No abstract

No abstract

The considerable year-to-year and seasonal variation in grassland production is of major importance to dairy farmers in Europe, as production systems must allow for the risk of unfavourable weather conditions. A large portion of the variability is caused by weather and its interaction with soil conditions and grassland management. The present study takes advantage of the interactions between weather, soil conditions and grassland management to derive a reliable grassland statistical model (GRAM) for grasslands under various management regimes using polynomial regressions (GRAM-R) and neural networks (GRAM-N). The model performance was tested with a focus on predicting its capability during unusually dry or wet years using long-term experimental data from Austrian sites. The GRAM model was then coupled with the Met&Roll stochastic weather generator to provide estimates of harvestable herbage dry matter (DM) production early in the season. It was found that, with the GRAM-N or GRAM-R methodology, up to 0AE78 of the variability in harvested herbage DM production could be explained with a systematic bias of 1AE1-2AE3%. The models showed stable performance over subsets of dry and wet years. Generalized GRAM models were also successfully used to estimate daily herbage growth during the season, explaining between 0AE63 and 0AE91 of variability in individual cases. It was possible to issue a probabilistic forecast of the harvestable herbage DM production early in the season with reasonable accuracy. The overall results showed that the GRAM model could be used instead of (or in parallel with) more sophisticated grassland models in areas or sites where complete data sets are not yet available. As the model was tested under various climatic and soil conditions, it is suggested that the proposed approach could be used for comparable temperate grassland sites throughout Europe.

No abstract

No abstract