Applicability of hybrid bionic optimization models with kernel-based extreme learning machine algorithm for...

Area of Science:

• Agricultural Science
• Environmental Science
• Data Science

Background:

• Accurate reference crop evapotranspiration (ET_O) prediction is essential for efficient agricultural water management, especially in data-scarce arid and semiarid regions.
• Existing models may lack accuracy in predicting ET_O due to complex environmental factors.

Purpose of the Study:

• To improve the accuracy of daily ET_O forecasts in data-deficient arid and semiarid regions of China.
• To identify key factors influencing ET_O and develop a hybrid prediction model.

Main Methods:

• Utilized the Classification and Regression Tree (CART) algorithm to determine the influence of various factors on ET_O.
• Developed a Kernel Extreme Learning Machine (KELM) model incorporating influential factors.
• Optimized the KELM model parameters using three bionic optimization algorithms: Sparrow Search Optimization (SSA), Harris Hawks Optimization (HHO), and Lion Swarm Optimization (LSO).

Main Results:

• Temperature (maximum or minimum) was identified as the most significant factor influencing ET_O (importance range: 0.399-0.554).
• Relative Humidity (RH) and Solar Radiation (Ra) were also identified as key influencing factors.
• The hybrid KELM models optimized with bionic algorithms outperformed the independent KELM model, with the SSA-KELM model demonstrating the highest accuracy (RMSE: 0.408-1.964, R²: 0.545-0.982).

Conclusions:

• The top five factors identified by CART should be used as inputs for the SSA-KELM model for ET_O estimation in arid and semiarid regions.
• The developed SSA-KELM model offers a reliable and accurate approach for daily ET_O forecasting in similar climatic regions.