...
Topic / title | Description | Related activities / comments | ||
|---|---|---|---|---|
General issues | ||||
| G1. Overly smooth forecasts. | A function of the error minimisation training of AIFS is to deliver increasingly smooth fields at longer lead times. Equally there is shortfall in energy at small scales that increases with lead time. One example is strong wind zones around deep cyclones, which get smoothed out, reducing peak values. Another example area would be objective fronts - identification of such requires the nearby thermal gradient to exceed a threshold; then in practice total front length will reduce with forecast lead time as gradient peaks get smoothed out. | Whilst this behaviour is also a well-known characteristic of an ensemble mean, the issue is less pronounced in AIFS. Plus, successive AIFS implementations have managed to further reduce the smoothing effect. | ||
| G2. Underestimation of small-scale extremes | AIFS resolution is ~28km. Where the spatial extent of extreme values is smaller the AIFS cannot and should not represent peak values. Examples include topographically- or convectively-forced localised rainfall extremes, low level wind extremes around tropical cyclones or extreme extra-tropical cyclones, localised temperature extremes in complex topography (e.g. in valleys or on mountain tops). | IFS output exhibits the same behaviour, but for the current medium range ensemble the issue is much less because gridlength is much smaller. In AIFS such issues are exacerbated by G1. | ||
| G3. Parameter consistency | In a non-physical model such as AIFS there is more scope for inter-parameter consistency to be lacking at specific locations at specific times. | Ordinarily this is not a major problem, but there have, for example, been cases of precipitation without cloud. T1 provides a more substantive example. | ||
Cloud cover | ||||
| C1. Under-dispersive distribution | In AIFS, cloud cover "extreme" values (zero and 100%) are systematically under-represented, whilst intermediate values are systematically over-represented. | For stratus cases, for example, where cover is commonly 0% or 100%, usually the AIFS value lies inbetween whilst the IFS value does not. | ||
| C2. Issue centred on the Horn of Africa | Whilst the verification of downward (shortwave) solar radiation shows AIFS to generally have smaller errors than IFS, we are aware of one particular problem area centred on the Horn of Africa (spanning parts of Somalia, Ethiopia and Yemen), where the AIFS errors looks to be much larger. | Issue over arabian pin | ? | |
2m temperature | ||||
T1. Consistency issues | AIFS is primarily trained using ERA5 data. For the 2m temperature component the offline land-data assimilation 2m temperature field is used. This uses 2m temperature observations, which can sometimes be inconsistent with the overlying atmosphere simulation in ERA5 which does not use those observations. One example would be challenging winter-time low cloud/fog. In such cases the cloud cover could be inconsistent with 2m temperature within AIFS output. | Despite this inconsistency, the AIFS 2m temperature fields in such situations are often much more accurate than those of IFS. | ||
Precipitation | ||||
P1. Convective precipitation extent | The areally-integrated amount of convective precipitation forecast is noticeably less in AIFS than in IFS. | This needs further checking. It is hard to verify what is the truth. | ||
P2. Small precipitation totals | Small amounts of precipitation look to be more commonplace in AIFS than they should be (and more common than in IFS). For example small totals (< ~0.1mm/6h) can be repeatedly predicted over arid areas like the Sahara when they look to be impossible. |