Radar imagery isn't a raw product so alot of features that might have helped like strong echos from the melting line have already been removed. That being said, I still think you can see some clues.

Clue 1: Snow tends to smear out, like a Gaussain blur has been applied compared to rain

Clue 2: Run the animation, snow has filaments of heavier precip appear and disappear, while rain has blobs of stuff that is neatly advected

Clue 3: Alot of heavy pixels is usually rain or hail. Snowfall is usually lighter than rain, on normal scales 2-3mm/h is moderate rain but really heavy snow. In showers the occasional red pixel I interpret as hail rather than snow as hail seems to behave like rain. 

Here is an example image from today showing where I think is the rain/snow boundary:

https://i.ibb.co/XZkmfp3y/Untitled.png 

It is worth familiarising yourself with where the metoffice radar stations are, this doesn't work well in areas of poor coverage where pixels tend to look 'larger'. In areas of high coverage you are also less likely to have the radar pointing at a cloud where snow is more likely to be seen. Naturally this is harder in showers compared to fronts. Showers are heavier and smaller which means a big intensity gradient which isn't great for this method. And in heavy snow showers pixels of hail are not uncommon. Nevertheless snow showers tend to have a largeish 2-3mm/h 'core' with hail pixels compared to rain. Also snow showers behave in the same erratic appear/disappear way with the different colours. 

I'm not sure how valid any of this is, but it seems to work reasonably well on fronts especially. You kind of get used overtime to see the subtleties. If the horizontal rain/snow interface on a front is rather clean then it becomes quite obvious, snow just 'moves' in such a different way compared to rain where any structures in the intensity tend to remain intact unlike with snow.  

I assume they use model data combined with intensity. There is probably an equation that links precip rate and wet bulb temperature to precip type.

I've tested out my method btw today and it genuinely works even for showers. Its quite tricky but snow does look different on the radar to rain. You have to look for 'smudging' and features that are poorly advected over time. On netweather heavy snow showers seems to usually take on the appearance of greens with yellow filament structures (aligned to the wind) and low intensity gradients in the core of the showers.

The method doesn't really work for lighter precip very well. 

I've also been looking at US radar that actually does use dual polarization to perfect the method. It really works, but its quite subtle. 

The other issue I seem to be running into is that UK radars are really quite low res, its much easier to identify the structures in the US (and we can then compare to the dual polarization precip type) because the data seems higher res.