Model initiation verification

[peteo]

So random thought that popped up reading about Irene:

I'm wondering if any of the models go back and compare forecast conditions with observations to decide whether certain initializations were credible or had some long-term bias that could be corrected in future model runs?

For example, it might be possible to determine if a given station is reporting temps that are not indicative of the general area if a +0.5 or +1C bias on that station during model initialization produces more accurate forecasts.