Advanced machine learning-based kharif maize evapotranspiration estimation in semi-arid climate.

Area of Science:

• Agricultural Meteorology
• Hydrology
• Machine Learning Applications

Background:

• Accurate crop evapotranspiration (ETc) estimation is vital for hydrological and agrometeorological processes.
• Existing numerical methods for ETc are limited by parameter complexity, data variability, and continuity issues.
• Developing robust models for daily ETc in semi-arid regions is essential for effective water management.

Purpose of the Study:

• To develop and evaluate machine learning (ML) models for estimating daily maize crop evapotranspiration (ETc).
• To assess the performance of various ML models using diverse weather inputs in a semi-arid climate.
• To identify the most influential meteorological parameters for ETc estimation.

Main Methods:

• Five ML models were developed: Category Boosting (CB), Linear Regression (LR), Support Vector Machine (SVM), Artificial Neural Network (ANN), and Stochastic Gradient Descent (SGD).
• Models were trained and validated using data from ICAR-IARI, New Delhi.
• Performance was benchmarked against the Penman-Monteith (PM) model, using metrics like R², MAE, RMSE, and MAPE.

Main Results:

• The SVM model demonstrated superior performance with the highest coefficient of determination (R² = 0.987) and lowest errors (MAE = 0.121 mm day⁻¹, RMSE = 0.172 mm day⁻¹, MAPE = 4.37%).
• The ANN model also yielded comparable and promising results.
• Wind speed was identified as the most significant input parameter influencing ETc estimation.

Conclusions:

• Support Vector Machine (SVM) and Artificial Neural Network (ANN) models are reliable alternatives for accurate daily maize crop evapotranspiration (ETc) estimation.
• These ML models can overcome limitations of traditional numerical methods in semi-arid environments.
• The findings support improved water resource management and agricultural planning in similar climatic conditions.