Estimating Daily Reference Evapotranspiration using Data Mining Methods of Support Vector Regression and M5 Model Tree

Evapotranspiration is one of the most important components of the hydrological circle and its proper determination is highly important in most researches such as water hydrological balance, design and management of irrigation systems, simulation of crop production and design and management of water resources. Nonlinear characteristic, uncertainty and needing for different climatological data in simulating evapotranspiration are the reasons that motivate researchers to investigate data mining methods such as M5 model trees and support vector regression. In the present study, the precision of mentioned methods in estimation of reference crop evapotranspiration in comparison with empirical methods such as Hargrivs and Torrent white equations was studied. For that purpose, using meteorological dataset of 1371-1394 years of Tabriz synoptic station, the daily values of reference crop evapotranspiration were computed by FAO-Penman-Monteith method. Then, using these computed values as target outputs, 17 various scenarios combining at last one to up to six meteorological parameters have been considered using mentioned methods. Finally, the capability of support vector regression and M5 model trees for estimation of evapotranspiration was analyzed using test data set. Results of statistical analysis and Taylor diagram showed that support vector regression and M5 model trees in a case of considering all meteorological parameters with root mean square of 0.398 and 0.44, respectively, provided precise results comparing with empirical methods such as Hargrivs and Torrent white.