Why do they always come down? I feel like if the models are predicting the most likely totals then there should be equal probability that it goes down OR up. However, it seems the actual totals are always lower than what the computers say days in advance.
Obviously they don't always come down. They can go up too, but as to why they spit out 99th percentile solutions? Many reasons:
All these maps that have been posted in this thread are raw snowfall totals, that don't take into account melting or snowfall compaction. I think in many cases above there's at least a 3"+ bias baked into it because of this alone.
Something that we see particularly in the summer, but can also apply here are feedback loops, convective feedback. Basically problems with how the model handles convection and how storm that the model has spun up is modifying the modeled environment...it can create a runaway train effect (in summer, it's going to storm cause its so humid, it's so humid cause of these storms, it's going to storm cause it's humid...).
Think Spiderman 2 (I think)...when the bad guy's "precious tritium™" sun energy whatsit becomes self-sustaining, more or less what's going on sometimes with the run away snow or rain totals.
There are other biases that are in the models, things they don't/can't handle well.
Plus, with the storm still way out in the Pacific, it obviously isn't getting sampled, so there's very little real information going into the starting point. Then you add in the fact that the model is running in an "idealized" environment, and the resulting error propagation and "compounding interest" errors make for some crazy results.
