Integrating Load-Cell Lysimetry and Machine Learning for Prediction of Daily Plant Transpiration

Area of Science:

• Agricultural Science
• Environmental Science
• Computer Science

Background:

• Accurate crop transpiration prediction is crucial for effective water management in agriculture.
• Traditional methods often lack the precision required for dynamic environmental conditions.
• Physiological traits and meteorological variables significantly influence plant water use.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting daily crop transpiration.
• To enhance the accuracy of transpiration estimates using extensive datasets.
• To identify key environmental factors influencing crop water loss.

Main Methods:

• Collected 7 years of high-resolution transpiration data from hundreds of plant specimens using gravimetric load cells and ambient sensors.
• Utilized Decision Tree, Random Forest, XGBoost, and Neural Network models for predictive analysis.
• Trained and validated models using ground truth physiological data and meteorological variables.

Main Results:

• Random Forest and XGBoost models achieved high predictive accuracy for whole plant transpiration (R²=0.89 on test set, R²=0.82 on holdout).
• Ambient temperature was identified as the most significant environmental factor impacting transpiration rates.
• The study demonstrated the effectiveness of machine learning in modeling complex plant physiological processes.

Conclusions:

• Machine learning offers a powerful tool for precise water management in agriculture.
• Accurate transpiration prediction can optimize irrigation scheduling and reduce water waste.
• Understanding the influence of environmental factors like temperature is key to improving crop water use efficiency.