Knowledge extraction from artificial neural networks: case study on reference evapotranspiration in southeastern of rolling pampas of Argentina

Abstract

Evapotranspiration is an important component of hydrologic balance and represent essential information for irrigation scheduling and water resources planning. The study aimed: a) to evaluate the performance of artificial neural networks (ANNs) with combinations of meteorological inputs for estimating reference evapotranspiration and b) to discuss the knowledge learned by the networks during the training process. Daily evapotranspiration values computed following the Penman Monteith equation (ET0PM), were used as target outputs for the implementation of the ANNs. Data of global radiation (Rg), net radiation (Rn) and extraterrestrial radiation (RTA) were alternated in combinations with air temperature (Ta), vapor pressure deficit (DPV) and wind (u) as inputs to networks. The ANNs with best performance for each combination of inputs were retained in order to evaluate the performance based on multi criteria analysis. According to the results, it can be concluded that it is possible to estimate accurately daily ET0PM values. A decomposition method based on Garson’s algorithm was applied to quantify the relative importance for each input variable. It was examined how model selection in ANNs can be guided by complementary procedures. The application of these methods in evaluation of ANNs models is discussed, paying attention especially on detection of the better predicting variables and analysis of errors.