Reference evapotranspiration estimation using genetic algorithm-optimized machine learning models and standardized Penman-Monteith Equation in a highly advective environment

Authors: KIRAGA, SHAFIK - Washington State University; PETERS, R - Washington State University; MOLAEI, BEHNAZ - Tennessee State University; Evett, Steven - Steve; Marek, Gary

Submitted to: Water
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 12/11/2023
Publication Date: 12/20/2023
Citation: Kiraga, S., Peters, R.T., Molaei, B., Evett, S.R., Marek, G.W. 2023. Reference evapotranspiration estimation using genetic algorithm-optimized machine learning models and standardized Penman-Monteith Equation in a highly advective environment. Water. 16(1). Article 12. https://doi.org/10.3390/w16010012.
DOI: https://doi.org/10.3390/w16010012

Interpretive Summary: Irrigation managers can schedule irrigations to replace the water used by crops if they know the crop water use. It is important to do so in order to maintain crop health and productivity. Unfortunately, measuring crop water use is too difficult and expensive for normal farming operations. However, crop water use can be accurately calculated during any growth stage of the crop using weather data and a crop coefficient specific to the crop grown and its growth stage, which farmers can easily observe. The weather data are used to estimate a value called a reference evapotranspiration (ET) and the crop water use is calculated as the reference value multiplied by the crop coefficient. USDA ARS scientists at Bushland, Texas had previously determined growth stage specific crop coefficients for the major crops grown in the Southern High Plains (alfalfa, cotton, corn, sorghum, soybean, sunflower, and winter wheat), but the reference ET calculation required difficult to find weather data. The Bushland ARS scientists worked with Washington State University scientists to develop a method to accurately estimate reference evapotranspiration from different kinds of weather data using a computer method called machine learning (ML). The ML method was as accurate as the standard reference evapotranspiration method but was more flexible in the kind of weather data needed, which can make it more useful to producers.

Technical Abstract: Accurate estimation of reference evapotranspiration (ETr) is important for irrigation planning, water resource management, and preserving agricultural and forest habitats. The widely used Penman-Monteith equation (ASCE-PM) estimates ETr across various time scales using ground weather station data. However, discrepancies persist between estimated ETr and measured ETr obtained from weighing lysimeters (ETr-lys), particularly in advective environments. This study assessed different machine learning (ML) models in comparison to ASCE-PM for ETr estimation in highly advective conditions. Various variable combinations, representing both radiation and aerodynamic components, were organized for evaluation. Eleven datasets (DT) were created for the daily timescale, while seven were established for hourly and quarter-hourly timescale. ML models were optimized by a genetic algorithm (GA) and included support vector regression (GA-SVR), random forest (GA-RF), artificial neural networks (GA-ANN), and extreme learning machine (GA-ELM). Meteorological data and direct ETr measurements of well-watered alfalfa, obtained from weighing lysimeters and a nearby weather station in Bushland, Texas (1996-1998), were used for training and testing. Model performance was assessed using metrics such as root mean square error (RMSE), mean absolute error (MAE), mean bias error (MBE), and coefficient of determination (R**2). ASCE-PM consistently underestimated ETr across all timescales (above 7.5 mm/day, 0.6 mm/hr, 0.2 mm/hr, at daily, hourly, and quarter-hourly, respectively). On hourly and quarter-hourly timescales, datasets predominantly composed of radiation components, or a blend of radiation and aerodynamic components demonstrated superior performance. Conversely, datasets primarily comprised of aerodynamic components exhibited enhanced performance on a daily timescale. Overall, GA-ELM outperformed the other models and was thus recommended for ETr estimation at all time scales. The findings emphasize the significance of ML models, in accurately estimating ETr across varying temporal resolutions, crucial for effective water management, water resource and agricultural planning.