Comparative Analysis of Estimating Monthly Reference Evapotranspiration Using Kernel and Tree-Based Data Mining Models Versus Empirical Methods

Abstract

Because direct measurement of evapotranspiration is costly and time-consuming, researchers have turned to the estimation of evapotranspiration via indirect approaches. The aim of this study is to investigate the capability of kernel-based, tree-based, bagging-based data-driven, and empirical models to estimate reference evapotranspiration. For this purpose, data related to meteorological parameters such as average temperature, hours of sunshine, maximum and minimum temperature, wind speed, precipitation, and relative humidity were collected over a period of 39 years. A correlation matrix, relief algorithm, and trial and error based on the author’s own experience were used to select input scenarios. The performance of these methods was evaluated using correlation coefficient (R2), root mean square error (RMSE), scattering index (SI), Nash Sutcliffe (NS), and Wilmot indexes (WI). Based on the results, scenario 13 includes maximum temperature and monthly time index based on the relief algorithm was selected as the best scenario, also on the other hand the random tree model with R=0.99, RMSE=0.04 mm/day, and SI=0.01 was selected as the superior method. Thus, the maximum temperature was defined as the efficient meteorological parameter for the reference evapotranspiration modeling.

Keywords

• data mining
• makkink modified
• random tree
• reference evapotranspiration