Integrating soil monitoring and machine learning to map historical tillage and stubble management across Australian cropping systems (2001–2023)

• ML predicts tillage and stubble practices from 22 years of soil data. • Random forest outperforms spatial baseline in predicting farm practices. • Year and latitude are key drivers of no-till and stubble grazing adoption. • Gridded maps reveal national trends in tillage and residue management. • Framework supports GHG accounting modelling and soil health policy development. Tillage and stubble management are key drivers of soil health, crop productivity, and greenhouse gas (GHG) emissions, yet long-term spatially explicit data on these practices remain scarce. Using data from two Australian national soil monitoring programs Soil Carbon Research Project (SCaRP) and Soil Organic Carbon Monitoring (SOC-M) we analyzed over two decades (2001–2023) of tillage and stubble management records collated across 300 Australian farmer paddock scale datasets. We developed machine learning models based on random forest to predict spatial and temporal trends in these practices, incorporating crop type, soil classification, climate variables, and spatial coordinates as predictors. Our models were benchmarked against multinomial logistic regression and a nearest-neighbour baseline, demonstrating that random forest consistently achieved higher accuracy, particularly for dominant practices such as no-tillage and stubble grazing. Variable importance analyses identified “Year” as the most influential predictors, capturing widespread no-tillage adoption post-2010, while latitude and crop type further explained regional variations. Partial dependence plots revealed sharp inflection points in adoption trajectories around 2010, coinciding with major policy and technological shifts. We applied the models to generate gridded predictions (0.5° resolution) across the Australian cropping zone, providing insights into the evolving distribution of tillage and stubble practices. These spatial products not only improve GHG accounting frameworks and Earth system models but also offer actionable intelligence for soil health policy and targeted extension programs. Our approach demonstrates how combining ground-based monitoring with machine learning can fill critical data gaps and provide a scalable framework for agricultural management assessment and climate mitigation strategies.