Estimation of Reference Evapotranspiration using Climatic Data

Abstract

In this study, daily reference evapotranspiration (ET0) was estimated from climatic data using Multivariate Adaptive Regression Splines (MARS), M5 Model Tree (M5MT), and Gene Expression Programming (GEP). These approaches were trained with climatic data from eight weather stations in Iran for years 2000-2007. Thereafter, they were tested with data from the same eight weather stations in Iran for year 2008 and fourteen weather stations in California for year 2015. Four data combinations were evaluated: daily mean air temperature, daily mean wind speed, daily mean relative humidity, and solar radiation (configuration 1); daily mean air temperature and solar radiation (configuration 2); daily mean air temperature and daily mean relative humidity (configuration 3); daily maximum, minimum, and mean air temperature, and extraterrestrial radiation (configuration 4). In the first part of the study, MARS, M5MT, and GEP models were tested with data from the same Iran stations they were trained with. In the second part of the study, these approaches were tested with data from stations in California. The performance of MARS, M5MT, and GEP models were evaluated using mean absolute error (MAE), root mean square error (RMSE), and the coefficient of determination (R2). For all approaches, configuration 1 produced the most accurate results. Configuration 4 was found to be region dependent and is suggested when region specific data is limited (i.e., only temperature data is available). Results indicated MARS, M5MT and GEP could successfully predict ET0 from climatic data. Comparison of these approaches showed that GEP yielded the most accurate results. Also, it was found that MARS outperformed M5MT.