Sound Science Bite: September 5. Toward a Better Weather Forecast

First, let me complain about current weather forecast policy. There is no sense as far as I'm concerned to give precipitation chances days in advance. If you follow forecasts, you will note how the chances for a particular day will often change significantly from one day to the next as the forecast day draws nearer in time. Surely, you wouldn't want to invest your money on that sort of information!

OK, beef over. Forecasts are nowadays largely based on the output of computer models of the atmosphere. The stars of this show are the GCMs (global circulation models). US weather scientists have been chafing for some time over the fact that the European model (ECMWF = European Center for Medium-range Weather Forecasts) is generally superior to the American model (GFS = Global Forecast System). Computer models are particularly critical for the forecast of damaging storms such as hurricanes, which are among the most difficult weather systems to forecast. Government officials need the best information as far in advance as possible to prepare and to take appropriate action. A forecast which is a dud can generate cynicism in a public that goes to a lot of trouble preparing for a storm that doesn't strike. A storm that strikes an area ill-prepared can be disastrous.

The effectiveness of GCMs depend, broadly, on three things. First, you have to have a mathematical model based on atmospheric physics expressed in computer code. Second, you have to have atmospheric data (temperature, pressure, dew point, wind, etc.) for the model to use. Finally, the faster the model can process the data, the more advanced the forecast. American hopes to match or better the Europeans resides presently in the FV3 (Finite-Volume Cubed-sphere Dynamical Core) model developed by Shian Jiann Lin and his team at the Geophysical Fluid Dynamics Laboratory (GFDL). This addresses the first and third items above. It is based on a different computational approach and more advanced mathematical techniques than the GFS. In tests it appears to at least match the ECMWF in predicting hurricane tracks. Right now it is considered experimental, but it is scheduled to become the operating model in 2019.