The concept is that the algorithm is not dealing with the input parameters in an appropriate way. The extremely localised nature of the very low minima suggest a direct and possibly flawed correlation between snow depth and 2m temperature to me. I would expect this may be an issue that will be tweaked in future. I'm surprised you don't think a model could be overdoing temperature extremes based on flaws in an algorithm.
In this case we are not dealing with any random perturbation, rather the operational run. If, since the last update/version, the model has not encountered such synoptics as shown it is perfectly valid to assume the model may be "overdoing it".
Originally Posted by: doctormog