Estimating Reference Evapotranspiration over Georgetown, Guyana, Using an Artificial Neural Network: A Comparative Analysis with 32 Empirical Methods

This study investigates the capability of an Artificial Neural Network (ANN) model to estimate daily ETo using several meteorological variables in a data-limited region along Guyana’s coast. The ANN was trained on historical data from 2001–2018 and independently evaluated over 2019–2022, with ETo computed by the FAO Penman–Monteith (PM-56) method serving as the reference benchmark. Model performance was assessed using standard statistical indicators, including the root mean square error (RMSE), mean absolute error (MAE), coefficient of determination (R 2 ), Nash–Sutcliffe efficiency (NSE), and index of agreement (IoA). In addition, the ANN model was compared against 32 commonly used empirical ETo estimation methods, encompassing temperature-based, radiation-based, and mass transfer-based approaches. Results indicate that the ANN reproduced PM-56 ETo estimates with high accuracy and minimal bias, achieving R 2 and NSE values exceeding 0.99 across the validation period. The ANN model also consistently outperformed all empirical methods across all performance metrics, demonstrating superior accuracy and robustness. Among conventional methods, the Hargreaves–Samani and Makkink approaches showed comparatively better performance, while mass transfer-based methods exhibited substantial deviations and poorer performance. These findings suggest that ANN-based models can serve as reliable alternatives for daily ETo estimation in regions where complete meteorological inputs for physically based methods are limited, thereby supporting improved water-resource and agricultural decision-making in Guyana and similar environments.