An accurate irrigation volume prediction method based on an optimized LSTM model.

Area of Science:

• Agricultural Science
• Data Science
• Machine Learning

Background:

• Accurate prediction of irrigation volumes is essential for sustainable agriculture and optimal water resource management.
• Traditional methods often struggle to capture complex spatio-temporal dynamics influencing crop water needs.

Purpose of the Study:

• To develop and validate an optimized irrigation volume prediction model leveraging deep learning techniques.
• To enhance the accuracy and reliability of irrigation predictions for improved agricultural decision-making.

Main Methods:

• Utilized a bidirectional Long Short-Term Memory (BiLSTM) network to model temporal features and sequential dependencies.
• Employed a Convolutional Neural Network (CNN) to extract spatial features and inter-feature correlations (temperature, precipitation, wind speed).
• Integrated an attention mechanism to prioritize critical features, further improving prediction performance.

Main Results:

• The proposed BiLSTM-CNN-Attention model demonstrated superior performance compared to RNN, CNN, LSTM, BiLSTM, and BiLSTM-CNN.
• Achieved significantly low error metrics: Mean Square Error (MSE) of 0.000004, Root Mean Square Error (RMSE) of 0.005968, and Mean Absolute Error (MAE) of 0.004599.
• Attained a high coefficient of determination (R²) of 0.9749, indicating strong predictive power for spring corn irrigation volumes.

Conclusions:

• The BiLSTM-CNN-Attention model offers a robust and accurate solution for predicting crop irrigation volumes.
• This approach facilitates more precise irrigation management, leading to efficient water utilization and enhanced crop production.
• The findings are particularly relevant for agricultural regions facing water scarcity or aiming for optimized resource allocation.