The value of the upper-air data in the Ernesto forecast

A few days ago I warned to not take much stock in the models until the Gulfstream-IV performed its first mission to collect upper-air observations to be fed to the models. Here’s a follow-up to show why that was the case, with a focus on the four models that make up the GUNA consensus, which was the best performing guidance last year. The members are the GFDL, UKMET, NOGAPS, and AVN models.

Let’s start with the last run of models without the data. As one can see in the graphic there was a good bit of spread in the models overall, with the AVN forecasting a south Florida hit, the GFS and NOGAPS forecasting west Florida, and the GFDL taking the storm out to the panhandle.

The Sunday night run with the upper air data did not feature such a spread. It isn’t until late in the forecast period that there’s an appreciable difference.

On the subsequent Monday morning run, some of the spread opened up again, with the UKMET putting its forecast back to the west some and though you can’t really see it on the graphic, NOGAPS moved Ernesto considerably faster to the north than the others.

Here’s how the 48 hour forecasts for these model runs turned out.

Time	GFDL error	UKMET error	NOGAPS error	GFS error	Avg error
12Z 27 Aug	86 nm	107 nm	68 nm	21 nm	70.5 nm
00Z 28 Aug	56	55	49	28	47
12Z 29 Aug	37	71	115	17	60

The upper-air data provided a 25-33% improvement to the average error of the four models.

As one can see though, the GFS was a significantly better performer than the other three members of the consensus and indeed as early as Saturday morning, its forecast was closer in the long term, though it was viewed as a “dubious solution”.