Predictability of High Impact
Weather during the Cool Season:
CSTAR Update
and the Development of a New
Ensemble Sensitivity Tool for the
Forecaster
Brian A. Colle, Edmund Chang,
and Minghua Zheng
Stony Brook University - SUNY
CSTAR Scientific Motivation
• Improve the understanding of high
•
•
•
impact weather predictability during
the cool season through objective
verification of cyclones and Rossby
wave packets (RWPs) in ensembles.
Integrate RWPs more in operations:
Better understanding of RWP
climatology, downstream impact of
targeted observations, and linkage
with extreme weather.
Better understanding of the
predictability of cyclones and some
mesoscale phenomena (e.g.,
snowbands).
Better ensemble construction and
post-processing.
From THORPEX International Science Plan
(Shapiro and Thorpe, 2004)
Forecasts from 24 Dec 2010
Courtesy: Dan Petersen
Forecasts from 25 Dec 2010
Courtesy: Dan Petersen
NCEP SREF Cyclone
Tracks
GDAS analysis 500 mb height/absolute vorticity 00z 24 Dec 2010
GDAS analysis 500 mb height/absolute vorticity 12z 24 Dec 2010
GDAS analysis 500 mb height/absolute vorticity 00z 25 Dec 2010
GDAS analysis 500 mb height/absolute vorticity 12z 25 Dec 2010
GDAS analysis 500 mb height/absolute vorticity 00z 26 Dec 2010
GDAS analysis 500 mb height/absolute vorticity 12z 26 Dec 2010
GDAS analysis 500 mb height/absolute vorticity 00z 27 Dec 2010
500 hPa difference between 06 UTC 24 Dec and
12 UTC 24 Dec runs starting at 18Z 24 Dec
From Richard Grumm
Objective: Can ensemble sensitivity analysis
(Torn and Hakim 2008; 2009) be used to enhance
forecaster awareness on how upstream uncertainty is
effecting a region of interest?
* Ensemble sensitivity is a correlation between a
forecast metric at the final forecast time and any
variable within the model state vector. It makes use
of the different evolution of the forecasts among the
different ensemble members to derive the sensitivity.
* For CSTAR: Want to use metrics useful to the
forecaster: cyclone location and intensity, or why a
shift in cyclone forecast between two model runs?
* Test for 26-27 Dec nor-easter and hurricane Irene
Initial time 12 Z 23 Dec, 84 hour ECMWF ensemble forecast
Ensemble mean
Dec 27 2010
00Z
(84-h fcst)
Variance
Sensitivity of EOF 1 (cyclone intensity) to 500
hPa height
0-lag
Projection of a pattern:
“sensitivity” carries the
dimension of the forecast metric
and describes the change in the
forecast metric corresponding to
one standard deviation change
in the initial condition
uncertainties in the selected
variable.
Pattern: pi
Ensemble member anomaly: xi
Projection of Pattern onto Ensemble member:
S pi xi
Basically value of projection is large when the
cov(J , xi )
" Sensitivity" 
 cor( J , xi )  var(J )
anomaly of the ensemble member resembles the
var(xi )
pattern
J is any forecast metric at the final forecast time
- xi is any variable within the model state vector
(500 hPa height in this example).
Here, J is the principal component (PC) of EOF.
PCs are the projections of the dominant EOF
patterns on each of the ensemble members.
EOF1 positive phase: weaker cyclone
Sensitivity of EOF 2 (SW-NE cyclone shift) to 500 hPa height
0-lag
00z 27 Dec
EOF2 positive phase:
southwest shift of cyclone
T-48 hr
(25/00Z)
Sensitivity
EOF1
EOF2
Why the shift in cyclone position between 24/00Z and 25/00z run cycles?
Using 50-member ECMWF, difference between MSLP ensemble mean
forecast at 12z Dec 27 2010. Obtain pattern within red box.
Initial time 2010Dec2500Z (60hr) – 2010Dec2400Z (84hr)
Here, J is the
projection of this
“shift” pattern
onto each
ensemble member
at final forecast
time (Dec27 12Z)
Ensemble sensitivity calculations
forecast valid time (Dec27 12Z)
positive areas: increasing z at that location is associated with enhancing
the corresponding EOF pattern
negative areas: decreasing z at that location is associated with enhancing
the corresponding EOF pattern
ECMWF (50 members): evolution of forecast difference between
DEC25Z00 and DEC24Z00
Valid DEC2712Z (0 hr)
MSLP (hPa)
Z500 (dm)
DEC25
00Z
* Results from sensitivity analyses suggest that:
– Cyclone E-W location at -0 hr (Dec 27 12Z) is
sensitive to conditions near the Gulf coast and
over northern Canada at -48 to -60 hr
– Forecast errors from Dec 24 00Z forecast
(comparing to Dec 25 00Z analysis and
forecast) also show errors over those sensitive
regions
– Can we assess the impacts of those short-range
(24-hr) forecast errors?
– Here, we introduce an alternative way of
ensemble sensitivity analysis to examine the
“forward” impact of forecast errors at early
time on forecasts at later time
Region near gulf coast at -60h
Z500 difference (m) between Dec2500Z analysis and the 24-h
ensemble mean forecast from Dec2400Z
Regression based on 500Z projection
in black boxed region using forecast
ensemble from Dec 24 00Z
Note:
Regression
based on
projection of
the pattern on
each
ensemble
member can
be considered
to be weighted
mean of
ensemble
members that
have
anomalies
resembling the
chosen pattern
MSLP
(unit mb)
Difference between Dec 25
and Dec 24 forecasts
Hurricane Irene
EOF 1: a deepening
of cyclone; explains
50.5% variance
Target time: 2011082800
(5-day forecast)
Target region: LON is from
278E to 294E; LAT is from
28N to 42N
EOF 2: slower moving
Cyclone; explains 24.1
% variance
EOF 3: western shifting
of cyclone; explains
11.0% variance
Ensemble mean SLP and variance
Sensitivity of PC1 (deeper cyclone)
to 500 hPa height for the target
time of 20110828/00Z
Sensitivity of PC2 (slower cyclone)
to 500 hPa height for the target
time of 20110828/00Z
Sensitivity of PC3 (western cyclone
shift) to 500 hPa height for the
target time of 20110828/00Z
Real-time Sensitivity Calculations
yyyymmddh
h
MAIN_CSH
GET_BIN
OUTPUT
SENSITIVI
TY
VARIANCE
EOF
• Code can run on any workstation (just need GNU Fortran and GrADS)
• Currently setting up at NCEP (so ECMWF ensemble can be combined with
NCEP)
Real-time Sensitivity Calculations
Default:
• Target time: 6 days forecast initialized either on
00hr or 12hr
• Target region: longitude is from 95W to 65W, and
latitude is from 30N to 50N
Optional run:
• target time: 0.5 days forecast to 6 days forecast
with an increment of 12h
• target region: Forecaster Defined
Ensemble Members:
• NCEP+CMC combined 40 members ensemble;
however. it can also compute sensitivity based on
individual NCEP (20 members) or CMC (20members)
ensemble. Will be setup at NCEP to include ECMWF
ensemble.
One Approach
1. There is a large spread is the forecast cyclone position
in the ensembles. Forecaster asks: Why are some
members closer to the coast and deeper than other
members?
2. Forecaster defines his/her sensitivity box around the
forecast cyclone in question (forecast hour of interest.
3. Using sensitivity analysis, Forecaster can get some
idea of the upstream source of the cyclone spread
(Pacific Rossby wave packet in medium range?; shortwave from more data sparse Canada?, …?)
4. If the next model cycle, the storm is closer to the
coast, forecaster can confirm whether it was because of
changes in upstream flow that sensi analysis suggested.
Summary
• Ensemble sensitivity approach has been constructed using
•
metrics useful for the forecaster (e.g., cyclone properties).
Can be expanded to other features (300Z jet, snowband
position, etc…).
The ensemble sensitivity will be run at SBU or NCEP in realtime with results on the SBU CSTAR page:
http://dendrite.somas.stonybrook.edu/CSTAR/cstar.html
• Results from 26-27 Dec 2010 sensitivity analyses suggest
that:
– Cyclone location at Dec 27 12Z sensitive to prior conditions (at -48hr to -60hr)
near the gulf coast as well as over northern Canada west of Hudson Bay
– Short range (24-hr) forecast errors over those locations from forecasts based
on Dec 24 00Z apparently led to cyclone being forecast too far east
– Hypothesis: Forecasts based on Dec 25 00Z are better because of
observations over those regions? Need to confirm with obs denial studies.
• Results from hurricane Irene suggest:
-- The intensity and speed of Irene has sensitivity patterns suggesting relatively
large regional influences from the cyclone itself.
-- The western shift of the cyclone for day 5 h suggests some influences from
the northern Pacific and downstream development.
Download

Predictability of High Impact Weather during the Cool Season: CSTAR