Comparison of Performance of GLM, RF and DL Models in Estimation of Reference Evapotranspiration in Zabol Synoptic Station

Evapotranspiration is one of the most important components of the hydrology cycle for planning irrigation systems and assessing the impacts of climate change hydrology and correct determination is important for many studies such as hydrological balance of water, design of irrigation irrigation networks, simulation of crop yields, design, optimization of water resources, nonlinearity, inherent uncertainty, and the need for diverse climatic information in estimating evapotranspiration have been the reasons why researchers have used artificial intelligence-based approaches. In this study, to estimate accurately the daily reference evapotranspiration between 2009-2018 in Zabol city, north of Sistan and Baluchestan province,  first was used a standard FAO-Penman-Montith method and Zabol synoptic station meteorological data- the ETo reference transpiration is calculated and then presented by various scenarios of meteorological parameters including: maximum, minimum and mean temperature, maximum, minimum and mean humidity, precipitation, sunshine, wind speed and evaporation as inputs for deep learning models, Random forest and generalized linear model were attempted on a daily time scale More accurately. In estimating daily evapotranspiration in these models, 25 scenarios were selected from meteorological data combination and FAO-Penman-Monteith method was used to evaluate the models. Among the investigated scenarios, the M5 scenario (maximum, minimum and mean temperature, maximum, minimum and mean humidity, wind speed, pan evaporation) for deep learning model with minimum error (0.517) and highest correlation coefficient (0.517). 0.996 had the best performance among the above models. The deep learning model showed more accuracy and stability than other models. Hence, this study is recommended a deep learning model for estimating reference plant evapotranspiration in Sistan plain.