Predicting evapotranspiration using IoT sensor node and Support Vector Machine

A Support Vector Machine learning algorithm has been created to predict the evapotranspiration rate (mm/day) for a pilot farm that showcases diverse seasonal conditions. An earlier developed IoT-based soil sensor node was utilized to acquire real-time field data vital for building an adaptable prediction model. The developed model has been validated against theoretical model estimates and actual evapotranspiration (ET) measurements essential for enhancing the model's reliability and utilization in the field. The parameters employed in the algorithm include air temperature, wind speed, relative humidity, rainfall, solar radiation, soil temperature, and soil moisture. The model evaluation with different data partitions has shown good consistency in predicting water loss and retention in the form of evapotranspiration across all four seasons of the year. It achieved R² values consistently above 0.96, with minimal prediction errors, RMSE = 0.17 to 0.21 and MAE = 0.13 to 0.18. However, some variances are recorded in the hot summer season suggesting the optimization of the model, which can then be employed as a powerful tool for decision-making in water resource management and agricultural planning.