Chapter 20 - Wavelet decomposition based on Gaussian process regression and multiple linear regression: Monthly reservoir evaporation prediction

Abstract

Evaporation from dam reservoir surfaces is crucial in terms of water resource management and planning. In this study, the new hybrid models of wavelet-Gaussian process regression (WGPR) and wavelet-multiple linear regression (WMLR) combining wavelet transform with the Gaussian process regression (GPR) model and conventional multi-linear regression (MLR) were used to predict reservoir evaporation (RE) in Amirkabir Dam, Iran. Having decomposed the input variables by wavelet transform based on the feature selection using correlation (CFS), the important components were detected and inserted into the GPR and MLR models, thereby creating WGPR-CFS and WMLR-CFS models. The results were then compared with the GPR and MLR models, and the effect of applying each of these models using correlation coefficient (R), root mean square error (RMSE), and mean absolute error (MAE) was evaluated, as well. As the findings showed, the MLR model had better performance with R=0.960, MAE=25.45 (mm), and RMSE=30.45 (mm) values for the variable of RE compared to the GPR model with R=0.863, MAE=45.30 (mm) and RMSE=57.61 (mm) values. In addition, comparing MLR and hybrid models, the proposed hybrid models were considered as viable options to improve the prediction accuracy of RE. The results also showed that, among the 32 hybrid models, the WGPR-CFS model with R=0.979, MAE=18.61 (mm), and RMSE=23.96 (mm) values was the best model for the prediction of RE values.