Implications of input estimation, residue quality and carbon saturation on the predictive power of the Rothamsted Carbon Model