Optimization of water-saving irrigation decision using dynamic spatiotemporal graph neural networks

Area of Science:

• Agricultural Science
• Hydrology
• Artificial Intelligence

Background:

• Accurate farmland water demand prediction is vital for water-saving irrigation and sustainable food production.
• Predicting water demand is challenging due to complex interactions between soil, crop, and meteorological factors across space, depth, and time.

Purpose of the Study:

• To develop an AI-STGNN-based framework for accurate farmland water demand prediction and irrigation scheduling.
• To enhance water use efficiency and crop yield through data-driven irrigation strategies.

Main Methods:

• Developed a novel AI-STGNN (Artificial Intelligence-Spatio-Temporal Graph Neural Network) model.
• Incorporated a dynamic 3D graph representing soil heterogeneity and hydrological connectivity.
• Integrated channel-wise meteorological attention and multi-scale temporal memory for agro-hydrology-specific enhancements.

Main Results:

• AI-STGNN achieved high accuracy (RMSE: 3.63 vol%, MAE: 2.88 vol%, R²: 0.89) on test data, outperforming STGCN and LSTM.
• The model showed improved performance under heavy rainfall conditions compared to baseline models.
• Field trials demonstrated reduced irrigation (12.3-15.7%), increased yield (3.5-4.2%), and improved water use efficiency (13.7-17.8%).

Conclusions:

• The AI-STGNN framework offers a physically informed, data-driven approach for dynamic water demand prediction and irrigation scheduling.
• The proposed method significantly enhances water management in Midwest cropping systems.
• Further validation across diverse conditions is recommended to confirm practical gains.