Abstract:

Accurate estimation of reference evapotranspiration (ET₀) is crucial for optimizing agricultural irrigation and managing water resources. However, the application of traditional empirical models is often limited due to missing meteorological data. This study investigated the applicability of a Genetic algorithm−optimized backpropagation neural network (GA−BP) model for ET₀ simulation in the Three Gorges reservoir area. Using daily meteorological data from six observation stations between 1980 and 2023, 16 kinds of GA−BP−based ET₀ estimation models were developed and compared with three traditional empirical models: Hargreaves−Samani (H−S), Makkink and Irmak−Allen (I−A). The results showed that the GA−BP models significantly outperform traditional models across different regions. When only temperature and solar radiation data were used, the GA−BP2 model improved R² by 15.69%, reduced MAE by 35.36%, and decreased RMSE by 40.63% compared to the H−S model. With the addition of sunshine duration, the GA−BP6 model improved R² by 31.20%, reduced MAE by 27.33% and lowered RMSE by 23.47% compared to the Makkink model. After incorporating relative humidity, the GA−BP12 model outperformed the I−A model, with a 15.10% increase in R², a 57.88% reduction in MAE, and a 43.92% decrease in RMSE. Therefore, the GA−BP model is recommended for ET₀ estimation in the Three Gorges reservoir area, especially under conditions of limited meteorological data.

Key words:

GA?BP model, Reference crop evapotranspiration, Three Gorges reservoir area, ET₀ , estimation, Meteorological data deficiency