mpo 674 lecture 2 1/20/15. timeline (continued from class 1) 1960s: lorenz papers: finite limit of...
DESCRIPTION
Ensemble Forecasts Epstein (1969), Leith (1974) suggested that instead of performing “deterministic” forecasts, stochastic forecasts providing an estimate of the skill of the prediction should be made. Several model forecasts with perturbations in the initial conditions or in the models themselves (will review perturbation methods later in the course)TRANSCRIPT
MPO 674 Lecture 2
1/20/15
Timeline (continued from Class 1)
• 1960s: Lorenz papers: finite limit of predictability?• 1966: First primitive equations model (6 layers)• 1971: First regional system (limited fine mesh
model)• 1978: Optimal Interpolation• 1980: Global Spectral Model• 1991: 3d-Var introduced at NCEP (Parrish and
Derber 1992)• 1993: First ensemble forecast systems at NCEP and
ECMWF
Ensemble Forecasts
• Epstein (1969), Leith (1974) suggested that instead of performing “deterministic” forecasts, stochastic forecasts providing an estimate of the skill of the prediction should be made.
• Several model forecasts with perturbations in the initial conditions or in the models themselves (will review perturbation methods later in the course)
Ensemble Forecasts
• Goals:– To provide an ensemble average that is more
accurate than individual forecasts, especially beyond the first few days. Components of forecast that are less predictable tend to be averaged out.
– To provide forecasters with an estimation of the reliability of the forecast
– Data assimilation– Adaptive Observations– Sensitivity Analysis
Ensemble Forecasts
• Can extend forecasts beyond Lorenz’s 2-week limit of weather predictability
• ENSO should be predictable a year or more in advance, since slowly varying surface forcing (from SST and land surface) should produce atmospheric anomalies that are longer lasting and more predictable than weather patterns
• Cane et al. (1986): first experiments
Ensemble Spaghetti Diagrams
108 h ECMWF Ensemble Forecast of pre-Karl, init. 0000 UTC 10 Sept 2010
CIRC
THICK
MSLPProbability
(TC at 108 h) = 68%
CIRC (x 10-5 s-1)TH
ICK
(m)
Ensemble prediction skill
Timeline (since 1993)
• 1993: Nonhydrostatic mesoscale models: MM5, CAPS, RAMS etc.
• 1997: ECMWF introduced 4d-Var operationally, most other centers (except NCEP) followed
• 2005: Canadian Meteorological Center introduced Ensemble Kalman Filter (EnKF) into operations
• 2012: NCEP introduced Hybrid 3d-Var / EnKF• 2015 (Jan 14): NCEP GFS 13 km resolution
Current and Future
• Detailed short-range forecasts (severe weather, rain and snow bands etc)
• Sophisticated, flow-dependent, continuous DA• Adaptive observing systems• Improving medium- and long-range forecasts,
primarily through ensembles• Fully coupled systems (atmosphere-ocean-
wave-land-ice-hydrology)• Public guidance: air pollution, UV radiation,
flooding levels, local winds, fires etc.