Can hybrid models predict actual evapotranspiration instead of weighting lysimeters?

Abstract

Actual daily evapotranspiration (ET) can be considered as one of the most important meteorological parameters. One of the main ways to measure ET is using lysimeters, which are expensive tools, and the data obtained from them are not available in most parts of the world. Therefore, the purpose of this research is to provide an intelligent model that can predict ET using data obtained from cheap and available tools. To do this, data from two NE and SE lysimeters located in Potter County, Texas between 1996-1999 were used as input for SVR and SVR-FFA models in 13 scenarios. The results of this study showed that the SVR-FFA model with an error value of 1.22 mm/day for 1996, 1.14 mm/day for 1997, 1.56 mm/day for 1998, and 1.54 mm/day for 1999 has the highest accuracy for all combinations. Among the standalone SVR models, the SVR-13 model has performed better than other SVR combinations for all years with a Willmott's index of agreement above 0.87. The comparison between the inputs used for the models showed that the Rn and PAR parameters had the greatest impact on the accuracy of the SVR and SVR-FFA models, so they increased the accuracy of the models in every four years. One of the limitations of this study is the lack of access to some parameters such as minimum and maximum temperature in the study area. Therefore, it is suggested to measure these parameters and consider them as other inputs to measure the performance of the models.