Lennart BengtssonESSC, Uni. Reading
THORPEX Conference December 2004
Predictability and predictive skill of weather systems and atmospheric flow patterns
-25 years of progress--The legacy of FGGE-
Professor Lennart Bengtsson
ESSC, University of Reading, UK
MPI for Meteorology, Hamburg
Many thanks to colleagues at ESSC and ECMWF
and especially to Kevin Hodges, ESSC
Lennart BengtssonESSC, Uni. Reading
THORPEX Conference December 2004
Predictability and predictive skill of weather systems and atmospheric flow patterns
The improvements in NWP over the last 25 years
• The impact of observations on forecast skill
• The importance of data-assimilation
• How much better can the forecasts be?
• The need for a new initiative in weather forecasting
Lennart BengtssonESSC, Uni. Reading
THORPEX Conference December 2004
Improvements in NWP from Miyakoda (1972) to 2002. Courtesy ECMWF
How long to get to D+10 in winter?
Lennart BengtssonESSC, Uni. Reading
THORPEX Conference December 2004
What are the causes of the improvements?
• More accurate models
• Advanced data-assimilation
• Better observational coverage
Lennart BengtssonESSC, Uni. Reading
THORPEX Conference December 2004
Lennart BengtssonESSC, Uni. Reading
THORPEX Conference December 2004
The principle of error reduction in data assimilation
Lennart BengtssonESSC, Uni. Reading
THORPEX Conference December 2004
Same model and data- assimilation but different observations. Results from ERA40 Courtesy ECMWF
Lennart BengtssonESSC, Uni. Reading
THORPEX Conference December 2004
Note that the error growth is virtually unchanged between the time of a good forecast (r=0.90) and the time of a useful
forecast (r=0.60). This fall in skill takes ca. 3.5 days
Lennart BengtssonESSC, Uni. Reading
THORPEX Conference December 2004
Possible causes?
• When errors have reached a certain size the growth rate is determined by internal dynamics. If so the skill can only be increased by a more accurate initial state.
• Or, there are errors due to down-scale cascade, say from tropical forcing, which not yet is properly handled by models
• Or, perhaps errors, due to aliasing or to incorrect handling of the up-scale cascade of physical processes, such as convection, projecting the errors on synoptic scale modes leading to an overly rapid growth?
Lennart BengtssonESSC, Uni. Reading
THORPEX Conference December 2004
There are large variations in forecast skillECMWF 1000 hPa height anomaly correlation
Cumulative frequency distribution for Europe. Winter 1988 and 1998
Lennart BengtssonESSC, Uni. Reading
THORPEX Conference December 2004
0
10
20
30
40
50
60
70
80
90
100
1 2 3 4 5 6 7 8 9 10
Forecast days
Anomaly correlation
There are large variations in forecast skillECMWF 500 hPa height anomaly correlation, Europe. Winter 1998 and 2004. Courtesy H Böttger, ECMWF
95% 75%
50%
25%
5%
Jan-Mar 1998
Jan-Mar 2004
Lennart BengtssonESSC, Uni. Reading
THORPEX Conference December 2004
Improvements in predictive skill in atmospheric flow patterns since FGGE
• NH extra-tropics has improved by more than three days• SH extra-tropics is as good as the NH• Five day forecasts in the tropics are as good as a one day
forecast 25 years ago But There are large differences in skill from day to day The time it takes for a forecast at 90% correlation to reach
60% correlation in the extra-tropics is about 3.5 days. This has not changed. Forecast improvements is due to more accurate initial states.
Lennart BengtssonESSC, Uni. Reading
THORPEX Conference December 2004
The impact of observations on forecast skillExperiments with different observing systems
DJF 1990/91 using ERA40 observations
• The control system (using all observations)
• A terrestrial based system ( radio-sondes and aircraft obs.)
• A satellite based system ( satellites and surface pressure)
• A surface based system ( surface observations)
Lennart BengtssonESSC, Uni. Reading
THORPEX Conference December 2004
Global forecasts DJF 90/91
• 7- day forecasts, every 6hr.• Later ECMWF model T159/L60
• Extra-tropics 20-90N and 20-90S• 500 hPa Z, normalized SD for the period
• Tropics 20N-20S
• Wind vector field 850 and 250hPa
Lennart BengtssonESSC, Uni. Reading
THORPEX Conference December 2004
Observing systems and predictive skillNorthern Hemisphere extra-tropics
Bengtsson and Hodges, 2004
Lennart BengtssonESSC, Uni. Reading
THORPEX Conference December 2004
Observing systems and predictive skillSouthern Hemisphere extra-tropics
Bengtsson and Hodges, 2004
Lennart BengtssonESSC, Uni. Reading
THORPEX Conference December 2004
Observing systems and predictive skillTropics V 250 hPa
Bengtsson and Hodges, 2004
Lennart BengtssonESSC, Uni. Reading
THORPEX Conference December 2004
Observing systems and predictive skillNorthern Hemisphere Z 500 hPa
DJF 1990/91 (full line) and DJF 2000/01 (dashed))
Lennart BengtssonESSC, Uni. Reading
THORPEX Conference December 2004
Observing systems and predictive skillSouthern Hemisphere Z 500 hPa
DJF 1990/91 (full line) and DJF 2000/01 (dashed))
Lennart BengtssonESSC, Uni. Reading
THORPEX Conference December 2004
Forecast skill and observations
• The terrestrial system is the best at the NH extra-tropics
• The satellite system is crucial for the SH extra-tropics
• In the tropics the terrestrial system and the satellite system are equally useful and highly complementary
• Between 1990 and 2000 the satellite system has increased its information content and the terrestrial system has decreased it
Lennart BengtssonESSC, Uni. Reading
THORPEX Conference December 2004
The impact of observations on forecast skillExperiment with ERA40
The impact of observations of humidity on NWP
360 global forecasts DJF 1990/91
Lennart BengtssonESSC, Uni. Reading
THORPEX Conference December 2004
Impact of humidity observationsNH Z 500 hPa
Full observing systemNo humidity observations in data-assimilation
Lennart BengtssonESSC, Uni. Reading
THORPEX Conference December 2004
Impact of humidity observationsSH Z 500 hPa
Full observing systemNo humidity observations in data-assimilation
Lennart BengtssonESSC, Uni. Reading
THORPEX Conference December 2004
Impact of humidity observationsTropics wind at 850 hPa
Full observing systemNo humidity observations in data-assimilation
Lennart BengtssonESSC, Uni. Reading
THORPEX Conference December 2004
Impact of humidity observationsTropics wind at 850 hPa (mean error at day 5)
Full observing systemNo humidity observations in data-assimilation
Control
No humidity
Lennart BengtssonESSC, Uni. Reading
THORPEX Conference December 2004
Humidity observationsin present data-assimilation
• Moisture observations have no detectable influence on the overall large scale predictive skill
• During the cause of the data-assimilation the large scale moisture field is mainly controlled by the model dynamics
• There is an urgent need to develop techniques for a better assimilation of humidity observations
Lennart BengtssonESSC, Uni. Reading
THORPEX Conference December 2004
Assessment of predictability
• Growth of small perturbations inserted in a model
• Comparing how consecutive forecasts separate from each
other ( Lorenz, 1982)
Lennart BengtssonESSC, Uni. Reading
THORPEX Conference December 2004
Lennart BengtssonESSC, Uni. Reading
THORPEX Conference December 2004
Estimation of predictability ( Lorenz, 1982)
The error between consecutive forecasts separated by a day ( or shorter) is a suitable expression for predictability. The initial error is then the difference between the analysis an day 1 and the forecast from the day before
Lennart BengtssonESSC, Uni. Reading
THORPEX Conference December 2004
Predictive skill ( Z 500 hPa) for the NHand predictability estimates ( for 6 ( red)
and 24 hr (blue) increments)
Lennart BengtssonESSC, Uni. Reading
THORPEX Conference December 2004
Predictive skill ( Z 500 hPa) for the SHand predictability estimates ( for 6 ( red)
and 24 hr (blue) increments)
Lennart BengtssonESSC, Uni. Reading
THORPEX Conference December 2004
Predictive skill ( V 850 hPa) for the Tropicsand predictability estimates ( for 6 ( red)
and 24 hr (blue) increments)
Lennart BengtssonESSC, Uni. Reading
THORPEX Conference December 2004
Lennart BengtssonESSC, Uni. Reading
THORPEX Conference December 2004
What are the predictive skill and predictability of storm-tracks?
• Storm- tracks as a proxy for predicting the transient weather
• Predictive skill of storm-tracks as a function of the observing system
• Estimation of predictability following Lorenz (1982)
Lennart BengtssonESSC, Uni. Reading
THORPEX Conference December 2004
NH, MSLP, Cyclones
Tracks Intensities
Lennart BengtssonESSC, Uni. Reading
THORPEX Conference December 2004
NCEP Ensemble Storm TracksCourtesy Z Todt, NCEP and L Froude, ESSC
Lennart BengtssonESSC, Uni. Reading
THORPEX Conference December 2004
Predictive skill and predictability of storm tracksfor different observing systems
NH
SH
Lennart BengtssonESSC, Uni. Reading
THORPEX Conference December 2004
Conclusions
• Major progress have been achieved following the FGGE in 1979 but significant improvements are still feasible
• Predictability estimates indicate a possible gain of several days compared to the present best forecasts
• The largest potential improvements are in the tropics
• The Thorpex program is a timely initiative and should be strongly supported
Lennart BengtssonESSC, Uni. Reading
THORPEX Conference December 2004
What is needed?
• Weather independent observations of wind and temperature profiles
• Research to better identify priorities of the observing system
• Higher resolution in model integrations to reduce numerical errors and representation problems and to simplify the parameterization of physical processes
• Continue dedicated process studies and associated field experiments
Lennart BengtssonESSC, Uni. Reading
THORPEX Conference December 2004
END