WRFDA Overview

WRFDA Overview
Zhiquan (Jake) Liu

NCAR/MMM
WRFDA is a Data Assimilation system built within the WRF
software framework, used for application in both research and
operational environments.
WRFDA Tutorial August 2016
Outline
What is data assimilation
Scalar case
Two state variables case
General case
Introduction to WRF Data Assimilation
Outline
Scalar case
General case
What is data assimilation?

A statistical method to obtain the best
estimate of state variables
In the atmospheric sciences, DA involves
combining model forecast (prior) and
observations, along with their respective
errors characterization, to produce an
analysis (Posterior) that can initialize a
numerical weather prediction model (e.g.,
WRF)
Scalar Case
State variable to estimate x, e.g., consider
todays temperature of Boulder at 12 UTC.
Now we have a background (or prior)
information xb of x, which is from a 6-h GFS or
WRF forecast initiated from 06 UTC today.
We also have an observation y of x at a surface
station in Boulder
What is the best estimate (analysis) xa of x?
Scalar Case
1
x
=
We can simply average them: a 2 (xb + y)
This means we trust equally the background

and observation.
But if their accuracy is different and we

have some estimation of their errors
e.g., for background, we have statistics (e.g.,
mean and variance) of xb y from the past
For observation, we have instrument error
information from manufacturer
Scalar Case
Then we can do a weighted mean: xa = axb + by in
a least square sense, i.e.,
Minimize
J(x) =
dJ ( x )
Requires
dx
2
1 ( xxb )
2 b2
( xxb )
b2
Then we can easily get
xa =
o2
b2 + o2
xb +
2
1 ( xy)
2 o2
( xy)
o2
b2
b2 + o2
=0
We can also write in the form of analysis increment
xa xb =
b2
b2 + o2
(y xb )
Scalar Case
Minimize
J(x) =
2
(
xx
)
b
1
2 b2
2
(
xy)
1
2 o2
Is actually equivalent to maximize a

Gaussian PDF
ce
J (x)
Assume errors of Xb and y are unbiased

Background error: N(0, b=1.5)

Observation error: N(0, o=1)
Analysis error: N(0, a=0.832)
1
1
1
= 2+ 2
2
a b o
a is always smaller than
b and o.

Consider two state variables to estimate: Boulder
and Denvers temperatures x1 and x2 at 12 UTC
today.
Background from 6-h forecast: x1b and x2b
and their error covariance with correlation c
!
2
1
#
B=
# c 1 2
"
c 1 2
22
$
&
&
%
We only have an observation y1 at a Boulder

station and its error variance o2
10
Analysis increment for two variables

2
1
b
x x = 2
(y
x
1
1)
2
1 + o
c 1 2
a
b
b
x2 x2 = 2
(y
x
1
1)
2
1 + o
a
1
b
1
Unobserved variable x2 gets updated through the error

correlation c in the background error covariance.
This correlation can be correlation between two
locations (spatial), two variables (multivariate), or
two times (temporal).
11
General Case
Observations
yo, ~105-106
Model state
x, ~107
12
General Case: vector and matrix notation

state vector
observation vector
y
1
y2
y=
!

y
n
x1

x = x2
!

xm
background error covariance
!
2
1
#
# c12 1 2
B =#
#
#
"
$
c12 1 2 &
22
&
&
! &
m2 &%
Observation error covariance
!
#
#
R =#
#
#
"
o12
2
o2
!
0
$
0 &
0 &
&
" ! &
2 &
on
%
13
General Case: cost function
1
1
b T 1
b
J(x) = (x x ) B (x x ) + [Hx y]T R 1[Hx y]
2
2
H maps x to y space, e. g., interpolation.
Terminology in DA: observation operator
Minimize J(x) is equivalent to maximize a
multi-dimensional Gaussian PDF
Constant *
J (x)
14
General Case: analytical solution

Again, minimize J requires its gradient (a vector) with respect to x equal to zero:
J x (x) = B1 (x x b ) HT R1[ y Hx] = 0

This leads to analytical solution for the analysis increment:
a
x x = BH (HBH + R) [y Hx ]
HBHT : projection of background error covariance
in observation space
BHT : projection of background error covariance
in background-observation space
15
Precision of Analysis with optimal B and R

1
T 1
=
+
A B H R H
1
Generalization of scalar case
Or in another form:
1
1
1
= 2+ 2
2
a b o
A = (I KH)B
With
T
K = BH (HBH + R)
called Kalman gain matrix
16
Precision of analysis: more

general formulation
A = (I KH)Bt (I KH) + KR t K
T
where Bt and Rt are true background and observation

error covariances.
This formulation is valid for any given gain matrix K,
which could be suboptimal (e.g., due to incorrect
estimation/specification of B and R).
17
Analysis increment with a single humidity observation

x a x b = BH T (HBH T + R)1[y Hx b ]
clk l k
b
x x = 2
(yk xk )
2
k + ok
a
l
b
l
It is generalization of previous
two variables case:
12
b
x x = 2
(y
x
1
1)
2
1 + o
c 1 2
a
b
b
x2 x2 = 2
(y
x
1
1)
2
1 + o
a
1
b
1
cv_options=6 in WRFDA
18
Other Remarks
Observation operator can be non-linear and thus
analysis error PDF is not necessarily Gaussian
J(x) can have multiple local minima. Final solution of
least square depends on starting point of iteration, e.g.,
choose the background xb as the first guess.
19
Other Remarks
B matrix is of very large dimension, explicit inverse of B
is impossible, substantial efforts in data assimilation
were given to the estimation and modeling of B.
B shall be spatially-varied and time-evolving according
to weather regime.
Analysis can be sub-optimal if using inaccurate
estimate of B and R.
Could use non-Gaussian PDF
Thus not a least square cost function
Difficult (usually slow) to solve; could transform into Gaussian
problem via variable transform
20
Outline
Scalar case
General case
21
WRFDA is a Data Assimilation system

built within the WRF software framework
22
What WRFDA can do?
Provide Initial conditions for the WRF model forecast

Verification and validation via difference b.w. obs and model
Observing system design, monitoring and assessment
Reanalysis
Better understanding:
Data assimilation methods

Model errors
Data errors
23
DA algorithms currently available in WRFDA

3DVAR and FGAT
Different options for choice of control variables (e.g., Psi/Chi or U/
V) and background error covariance modeling (e.g., vertical EOF
or vertical recursive filter)
4DVAR
TL/Adjoint (i.e., WRFPlus code) of WRF up-to-date with WRF
Allow LBC control variable and Jc-DFI
Hybrid-3DEnVar
Can run in dual-resolution mode
Can ingest ensemble from global or regional sources
ETKF: for generating ensemble analysis
24
WRFDA Observations
In-Situ:
-
-
-
-
-
-
-
-
Bogus:
-
-
Remotely sensed retrievals:
SYNOP
-
METAR
-
SHIP
-
BUOY
-
TEMP
-
PIBAL
AIREP, AIREP humidity -
-
TAMDAR
TC bogus
Global bogus
Atmospheric Motion Vectors (geo/polar)

SATEM thickness
Ground-based GPS TPW or ZTD
SSM/I oceanic surface wind speed and TPW
Scatterometer oceanic surface winds
Wind Profiler
Radar data (reflectivity/retrieved rainwater, and radial-wind)
- Satellite temperature/humidity/thickness profiles
- GPS refractivity (e.g. COSMIC)
- Stage IV precipitation/rain rate data (4D-Var only)
Radiances:
HIRS
AMSU-A
AMSU-B
MHS
AIRS
SSMIS
IASI
ATMS
MWTS
MWHS
SEVIRI
AMSR2
NOAA-16, NOAA-17, NOAA-18, NOAA-19, METOP-A

NOAA-15, NOAA-16, NOAA-18, NOAA-19, EOS-Aqua, METOP-A, METOP-B
NOAA-15, NOAA-16, NOAA-17
NOAA-18, NOAA-19, METOP-A, METOP-B
EOS-Aqua
DMSP-16, DMSP-17, DMSP-18
WRFDA is flexible to allow assimilation
METOP-A, METOP-B
of different formats of observations:
Suomi-NPP
Little_r (ascii), HDF, Binary
FY-3
NOAA MADIS (netcdf),
FY-3
METEOSAT
GCOM-W1 (new in V3.8)
NCEP PrepBufr,
NCEP radiance bufr
25
WRFDA
Radiance Assimilation
Two RTM interfaces
RTTOV or CRTM
NOAA (HIRS, AMSU)
Variational Bias Correction

Modular code design to ease adding new
satellite sensors
Aqua (AMSU, AIRS)
Capability for cloudy radiance DA
DMSP(SSMI/S)
26
New: all-sky radiance DA: AMSR2

Channel
Frequency
(GHz)
Polarization
Footprint
(along scan*
along track)
1,2
6.925
V,H
35*61 km
3,4
7.3
V,H
35*61 km
5,6
10.65
V,H
24*41 km
7,8
18.7
V,H
13*22 km
9,10
23.8
V,H
15*26 km
11,12
36.5
V,H
7*12 km
13,14
89.0
V,H
3*5 km
27
28
3DVAR (Barker et al. 2004)

1
1
J(x) = (x x b ) T B1 (x x b ) + [H(x) y]T R 1[H(x) y]
2
2
29
the NWP model into the cost function, we obtain
4DVAR (Huang et al. 2009)
1
J(x0 ) = (x0
2
2.2
xb0 )T B 1 (x0
xb0 )
N
1X
+
[Hi (Mi (x0 ))
2 i=1
yi ]T Ri 1 [Hi (Mi (x0 ))
yi ] (1
Incremental 4DVAR Formulation
Linearization, let x0 = x0
1
J( x0 ) = ( x0
2
xg0 )T B 1 (
xg0 and xg0 = xb0

x0
1
xg0 )+
N
X
2 i=1
xg0 , thus x0 = x0 + xg0 , we have
[Hi (Mi ( x0 +xg0 ) yi ]T Ri 1 [Hi (Mi ( x0 +xg0 ))
(1
Do Taylor Expansion for observation term

1
J( x0 ) = ( x0
2
where di = yi
2.3
xg0 )T B 1 (
x0
xg0 ) +
N
1X
(Hi Mi x0
2 i=1
di )T Ri 1 (Hi Mi x0
di ) (1
Hi [Mi (xg0 )].
Control Variable Transform (CVT)
Again, control variable transform x0 = Uv and xg0 = Uvg . x0 indicates that analys
increment is valid at the beginning
of the 4DVAR time window. Then the cost30functio
x1a
x1f
x1a
x1a
x a2
x a2
H(x 2f ), o
.
.
.
.
.
.
.
.
.
x1f
H(x1f ), o
x a2
.
.
.
a
N
x 2f
.
.
.
x 2f
.
.
.
x Nf
x Nf
x aN
x aN
H(x Nf ), o
Hybrid
Var+ETKF
xf
a
xVar
yo
(Wang et al. 2008)

W
3/4D-Var
n
W T 1 1
T
J=
v v+
a A a + [ d i Hi Mi Uv ] R i1 [ d i Hi Mi Uv ]
2
2
2 i=0
b
31
Beijing Met Bureau - RUC
30/10/3.3/1.1-km
9km
3km
Antarctic Mesoscale Prediction SystemAMPS
15km
synop
+ metar
ship
buoy
* sound
gpsref
profiler
airep
quikscat
SatWind
3km
Taiwan Central Weather Bureau

Hybrid-3DEnVar
Arctic System Reanalysis
32
12km
12km
4km
4km
Operational domains by PWS (hybrid-3DEnVar)

Ensemble covariances from global GFS ensemble
15km
4km
12km
33
Radar DA for hydrological application

STEP Hydromet Real Time Exp. during spring time
15KM
3KM
The goal is to improve lo

cal-scale QPF in coupled
hydromet system
OBS
< 1 h rapid update

Radar radial velocity an
d reflectivity assimilation
High resolution vs. ense
mble
Impact of terrain
WRFDA
RADAR
Improved results in capt

uring localized storms
34
Real-Time WRF/WRFDA-hybrid analysis/forecast over CONUS
35
36
37
Real-time WRF/WRFDA forecast
38
Ongoing work: Variational Bias

Correction of Aircraft T
J ( x, ) = ( x x b )T x1 ( x x b ) + ( b ) T 1 ( b ) + ( y H [x, y, ]) T R 1 ( y H [ x, y, ])
H%(x, y, ) = H (x) b( y, )
0+ + 1+ w if w > 0
b( y, ) = n pn ( y ) =
n =0
0 + 1 w if w < 0
dp
w = , is updated in cost function each cycle and written in parameter table.
dt
N
39
Impact of Aircraft T VarBC on rainfall forecast
Valid at 0000 UTC 14 Sep, 2016

TS
0.8
0.7
0.6
CNT
TS
0.5
VarBC
0.4
0.3
0.2
0.1
0
0.05
0.1
0.25
0.5
1.25 Tutorial
1.5 August
1.75 2016 2
WRFDA
Precipitation (inch)
2.5
3.5
40
GEOS imager radiance DA at convection-permitting

scale (4km, hourly-cycling, hybrid-3DVAR)
obs
No-GOES DA
GOES-DA
41
Other ongoing work

Implemented Hybrid-4DEnVAR
Improving computing efficiency
Continue developing Multi-Resolution

Incremental 4DVAR (MRI-4DVAR)
Continue developing cloudy radiance/product DA
Improving surface data assimilation
Improving radar DA
Adding divergence constraint
Assimilation of non-rain data
WRFPlus-Chem & WRFDA-Chem

CU Boulder.
42
Last Remarks
We welcome contributions from external users/
developers.
Contact wrfhelp@ucar.edu or directly email to me
liuz@ucar.edu for contributing back your code
We maintain a WRFDA-related publications list,

please inform us your papers to be included
http://www2.mmm.ucar.edu/wrf/users/wrfda/
publications.html
43

WRFDA Overview

Diunggah oleh

Informasi Dokumen

Hak Cipta

Format Tersedia

Bagikan dokumen Ini

Bagikan atau Tanam Dokumen

Opsi Berbagi

Apakah menurut Anda dokumen ini bermanfaat?

Apakah konten ini tidak pantas?

Hak Cipta:

Format Tersedia

WRFDA Overview

Diunggah oleh

Hak Cipta:

Format Tersedia

WRFDA Overview

Zhiquan (Jake) Liu

WRFDA Tutorial August 2016

Introduction to WRF Data Assimilation

WRFDA Tutorial August 2016

Introduction to WRF Data Assimilation

WRFDA Tutorial August 2016

What is data assimilation?

This means we trust equally the background

But if their accuracy is different and we

Then we can easily get

We can also write in the form of analysis increment

WRFDA Tutorial August 2016

Is actually equivalent to maximize a

Assume errors of Xb and y are unbiased

Background error: N(0, b=1.5)

WRFDA Tutorial August 2016

Two state variables case

We only have an observation y1 at a Boulder

Analysis increment for two variables

Unobserved variable x2 gets updated through the error

WRFDA Tutorial August 2016

General Case: vector and matrix notation

Observation error covariance

WRFDA Tutorial August 2016

General Case: cost function

WRFDA Tutorial August 2016

General Case: analytical solution

J x (x) = B1 (x x b ) HT R1[ y Hx] = 0

Precision of Analysis with optimal B and R

Generalization of scalar case

Precision of analysis: more

where Bt and Rt are true background and observation

WRFDA Tutorial August 2016

Analysis increment with a single humidity observation

WRFDA Tutorial August 2016

WRFDA Tutorial August 2016

Introduction to WRF Data Assimilation

WRFDA Tutorial August 2016

WRFDA is a Data Assimilation system

What WRFDA can do?

Provide Initial conditions for the WRF model forecast

Data assimilation methods

WRFDA Tutorial August 2016

DA algorithms currently available in WRFDA

ETKF: for generating ensemble analysis

WRFDA Tutorial August 2016

Remotely sensed retrievals:

Atmospheric Motion Vectors (geo/polar)

NOAA-16, NOAA-17, NOAA-18, NOAA-19, METOP-A

WRFDA Tutorial August 2016

NOAA (HIRS, AMSU)

Variational Bias Correction

Capability for cloudy radiance DA

WRFDA Tutorial August 2016

New: all-sky radiance DA: AMSR2

WRFDA Tutorial August 2016

WRFDA Tutorial August 2016

3DVAR (Barker et al. 2004)

WRFDA Tutorial August 2016

the NWP model into the cost function, we obtain

4DVAR (Huang et al. 2009)

yi ]T Ri 1 [Hi (Mi (x0 ))

Incremental 4DVAR Formulation