Page 1© Crown copyright 2006ESWWIII, Royal Library of Belgium, Brussels, Nov 15 th 2006 Forecasting uncertainty: the ensemble solution Mike Keil, Ken Mylne, Richard Swinbank and Camilla Mathison Data Assimilation and Ensembles, Met R&D, Met Office ESSWIII, November 2006, Royal Library of Belgium, Brussels.
Page 2© Crown copyright 2006ESWWIII, Royal Library of Belgium, Brussels, Nov 15 th 2006 Outline Introduction to Ensemble Forecasting Perturbing analyses/models Examples of probability forecasts Application to space weather
Page 3© Crown copyright 2006ESWWIII, Royal Library of Belgium, Brussels, Nov 15 th 2006 Weather forecasting Today’s NWP systems are one of the great scientific achievements of the 20 th Century, but… Forecasts still go wrong! Oct '87 – still difficult with today’s systems Less severe errors are much more common, especially in medium-range forecasts What causes errors in forecasts? Analysis errors Model errors and approximations Unresolved processes
Page 4© Crown copyright 2006ESWWIII, Royal Library of Belgium, Brussels, Nov 15 th 2006 Ensemble Forecasts Small errors grow and limit the useful forecast range. By running an ensemble of many model forecasts with small differences in initial conditions and model formulation we can: take account of uncertainty sample the distribution of forecast states estimate probabilities Ensembles turn weather forecasts into Risk Management tools
Page 5© Crown copyright 2006ESWWIII, Royal Library of Belgium, Brussels, Nov 15 th 2006 Ensemble forecasts time Forecast uncertainty Climatology Initial Condition Uncertainty X Deterministic Forecast Analysis X Deterministic Forecast Forecast uncertainty
Page 6© Crown copyright 2006ESWWIII, Royal Library of Belgium, Brussels, Nov 15 th 2006 Adding perturbations
Page 7© Crown copyright 2006ESWWIII, Royal Library of Belgium, Brussels, Nov 15 th 2006 IC perturbations: ensemble spread Time Deterministic forecast, with increments each analysis cycle Ensemble forecast - spread increases, reflecting chaotic dynamics and model error Ensemble spread is a measure of forecast error After each analysis, spread is reduced, because of new information from observations data assimilation creates a new analysis Forecast phase
Page 8© Crown copyright 2006ESWWIII, Royal Library of Belgium, Brussels, Nov 15 th 2006 The Met Office has three schemes to address different sources of model error: Error due to approximations in parameterisations Random Parameters (RP) Unresolved impact of organised convection Stochastic Convective Vorticity (SCV) Excess dissipation of energy at small scales Stochastic Kinetic Energy Backscatter (SKEB) Model perturbations: stochastic physics
Page 9© Crown copyright 2006ESWWIII, Royal Library of Belgium, Brussels, Nov 15 th 2006 Examples
Page 10© Crown copyright 2006ESWWIII, Royal Library of Belgium, Brussels, Nov 15 th 2006 Ensembles – estimating risk By running models many times with small differences we can: take account of uncertainty estimate probabilities and risks eg. 10 members out of 50 = 20%
Page 11© Crown copyright 2006ESWWIII, Royal Library of Belgium, Brussels, Nov 15 th 2006 Example: Early Warnings of Severe Weather Met Office issues Early Warnings up to 5 days ahead - when probability 60% of disruption due to: Severe Gales Heavy rain Heavy Snow Forecasters provided with alerts and guidance from ensembles Challenges: Severe events not fully resolved Few events so difficult to verify
Page 12© Crown copyright 2006ESWWIII, Royal Library of Belgium, Brussels, Nov 15 th 2006 Katrina – from “operational” system
Page 13© Crown copyright 2006ESWWIII, Royal Library of Belgium, Brussels, Nov 15 th 2006 Katrina – NHC warning
Page 14© Crown copyright 2006ESWWIII, Royal Library of Belgium, Brussels, Nov 15 th 2006 Courtesy of Robert Mureau, KNMI.
Page 15© Crown copyright 2006ESWWIII, Royal Library of Belgium, Brussels, Nov 15 th 2006 End-to End Outcome Forecasting An ensemble weather forecast can be used to drive an ensemble of outcome models, eg: Wind power output Energy demand Hydrology – flood risk Ship or aircraft routes
Page 16© Crown copyright 2006ESWWIII, Royal Library of Belgium, Brussels, Nov 15 th 2006 Application to space weather
Page 17© Crown copyright 2006ESWWIII, Royal Library of Belgium, Brussels, Nov 15 th 2006 Application to SW: power supply Forecasts of disruption to power distribution High degree of uncertainly Longer timescales Ensemble thinking can help! A variety of perturbations can be applied to models Inputs – the behaviour of the sun Model parameters – known weaknesses
Page 18© Crown copyright 2006ESWWIII, Royal Library of Belgium, Brussels, Nov 15 th 2006 Power disruption probability Information of this kind can be useful to customers Critical thresholds can aid planning decisions: rescheduling grid maintenance load reduction
Page 19© Crown copyright 2006ESWWIII, Royal Library of Belgium, Brussels, Nov 15 th 2006 There’s a 50% prob of snow in London tomorrow 50% ? You mean you don’t know what will happen! Probabilities in context - a warning Probabilities need to be explained properly Normally it only snows one day in 50 at this time of year - so 50% is a strong signal. Probabilities must be unambiguous and relevant to the end user When’s this talk going to end?
Page 20© Crown copyright 2006ESWWIII, Royal Library of Belgium, Brussels, Nov 15 th 2006 Summary Utilising ensembles is now a mature tool in operational weather forecasting Ensembles provide extra information on Uncertainty Risks, particularly for high impact weather We are learning how to use probability forecasts for improved decision-making These ideas are being now considered in space weather forecasting Power supply disruption Applicable to other areas
Page 21© Crown copyright 2006ESWWIII, Royal Library of Belgium, Brussels, Nov 15 th 2006 Questions
Page 22© Crown copyright 2006ESWWIII, Royal Library of Belgium, Brussels, Nov 15 th 2006 Met Office Operations Centre Ops Centre forecaster uses the ensemble to assess the most probable outcome before creating the medium-range forecast charts… …and assess risks