Abstract
Forage quality is known to change with the phenological development of plants. A large data set, encompassing a wide range of climatic conditions and management practices was obtained in permanent meadows in South Tyrol (NE Italy) by means of sequential sampling and analysis of fresh-cut forage for a period of seven weeks starting at the phenological stage of stem elongation. Sampling was performed at 202 environments from 2003 to 2014 at altitudes between 666 and 1,593 m a.s.l. Statistical predictive models, taking into consideration meteorological and climatic variables (primarily growing degree days), as wells as variables related to geomorphology, botanical composition, soil and agronomic management were developed for crude protein (CP) and K by means of mixed models. ey were optimised using a stepwise forward selection and subsequent ve-fold cross-validation. Four models per each parameter, based on diierent combinations of available predictor variables, were developed. Higher predictive accuracy was found for the models taking the entire set of independent variables into account. e best predicted quality parameter was CP, while lower prediction accuracy was found for K. Lack of correlation was the most relevant component of the mean squared deviation of the models.