Estimating reference evapotranspiration using a machine learning approach.

Area of Science:

• Agricultural Meteorology
• Artificial Intelligence in Hydrology

Background:

• Evapotranspiration (ET0) is crucial for crop water management.
• Accurate ET0 estimation relies on key weather parameters like temperature, solar radiation, humidity, and wind speed.

Purpose of the Study:

• To develop and compare machine learning models for improved ET0 estimation during wheat growth.
• To identify optimal weather input combinations for accurate ET0 prediction.

Main Methods:

• Collected daily weather data (1970-2018) for Amritsar, India.
• Developed Random Forest (RF), Support Vector Machine (SVM), and Artificial Neural Network (ANN) models.
• Evaluated models using statistical criteria for calibration and validation.

Main Results:

• RF model demonstrated the best performance, followed by SVM and ANN.
• The model using Tmax, Tmin, RHM, RHE, and Rs achieved the highest rank.
• Specific input combinations (Rs, Tmax; Rs, Tmin; Tmax, Tmin) showed excellent performance with RF, SVM, and ANN, respectively.

Conclusions:

• Machine learning techniques can accurately estimate ET0 with fewer data points.
• Selected input combinations are suitable for ET0 estimation when data is limited.
• This approach enhances the practicality of ET0 estimation in data-scarce scenarios.