Development of long short-term memory models using rainfall and soil moisture to predict soil moisture dynamics

Area of Science:

• Agricultural Science
• Environmental Science
• Data Science

Background:

• Soil moisture dynamics are crucial for precision agriculture.
• Accurate soil moisture prediction supports efficient irrigation and crop management.
• Long Short-Term Memory (LSTM) networks offer potential for time-series soil moisture forecasting.

Purpose of the Study:

• To develop and evaluate LSTM models for predicting soil moisture at various depths in an Asian pear orchard.
• To compare the effectiveness of hourly versus event-based rainfall data as model inputs.
• To assess the impact of prediction time horizons on model accuracy.

Main Methods:

• LSTM models were trained using hourly and event-based rainfall data alongside soil moisture measurements.
• Soil moisture was monitored at 20, 40, and 60 cm depths using frequency domain reflectometry.
• Model performance was assessed using Mean Absolute Error and Root Mean Square Error (RMSE).

Main Results:

• Short-term soil moisture predictions were more accurate than long-term predictions.
• Soil moisture content increased with rainfall and depth, with more gradual responses at deeper levels.
• Event-based rainfall input significantly reduced prediction errors compared to hourly rainfall data.

Conclusions:

• Event-based rainfall data is superior for LSTM-based soil moisture prediction in Asian pear orchards.
• LSTM models utilizing event-based rainfall can enhance precision irrigation systems.
• Optimized water management through accurate soil moisture forecasting leads to improved water use efficiency.