Study of ET0 by using Conventional and Soft Computing Techniques in the Eastern Gandak Project in Bihar, India – A Case Study

Abstract

Reference evapotranspiration (ET₀) is a key parameter for irrigation planning, water resource management, and sustainable agricultural development. Because direct field measurement of evapotranspiration is difficult, time-consuming, and costly, ET₀ is commonly estimated using empirical and data-driven approaches based on meteorological variables. In this study, reference evapotranspiration was estimated for the Eastern Gandak Project region of Bihar, India, using both conventional empirical methods and soft computing techniques. Long-term climatic data recorded at the IARI regional station, Pusa, for the period 1981–2024 were used for analysis.

Four conventional methods—Hargreaves, Pan Evaporation, FAO-24 Modified Penman, and FAO-56 Penman–Monteith—were employed to compute ET₀, with the FAO-56 Penman–Monteith method adopted as the benchmark. In addition, three soft computing approaches, namely Artificial Neural Network (ANN), Support Vector Machine (SVM), and Gaussian Process Regression (GPR), were developed to estimate ET₀ using different combinations of meteorological inputs. Model performance was evaluated using the coefficient of determination (R²) for both training and testing datasets.

The results indicate that among the conventional methods, the FAO-24 Modified Penman method shows the strongest agreement with the FAO-56 Penman–Monteith method. Among the soft computing techniques, ANN demonstrates superior predictive capability on the testing dataset (R² up to 0.98) compared to SVM and GPR, indicating its robustness in capturing the nonlinear relationship between climatic variables and reference evapotranspiration. The findings highlight the potential of soft computing techniques, particularly ANN, as reliable alternatives for ET₀ estimation in data-rich and data-limited conditions.