Each model climate is derived from a number of perturbed ENS forecasts using the same model techniques and physics as the current model, and the system uses historical runs on dates in past years relating to the date (i.e. day-of-month and month) of the current ENS run. The re-forecasts differ in number and detail and are described in the relevant section for M-climate, ER-M-climate, and S-M-climate.
Each re-forecast starts from a reanalysis. The Control run proceeds on from this, whilst the other members have perturbations applied in a way that is similar to the operational ENS, but that does not involve any data assimilation. The initial perturbations come from singular vectors, and by adding geographical averages of EDA perturbations that have been computed operationally in the most recent 12 months. This approach means that the flow-dependence inherent in operational EDA perturbations is missing in the re-forecasts. Stochastic physics are also used during the re-forecast runs, as in operational ENS runs.
The procedures adopted for using re-forecasts (to help create real-time forecast products) allow for seasonal variations and model changes to be taken into account. But it should be noted that the model climates (M-climate, ER-M-climate, or S-M-climate) can nevertheless be different from the observed climate.
Some Limitations of Re-Forecasts
Impact of differences between Reanalysis and Re-forecast systems
The model climate is generally compatible with model forecast output but there are still some local inconsistencies. In particular, the land surface scheme used by ERA-Interim differs. An attempt to address this inconsistency is by replacing the ERA-Interim land surface data with the contents of a separate but equivalent "offline land surface Reanalysis" that is compatible with the new model version. However, some inconsistencies remain - notably, for open water surfaces, a Reanalysis is not performed and climatological reference data is used instead. Also the real-time forecasting system contains a lake model (FLake) while ERA-Interim did not. Thus the representation of temperature in the model climate may unfortunately diverge in a systematic way from the actual changes within the forecast period. These effects are particularly important over and around large lakes or inland waters (e.g. the Great Lakes - notably Lake Superior, Aral Sea, Caspian Sea and some others). Users should therefore consider the possible deficiencies in model climates when considering EFI data. For an example of the effect, see Fig5.3.1 and Fig5.3.2. Such effects have also appeared in the Extended Range ENS and Seasonal forecasts.
Fig5.3.1(left): Extreme Forecast Index (EFI) for 2m temperature for Days10-15, ENS forecast run data time 00UTC 20 June 2017.
Fig5.3.2(right): Cumulative Distribution Function (CDF) for 2m temperature for Days10-15 in the middle of Lake Superior (red), with M-climate (black). The initialisation techniques are different for real-time forecasts (which make use of lake surface temperature information observed by satellites) and for the reforecasts (for which this information is not available). This can lead to the model climate developing anomalously warm or cold lake surfaces with consequent anomalously warm or cold 2m CDF temperature curve (black). This will then affect subsequent EFI and SOT fields. Here the realistic real-time forecast of 2m temperature CDF (red) over Lake Superior is thus incorrectly flagged as having a strongly negative EFI value in Fig5.10.
Use of Reanalyses by Re-forecasts
Re-forecasts start out using re-analysis fields for their initialisation, as described above. To this end it is important to understand the structure of reanalyses, and how they may differ from the model version used to create the re-forecasts, not least because spatial resolutions are generally different.
Reanalyses deliver a numerical description of the recent climate, produced by combining models with observations. They contains estimates of atmospheric parameters such as air temperature, pressure and wind at different altitudes, and surface parameters such as rainfall, soil moisture content, and sea-surface temperature. The estimates are produced for all locations on earth (albeit as gridbox averages), and they span a long time period that can extend back decades or more. ERA5 is an improved and more comprehensive ECMWF climate reanalysis that became available in early 2019, and that will replace ERA-Interim as the fundamental initialising analysis for re-forecasts with the release of Cycle 46r1 in mid-2019. The main differences of ERA5 from ERA-interim are
- higher spatial resolution (137 levels, 31km; 62km for EDA),
- higher output frequency of analysis fields (hourly, 3-hourly for EDA),
- introduction of uncertainty estimates,
- additional input observation types,
- many more output parameters,
- use of a much longer period of historical data (back to 1950, to become available by the end of 2019),
ERA5 also provides:
- a much improved representation of the troposphere,
- an improved representation of tropical cyclones,
- better global balance of precipitation and evaporation,
- better precipitation over land in the deep tropics,
- better soil moisture,
- more consistent sea surface temperatures and sea ice.
ERA products are normally updated once per month and within three months of real-time after quality assurance processing to ensure consistency by removal of biases in models and observations. Preliminary daily updates of the dataset can be available to users within 7 days of real time.
For more information regarding differences between ERA-Interim and ERA5 see:
For documentation and a full description of ERA5 see
Additional Sources of Information
(Note: In older material there may be references to issues that have subsequently been addressed)
- Watch a comprehensive lecture on estimation of the model climate (re-forecasts).
- Read about the importance of reanalysis for climate monitoring (page 2).