Page History
| Table of Contents |
|---|
Machine Learning Process for AIFS ENS
The algorithms for AIFS ENS are developed rather differently from those for use by AIFS single.
...
A small ensemble group (ECMWF uses a group of 4) is used that give information on the variability of these independent results and to introduce a measure of model uncertainty. At each grid point the set of observed data is processed using four different sets of random weighting functions for each parameter. The four forecast results are then compared with verifying data. The error metric (loss function) is the "almost fair CRPS" which measures how good forecasts are. CRPS is evaluated for the results of the four forecasts. The CRPS influences what is fed back to the ML processors (back propagation). This induces modification of the weighting functions for each parameter and the resulting forecasts are compared with the verifying data giving new CRPS values. This process is repeated many times with the aim to progressively minimise the error metric (CRPS). When this minimisation is reached the set of weights for each parameter forms the algorithm for the ensemble forecasting process (See Fig2B.4-1). Using CRPS as a loss function accounts for the limitations of using a finite number of ensemble members and ensures an accurate and well-calibrated distribution. Model uncertainty is incorporated as a learnt aspect due to the insertion of white noise.
Fig2B.4-1: Machine learning training process in deriving an algorithm for use in AIFS ENS. A range of observed data is processed using four different sets of random weighting functions for each parameter. White noise is introduced to each iteration to emulate model uncertainty. The four forecast results are then compared with verifying data. CRPS, which measures how good forecasts are, is evaluated for the results of the four forecasts. The CRPS influences what is fed back to the processor (back propagation). This induces modification of the weighting functions for each parameter and the resulting forecasts are compared with the verifying data again giving new CRPS value. Iteration continues until the CRPS is minimised and the set of weights for each parameter becomes the algorithm used by all ensemble members for the forecasting process. Note: In the diagram "other parameters" includes 6hr rainfall.
Forecasting process of AIFS ENS
Ensemble forecasting (AIFS ENS) brings an ability to quantify uncertainty and to identify possible extremes in the evolution. An ensemble of the AI forecast procedure can be executed because AI forecasting is rapid and cheap to run. At present AIFS ENS has 50 members and is run at six hour intervals. However there is potential for much larger and/or more frequent ensembles. Essentially each ensemble member is similar to AIFS single.
...
Making a forecast with AI is very efficient. It requires only a single Graphics Processing Unit (GPU), takes less than a minute to run, and consumes a tiny fraction of the energy required for an IFS forecast. This brings the prospect of more frequent and/or quite large ensembles of AI forecasts.
Fig2B.4-2: Schematic illustrating AIFS ENS. 50 members have perturbations based on Ensemble Data Assimilation (EDA) with singular vectors as with IFS ENS. The control is not truly unperturbed, as the algorithms have been generated probabilistically .
Points to consider when using AIFS ENS output
In addition to points for consideration outlined in the AIFS single section
AIFS ENS forecasts:
- High-quality ensemble forecasts that outperform ECMWF’s traditional physics-based ensemble system in medium-range forecasts on most accuracy and reliability scores.
- Very fast and economical to produce a forecast sequence.
- The principle for construction of meteograms and charts is very similar to that used by IFS ensembles.
- Plots of TCs probability, wind speed and central pressure can show disconcerting gaps. This is being investigated.
- Individual ensemble members do not exhibit the smoothing seen in AIFS SIngle and have similar levels of forecast activity to IFS. This is because the CRPS loss function does not encourage smoothing.
- AIFS ENS control verifies better than AIFS ENS members (possibly with 6-12h gain).
- Are currently little bit over-dispersive.
- Very small totals can appear in the precipitation fields far too often. This will be fixed in upcoming cycles.
- Physical consistency in the output is unlikely to be as good as with IFS. This is being investigated.
- Cloud fields appear very "blocky" after T+0. This is being investigated.
Practical use of AIFS ENS output
Forecasters should consider:
...
- AIFS schedules and products.
Fig2B.4-3: An example of an AIFS ENS meteogram for Riga DT06UTC 02 Oct 2025. The plots are in standard box and whisker form. The AIFS ENS control (Red) and AIFS Single (Blue) are shown as continuous lines. Note AIFS single departs largely outside the AIFS ENS box and whisker in places (e.g. precipitation). This may be due to the difference in the ML derivation of the algorithms for AIFS single and AIFS ENS.
FIG2B.4-4: An example of AIFS ENS chart output from AIFS ENS mean and spread VT12UTC 30 Jun 2025, DT12UTC 24 Jun 2025. Large uncertainty over the southern Andes as high as 8-14C locally. This can be due to 1000hPa contour lying below the model surface in mountainous areas.
...
Overview
Community Forums
Content Tools