mswx//nerds Monte Sano weather ops
forecast bus

Models seven forecasts, none trusted alone

No single weather model is reliably best, so we pull seven and play them off each other. Here's what each one actually is, how often it updates, and how much weight it carries — ranked by how much it shapes the number you see on the front page. New to the acronyms? They all link into the decoder.

NWS_GRID NWS gridpoint forecast core US · National Weather Service
The official forecast a Huntsville-office meteorologist publishes for our grid square — human judgment layered on model guidance. It's our reference line, and the one we bias-correct into the headline mountain number (*_MSWX). If you only watched one line, watch this one.
weight refresh ~hourly range 7 days
PIRATE National Blend of Models (via Pirate Weather) core US · NOAA NBM
NBM is NOAA statistically blending dozens of models into one calibrated forecast — the modern US backbone. We read it through the Pirate Weather API. It feeds the rain-chance blend and is our source for the UV index.
weight refresh hourly range 7+ days
OM_ECMWF ECMWF (the European model) support Europe · ECMWF · via Open-Meteo
Widely the most accurate global model at medium range. When the Euro and the American models disagree two or three days out, it usually wins. Carries a lot of weight in the multi-day temperature picture on the graph.
weight refresh 4×/day range 10+ days
OM_HRRR HRRR — High-Resolution Rapid Refresh support US · NOAA · via Open-Meteo
A 3 km US model that re-runs every hour and is the best tool for the next few hours — storms firing, a front arriving, this afternoon's cloud break. We poll it hourly for the short fuse; it fades to irrelevance past ~18 h.
weight refresh hourly range 0–18 h
OM_GFS GFS — Global Forecast System support US · NOAA · via Open-Meteo
America's flagship global model. A dependable workhorse, rarely the sharpest. We average it into the pool ensemble, where being independent of ECMWF matters more than being individually best.
weight refresh 4×/day range 16 days
OM_ICON ICON (German DWD model) support Germany · DWD · via Open-Meteo
Germany's global model. We include ICON precisely because it's built independently of the American and Euro models — when three independent models agree, that's real signal. Pool-ensemble member.
weight refresh 4×/day range 7.5 days
OM_BEST Open-Meteo "best match" support blend · via Open-Meteo
Open-Meteo auto-picks the highest-resolution model available for our spot at each lead time and stitches them together. A solid general-purpose line and a pool-ensemble member.
weight refresh continuous range 7 days
OM_HIST_* Historical-forecast archive archive archive · via Open-Meteo
Not a live forecast — a record of what each model said in the past (OM_HIST_GFS, OM_HIST_ECMWF, …). It's how the Verify page grades every source against what actually happened.
weight refresh nightly refill range past only

what you actually see the headline is never one raw model

The lines above are raw ingredients. The numbers on the front page are derived products built from them:

Mountain temperature (*_MSWX)

We take the raw forecast for the neighborhood point and apply bias correction — a five-year lookup of how much cooler or warmer the plateau runs than the airport at this hour, wind, and sky. That correction buys about 1.2°F of accuracy at short range over the raw model.

Rain chance (POP_BLEND_RADAR)

A lead-time-weighted blend of every source's rain probability, then bent toward the live radar nowcast inside the next hour — radar beats any model at "is it raining right now."

Pool temperature (POOL_ENSEMBLE)

We run the pool physics model separately against GFS, ICON, and Open-Meteo's blend — the three models that publish usable solar forecasts — and average the resulting water curves. Independent errors cancel; the mean beats any single model past day two. See it graded on the pool page.

Want to grade any of these yourself? The Verify page lays every source over reality and scores it.