Forecasts from the ECMWF Artificial Intelligence/Integrated Forecast System (AIFS) are now available on weather.utah.edu. Look for ECMWF AIFS in the left-hand nav bar. An example forecast is below. Based on the data available, most of our favored four-panel variables are available, but due to limitations in the data provided, there is no forecast of outgoing longwave radiation (used for a forecast infrared satellite image) in the upper-right panel and the precipitation is six hourly, which is a longer accumulation period than many of our products.
As illustrated in the link at the top of this post, AIFS forecast skill at 500 mb is roughly the equivalent of the ECMWF IFS (called HRES on weather.utah.edu) through about day four. At longer lead times, it exhibits somewhat better skill. The AIFS forecasts after 9 days are about as good as the IFS at 8.5 days, so they are basically giving you an extra half day of skill at longer lead times. That's not enough to abandon looking at the IFS, GFS, or ensembles, but it certainly makes sense to make the AIFS an important edition to your forecast workflow.
The AIFS forecasts are produced at ~31 km grid spacing and provided on at 0.25° grid. The latter is similar to what we use for the GFS-0p25deg and ECMWF HRES products on weather.utah.edu. However, the AIFS forecasts are smoother, especially for precipitation, which is a reflection of the training dataset and techniques.
In the long run, I am interested in adding the AIFS ensemble to the Utah Snow Ensemble. That is a 51 member ensemble run at ~31 km grid spacing and available at 0.25 deg. We have put in for computer time and engineering support on the new University of Utah Redtail supercomputer, but a lot has to fall into place for this to happen. It's unclear if I have enough bandwidth and compute power to make this happen otherwise, although one never knows.
No comments:
Post a Comment