Transcript CliffStat - Atmospheric Sciences

Introduction to Weather
Forecasting
Cliff Mass
Department of
Atmospheric
Sciences
University of
Washington
The Stone Age
• Prior to approximately 1955, weather
forecasting was basically a subjective art, and
not very skillful.
• The technology of forecasting was basically
subjective extrapolation of weather systems,
in the latter years using the upper level flow
(the jet stream).
• Local weather details—which really weren’t
understood-- were added subjectively.
Upper
Level
Chart
The Development of Numerical
Weather Prediction (NWP)
Vilhelm Bjerknes in his landmark
paper of 1904 suggested that NWP-objective weather prediction-- was
possible.
– A closed set of equations existed
that could predict the future
atmosphere
– But it wasn’t practical then
because there was no reasonable
way to do the computations and a
sufficient 3-D description of the
atmosphere did not exist.
Numerical Weather Prediction
One such equation is Newton’s Second Law:
F = ma
Force = mass x acceleration
Mass is the amount of matter
Acceleration is how velocity changes with time
Force is a push or pull on some object (e.g.,
gravitational force, pressure forces, friction)
Using observations we can determine the
mass and forces, and thus can calculate the
acceleration--giving the future
NWP Becomes Possible
• By the 1940’s an extensive upper air network
was in place, plus many more surface
observations. Thus, a reasonable 3-D
description of the atmosphere was possible.
• By the mid to late 1940’s, digital
programmable computers were becoming
available…the first..the ENIAC
The Eniac
1955-1965:
The Advent of Modern Forecasting
• Numerical weather prediction became the
cornerstone.
• New observing technologies also had a
huge impact:
– Weather satellites
– Weather radar
Satellite and Weather Radars Provides a More
Comprehensive View of the Atmosphere
Camano
Island
Weather
Radar
Weather Prediction Steps
• Data collection and quality control
• Data assimilation: creating a physically realistic
3-D description of the atmosphere called the
initialization.
• Model integration. Solving the equations to
produce future 3D descriptions of the atmosphere
• Model output post-processing using statistical
methods
• Dissemination and communication
Initialization
Using a wide range of weather observations we
can create a three-dimensional description of the
atmosphere…
Numerical Weather Prediction
• The observations are interpolated to a 3-D grid where they are
integrated into the future using a computer model--the
collection of equations and a method for solving them.
• As computer speed increased, the number of grid points could
be increased.
• More (and thus) closer grid points means we can simulate
(forecast) smaller and smaller scale features. We call this
improved resolution.
Model Postprocessing in the U.S.:
Model Output Statistics (MOS)
• Main post-processing approach used by the
National Weather Service
• Based on linear regression: Y=a0 + a1X1 +
a2X2+ a3X3 + …
• MOS is available for many parameters and
time and greatly improves the quality of
most model predictions.
Prob. Of Precip.– Cool Season
(0000/1200 UTC Cycles Combined)
0.7
Brier Score Improvement over Climate
0.6
Guid POPS 24 hr
Local POPS 24 hr
Guid POPS 48 hr
Local POPS 48 hr
0.5
0.4
0.3
0.2
0.1
0
1966
1969
1972
1975
1978
1981
1984
Year
1987
1990
1993
1996
1999
2002
Major Improvement
Weather forecasting skill has substantially
improved over the last 50 years. Really.
P
Forecast Skill Improvement
NCEP operational S1 scores at 36 and 72 hr
over North America (500 hPa)
National Weather Service
75
S1 score
65
"useless forecast"
55
36 hr forecast
72 hr forecast
45
Forecast
Error 35
10-20 years
Better
"perfect forecast"
25
15
1950
1960
1970
Year
1980
Year
1990
2000
Why Large Improvement in Weather
Forecast Skill?
•As computers became faster, were able to
solve the equations at higher resolution
•Improved physics
•New observational assets allowed a better
initialization
A More Basic Problem
• There is fundamental uncertainty in
weather forecasts that can not be ignored.
• This uncertainty has a number of causes:
–
–
–
–
Uncertainty in initialization
Uncertainty in model physics
Uncertainties in how we solve the equations
Insufficient resolution to properly model
atmospheric features.
The Atmospheric is Chaotic
• The work of Lorenz (1963, 1965,
1968) demonstrated that the
atmosphere is a chaotic system, in
which small differences in the
initialization…well within
observational error… can have
large impacts on the forecasts,
particularly for longer forecasts.
• Not unlike a pinball game….
Probabilistic Prediction
• Thus, forecasts must be
provided in a probabilistic
framework, not the
deterministic single answer
approach that has dominated
weather prediction during the
last century.
• Interestingly…the first public
forecasts were probabilistic
“Ol Probs”
Cleveland
Abbe
(“Ol’
Probabilities”), who led the
establishment of a weather
forecasting division within the
U.S. Army Signal Corps.
Produced the first known
communication of a weather
probability to users and the public
in 1869.
Professor Cleveland Abbe, who issued the first public
“Weather Synopsis and Probabilities” on February 19,
1871
Ensemble Prediction
• The most prevalent approach for producing
probabilistic forecasts and uncertainty
information…ensemble prediction.
• Instead of making one forecast…make
many…each with a slightly different initialization
or varied model physics.
• Possible to do now with the vastly greater
computation resources that are now available.
The Thanksgiving Forecast 2001
42h forecast (valid Thu 10AM)
SLP and winds
1: cent
Verification
- Reveals high uncertainty in storm track and intensity
- Indicates low probability of Puget Sound wind event
2: eta
5: ngps
8: eta*
11: ngps*
3: ukmo
6: cmcg
9: ukmo*
12: cmcg*
4: tcwb
7: avn
10: tcwb*
13: avn*
Ensemble Prediction
•Can use ensembles to provide a new generation
of products that give the probabilities that some
weather feature will occur.
•Can also predict forecast skill.
•It appears that when forecasts are similar, forecast
skill is higher.
•When forecasts differ greatly, forecast skill is less.
Ensemble-Based Probabilistic Products
Ensemble Post-Processing
• To get the maximum benefits from
ensembles, post-processing is needed, such
as:
– Correction for systematic bias
– Optimal weighting of the various ensemble
members--e.g., Bayesian Model Averaging
The UW-MURI Project
• Possibility the most advanced
ensemble/postprocessing system in the world has
been developed at the UW
• Includes UW Atmospheric Sciences, Statistics,
Psychology, and Applied Physics Lab
• Remaining talks will describe some of the
research and development completed by this
effort.
Providing forecast uncertainty information is good….
But you can have too much of a good thing…
The END