Known AIFS ENS Forecasting Issues

At long lead times, particularly beyond 10 days, unphysical artefacts can emerge in the model, particularly under areas of high elevation. These are particularly visible for mean sea level pressure and temperature at 1000hPa.

1000hPa temperature artefact beneath the surface of the Andes at 360 hours into the forecast.

Beyond about day 7, and more especially beyond day 10, unrealistic 2m temperatures have developed in some members, in a few regions around the world. Mostly the locations are high altitude, although the Red Sea has also been affected. Whilst there is intermittency in regions affected, some regularly feature. The unrealistic values tend to be too cold.

The difference, in 2m temperature standard deviation, at T+360h, in matching sets of AIFS ENS and IFS ENS 12UTC forecasts (AIFS minus IFS) (left map). Right map shows the same but for mean 2m temperature. Note how the high spread "hotspots" on the left panel have a lower mean.

Example AIFS-ENS meteogram, showing erroneously cold members developing towards day 10. The 12UTC (daytime) value of 15C on day 10 is below even the lowest night time minimum in IFS reforecasts for this site.

Very small values (e.g. <0.1mm/6h) of precipitation can occur in arid regions, in what appears to be an unphysical way. This is particularly noticeable when looking at long accumulation periods. These anomalies can also be seen in snowfall.

240h accumulated precipitation

In general, AIFS ENS significantly improves on the representation of fine scale features compared with AIFS Single. However cloud fields display features at a lower spatial resolution than many of the other fields predicted by the AIFS ENS. Features have a resolution closer to the 1-degree resolution of the model processor grid. T+0 fields do not show this behaviour because they are imported from the IFS initialisation.

Example of reduced spatial resolution for cloud fields over Europe in an AIFS ENS forecast.

Energy spectra for total cloud cover in AIFS ENS at day 1 (blue line), showing reduced energy at small scales (=high wavenumbers, on the right of the plot) compared with the IFS analysis (black). Energy can be loosely interpreted as frequency of occurrence.