The National Blend of Models (NBM) is a National Weather Service system that combines many computer models, from the NWS and from other centers, into one calibrated set of forecast guidance. Its goal is an accurate, nationally consistent starting point that forecasters then refine into the official gridded forecast.
How the blend is built
A 2023 overview from its developers lists inputs such as the Global Forecast System (GFS), its ensemble and the European ensemble, along with rapid-update models such as the High-Resolution Rapid Refresh (HRRR).
The same overview describes bias-correcting each input separately for every grid point, forecast hour and weather element, and tracking each input's recent errors the same way. The Meteorological Development Laboratory built it, and that lab has since moved into a new National Weather Service modeling office.
What it produces
The NBM runs every hour and produces output out to 264 hours, about 11 days. It covers the contiguous United States, Alaska, Hawaii, Puerto Rico, Guam and oceanic areas.
It offers both single best-estimate values and probabilistic guidance, such as percentiles and the chance of exceeding a threshold. The Weather Prediction Center's Probabilistic Precipitation Portal, operational on or about 10/01/26, draws its ensemble information from the NBM.
Version 5.0
As of October 2026 the NBM is at version 5.0, which took effect in spring 2026. It added higher-resolution European and Canadian model data, two AI-based models, probabilistic CAPE guidance and a new method for wind speed and gust forecasts.
Version 5.0 also switched to a quantile-mapping method for percentiles and exceedance values, aiming for better consistency between its deterministic and probabilistic output.
Reading it with care
A blended value works like a well-calibrated consensus, not a forecast of the extreme case. Like an ensemble mean, blending smooths out small, uncertain features, so local NWS forecasts and the spread in its probabilities are worth checking alongside the headline number.