Forecasting Weekly Evapotranspiration with ARIMA and Artificial Neural Network Models

Abstract

Information about the parameters defining water resources availability is a key factor in their management. Reference evapotranspiration (ET0) prediction is fundamental in planning, design, and management of water resource systems for irrigation. The application of time series analysis methodologies, which allow evapotranspiration prediction, is of great use for the latter. The objective of the present study was the comparison of weekly evapotranspiration ARIMA and artificial neural network (ANN)-based forecasts with regard to a model based on weekly averages, in the region of Álava situated in the Basque Country (northern Spain). The application of both ARIMA and ANN models improved the performance of 1week in advance weekly evapotranspiration predictions compared to the model based on means (mean year model). The ARIMA and ANN models reduced the prediction root mean square differences with respect to the mean year model (based on historical averages) by 6–8%, and reduced the standard deviation differences by 9–16%. The variations in the performances of the prediction models evaluated depended on the weekly evapotranspiration patterns of the different months.