Kavina Dayal1, Tim Weaver2, Michael Bange2, CSD Ltd. Extension & Development Team3
2 CSIRO Agriculture & Food, 21888 Kamilaroi Highway, Narrabri, NSW, 2390
3 Cotton Seed Distributers Ltd., 2952 Culgoora Road, Wee Waa, NSW 2388
The ability to understand the impact of genetics x environment x management (GxExM) influences at a farm and paddock scale offers significant opportunities for informing management interventions to raise crop productivity. New means of agronomic data collection and collation, along with machine learning statistical approaches can help realise these opportunities. Cotton Seed Distributors Ltd. agronomy and extension team collect a large number of crop physiological and agronomic characteristics every year in their key varieties across the whole industry. Over the past four seasons this has resulted in the collection of a significant dataset for in-depth modelling. A machine learning algorithm (i.e., Random Forest) has been applied to understand which measured variables affect yield which can be used to identify management interventions. To evaluate the approach, the Random Forest method was applied to the dataset using key variables only during first flower. Variables were then used to predict yield at this stage (r2 = 0.74). The machine learning algorithm is intended to form the back bone of a decision tool so that crop managers can access the insights being generated from the dataset in real time and project current crop performance, giving them the ability to investigate the consequences of management interventions.