Building on close to two decades of remote sensing-based forecasting of sugarcane, this study explored a newly derived machine learning (ML) approach for sugarcane yield prediction at the block level. Time series (2014–2024) Landsat 7 (L7 ETM+) and Landsat 8 (L8 OLI) derived vegetation indices (VIs), along with historical yield and agronomic data (for each of the modelled blocks) were sourced from the Condong growing region of Australia. This data were used as inputs into three well-performing ML models: random forests (RF), support vector regression (SVR) and extreme gradient boosting (XGBoost). RF was identified as the best-performing model for forecasting individual block yield achieving the mean absolute error (MAE) of 10.3 (T/Ha) and a root mean square error (RMSE) of 12.6 (T/Ha). This predictive model demonstrates higher accuracy compared to previous studies and provides valuable insights for harvest and forward selling planning in the sugarcane industry. Additionally, the model can be scaled to deliver block level predictions for all sugarcane blocks within a region or the entire industry.