PubAg

Main content area

On the applicability of maximum overlap discrete wavelet transform integrated with MARS and M5 model tree for monthly pan evaporation prediction

Author:
Ghaemi, Alireza, Rezaie-Balf, Mohammad, Adamowski, Jan, Kisi, Ozgur, Quilty, John
Source:
Agricultural and forest meteorology 2019 v.278 pp. 107647
ISSN:
0168-1923
Subject:
algorithms, evaporation, models, prediction, relative humidity, solar radiation, temperature, water utilization, wavelet, wind speed
Abstract:
Accurate pan evaporation (Epan) prediction is a critical issue in water resources management, particularly when designing and managing rural water resource systems, and when assessing water utilization and demand. In this study, Multivariate Adaptive Regression Spline (MARS) and M5 Model Tree (MT) models were coupled with a maximum overlap discrete wavelet transform (MODWT) to create MARSMODWT and MTMODWT models for the prediction of monthly pan evaporation for Turkey’s Siirt and Diyarbakir meteorological stations. The performance of the standalone MARS and MT models was compared to the corresponding MODWT-based hybrid models. Furthermore, the developed hybrid models were combined with (Epan) Mallow’s coefficient (Cp) to minimize the number of predictor variables needed to predict monthly Epan. The models used preprocessed input data, including temperature (T), wind speed (W), relative humidity (RH), and solar radiation (SR). The performance of each approach was evaluated using standard statistical measures (i.e., correlation coefficient (R), root mean square error (RMSE), Nash-Sutcliffe efficiency (NSE) and mean absolute error (MAE)). The results showed that the MARSCpMODWT model improved the MARS accuracy with respect to lower percentages of RMSE (29.46%) and MAE (24.53%) in the validation phase for the Siirt station. In case of the Diyarbakir station, the MARSCpMODWT improvements decreased the RMSE (17.91%) and MAE (16.49%) values in comparison to standalone MARS model. The overall results indicated that the use of both MODWT and Cp as pre-processing techniques improves prediction accuracy, and thus, they are both recommended for use in further studies.
Agid:
6494910