The forecast API carries two models side by side. The
National Blend of Models (NBM) fills hourly and daily. The HRRR fills
nowcast and current. The resolution, horizon and update rate of each are in
the reference.
Why they are never blended
They are kept apart on purpose. When they disagree about the timing of rain,
that disagreement is information, and a blend would hide which model said
what. The models block names each run, its age and how far it reaches, so you
can tell which model said what and how fresh it is.
What the probabilities say
The NBM probabilities in hourly are the highest value anywhere in the 10 km
cell, so they answer “anywhere near here”, not “on average across it”.
The four precipitation type probabilities (rain_probability,
snow_probability, freezing_rain_probability and ice_pellet_probability)
are four independent numbers, not a split of 100. An uncertain hour can be
plausibly both rain and snow.