IoT-Based Estimation of Daily Evapotranspiration: Performance Evaluation and Optimization for Sustainable Crop Water Management

Evapotranspiration (ET) is pivotal in energy and hydrological cycles. Accurately modeling actual evapotranspiration (AET) holds significant importance for ecological restoration, agricultural practices, and the effective management of water resources. It represents the portion of potential evapotranspiration (PET) that is limited by the moisture content in the soil. Potential evapotranspiration (PET) estimation is achievable using the extensive meteorological data available in FLUXNET observations. Nevertheless, the difficulty lies in precisely modeling daily evapotranspiration based on PET. This investigation compares a mathematical approach and a machine learning approach to simulate AET values accurately. Among the mathematical approaches for computing AET from PET, nonlinear methods exhibit superior performance compared to linear and complementary relationship methods. Lately, numerous machine learning methodologies have emerged for calculating AET values. This research also appraises two categories of machine learning algorithms—regression algorithms and artificial neural networks—to predict daily AET values. The regression algorithms help to extract critical parameters that significantly influence evapotranspiration values. All the machine learning models are evaluated based on their root-mean-square error (RMSE) and mean square error (MSE). Additionally, this study assesses both mathematical and machine learning models. Despite the inherent challenge in accurately estimating AET values, our findings suggest that precise simulation of daily evapotranspiration is attainable by leveraging meteorological data alongside soil water content (SWC) data.