Reference evapotranspiration estimation using adaptive neuro-fuzzy inference system with limited meteorological data

Machine learning tools are extremely useful for the estimation and modelling of hydrological processes such as evapotranspiration (ET). In this study, reference evapotranspiration (ET0) in Labuan located in the East Malaysia was estimated using an artificial neuro-fuzzy inference system (ANFIS). In order to investigate the feasibility of the ANFIS model for a wide temporal range, daily meteorological data collected at Station 96465 (Labuan) from year 2014 to 2018 were divided on an annual basis. ANFIS models were trained using data from different years as well as varying combinations of one climatic parameter with solar radiation. The study revealed that the ANFIS model was capable of performing accurate estimation when only one year of training data were used where errors of less than 5 % and NSE above 0.950 were achieved. This finding could be useful for new meteorological stations where data are limited. Furthermore, solar radiation and minimum temperature were deemed to be the best input combination because of their distinguishable characteristics. Maximum temperature which highly overlaps solar radiation in nature was found the worst complementary input. However, it is important to note that the importance of climatic parameters could be affected by extreme weather conditions.