NOTES FROM THE ENGINE ROOM
Why the maps here are drawn the way they are — with the numbers, and the reasoning.
Each of these started life as a section of the engineering log and has been rewritten for a reader rather than a maintainer. The implementation detail is gone; the measurements, the trade-offs and the opinions are not. If you have ever wondered why a field stops where it does, why a number under the cursor is a third of a degree off the city label beside it, or why the GFS here looks sharper than the GFS elsewhere, the answer is in one of them.
The notes
ABOUT FIVE MINUTES EACHWhy the GFS here is 13 km, not 25
The obvious file is an interpolation of the model onto a grid twice as coarse. The right file is missing two fields, so they are derived — and checked against NCEP's own to a fifth of a degree.
Fifteen-minute steps, and the end of the 3-hourly barrier
Two cycles in three publish no hourly file at all, which is why the explorer was stale — and why the sub-hourly files fix freshness and time step at once.
The coordinate caveat: why a Lambert grid can't be read as lat/lon
The archive's values are right and its axes are wrong — by 413 km on average and 848 km at the corners. The errors that matter are the ones that still draw a plausible map.
The analysis, not a forecast: RTMA's cycle is a day
One field set an hour with nothing after it, a picker that offers days, and a "feels like" that steps seven degrees along a contour because the NWS's own definition does.
Absolute colour scales, and the probe that reads a pixel back
A map is a picture of numbers and should give them back. Why the ramps never re-fit, how the pointer inverts the very bytes that were drawn, and what ±0.63 °F is the price of.
Looking for the comparison instead?
Which model should I look at? puts the six live products — RRFS, HRRR, RTMA, GFS, NDFD and NBM — in one table: resolution, cycles, horizon, domains, fields, sources, what each is good for and what it is not.