Global Modeling Systems Upgrades

The European Center for Medium Range Weather Forecasting (ECMWF) and National Centers for Environmental Prediction (NCEP) are substantially upgrading their global medium range forecast models this week.

The ECMWF upgrade of its Integrated Forecast System (IFS) occurred with yesterday's (11 June) 1200 UTC run.  Perhaps the most important change is an upgrade to the data assimilation system to make it more continuous.  This enables more accurate initial conditions and ultimately better forecasts.  Additional changes are described here.  ECMWF reports an improvement in midlatitude precipitation skill of 0.5-1% in their high-resolution deterministic run (HRES) and about 1% in their ensemble modeling system.  Those numbers seem small, but remember that they upgrade the IFS once or twice a year, so these incremental improvements add up.

NCEP is planning to upgrade their Global Forecast System (GFS) with the 1200 UTC run today (12 June.  This is a major upgrade as although though the modeling system name isn't changing, the "guts" are. 

The guts of a numerical modeling system are the dynamical core.  The dynamical core is basically the part of the model that integrates key equations describing atmospheric fluid dynamics and thermodynamics forward.  For a global model, this involves taking a complex set of partial differential equations and writing them in a way that they can be solved on a sphere.  This is easier said than done, and there are many approaches to doing this, with various strengths and weaknesses.  Typically there are tradeoffs made between numerical accuracy and computational efficiency. 

For the upgrade, NCEP is replacing the old GFS spectral dynamical core with a new finite volume cubed sphere dynamical core known as FV3 and visualized below.  Gory details available here. 

Source: GFDL

The new GFS, known as version 15.1 (the previous operational version was version 14, with no decimal) also includes some additional upgrades.  For precipitation prediction over the western U.S, the most significant is an upgrade to the cloud microphysics parameterization, which simulates cloud processes.  Resolution is comparable between the old and new operational versions. 

For the upgrade, NCEP performed 3 years of retrospective model simulations and real-time parallel simulations using the FV3-based GFS version 15.0.   Version 15.1, which is becoming the operational version, includes some tweaks to improve performance based on analysis of those retrospective model simulations and real-time parallel simulations. More information is available here. 

One of my students, Marcel Caron, has been examining the performance of GFS cool-season precipitation forecasts over the western U.S. Preliminary results suggests that version 15.0 produces improved forecasts compared to version 14, but still does not reach the level of the previous version of the ECMWF IFS.  I don't expect the tweaks made for version 15.1 to change these results.  Thus, I anticipate the new GFS will perform slightly better than the old across the western U.S. next winter, but that the ECMWF will remain in front.  Note, however, that this is a generalization for the entire western U.S.  The performance at individual sites may vary.

Ultimately, the success of the GFS upgrade and the FV3 dynamical core will be determined in the coming years.  The FV3 dynamical core is part of the NOAA/NWS Next Generation Global Prediction System, an effort to build a unified modeling framework for NCEP operational models.  For example, the HRRR would eventually move to the FV3 dynamical core.  Efforts are also underway to engage the broader research community.  Time will tell if these activities prove successful. 

Because of the GFS upgrade, we will no longer be providing the experimental FV3 runs on weather.utah.edu.