Can data series from variety testing be used to predict yields in leys?

Data series from official variety trials of grass leys harvested three times a year (eight locations in Sweden during 1999-2011) were used for calibration of the semi-mechanistic growth FOPROQ model. Observations were available only at harvest. The linear fit between simulated and observed data was poor (R2 ≤0.33) and relative RMSE ranged between 20.7 and 38.1 %. It was concluded that current final harvest data is not enough for model calibration and extra samplings would be required to improve data quality for prediction of differences in variety performance in a variable climate