Estimation of Reference Crop Evapotranspiration Under Different Combination of Meteorological Elements Using Multivariate Adaptive Regression Splines

  Objectives  Building a calculation model for accurate reference crop evapotranspiration (ET₀) prediction is important for regional water resource planning and irrigation scheduling design. This study focused on evaluating the performance of the multivariate adaptive regression spline model (MARS) in calculating the daily ET₀.

  Methods  Firstly, ET₀ calculated by the Penman-Monteith equation was used as the standard value. Secondly, daily meteorological data of the Yining station, in Yili Area, the Xinjiang Uygur Autonomous Region, China from 1996 to 2015 were adopted to construct 14 MARS models under different combinations of meteorological parameters and calculate ET₀. The calculation results were compared with those of the generalized regression neural network (GRNN), support vector machine (SVM), and 10 empirical equations based on temperature, mass transfer, radiation, and meteorological parameters.

  Results  The calculation accuracies of MARS, GRNN, and SVM were higher than that of the empirical equations. Overall, MARS had the best performance and the highest accuracy, and SVM was slightly better than GRNN.

  Conclusion  MARS was the best model for estimating ET₀ in this study.