Page History
...
A single execution of the model does not give definitive results. Any individual forecast may or may not be skilful and it cannot provide an estimate of uncertainty, or the confidence to be placed on the forecast. To address this problem an ensemble of forecasts is executed, each starting from a slightly perturbed analysis. The perturbations are not arbitrary but are designed to be what are believed to be possible truths. The members of the ensemble then represent the impact that changes to the initial conditions and physical parameterisations would actually have on the atmospheric evolution. This enables assessment of the uncertainty in the forecast, and also gives an indication of the predictability of the future evolution of weather systems.
Other ECMWF experimental models
- Destination Earth (DestinE): This is an experimental IFS model with finer resolution of ~4.4km. It is currently non-operational.
- Artificial Intelligence models: These are trained on historical weather data, usually a subset of ECMWF’s ERA5 reanalysis dataset. They rely on traditional NWP analyses as initial conditions when producing a forecast. Currently horizontal resolution is about 25km and vertical resolution is at standard pressure levels. Artificial Intelligence models currently available on Opencharts and ecCharts are:
- ECMWF Artificial Intelligence/Integrated Forecasting System (AIFS).
- Non-ECMWF AI models
- FourCastNet
- FuXi
- GraphCast
- Pingu-Weather
These AI models have proved skilful in broad scale pattern but local details are often missing due to resolution and many parameters are not produced or mot available. Section 2.1.6 Machine Learning models gives further information.
Changes and improvements are outlined in the latest IFS model upgrade Cy48r1 .
An informative videocast (from the Weather Pod) of a discussion on AIFS is available.
ML models are evolving quickly. Users should try to stay up to date with the latest AIFS developments.