Ensemble forecasting means producing multiple forecasts by making small alterations to the starting conditions or the forecast computer model.
Ensembles are seen as the future of Numerical Weather Prediction - NWP; they work in the same way as traditional NWP, but instead of one model running at a time, multiple models run simultaneously with slight perturbations.
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. We design the ensemble forecast system so that each member should be equally likely.
Currently the UK ensemble model, MOGREPS-UK, has a resolution of 2.2 km with 18 members and another 18 time-lagged members (from the previous run of the ensemble, given updated started conditions), giving 36 possible forecasts.
How ensembles improve forecasts
Traditional models only give you one forecast, ensembles will give multiple forecasts and better take into account variations in conditions. This allows meteorologists to see a variety of scenarios that may occur, then apply their expertise to decide on the most likely forecast. By comparing these different forecasts the forecaster can decide how likely a particular weather event will be. If the forecasts vary a lot then the forecaster knows that there is a lot of uncertainty about what the weather will actually do, but if the forecasts are all very similar they will have more confidence in predicting a particular event.
As the atmosphere is a chaotic system, very small errors in its initial state can lead to large errors in the forecast. This means that we can never create a perfect forecast system, because we can never observe every detail of the atmosphere's initial state. Tiny errors in the initial state will be amplified, so there is always a limit to how far ahead we can predict any detail. Ensembles thus provide the advantage that any errors there may be in a single traditional forecast are more likely to be discovered.
Ensembles can aid decision-making for those who are sensitive to the occurrence of certain weather events. However, this also presents its own set of problems, particularly when communicating uncertainty and probabilities.
For example, the forecast tells you there is a 60% chance of a shower in London at 2pm, what does that mean? Does this mean anywhere in London or the whole of London? Does it mean the shower will be exactly at 2pm? And does it mean there is no risk of a shower before or after 2pm?
This is where the expertise of the meteorologist can be applied, so in this case it could mean that of the 36 members, 22 have a shower somewhere over London at 2pm and 14 do not; they may have the shower later, earlier or not at all.
Being able to look at all the members doesn't just provide the percentage risk of an event, but also where there is uncertainty; where this forecast could be different and why. In this example, it could be that the other 14 members do not have the shower, because those members do not have high enough day time temperatures to produce showers.
This extra information allows people to make better decisions; if the temperature needs to be over a set number for showers then people know that if this temperature hasn't been reached by 2pm, then there will not be any showers and if it has, there will likely be showers.
European Centre for Medium-Range Weather Forecasts (ECMWF) is an international organisation supported by many European states, including the UK, which specialises in numerical weather prediction for medium-range prediction.
The ECMWF EPS consists of 51 forecasts run using the ECMWF global forecast model with a horizontal resolution of around 32 km. One member, called the control forecast, is run directly from the ECMWF analysis - our best guess at the initial state of the atmosphere.
Initial conditions for the other 50 members are created by adding small perturbations to the original ECMWF analysis, to represent uncertainties in the initial state. Small random variations are also made to the forecast model, to represent uncertainties in how the forecast model represents atmospheric processes.
We use the ECMWF ensemble, along with our own model and models from other forecast centres, to assess the most likely outcome in the medium-range forecast, plus the uncertainty in that forecast. ECMWF ensemble forecasts are then processed further to generate a range of probability forecasts for our forecasters and for customers.
The future of ensembles
A lot of work still needs to be done on ensembles before they can replace traditional models, for example, developing realistic tweaks between each member of the ensemble. But also, our current models are so reliable in the short range that ensembles do not add much value until the medium to longer range, so will need significant improvements in resolution and other properties before replacing traditional models in the short range.