often fail to respond daily weather dynamics, resulting in inefficient water use. This study analyzes reference evapotranspiration (ET₀) as a climate-based indicator of irrigation water requirements and predicts daily ET₀ values using deep learning approaches. ET₀ was calculated using the FAO Penman-Monteith method based on daily meteorological variables, including air temperature, relative humidity, wind speed, surface pressure, and solar radiation. Daily climate data for the 2018-2024 period were obtained from the NASA POWER database, comprising 2,892 observations, and processed using Min-Max normalization, a 30-day sliding window scheme, and chronological data partitioning. Long Short-Term Memory (LSTM) and Bidirectional LSTM (BiLSTM) models were trained using the Adam optimizer to perform daily ET₀ prediction. The results show that ET₀ exhibits a distinct seasonal pattern, with higher values during the late dry season and lower values in the rainy season. Model evaluation indicates that BiLSTM slightly outperforms the LSTM model, achieving an MAE of 0.4185 mm/day, an RMSE of 0.5369 mm/day, and a MAPE of 10.42%, compared to the LSTM model with an MAE of 0.4205 mm/day, an RMSE of 0.5389 mm/day, and a MAPE of 10.44%. Medium-term ET₀ projections for the 2025-2027 period demonstrate consistent seasonal patterns relevant for irrigation planning. This study contributes to informatics by providing a reproducible deep learning-based time-series prediction framework to support adaptive irrigation management and sustainable agriculture.]]>
