NCEP Model Output
Brian Tang’s Page (NAM, GFS)
Nice Visualizations of Outputs:
NCEP Model Guidance
Penn State e-Wall
University of Wyoming NAM, GFS, and RAP
*Watch out for models predicting huge areas of “light rain,” this sometimes means you should expect scattered showers instead (GFS especially tends to spread precipitation out over large regions). For Tropical Tidbits and Pivotal Weather, click anywhere to see a sounding.
Numerical MOS – here’s what the symbols are and here are the symbols for extended forecast.
Can also get MOS outputs from a long time ago here.
NCEP Model Diagnostic Discussions – see here for model biases
Vince Agard’s MOS verification page
Other Model Output
NWS Forecast and Forecast Discussion
NCAR Ensemble Forecasts:
NCAR/WRF-ARW Ensemble Models
NOAA ESRL High-Resolution Rapid Refresh Output:
Real-Time HRRR Maps – Philippe Papin
Experimental HRRR Ensemble
NOAA Storm Prediction Center Short-Range Ensemble Forecasts:
SREF Plumes – use dProg/dt “the trend is your friend”
SREF Precipitation Viewer
Other People’s Useful Pages
Notes on Weather Models
North American Mesoscale (NAM) Forecast System is just for North America and is higher resolution than the Global Forecast System (GFS). NAM is best within 48 hours or so; GFS model goes out two weeks or so, but the first week is much better resolved, second week is just for trends. GFS is spectral model, whereas NAM is run on a grid. GFS has slightly more vertical resolution (NAM is sigma vertical scale, GFS transitions from sigma to isobars). GFS uses longer timesteps than NAM. GFS uses slightly more complex convection scheme (Arakawa) than NAM (Betts/Miller). But parametrized convective schemes are generally bad for precipitation. Spectral models (such as GFS) better for dry forecasts? GFS tends to spread precipitation over a large area.
Uncoupled Surface Layer (USL): doesn’t simulate anything about the atmosphere, but rather uses other models’ atmosphere output for its boundary conditions. Works with a thin surface boundary layer without turbulence. However, is trained against a lot of previous data to optimize output based on the diurnal cycle, soil temperature, etc., so generally better for temperature than for precipitation-related things. Precipitation is based on some kind of averaging of what other models say (RAP, SREF, …). Probably better for stations where convective showers aren’t the dominant mode of precipitation (e.g. Key West). 2016, creator of USL said it should be best for Harrisburg and Reno. USL has pretty much no resolution through the boundary layer and uses other models.