The impact of uncertainties on predicted GHG emissions of dairy cow production systems

Abstract Dairy farms produce significant greenhouse gas (GHG) emissions and are therefore a focal point for GHG-mitigation practices. To develop viable mitigation options, we need robust (insensitive to changes in model parameters and assumptions) predictions of GHG emissions. To this end, we developed a stochastic model to estimate the robustness of predictions based on input parameters (GHG emission factors and production traits) and their uncertainties. In our study we explored how sensitive predictions of GHG emissions are to three factors: (1) system boundaries of the emission model (2) the uncertainty of input parameters due to quality of data or methodological choices (epistemic uncertainty) and (3) inherent variability in input parameters (variability uncertainty). To assess the effect of system boundaries, we compared two different boundaries: the “dairy farm gate” boundary (all GHG emissions are allocated to milk) and “system expansion” (the model gives a GHG credit to beef derived from culled cows and bull, heifer and calf fattening of surplus dairy calves outside the farm). Results using the farm-gate boundary provide guidance to dairy farmers to reduce GHG emissions of milk production. The results using system expansion are important for defining GHG abatement policies for milk and beef production. We found that the choice of system boundary had the strongest impact on the level and variation of predicted GHG emissions. Model predictions were least robust for lower-yielding dairy cow production systems and when we used system expansion. We also explored which GHG-abatement strategies have the most leverage by assessing the influence of each input parameter on model predictions. Predicted GHG emissions were least sensitive to variability-related uncertainty in production traits (i.e. replacement rate, calving interval). Lower-yielding production systems had the highest variation, indicating the highest potential for GHG mitigation of all production systems studied. Variation in predicted GHG emissions increased substantially when both epistemic and variability uncertainty in emission factors and variability uncertainty in production traits were included in the model. If the system boundary was set at the farm gate, the emission factor of N 2 O from nitrogen input into the soil had the highest impact on variation in predicted GHG emissions. This variation stems from uncertainties in predicting N 2 O emissions (epistemic uncertainty) but also from inherent variability of N 2 O emissions over time and space. The uncertainty of predicted GHG emissions can be reduced by increasing the precision in predicting N 2 O emissions. However, this additional information does not reduce GHG emissions itself. Knowing site specific variability of N 2 O emissions can help reduce GHG emissions by specific management (e.g. reduce soil compaction, adopted manure management, choice of suitable crops). In case of system expansion, uncertainty in GHG emission credit for dairy beef contributed the most to increasing the variation in predicted GHG emissions. The stochastic-model approach gave important insights into the robustness of model outcomes, which is crucial in the search for cost-effective GHG-abatement options. Despite the high degree of uncertainty when using system expansion, its results help identifying global GHG mitigation options of combined milk and beef production.