Allgemein

1
Status Report PP KENDA
Christoph SchraffDeutscher Wetterdienst,
Offenbach, Germany
Contributions / input by Hendrik Reich, Andreas
Rhodin, Annika Schomburg, Ulrich Blahak, Yuefei
Zeng, Roland Potthast Yuefei Zeng, Klaus Stephan,
Africa Perianez, Michael Bender (DWD) Chiara
Marsigli, Tiziana Paccagnella (ARPA-SIM) Lucio
Torrisi (CNMCA) Daniel Leuenberger, Luca Weber
(MeteoSwiss) Mikhail Tsyrulnikov, Igor Mamay
(HMC) Amalia Iriza (NMA)

• general overview
• assimilation of SEVIRI-derived cloud top height
  in LETKF

2
LETKF implementation

• experiment chain
• in NUMEX set up
• GME LETKF exp. (Nens 40)
• June 2011 for lateral BC
• somewhat too little spread
• particularly over Europe,
• but LBC spread at least as
• large as from COSMO-SREPS (Nens 12)

ana. spread
ana. rmse

• for COSMO-LETKF .
• lateral BC by direct interpolation
• from 60 km to 2.8 km
• (moderate noise at model top
• surface, acceptable)

• MCH / ARPA-SIM
• resolution gap IFS-EPS
• (32 km to 2.8 km) tested ok

3
LETKF implementation

• experiment chain in NUMEX set up
• lateral BC by direct interpolation from 60 km to
  2.8 km
• KENDA
• 1-hourly cycling, radiosonde, aircraft, wind
  profiler, synop 40 ens. members
• assimilation only, optimally takes 1 real day
  for 1 day of assimilation,
• but in fact 1 4 real months for 1 week
  of assimilation !
• (without forecasts !!)
• ? only 3 experiments so far

• Hendrik new flexible stand-alone scripts to run
  LETKF experiments
• without using NUMEX / archive ? very
  limited disk space
• ? 1 real day for 1 day of LETKF assimilation
• to do implement evaluation / verification tools
  in script suite
• may become very suitable tool for users outside
  DWD (academia)

• offline adaptive estimation of obs errors in
  observation space
• multi-step analysis approach (different
  localization radii for different sets of obs)

4
LETKF ensemble forecasts (20 ens. members)
4 12 June 2011
KENDA / COSMO-DE-EPS
precipitation
max. 10-m wind gusts
2-m temperature
RMSE
spread
? results of LETKF (without explicit surface
/ soil perturbations)
larger rmse, larger spread
larger rmse, initially larger spread
larger rmse, equal spread
5
LETKF at MCH, compare det. LETKF analysis with
nudging

• COSMO-2, Nens 40, LETKF as at DWD
• lateral BC from IFS (det./EPS) soil moisture
  from COSMO-2

• LETKF technically works
• comparison with surface observations suggest
• generally LETKF better than NO_OBS, but worse
  than nudging
• assimilation of surface pressure appears to work
  particularly well
• LETKF_DET analysis very close to LETKF ensemble
  mean
• SPPT has only small but positive impact

6
LETKF implementation of verification

• production of full NetCDF feedback files
• done COSMO observation operators (conventional
  obs)
• integrated in 3DVAR package
• to be done (this autumn !) extend flow
  control (read correct Grib files etc.)
• ensemble-related diagnostic verification tool,
  using feedback files
• (Iriza, NMA)
• ensemble scores implemented, further testing
  required

7
accounting for model error

• stochastic perturbation of physics tendencies
  (SPPT)
• (Torrisi)
• implemented in (private) V4_26
• tests at CNMCA / MCH / ARPA-SIM (WG7)

• Pattern Generator (for random fields with
  prescribed correlation scales)
• (Tsyrulnikov et al.)
• based on a stochastic partial differential
  equation approach
• basic version developed, being revised to make it
  efficient
• being embedded in COSMO code
• 2-D version planned for next year

8
high-resolution obs

• radar
• obs operators finished, assimilation works
  technically
• radial winds vr in LETKF Yuefei Zeng (DWD,
  until summer 2014)
• ? need to test thinning / superobbing
  strategies
• ? 3-hour assimilation with 1-hrly cycle done
  (different localization radii)
• vr reflectivity Z in LETKF Theresa Bick
  (HErZ-I Bonn, until end 2014 at least)

• GPS slant path delay
• obs operator (incl. TL / adjoint) implemented in
  3DVar, approximations tested
• implementation in COSMO should start soon

9
high-resolution obs

• (SEVIRI-based, radiosonde-corrected) cloud top
  height see next slides
• (Schomburg)

• direct assimilation of SEVIRI radiances (window
  channels for cloud info)
• (Perianez)
• technically implemented (obs operator (RTTOV),
  reading / writing)
• work on monitoring / assimilation start in Nov.

• new task microwave radiometer Raman lidar
  T- , q- profiles
• (Haefele, MCH)

10
use of (SEVIRI-based) cloud top height (CTH)
observations in LETKF method
if cloud observed with cloud top height CTHobs
, what is the appropriate type of obs increment ?
Z km

• avoid too strong penalizing of members with high
  humidity
• but no cloud
• avoid strong penalizing of members which are dry
  at CTHobs but have a cloud or even only high
  humidity close to CTHobs
• ? search in a vertical range ?hmax around
  CTHobs for
• a best fitting model level k, i.e. with
  minimum distance d

model profile
Cloud top
CTHobs
(if above a layer with cloud fraction gt 70 ,
then choose top of that layer)

• use f (RHobs1) f (RHk)
• and CTHobs hk
• as 2 separate obs increments in LETKF

RH
11
use of (SEVIRI-based) cloud top height (CTH)
observations in LETKF method
Z km
type of obs increment , if no cloud observed ?
9

• assimilate cloud fraction CLCobs 0
• separately
• for high, medium, low clouds
• model equivalent
• maximum CLC within vertical range

no high cloud
6
model profile
no medium cloud
3
no low cloud
CLC
12
CTH single-observation experiments

• 1 analysis step , 17 Nov. 2011, 6 UTC
  (wintertime low stratus)

• example missed cloud event

vertical profiles
relative humidity cloud cover cloud water cloud
ice observed cloud top
3 lines on one colour indicate ensemble mean and
mean /- spread
13
CTH single-observation experiments

• example missed cloud event

cross section of analysis increments for ensemble
mean
specific water content g/kg
observation location
relative humidity
observed cloud top
14
CTH single-observation experiments

• example missed cloud event

temperature profile (mean /- spread)
3000 m
first guess
analysis
2000 m
observed cloud top
1000 m
270 K
280 K
290 K
270 K
280 K
290 K

• LETKF introduces inversion due to RH(CTH) ? T
  cross correlations
• in first guess ensemble perturbations

15
CTH single-observation experiments

• example false alarm cloud ? assimilated
  quantity cloud fraction ( 0)

vertical profiles
relative humidity cloud cover cloud water cloud
ice observed cloud top
3 lines on one colour indicate ensemble mean and
mean /- spread
16
CTH single-observation experiments

• example false alarm cloud ? assimilated
  quantity cloud fraction ( 0)

observation increments - histogram over
ensemble members
low cloud cover octas
cover
17
cycled assimilation of dense CTH obs
1-hourly cycle over 21 hours, 13 Nov., 21 UTC
14 Nov. 2011, 18 UTC (wintertime low stratus)
observed cloud top height (CTH)

000 UTC
600 UTC
1200 UTC
1700 UTC
18
cycled assimilation of dense CTH obs LETKF setup

• thinning use obs at every 5th grid pt.
• adaptive covariance inflation, adaptive
  localisation scale ( ? 35 km)
• Observation error variances relative humidity
  10
• cloud cover 3.2 octa
• cloud top height m ?

000 UTC
600 UTC
1200 UTC
1700 UTC
19
cycled assimilation of dense CTH obs
time series of first guess errors of ensemble
mean / spread of ensemble
averaged over cloudy obs locations
RMSE

• underdispersive,
• but no trend
• for reduction
• of spread

spread
averaged over cloud-free obs locations
20
cycled assimilation of dense CTH obs
time series of first guess errors of RH at
observed CTH (det. run), averaged over cloudy
obs locations
no assimilation with cloud assimilation
RMSE
bias

• CTH assimilation reduces RH (1-hour
  forecast) errors

21
cycled assimilation of dense CTH obs
time series of first guess errors of RH at
observed CTH (det. run), averaged over cloudy
obs locations
no assimilation with cloud assimilation
assimilation of conventional obs only
assimilation of conventional cloud obs
localization scale adaptive / 20 km
RMSE
bias

• CTH assimilation reduces RH (1-hour
  forecast) errors

22
cycled assimilation of dense CTH obs
time series of first guess errors, averaged over
cloud-free obs locations (errors are due to
false alarm cloud)
mean square error of cloud fraction octas

• error reduced
• (almost) everywhere

23
cycled assimilation of dense CTH obs
cloud assimilation
satellite obs
no assimilation

conventional only
conventional cloud
CTH obs
No assim
total cloud cover of first guess fields after 20
hours of cycling
24
use of (SEVIRI-based) cloud top height (CTH)
observations in LETKF

• Summary
• assimilation of CTH by LETKF reduces errors of
  first guess (1-h forecast)
• tends to introduce humidity / cloud where it
  should ( temperature inversion)
• tends to reduce false-alarm clouds
• despite non-Gaussian pdfs
• no sign of filter collapse (decrease of spread)
• next evaluate forecast impact

25
Status of PP KENDA
Thank you for your attention Questions ?
26
LETKF implementation
GME COSMO
ensemble members 40 1 (3DVar) 40 1 (det. Run)
horiz. resolution (ens.) 60 km (ni128) 2.8 km
horiz. resolution (det.) 30 km (ni256) 2.8 km
vertical localis. length scale (ln p) 0.3 (fut. 0.075 - 0.5) 0.3 0.075 0.5
horiz. localisation length scale 300 km 100 km
adaptive horiz. localisation - (not used) used in latest exp.
additive covariance inflation 3DVar - B -
adaptive multiplicative cov. infl. yes yes
conventional obs yes yes
satellite radiances AMSU-A -
GPS-RO new exps -
radar data - passive
update frequency 3 h 1 h (? 30 / 15 min)
27
LETKF implementation

• multistep analysis (batch assimilation)
  implemented ? motivation
• local / nonlocal observations (e.g. radiances)
• different observation errors ? better use
  different localization scales
• in view of adaptive localization different obs
  densities (conventional / radar)
• (Perianez et al. work on paper with theoretical
  concept toy model/ idealised experiments)
• next step test with radar / SEVIRI CTH data

• Hendrik new flexible stand-alone scripts to run
  LETKF experiments
• without using NUMEX / archive ? very
  limited disk space
• ? 1 real day for 1 day of LETKF assimilation
• to do implement evaluation / verification tools
  in script suite
• may become very suitable tool for users outside
  DWD (academia)

28
cycled assimilation of dense CTH obs
false alarm cloud cover (after 20 hrs cycling)
high clouds
mid-level clouds
low clouds
conventional cloud
conventional obs only
29
Low cloud cover (COSMO)
1700 UTC
Cloud assim

No assim
Cloud conv
conv