MOGREPS-G is a global ensemble that produces weather forecasts for the whole globe up to a week ahead at 20km resolution, and is comprised of 36 separate model forecasts, or ensemble members. Ensemble forecasts quantify uncertainties in weather prediction and estimate risks of particular weather events to enable the user to assess the risks more accurately. Instead of running just a single forecast, the computer model is run a number of times from slightly different starting conditions. The complete set of forecasts is referred to as the ensemble, and individual forecasts within it as ensemble members.
MOGREPS-G model data is now available on Met Office Weather DataHub in a selectable format via our established API service to our users. Data is available for all 18 members in GRIB2 format and users will be able to subset by:
- Geographical region
- Time step
- Model run
With global coverage, MOGREPS-G is produced 4 times a day with weather forecasts to 198 hours at 0.18 degree horizontal resolution.
MOGREPS-G is run four times per day with 18 members (1 control + 17 perturbed) and the Met Office strongly recommends the combination of the latest 2 runs to provide a 36 member time-lagged ensemble.
|Grid length in mid-latitudes||Grid points||Vertical levels||Ensemble members||Forecast length||Run times (UTC)|
|20 km||1280 x 960||70 (lid ~80 km)||Control member + 17 perturbed members to 7 days
(36 members time-lagged: By combination of latest two runs)
|7 days 198 hours||00, 06, 12, 18|
|Initial conditions||Stochastic Physics|
|Global analysis with 4D-ensemble-Var perturbations||Model perturbations from Additive Inflation from historical analysis increments, Stochastic Kinetic Energy Backscatter (SKEB), Stochastic Perturbation of physics Tendencies (SPT), SST and soil-moisture perturbations.|
Starting Conditions Perturbation Methods
The future evolution of the atmosphere is very sensitive to small errors in the analysis that we use to start the forecast. To start an ensemble forecast we first make a set of small changes (or perturbations) to the analysis, which are consistent with the uncertainties in the starting conditions.
Stochastic Physics Perturbations
The forecast model tries to replicate the complex dynamics of the atmosphere and it does this by including many equations and approximations. These approximations will not always adequately represent the processes taking place and this can lead to errors in the forecast.
To account for as many different causes of forecast error as possible, MOGREPS makes small random variations to the forecast model itself, referred to as stochastic physics perturbations, as well as changes to the initial state. 1
- Randomly selected historical analysis increments which create ensemble spread and have the added benefit of reducing systematic model bias.
- Stochastic Kinetic-Energy Backscatter (SKEB). To backscatter (stochastically) into the forecast model some of the energy excessively dissipated by it at scales near the truncation limit. 2
- Stochastic Perturbation of Physics Tendencies (SPT). This scheme makes small random changes to the total tendencies from model physics routines returned to the model grid variables, recognising the uncertainties due to approximations inherent within the physics schemes.
- Perturbations are also applied to the Sea Surface Temperature (SST)3 and the Soil Moisture content, two surface parameters which have a significant impact on the evolution of the forecast for the atmosphere.
Each time we run an ensemble forecast, we use 17 of those perturbations, plus the 1 unperturbed analysis, as starting conditions for an ensemble of 18 different forecasts. We strongly recommend users to combine the previous 18 ensemble members with the current ensemble members to create a 36 member time-lagged forecast.
- 18 members – 1 control member and 17 perturbed members forecast to 7 days (198 hours).
- 36 time-lagged members – combining these 18 members with the 18 members from the previous cycle (6 hours earlier) to give a 36 member ensemble provides a much better representation of the spread of possible evolutions. The inclusion of the 2 analysis cycles improves the quality of the total ensemble spread. (Depending on the time of day at which the forecast is being used, the most recent members may be members 0-17 or 18-35).
Threshold and Probability Scenarios
What are the chances it will be warm enough to commence a garden party at 1pm?
- Assuming 20 Celsius is warm enough and based on 12 members predicting a temperature of less than 20 Celsius and 24 predicting a temperature of 20 Celsius or greater, the probability that the temperature shall rise above 20 Celsius is (24/36) x 100 = 67%.
Will it rain during an air show tomorrow afternoon?
- Based on 18 members predicting dry weather and 18 predicting precipitation, the probability that precipitation shall take place tomorrow afternoon is (18/36) x 100 = 50%.
Will the wind be safe enough to fly in a hot air balloon tomorrow?
- Assuming a wind speed below 3 m/s is acceptable for balloon flight and based on 6 members predicting a wind speed of higher than 3 m/s and 30 members predicting a wind speed of less than 3 m/s, the probability that the wind speed shall be below 3 m/s tomorrow is (30/36) x 100 = 83%.
- A Stochastic Kinetic Energy Backscatter Algorithm for Use in Ensemble Prediction Systems – Shutts, 2004. European Centre for Medium Range Weather Forecasts.