Scale dependent prediction of reference evapotranspiration based on Multi-Variate Empirical mode decomposition

This study proposes a novel method for estimation of reference evapotranspiration ( ET o ) by accounting the time scale of variability using the Multivariate Empirical Mode Decomposition (MEMD). First the ET o and the four predictor variables such as solar radiation, air temperature, relative humidity and wind velocity are decomposed into different intrinsic mode functions (IMFs) and a residue using MEMD. To model ET o , first the modes are modeled separately using the Stepwise Linear Regression (SLR) after identifying the significant predictors at different time scales based on the p -value statistics. Subsequently, the predicted modes are recombined to obtain ET o at the observation scale. The method is demonstrated by predicting the monthly ET o from Stratford station in United States. The results of the study clearly exhibited the superior performance of the proposed MEMD-SLR model when compared with that by M5 model tree, SLR and the EMD-SLR hybrid model.