There's always some spread between what models show and which forecast a meteorologist will put up at the time.
For example: Today the low level clouds started to move north about 6-7 hours ahead of what the high-res rapid refresh CAMs (Convection Allowing Models) were predicting. This indicates that even in the short range period, those models are off a bit. Likely due to small variances in the data that was used to initialize the model run vs the actual conditions.
This provides a clue that the models are "off" a bit, maybe only in timing, but likely more than that. Keep in mind that the lowest level of the atmosphere in the western GoM (Gulf of Mexico) is undergoing rapid changes due to dynamics taking place up stream and at higher levels above it. There's a layer of dry air above and the precip is starting from clouds that are a bit elevated. So most of the early radar returns are of virga where the falling precip will have to have some time to moisten and cool the lower levels that it's falling into, this wet-bulbing effect takes a bit of time to occur. This is one of many variable that a meteorologist must contend with besides just taking any model as "gospel". Other uncertainties have are around the thermodynamics within the column of air relate to the layers in which dendritic ice crystals (snow) formation mostly occurs ... that layer is rather dry right now, but moistening up. How fast can it moisten ?
Other variables (not involving cloud physics or storm precipitation efficiencies) around snowfall accumulation amounts have to do with complex ground level micro-physics involving temperatures, moisture, sublimation, compaction, melting, etc.
In the end every met can go with the model verbatim, or trust their instincts combined with their experiences. So you see large variance between a numerical model or an NWS met or the mets on TV.
The models, even this close to the event, are not in agreement with details on where, when and the amount of snowfall. Winter weather precip is always very hard to predict.
As an aside, a lot of people have most likely not seen this kind of data, but they have heard of the concept of Ensemble models. Take the GFS model as an example: It has 30+ members that are executed and each one will at arrive at it's own solution (forecast) for a given point of time in the future. This 30+ member run comprises an "Ensemble". Here's a precip forecast for Texas for next Monday (January 27th) .... you can see a huge variance in this single model (GFS) and its 30 members:
So from this; for next Monday what's the precip forecast for Austin TX ?
Here's what NWS Austin says:
No mention of precip or precip type.
-Matt