Ms Sally Poole1, Dr Patrick Filippi, Prof. Thomas Bishop, Dr Dhahi Al-Shammari
1Precision Agriculture Laboratory, School of Life and Environmental Sciences, Faculty of Science, Sydney Institute of Agriculture, The University of Sydney
Biography:
Sally Poole is an Agronomist and Digital Ag consultant working across southern QLD and northern NSW. She is also currently completing a PhD at the University of Sydney, exploring the use of digital tools to understand crop variability for improved agronomic management. Both her research and professional work focuses on the utilisation of spatial and on-farm data to gain a better understanding of crop production to develop more sustainable farming systems. Sally also serves on the board of directors for Crop Consultants Australia.
Abstract:
Agronomists and producers often inherently know the key drivers of seasonal and within-field crop variability. Yet with growing global demand for more sustainable and productive food systems, the ability to understand and quantify the key spatiotemporal drivers of crop production is critical to maximise input efficiencies and productivity potentials. This research focussed on a case study field of 1099 hectares located west of Moree, NSW, where 10+ seasons of yield data was available. Digital soil maps were produced of key soil properties and constraints (e.g. water-holding capacity, sodicity) using field-collected soil data to 1 m, and both proximally and remotely sensed spatial data. Elevation maps of the field were created using LiDAR data at 1 meter resolution. An XGBoost model was used to model yield for each season with the suite of soil and elevation predictors as variables. Interpretive machine learning techniques were then implemented on each seasons yield data to create maps of the most limiting variables by determining and mapping the most negative Shapley Additive exPlanation (SHAP) values of the predictors. The most limiting factor at each point was then determined for the amalgamation of the 10 seasons, by crop type, and by seasonally wet or dry years. The results show some consistent trends in the most limiting constraints on production. The ‘wet’ years produced the least consistent trends due to the varying impact and severity of the waterlogging events on different crop stages or crop types. The research shows that this methodology is useful for understanding and quantifying spatiotemporal influences on crop variability and can allow for improved management of crops in the future.