This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
1. If we consider local, instantaneous Z500 fcs, how long is the FiSH?
2. Does it make sense to talk very generally about a forecast ‘predictability limit’?
3. Can we develop a unifying framework that allows us to compare in a clear way the skill of forecasts of different variables at different scales and over different regions?
ENS fcs: bias‐corrected forecasts are from the ECMWF 51‐member ENS, the medium‐range/monthly forecasts (32km up do d10, 64km afterwards)
Verification: ERA‐I analyses CLI fcs: 100‐member climatological ensemble defined by ERA‐I 32‐d subsequent
analyses Accuracy metric: Continuous Ranked Probability Score (CRPS) Skill: CRPS(ENS) vs CRPS(CLI) Cases: 141 (2 per week, for 16m from 2/7/12 to 4/11/13)
Forecast skill depend on the spatial‐temporal scale
Large‐scale, time‐average features are more predictable than instantaneous, grid‐point values, and certain phenomena are known to be predictable weeks and months ahead.
Forecast skill depend on the spatial‐temporal scale
All forecasts represent average values over a space‐time volume: even a instantaneous, local values represents an implicit average.
Large‐scale, t‐average features are more predictable than instantaneous, local values. Unpredictable “noise” can be removed by averaging to isolate the predictable signal.
We have applied the same metric to differently averaged (in 4D) forecasts and asked:a) Does FiSH depend on the spatial‐temporal average (and on the variable)?b) Does it make sense to talk very generally about a forecast ‘predictability limit’?
Suppose that we have a good system that can simulate all scales relevant to predict phenomena with a scale (X,T), and initialise them properly. The skill of the phenomena depends on the competition between:• Errors propagating from the smaller scales, i.e. noise destroying the signal, and• Predictive signal propagating from the wider, longer‐range scales
‐ The MJO can affect extra‐tropical, low‐frequency phenomena such as blocking‐ Diurnal tropical convection influences organized convection and the MJO‐ The MJO propagates interacting with El Nino‐ El Nino and the MJO are affected by variations in solar radiation and greenhouse gases‐ Blocking influences and is influenced by synoptic scales, fronts‐ …..
Blocking
fronts
Solar radiationGreenhouse gases
organizconvec
MJO, El Nino
convec
(XS,TS) (X,T) (XL,TL)
Free smallerscales
An example: blocking over the Euro‐Atlantic sector
Lorenz (1969): ‘… one flap of a sea gull’s wing would forever change the future course of the weather .. Such a change would be realized within about 17 days ..’
We showed that there is not a unique definition of predictability and that the Forecast Skill Horizon, say the FiSH length, depends on the forecast field (scale, variable, region).The forecast skill horizon is well beyond 2 weeks even for local, instantaneous fields, thus confirming results published in literature that certain phenomena (MJO, NAO, blocking, ..) can be predicted beyond 2 weeks using a unifying, coherent framework.
• Reduced initial errors• More complete models (coupling to land and ocean)• Better models (improved moist processes, ..)• New methods (ensembles, ..)• Understanding of sources of predictability• Scale analysis
In other words .. 1970s: results based on atmosphere‐only models suggested that a sea‐gull wing could affect the weather anywhere after ~ 2 weeks2010s: results based on more accurate, higher resolution coupled ocean‐atmosphere models indicate that the limit is well beyond 2 weeks and that the predictability limit has not yet been reached