Comparison of using regression modeling and an artificial neural network for herbage dry matter yield forecasting

This study presents an application of artificial neural network and regression modeling techniques for forecasting grassland dry matter yield. Using data from a field plot experiment on semi-natural grassland in Maribor (Slovenia), the multiple regression and artificial neural network methodologies were employed to explain the patterns of dry matter yield during a 6-year period. On the basis of the two proposed approaches forecasts were conducted for the independent, validation year (6). The results in terms of Theil inequality coefficient, mean absolute error, and correlation coefficient show a better forecasting performance for the artificial neural network (likely due to the non-linear relationships prevailing among regressors and regressand) while relationships between observables can be better explained by regression modeling results. Copyright (c) 2016 John Wiley & Sons, Ltd. The application of artificial neural network and regression modeling techniques for forecasting grassland dry matter yield is presented. Using data from a field plot experiment on semi-natural grassland in Slovenia, the multiple regression and artificial neural network methodologies were employed to explain the patterns of dry matter yield during a 6-year period. The results show a better forecasting performance for the artificial neural network while relationships between observables can be better explained by regression modeling results.