Application of Support Vector Machine, CHAID and Random forest Models, inEstimated Daily Reference Evapotranspiration in Northern Sistan andBaluchestan Province

Evapotranspiration as an important component of the hydrological cycle plays a very important role in the planning and management of water resources in dry and hyper dry areas. Accurate estimation of evapotranspiration requires a costly tool which can not be used anywhere. Hence, researchers are always looking for applied relationships and practices that are low-cost and accurate for the correct estimation of the value of this parameter. Several methods have been developed for the accurate estimation of evapotranspiration throughout the world. Among of these methods can be pointed out empirical equations, Artificial Neural Network, support vector machin, and tree models. Therefore, the purpose of this study was to investigate the accuracy and the capability of linear supporting vector machine models, the decision tree of the type of chiad and the random forest model in the estimation of reference evapotranspiration.The data used in this study include maximum temperature, minimum temperature, average temperature, maximum moisture content, minimum humidity, average humidity, precipitation, sunny hours, wind speed, and a shift from the meteorological station of Sistan Plain between2009-2018. In this study, using meteorological data and the FAO Penman-Monteith model, the values of evapotranspiration were calculated and then by providing different combination scenarios of the meteorological parameters as inputs of the studied models on a daily basis, an attempt was made to find a more accurate estimate of the refrence evapotranspiration as the output of the models. In this research, correlation coefficient (R) and Mean Absolute amount of Error (MAE) were used to compare different model. The results showed among the support model carriers, the random knife, the random forest model with M7 patterns has the highest accuracy with the correlation coefficient (R = 0.983) and the lowest mean error magnitude (MAE = 0.798). Therefore, this research recommends a random forest model for estimating evapotranspiration in the area of Sistan plain.