Downscaling weather forecast models has been done on multiple systems in the past, they are to an extent compute intensive task. At ArabiaWeather the Weather Technology division has managed to downscale multiple weather models (GFS, UKMET, GEM and others) to a 500m resolution for temperature. This is possible with a unique system (ModMS) that can do large scale storage and computations on grids in realtime and weather related algorithmic.
The algorithm combines multiple models, does some secret sauce algorithms by interpolation and lapse rate to compute the temperature down to 500m. We are currently working on going down to 100m in the next releases.
Why does it matter ? With the current deployments of the weather stations and the downscaled equation, error correction and model weighing will be a big thing giving rise to a new method of weather models post processing.