Statistical modelling of nitrogen use efficiency of dairy farms in Flanders

In the past decade it has repeatedly been shown that agriculture is a significant source of ground-and surface water pollution. Nitrogen losses and nitrogen use efficiency (NUE) are major concerns in agricultural practice and of policy-makers. Rapid intensification of livestock production, a result of the focus on increasing productivity from the 1950s onwards, has contributed to a large increase in nutrient surpluses. Here, we performed a quantitative analysis of the variables influencing the nitrogen use efficiency in Flemish grassland-based farming systems. The analysis was based on the large dataset of the Farm Accountancy Data Network, holding technical and economic data of Flemish farms. A statistical model is proposed by performing multiple regression with several variable selection procedures. Many combinations of variables were studied in 92 models and different criteria were taken into account to select the most adequate combination of variables. This approach focuses on a deep statistical analysis and interpretation of the model. The novelty of this research is the quantification and comparison of the influence of different inputs and other variables in nitrogen use efficiency at the farm level. Our results show that, contrary to current knowledge, a higher nitrogen use efficiency was observed for farms with a higher number of 'Dairy cows ha(-1)', holding the rest of the N inputs constant. A higher stocking density is compatible with a higher agricultural sustainability. It is demonstrated that the amount of milk N produced by added cows is higher than the decrease in milk N produced by each single cow due to a reduced input of feed N per cow. The dairy cow becomes more efficient in the use of N, increasing the farm-gate NUE and the farm sustainability. In the dataset of this study, the variable 'Dairy cows ha(-1)' is more relevant than suggested by previous studies: 1.4 times more relevant than the variable 'kg of N in fertilisers ha(-1)', which is 1.5 times more relevant than 'kg of N in concentrates ha(-1)'. According to previous knowledge, the N input variables present a negative sign. Decreasing the N input in fertilisers, concentrates and by-products are recommended actions to increase the NUE. Unexpected interaction effects were found