Extended Data Fig. 2: Learning curves for models constructed using progressive seasonal data.

Models constructed to predict yield from a given point in the season relative to sowing (x axis), in a wheat and b canola, trained on 2008–2017 data and used to predict 2018 season data. As the season progresses, later-season data is included and models improve in forecast accuracy (y axes) with xvBCRF models (black) outperforming PLSR models (blue) and LSVMs (orange) over almost all forecast horizons.