A comparative study on daily evapotranspiration estimation by using various artificial intelligence techniques and...

Area of Science:

• Hydrology and Water Resources Engineering
• Environmental Science
• Computational Intelligence

Background:

• Evapotranspiration (ET) is a key hydrological parameter influencing water resource management and infrastructure design.
• Accurate ET estimation is vital for optimizing water structure efficiency and ensuring design safety.
• Factors influencing ET include temperature, humidity, wind speed, and solar radiation.

Purpose of the Study:

• To develop and compare various models for estimating daily evapotranspiration (ET).
• To evaluate the performance of machine learning techniques against traditional methods for ET prediction.
• To identify the most accurate models for hydrological applications.

Main Methods:

• Developed models using fuzzy rules generation technique (fuzzy-SMRGT), multivariate regression (MR), artificial neural networks (ANNs), adaptive neuro-fuzzy inference system (ANFIS), and support vector regression (SMOReg).
• Utilized daily data on air temperature, wind speed, solar radiation, and relative humidity from Lake Lewisville, Texas.
• Employed the Penman-Monteith (PM) method as the reference equation for empirical ET calculation.

Main Results:

• Artificial Neural Networks (ANNs) achieved the highest accuracy with R²=0.998, RMSE=0.075, and APE=3.361%.
• Adaptive Neuro-Fuzzy Inference System (ANFIS) also showed excellent performance with R²=0.996, RMSE=0.103, and APE=4.340%.
• Quadratic-Multivariate Regression (Q-MR) performed well, outperforming other regression and support vector methods.

Conclusions:

• ANN and ANFIS models provide highly accurate daily evapotranspiration estimations.
• These advanced computational models offer significant improvements over traditional methods for hydrological studies.
• Accurate ET prediction using ANNs and ANFIS can lead to more efficient and safer water resource management.