Assessing the validity of sensor-based predictions of post-grazing residual in dairy systems

Knowledge of post-grazing residuals are crucial for dairy farmers to adjust feed inputs and optimise pasture utilisation. However, many farmers rely on subjective methods, like visual assessment, to make grazing decisions. This case study evaluation applied a predictive model for post-grazing residuals, based on previous research, to two farms: a research dairy farm with smaXtec rumen boluses divided into three farmlets (alternative pastures, current and low emissions) and a commercial dairy farm with cows equipped with AfiCollars. The aim was to assess the model’s performance in an uncontrolled environment. There was some alignment between the predicted post-grazing residuals from the sensor-based model and the farm-walk data. However, the model’s explanatory ability was poor, with R² values calculated as the coefficient of determination ranging from -1.31 for the alternative pastures farmlet to 0.05 for the current farmlet and 0.15 for the low emissions farmlet. On the commercial farm, the R² was -1.12. While previous studies have demonstrated the potential of predicting pasture residuals from animal sensors, our study identified some challenges that would need to be overcome for broader application, including variations in on-farm management practices (e.g., supplement usage, frequency of pasture allocation, and mulching) and pasture species diversity. Given the infancy of this approach, further research is necessary to refine the predictive capabilities and clarify the specific contexts where its use could benefit New Zealand farmers.