Real-time Ensemble Display and Analysis System: An Academic`s

Transcription

Real-time Ensemble Display and Analysis System: An Academic`s
Real-time Ensemble Display and Analysis
System: An Academic’s Vision
Steven Mullen
Basics Overview
http://www.raytheon.com/
• Wide range of user
configurable functions
• Reliable under heavy
loads
• Intuitive graphics
• Simple to add new code
• Fast data access
Where we are: ECMWF
:50
www.ecmwf.int
Where we are: TIGGE
:250
www.ecmwf.int
Where we will be in 2020?
:1000
www.ecmwf.int
Must reduce dimensionality
Well know example of reducing the
dimensionality of the ensemble
1
2
N1
N2
Clusters
Ensemble Mean
500 mb Hght
3
4
N3
N4
Ensemble Mean
SFC Temp
Member
Number
Active
Links
Active
Links
ButDefine
ensemble
Ensemble
meanFlow
lacksRegimes
specificity
Specified Algorithm & Distance Metric
Specified Max/Min # Regimes
Specified Temporal Interval
http://www.meteo.psu.edu/ewall/ewall.html
1
2
N1
N2
Regime
Members
3
4
N3
N4
Cluster
Members
Cluster
or Members
Sub-Clusters
http://www.meteo.psu.edu/ewall/ewall.html
1
2
N1
N2
Regime
Members
3
4
N3
N4
Each panel has
mouse over zoom
and/or animation
http://www.meteo.psu.edu/ewall/ewall.html
1
2
N1
N2
Regime
Members
Each grid point
has point-n-click
meteogram or
time-hght section
3
4
N3
N4
http://www.meteo.psu.edu/ewall/ewall.html
1
2
N1
N2
Regime
Members
3
4
N3
N4
Click-n-hold gives
cross section
along the line
http://www.meteo.psu.edu/ewall/ewall.html
Q
1
2
N1
N2
Multivariate
Grouping
3
4
N3
N4
Cluster Members
Sibling Clusters
or Sub-Clusters
http://www.meteo.psu.edu/ewall/ewall.html
Multiple Options
• Need to have multiple clustering algorithms
and distance/similarity metrics
Ensemble Flow Grouping
(Atger 1999)
Ensemble Flow Grouping
Circled Pattern Not
Identified by Ward’s
Need Multiple Algorithms
and Distance Metrics too
(Atger 1999)
• Some comments about the ubiquitous
spaghetti chart
Typical Spaghetti Chart
Adequate for 540/546 dm thickness line
Allows for “visual clustering” of members
Provides some info on spatial correlations
Can be misleading in areas of weak gradients
Contour probability maybe the most informative of spaghetti family,
but…..ensemble mean is absent and users have no choice on contour to display
“Plume” plots for user selected
locations and thickness values
(Note poor selection of vertical range)
0
1
2
3
4
5
6
7
8
9
10
• Lets briefly consider error growth and
predictability estimates
FNMOC Ensemble Forecast System Basics
Means, Anomalies, Spread, Normalized Spread
25
• Suggestion: Make available reanalysis and
Reforecast fields in the format
summary statistics minimum
some subset of full fields.
Diagnostic Functions
• Envelopes and Wave Packets
Wave Packet diagnostics (Zimin et al. 2003)
See Zimin et al. (2006) for flow-dependent refinement
Wave Packet
Diagnostics
Space-Time Envelopes
Filtering to estimate
wave packets
Especially useful at
extended ranges…
After transient
synoptics have lost
deterministic skill
Zimin et al (2006)
Alternative Perspectives
Beyond Eulerian perspective
Lagrangian diagnostics
Quasi-lagrangian feature tracking
Quasi-Lagrangian Feature Tracking and
Verification of User Defined Patterns
MET tools feature tracking
and verification
Troughs, cyclones, frontal
discontinuities, precipitation
clusters, PV anomalies
(Morgan and Neilsen-Gammon 1998)
Lagrangian-Dispersion Tracking
• Trajectories
–
–
–
–
–
–
–
–
Air mass history
Sounding construction
Diabatic heating inference
Air pollution
Dispersion
Visibility
Forecast verification
Lagrangian predictability
Potential Vorticity Diagnostics
• Conservation of Ertel PV
Non conservation implies
diabatic processes
• Balance Constraint
• Invertibility Principle
• Piecewise PV Inversion
(Davis & Emanuel 1991)
– Provides a method to modify
NWP initial conditions
– Add, Remove or Modify
Circulation features
http://www.meteo.mcgill.ca/atoc541/index_files/potentialvorticity.ppt
Piecewise PV Inversion
and Modified Initial
Conditions
• McIntyre Vision “Even in the year 2020,
there are severe
limitations on the size
of the ensemble of
initial conditions … So
although a basic
ensemble of a few
thousand members is
always run...”
(McTaggart-Cowan et al. 2001)
Piecewise PV Inversion
and Modified Initial
Conditions
• McIntyre Vision “…there is also
provision for…forecast
runs, based on the
forecaster’s subjective
assessment of the most
sensitive locations for
varying the initial
conditions.”
(McTaggart-Cowan et al. 2001)
Piecewise PV Inversion
and Modified Initial
Conditions
Suspect analysis
• McIntyre Vision “The basic mode of
operation is…visual
inspection and
manipulation of
…thermodynamical and
dynamical fields…to
facilitate … rapid repairs
to the model (potential
vorticity) fields.”
Z on 1.5 PVU
Altered analysis
Santurette and Georgiev 2005
Z on 1.5 PVU
Perpetual Human
Role in the
Forecasting Process
Original forecast
• Tennekes comment “…only human
professionals can bare
the responsibility for life
and property in
emergencies.”
Altered forecast
24 h precipitation
Ensemble Synoptic
Analysis
-Provide guidance for
assessing “sensitivity”
points for initial
conditions during the
linear regime without
the need for adjoint
modeling
Hakim and Torn (2008)
Suggestions for other diagnostic packages
Think weather-climate connection
Think general circulation statistics
Consider one point teleconnection plots
http://www.nws.noaa.gov/om/csd/pds/
Whitaker and Sardeshmukh 1998
Intra-Ensemble “Teleconnections” Mapping
Size of Sign => Correlation Size
Intra-Ensemble “Teleconnections” Mapping
Size of Sign => Correlation Size
Intra-Ensemble “Teleconnections” Mapping
Size of Sign => Correlation Size
Atmospheric
Teleconnection
Patterns
Projections onto leading
teleconnection patterns at
extended ranges could …
-help assess the “stability”
of the forecast pattern,
-alter confidence in the
ensemble,
-enhance understanding
of model tendencies
-aid model verification.
PNA Pattern Negative Phase
Projection onto Atmospheric Teleconnection Patterns
Negative phase of PNA? Sort of? Not Really?
Functions: Shopping List
• User control of fields, region, time interval for output,
display and interrogation
• Input for output models
• Reduce dimensionality and multiple similarity options
• Categorization => clustering, rotated EOF’s, “Tubes”
• Complex EOF’s (Branstator 1987)
• Covariance/correlation/teleconnection projections
• Normalized variance=> loss of skill; nonlinear saturation
• Wave packets – envelopes
• Feature tracking => cyclones, MCS’s, boundaries, PV anomalies
• Lagrangian analysis => trajectories, dispersion, PV fields and
piecewise inversion
Model diagnosis; McIntyre’s (1976, 1988, 1996) vision
• Goals
– Deeper understanding of the physical connection between
observed and model variability, and error
growth/ensemble spread
– Better assessment of the unpredictable (in a rmse sense)
scales and their relation to the predictable features
Nonlinear Saturation
Limit of
Deterministic
Skill
Sweet Spot for
Ensemble Forecasts
Under Dispersion Typifies NWP Models
Where We Are
Where We Want to Be?
Where We Don’t Want to Be?