Bug in unit conversion for cascade-wise correction
The bug was discovered when running temperature which is corrected cascade-wise, but it could occur for other variables that run cascade-wise. For the case of temperature, the situation was that the reference data came in degC and the model data in K. Thus, the convert_units method kicks in at doy_statistics on l261. That works well for the multi-day cascade component, but fails for the daily cascade components. That is because for daily anomalies around the multi-day mean, since they are differences, they do not need any unit conversion. Nevertheless, MIdAS convertes the daily anomalies in the reference data from degC to K and by doing that, they are far too high.