Senani Karunaratne, Elizabeth Morse-McNabb, Anna Thomson, Dani Stayches, Joe Jacobs
Traditionally, quantification of dry matter (DM) yield at a paddock or farm scale is undertaken using a rising plate meter (RPM) which provides paddock scale estimates via calibrated equations to predict pasture DM yield. This approach ignores the inherent spatial variability within a paddock which may limit optimum utilisation. In this study, unmanned aerial vehicle (UAV) datasets were coupled with modern machine learning data analytical methods to model and map pasture DM yield variability across individual paddocks at 1 m spatial resolution. The results revealed that the near infrared spectral band had the highest influence in predicting the pasture DM yield. However, the use of additional UAV-derived data sources, such as digital surface and digital terrain models as proxies for pasture height, further improved the prediction. Height derived from the UAV datasets was identified as the second most important variable in prediction of the pasture DM yield. Derived models were cross-validated and also independently validated through data splitting which resulted in concordance values of 0.90 and 0.40 respectively. Model comparison with the calibration equation derived using a RPM revealed that both methods reported equal validation of the results based on the cross-validation. However, the RPM model surpassed the independent validation results of the UAV-machine learning modelling approach. There is potential to explore a wider spectral range and other ancillary datasets for model improvement, in order to improve these machine learning models for prediction of DM yield across the paddock scale.