Couldn't that also be because the snow had the entire winter to cool the ground it's sitting on? i.e. there'd be less melt of the snow that is in contact with the ground by spring time because the ground temperature (surface, and at depth) would be a lot lower by then, which might be effectively a lot like reducing snowpack temp gradients.
I'd be interested in using the model for an entire grid of meteorological values as the season progresses. The idea is to provide some kind of real time map, which could reduce the need for observations. You are right though, it may not provide very useful data, at least not initially. All weather models (even the high resolution ones) use a kind of 'smoothed' terrain as part of their lower boundary condition, which creates a mismatch between model elevations and real world elevations). I guess you could also use a statistical downscaling technique as well though...
Unfortunately I haven't got a full season worth of hourly data but I'd like to look into a snowpack model. You essentially just turn your snowpack model into a subroutine and call it form the main atmosphere model and you've got a nice tool to look at the snowpack where there are no observations. You could even make it a 2-way coupling and improve the output of the atmosphere model, although that probably wouldn't be worth it for the computational resources required.