Post on 19-Dec-2015
Description and Preliminary Evaluation of the Expanded UW
Short Range Ensemble Forecast System
Maj. Tony Eckel, USAF
University of Washington Atmospheric Sciences Department
Advisor: Prof. Cliff Mass
March 2002
Overview
• UW SREF Methodology
• Generating New Initial Conditions
• Case Study• Error Statistics• Individual SREF Members• Ensemble-Based Probability
• Summary
- Construct the initial state of the atmosphere with multiple, equally likely analyses, or initial conditions (ICs)
Ensemble Forecasting Theory
0
1
2
3
4
5
6
7
8
1 3 5 7 9 11 13 15 17 19
Fre
quen
cy
Initial State0 5 10 15 20
0.2
0.4
.5
7.6946e-023
dnorm ( ),,x 10 1
200 x
- From our point of view, truth is random sample from the pdf
0 5 10 15 20
0.2
0.4
.5
0.000514093
dnorm ( ),,x 10 3
200 x
0 5 10 15 20
0.2
0.4
.5
7.4336e-007
dnorm ( ),,x 10 2
200 x
0
1
2
3
4
1 3 5 7 9 11 13 15 17 19
0
1
2
3
4
5
1 3 5 7 9 11 13 15 17 19
24hr Forecast State 48hr Forecast State
- Let all ICs evolve to build PDF at future time (i.e., a forecast pdf)
- Error growth spreads out PDF as forecast lead time increases
0
1
2
3
4
1 3 5 7 9 11 13 15 17 190
1
2
3
4
5
1 3 5 7 9 11 13 15 17 190
1
2
3
4
5
6
7
8
1 3 5 7 9 11 13 15 17 19
Difficult to consistently construct the “correct” analysis/forecast pdf.Errors in mean and spread result from:
1) Model error
2) Choice of ICs
3) Under sampling due to limits of computer processing
Result: EF products don’t always perform the way they should. (especially a problem for SREF)
Limitations of EFF
requ
ency
Initial State0 5 10 15 20
0.2
0.4
.5
0.000514093
dnorm ( ),,x 10 3
200 x
0 5 10 15 20
0.2
0.4
.5
7.4336e-007
dnorm ( ),,x 10 2
200 x24hr Forecast State 48hr Forecast State0 5 10 15 20
0.2
0.4
.5
7.6946e-023
dnorm ( ),,x 10 1
200 x
analysis pdf
ensemble
phasespace
T
48hr forecast state (core)
48hr true state
Analysis pdf :
Forecast pdf :
7 “independent” atmospheric analyses, the centroid, plus 7 “mirrored” ICs15 divergent, “equally likely” solutions using the same primitive equation model, MM5
Forecast pdf
48hr forecast state (perturbation)
ngp
uk
eta
cmc
gsp
avn
Analysis pdfcwb
Cngp
eta
cmc
avn
gsp
cwb
uk
UW SREF Methodology Overview
A point in phase space completely describes an instantaneous state of the atmosphere.For a model, a point is the vector of values for all parameters (pres, temp, etc.) at all grid points at one time.
cmcg*
Filling in the Holes of the IC Cloud
STEP 1: Calculate best guess for truth (the centroid) by averaging all analyses.
STEP 2: Find error vector in model phase space between one analysis and the centroid by differencing all state variables over all grid points.
STEP 3: Make a new IC by mirroring that error about the centroid.
cmcgC cmcg*
Sea
Lev
el P
ress
ure
(mb)
~1000 km
1006
1004
1002
1000
998
996
994
cent
170°W 165°W 160°W 155°W 150°W 145°W 140°W 135°W
eta
ngps
tcwbgasp
avn
ukmo
cmcg
ICs: Analyses, Centroid, and Mirrors
Strengths• Good representation of analysis error
• Perturbations to synoptic scale disturbances• Magnitude of perturbation(s) set by spread among analyses
• Bigger spread Bigger perturbations• Dynamically conditioned ICs• Computationally affordable
Weaknesses• Limited by number and quality of available analyses
• May miss key features of analysis error• Analyses must be independent (i.e., dissimilar biases)• Calibration difficult; no stability since analyses may change techniques
CASE STUDY: Thanksgiving Day Non-Wind Event
eta-MM5 Initialized: 00z, 21 Nov 2001 (Tuesday evening)
39h, valid 22 Nov 15z 42h, valid 22 Nov 18z
(Thursday 7AM) (Thursday 10AM)
Note: This study used only 13 ensemble members since missing gasp grids.
ZCZC SEANPWSEAWWUS45 KSEA 212348 URGENT - WEATHER MESSAGENATIONAL WEATHER SERVICE SEATTLE WA344 PM PST WED NOV 21 2001 AN INTENSE LOW PRESSURE SYSTEM WILL MOVE ALONG THE NORTH WASHINGTON COAST EARLY MORNING THANKSGIVING DAY...AND MOVE INLAND OVER THE NORTH INTERIOR OF WESTERN WASHINGTON BY MIDDAY. STRONG SOUTH WINDS WILL DEVELOP ALONG THE COAST AFTER MIDNIGHT. AS THIS SYSTEM MOVES INLAND THANKSGIVING MORNING IT HAS THE POTENTIAL TO CAUSE HIGH WINDS ACROSS THE INTERIOR OF WESTERN WASHINGTON AFTER ABOUT 8 AM.
...HIGH WIND WATCH FOR THURSDAY MORNING THROUGH THURSDAY EVENINGREMAINS IN EFFECT...
Root Mean Square Error (RMSE) by Model VerificationR
MS
E o
f M
SL
P (
mb
)
36kmOuter
Domain
cmcg
cmcg
*av
nav
n*
eta
eta*
ng
ps
ng
ps*
ukm
ou
kmo
*tc
wb
tcw
b*
cen
t
12h
24h36h48h
12kmInner
Domain
cmcg
cmcg
*av
nav
n*
eta
eta*
ng
ps
ng
ps*
ukm
ou
kmo
*tc
wb
tcw
b*
cen
t
00h cmcg 21 Nov 00z 00h cmcg* 21 Nov 00z
00h cent 21 Nov 00z
Initialization
cmcg
C
cmcg
*
36h cmcg 22 Nov 12z 36h cmcg* 22 Nov 12z
36h cent 22 Nov 12z
36 Hour Forecast
00h cent 22 Nov 12z
“Verification”
42h (18z, 22 Nov) mslp and sfc wind.
avn*
ngps*
cmcg*
tcwb*
ukmo*
eta*
“Verification”
cent
avn
ngps
cmcg
tcwb
ukmo
eta
Ensemble-Based Probability of Wind Speed
Prob. of (sustained) Winds > 21 kt
39h, 22 Nov 15z 42h, 22 Nov 18z 39h, 22 Nov 21z
(Thursday 7AM) (Thursday 1PM) (Thursday 10AM)
10%30%50%10%
90%
Summary
• Set of 15 ICs for UW SREF are not optimal, but may be good enough to represent important features of analysis error
• The centroid may be the best bet deterministic model run, in the big picture
• Need further evaluation• How often does the ensemble fail to capture the truth?• How reliable are the probabilities?• Does the ensemble dispersion represent forecast uncertainty?