Neural network based system in evapotranspiration time series prediction

Evapotranspiration is a very important process of the water cycle. Thus, the ability to model and understand evolution of this process is very important for any field that is dealing with water management issues. This paper presents a model which is able to predict evapotranspiration value for one step ahead. The model includes external factors that have influence on evapotranspiration. Evapotranspiration is observed as a time series with monthly records for several years. The time series modeling is conducted by artificial neural network infrastructure. The model is composed of three neural networks with different type of features in the input layer. Two networks are feed-forward, and one is recurrent. The prediction model composed of the three artificial neural networks is able to predict evapotranspiration values quite close to the original ones. The right choice of features and network type made this model very useful. This paper introduces another aspect of using artificial neural network in order to increase the precision of predicting the evapotranspiration time series.