In the past year, CSD’s Commercial Research Manager Dr. Michael Bange and Digital & Data Manager Chris Teague have teamed up with Dr. Tim Weaver, Dr Stuart Gordon and the Information Management and Technology (IM&T) team at CSIRO to develop and refine a new and improved Micronaire prediction model.

Micronaire is an important fibre quality parameter that affects the bottom line of growers and spinners alike. Discounts can be painful with little insight often till after classing; the intent of this model is to offer crop managers a predictive indication of Micronaire while the crop is still growing. Being able to improve management practices e.g., time of sowing and guiding the timing of harvest aid application to affect a harvest avoiding Micronaire discounts has been a long-term ambition in the industry.

Leveraging the insights discovered via the application of machine learning algorithms while developing BARRY, we decided to explore this technology further to evaluate its ability to improve the prediction of Micronaire beyond what has been reported in the literature. The team used machine learning algorithms to explore the extensive CSD trials dataset. The resulting algorithms revealed more accurate predictions, but also aligned with previous research relating to the impact of temperature on Micronaire.

Key variables (capturing temperature dynamics) applied in the new model included latitude, daily average temperature and the number of days above 35oC that a crop experienced from date of seed imbibition to first defoliation. Interestingly, exploring these same variables from first flower to defoliation did not contribute to better predictions. So, we kept it simple and deferred to date of seed imbibition.

The most exciting aspect of this work to date, is that the accuracy of the model improves to a R-squared (R2) value of 0.83 (Figure 1), which means 83% of the variance in the predicted Micronaire value can be explained by the above independent variables. Accompanied by a root mean square error (RMSE) of 0.206, means that predictions are reliable.

The collaborative effort leading up to the deployment of this model is in the final stages of development, visualisation and testing of the online tool. We look forward to introducing this tool for the 2024-25 season.