Decoding spatio-temporal drivers of yield variability with interpretive machine learning
S. Poole, Thomas F. A. Bishop, Dhahi Al-Shammari, Patrick Filippi · 2025
Interpretive machine learning (IML) techniques were utilised to determine if the spatio-temporal drivers of yield could be identified, quantified and map. For two case study farms, in different agroecological zones of Australia, digital soil maps and elevation data were used in XGBoost predictive models for multiple seasons of yield data. The Shapley Additive exPlanations (SHAP) values provided a localised explanation within the field of each year’s predictive yield model. Temporally, the most commonly occurring yield limiting variable was determined for different crop types and rainfall seasonality. This study shows, the IML provides an interpretable way of determining the spatio-temporal drivers of yield that can vary in different agronomic contexts.