H166-0014
Forecasting daily reference evapotranspiration using the wavelet-Gaussian Process Regression (GPR) approach

Tuesday, 15 December 2020
Poster
Sayed M. Bateni1, Helaleh Khoshkam2, Masoud Karbasi3, Mohammad Valipour1, Tongren Xu4 and Essam Heggy5, (1)University of Hawaii at Manoa, Honolulu, HI, United States, (2)Karaj College of Environment, Civil and Environmental Engineering, Karaj, Iran, (3)University of Zanjan, Water Engineering Department, Faculty of Agriculture, Zanjan, Iran, (4)Faculty of Geographical Science, Beijing Normal University, Beijing, China, (5)University of Southern California, Electrical Engineering - Electrophysics, Los Angeles, CA, United States
Abstract:
The accurate reference evapotranspiration (ET) forecasts are required for the efficient water resources management and planning, and irrigation scheduling. In this study, the Gaussian Process Regression (GPR) and wavelet-GPR approaches are applied to ten sites in China to forecast daily ET for 1, 3, 7, 10, and 14 days ahead. These sites are chosen to cover different climatic and vegetative conditions. The daily ET data from 2010 to 2017 (8 years) are used for training the models. The trained models are used to forecast ET for 2018-2020 (3 years). The ET forecasts from both models agree well with the measurements in all the study sites. Also, the wavelet-GPR approach performs better than GPR. The ten-site average root-mean-square-error (RMSE) of ET forecasts from GPR for 1, 3, 7, 10, and 14 days ahead are 1.14, 1.25, 1.27, 1.36, and 1.36 (mm/day), respectively. Corresponding RMSEs from the wavelet-GPR are 0.36, 1.12, 1.16, 1.21, and 1.25 (mm/day), which are a 68.4%, 11.4%, 9.6%, 12.3%, and 11.4% reduction compared to those of GPR. Finally, performance of GPR and wavelet-GPR are compared to those of autoregressive moving average (ARMA) and autoregressive integrated moving average (ARIMA) methods. The outcomes show that the GPR and wavelet-GPP provide significantly more accurate ET forecasts.
<< Previous Abstract | Next Abstract >>