Pan evaporation modeling by three different neuro-fuzzy intelligent systems using climatic inputs

Modeling pan evaporation (Epan) estimation is a vital issue in water resources management because it directly affects water reservoir and water supply systems. In the developing countries (e.g., India), Epan data are generally limited, and in such a circumstance, theoretical estimates from available climatic data could be beneficial. The study investigates the capability of three adaptive neuro-fuzzy methods, adaptive neuro-fuzzy inference system (ANFIS)-embedded grid partition (GP), subtractive clustering (SC), and fuzzy c-means clustering (FCM), in estimation of monthly pan evaporation using climatic inputs of minimum and maximum air temperatures, wind speed, sunshine hours, and relative humidity obtained from two stations, Uttarakhand, India. Cross validation method is applied by dividing data into three equal parts, and methods are tested using each part. Methods are evaluated by applying various combinations of inputs and using root mean square error (RMSE), mean absolute error (MAE), Nash-Sutcliffe efficiency (NSE), and determination coefficient (R-2) criteria. The ANFIS-FCM is found to be superior to the ANFIS-GP and ANFIS-SC methods in Epan modeling. Cluster-based proposed neuro-fuzzy method increases performance of the best ANFIS-GP and ANFIS-SC models with respect to RMSE by about 9-14% for the both stations. The three ANFIS methods are also compared with each other and Stephen Stewart (SS) method by dividing data into three stages, training, validation, and test. The results indicate the superior accuracy of the ANFIS methods to SS for the same input variables. The ANFIS-FCM generally produces better Epan estimates than the other two ANFIS methods.