Four new models have been added to the LuckGrib suite of models today, the National Blend of Models (NBM), as discussed in this announcement.

One of the uses of the NBM is that it can be used by the forecasters who generate the National Digital Forecast Database (NDFD), as a starting point for their forecasts.

The NBM versions of the Hawaii and Oceanic domains appear to be superior to the NDFD versions, and those two NDFD models are being deprecated today. At some time in the future they will no longer be available on the LuckGrib server cluster.

Consider these two images comparing the NDFD and NBM Oceanic domains:

All of the pink areas in the NDFD image show areas where there is no data available. Notice that the NBM image has none of these missing areas - this is a very important distinction. The NDFD Oceanic domain is restricted to the boundaries where the USA has operational responsibility, where the NBM domain covers the entire area. This NDFD restriction has always made this model less useful that it could have been and the NBM domain is clearly superior.

The process of how the NDFD data is generated is also one of both its strengths and weaknesses. NDFD data is unique in that this GRIB data is created by professional forecasters. This should have made this data set one of the most desirable available. However in practice when using this model, there are frequent delays for large regions being updated and the boundaries between forecasters regions are also often very obvious.

It appears that the NBM has great promise. Additionally, there is a planned upgrade of this model planned for Sept/2018 which will improve its skill.

For now, the NDFD model covering the USA (note, not Canada or Mexico) remains. I plan to provide both the NBM and NDFD versions of this domain for a little while so that they can be evaluated. It is likely that NDFD Conus will soon be deprecated as well.

Consider two more images, the NDFD and NBM versions of the CONUS domain:

Again, notice that the NDFD domain does not contain any data in Canada or Mexico while NBM contains data across its full domain.

I welcome feedback on this action as well as feedback on both the NDFD and NBM models.