Predicting sub-continental fuel hazard under future climate and rising [CO2] - combining ground-based observations with optimal plant behaviour in a machine learning approach