Smart irrigation management: IoT-based RNN-LSTM model for soil moisture prediction in precision agriculture

This study presents an IoT-enabled smart irrigation management system utilizing subsurface soil moisture sensors and a recurrent neural network–long short-term memory (RNN-LSTM) model to predict soil moisture in real-time for precision agriculture. The proposed system was deployed in Malaysia for six months, achieving a root mean square error (RMSE) of 1.222, a mean absolute error (MAE) of 0.6374, and a coefficient of determination (R²) of 0.6723, explaining approximately 67% of the variance in the observed data. Additionally, 95.49% of predictions fell within ±5% of actual measured values, a tolerance-based metric distinct from classification accuracy. Outlier analysis revealed that the largest residuals occurred during heavy rainfall events, and adopting a robust Huber loss function improved R² to 0.70. The results indicate that the system can effectively support irrigation scheduling, although future work should extend seasonal coverage and address spatial variability in larger fields.