Klaus Stephan , Stefan Klink and Christoph Schraff and Daniel Leuenberger

05.08.2005 - 1 -COSMO General MeetingCOSMO General Meeting

Bucharest, 18 - 21 Sept 2006 [email protected]

Klaus Stephan , Stefan Klink and Christoph Schraff

and Daniel [email protected]

[email protected]

[email protected]

[email protected]

Recent developments in Latent Heat Nudging at DWD

• revision to grid point search and its impact

• problem case with strong advection

• latest test suite June – July 2006:

impact of LHN, comparison to LME

• some conclusions



Adaptations for applying LHN with prognostic precipitation

• use of a reference precipitation: vertically averaged precipitation flux

• apply LHN-increments only where latent heat rates are positive

• apply (upper and lower) limits to scaling factor ,logarithmic scaling (replace ( –1) by ln() effective limits: [0.3, 1.7] )

• impose absolute limits to LHN-increments

• search for nearby grid points, if model precipitation rate is too low but to use a moderate forcing of precipitation at these points

mo

obsLHmoLHN RR

RRTT , 1

Other options used

• adjustment of specific humidity in order to maintain relative humidity

• vertical filtering of profiles of LHN-increments

• horizontal filtering of incoming variables (of small extent)



LHN: Grid-point search

0 0 0 0 0

1 0 0 0 0

1,7 0 0,4 0,3 0

2,1 0,2 0,1 0,2 0

1,5 0,8 0 0 0

Example:

RRobs = 3.1 mm/h

RRref = 0.4 mm/h

RRsearch = 2.1 mm/h

LHserach

z

*

z

Tinclhn

but, how large should be the scaling factor ?(want to add ‘0.7 * RRref = 0.28 mm/h)

17.0,min,7.1max

1

obsrefmax

search

refmax

RRRRRR

RR

RRRR

(revised version of RRmax ; in the old version,

RRmax and could become very large)

conditions: RRsearch close to RRobs

LHref small enough

LHsearch large enough



radar weaker forcing stronger forcing

spatial average of actual precipitation rate:old stronger forcing overestimates precipitation

radarweaker

stronger

UTC

impact of new version with revised (‘weaker forcing’) grid point search compared to old version (‘stronger forcing’)

case study: assimilation at 21 May 2005

4 U

TC

4 U

TC

6 U

TC

6 U

TC

mm/h

hourly precipitation (4, 6 UTC):old stronger forcing produces strong gravity waves



ETS

ETSAssimilation

FBI

FBI

0.1 mm/h

2.0 mm/h

impact of revised (‘weaker forcing’) grid point search compared to old ‘stronger forcing’

test period: 8 – 20 July 2004 , comparison to control without LHN (older LMK version)

underestimation of model (control) largely but not fully corrected

almost no bias any more for strong precip

higher ETS despite lower FBI:better match of precip patterns



free forecasts Threshold = 2.0 mm/h

impact of revised (‘weaker forcing’) grid point search compared to old ‘stronger forcing’

test period: 8 – 20 July 2004 , comparison to control without LHN (older LMK version)

ETS

positive forecast impact for 4 – 6 h

FBI

undershooting delayed and strongly reduced

12 UTC



ETS

free forecasts

FBI

threshold = 2.0 mm/h

48 forecasts, different convective cases

scores for hourly precipitation : with latent heat nudging / without latent heat nudging


+0h +0h +6h+6h

in 80 stratiform cases, LHN has less impact (3 – 4 h)



radar LMK ass without LHN LMK ass with LHN

hourly precipitation on 21 July 2005, 11 UTC: problem case with very strong low-level winds




strong low-level flow

no precipitation simulated

build-up of LH,low pressure

rain,high pressure

radar sees precipitation → constant input of latent heat by LHNbut takes time to

produce rainadvection of LH → influence

of LHN too fardownstream

flow slowed down,positive feedback

spurious small-scale pressure disturbance

and heavy rain

system eventually propagating upstream

and producing strong gravity waves



LMK ass with LHN


duplicating LH incr. weighting of LH incr.

‘weighting’:LH increments decreased linearly from 1 to zerowhen low-level wind speed increases from 20 to 30 m/s(low-level wind speed vll := ½ v950 + ¼ v850 + ¼ v700hPa )

radar



ETS

assimilation

FBI


16 June – 30 July 2006 (45 days)



overestimationin early morning

well balanced

new LMK version with revised droplet size distribution,

reducing evaporation of precip weak precipitation enhanced

drop

rise

drop

risenumber ofobserved

events

smalldrop



ETS

free forecasts

FBI


16 June – 30 July 2006 (45 days)

00 + 12 UTC runs



same ETS despite

smaller FBI

+0h +0h +4h+4h

higher ETS

undershooting(w. resp. to no-

LHN)

LMK: too little precip



ETS

free forecasts

FBI

threshold = 0.5 mm/h 12 UTC runs


+0h +0h +4h+4h

00 UTC runs

16 June – 30 July 2006 (45 days)



ETS

free forecasts

FBI


16 June – 30 July 2006 (45 days)

00 + 12 UTC runs

scores for hourly precipitation : comparison LME LMK with LHN, LMK without LHN


higher ETS despite

smaller FBI

+0h +0h +4h+4h

LMK: precip areas too small

all models: strong precip underestimated



Conclusions & outlook

• due to revised ‘weaker forcing’ grid point search(and using all the other adaptations of LHN to prognostic precipitation):– FBI close to 1 during assimilation, (much) less undershooting in forecasts– LHN better balanced, less gravity waves (but still too much, too strong gusts, etc.)– duration of positive forecast impact enlarged ( ~ 4 hours)

• however: Still rapid loss of benefit

need for better understanding of convection, in particular how the modeldevelops convection, role of environment, what kind of information is

required

further improve LHN, vertical distribution of LH (3D reflectivity ?), horizontal filtering, use of cloud info

need for use of radar radial velocity, GPS tomography, Ensemble DA ?

• model bias: model produces too little precipitation by itself, wrong diurnal cycle LHN able to compensate this during the assimilation by activating the model

to produce more rain, i.e. pushes model away from its climate, butat the price of: – cooling and drying of PBL

– increasing mid-tropospheric stability– undershooting of precipitation in forecast

– stronger limitation to duration of forecast benefit from LHN

need for improving model (particularly diurnal cycle and bias of precipitation)



ECMWF EPS

COSMO LEPS (7km)

HIRES LEPS (2.2km) ANA FC

Radar Rainfall Assimilation and Short-Range QPF in a High-Resolution EPS: A Case Study (Daniel Leuenberger, Marco Stoll, MeteoSwiss)

-28h-28h +8h+8h-10h-10h 00

12UTC12UTC 06UTC06UTC 16UTC16UTC 00UTC00UTC

Nested high-resolution EPS: role of convective environment for LHN,investigate on nested EPS with best-member selection based on satellite + radar data



Meteosat 7 IR 16:00 UTC

ENS mean & spread

1 2 3 4 5

6 7 8 9 10

SAT

Ch. Keil, DLR

COSMO LEPS (7km)



det 21RADAR

3 54 6

7 98 10

Precipitation at 18UTC: Forecast (+2h)



RAD 1 2 3 4 5 6 7 8 9 10 det

Mean Area Precipitation (Bavaria)

convective environment matters a lot



Best member selection possible ? only to a limited degree in current case

0

1

2

3

4

5

6

7

8

9

10

0 1 2 3 4 5 6 7 8 9 10

Ranking Satellite (non-linear pattern recognition)

Ra

nk

ing

QP

F



RadarWith LHN Without LHN (dashed: determininistic)

Benefit of LHN ? in all cases very significantly



Findings

• Substantial spread in QPF among fine-scale members during first 4 hours

• Large benefit of radar assimilation with LHN

• Ranking in QPF does not correspond well with that using satellite data of driving members (convective environment is not explained with the cloud structure alone!)

• Large spread in humidity among coarse members, smaller in temperature and wind

• Some spread in CAPE („good“ members with higher CAPE)

• Some difference in upper-level flow (some of the „bad“ members exhibit upper level convergence->subsidence in lower levels)

• Cloud-based best-member selection does not work well for this case



radar reflectivity data for LHN at DWD

• reflectivity from „precipitation scan“ (lowest elevation angle between 0.5° and 1.8)– spatial resolution: 1 km x 1°, max. range 120 km, time resolution: 5’

• data processing:– correction of ground clutter by doppler filter– correction of orographic attenuation– use of a variable Z-R-relation to get precipitation rate

• quality product of „precipitation scan“,

detection of non-rain echoes (by K. Helmert and B. Hassler):– corrupt image– ‘German Pancake’– anomalous propagation– spokes (of positive or negative attenuation)– circular arcs (of positive or negative attenuation) – echos of small extension (< 9 pixels) caused by wind energy plants etc.

to be done: detection of other errorsnon-rain echoes– precipitation and radome damping– bright band

• compositing of the 16 German doppler radars:precipitation using quality information, then quality product, 1 x 1 km

• gribbing: use quality product to mask precip, interpolate to 2.8 x 2.8 km

• use of blacklist



original (Emden, 19 July 2005, 23 UTC) after detection of spokes + clustersquality product

anaprop‘German Pancake‘ arcs



precipitation damping bright band

much more obvious in 24-hour precipitationthan in reflectivity / precipitation rate

obs 24-h precip radar

not yet done




strong low-level flow

no precipitation simulated

LH inputat beginning

idea:duplicate LH increments

near inflow border of radar domain, depending on wind vectors

(average at low levels)

rapidly,(areas of) LH inputare significantly

reduced

hardly anypositive feedback

effects and pressure disturbances

however:problems further

downstream

rain produced closer to

radar border

LH inputlater on




new LMK / LHN with weighting LMK ass with old LHN LME ass (without LHN)

mm/h



ETS

free forecasts

FBI

threshold = 0.5 mm/h 12 UTC runs

+0h +0h +4h+4h

00 UTC runs

16 June – 30 July 2006 (45 days)

scores for hourly precipitation : comparison LME LMK with LHN, LMK without LHN

LMK better than LME for 12 UTC runs

LMK: too weak diurnal cycle, too little precip in afternoon, less bias at night



+

-from: R. A. Houze, Jr.: Cloud DynamicsInternational Geophysics Series Vol. 53

• main part of positive latent heat release occurs in updrafts, strong precipitation rates are often related to downdrafts

• at x < 3 km , with prognostic treatment of precipitation (model resolves large clouds): model is able to distinguish between updrafts and downdrafts inside convective systems

horizontal displacement of areas with strong latent heating resp. to surface precipitation,

modified spatial structure of latent heat release in the model

scheme will notice only with temporal delay if precipitation already activated by LHN

• LHN-Assumption: vertically integrated latent heat release precipitation rate



possible adaptations II

• change of the spatial structure of latent heat release in the model:

– updraft regions (at the leading edge of a convective cell): very high values of latent heat release TLHmo, little precipitation RRmo

higher values of the scaling factor and of LHN increments often occur

mo

obsLHmoLHN RR

RRTT , 1

reduce upper limit of the scaling factor

adapt grid point search routine

– downdraft regions (further upstream): high precipitation rate, weak latent heat release (often negative in most vertical layers)

LHN increments are inserted only in the vertical layers where the model latent heating rates are positive (approx. in cloudy layers)

(to avoid e.g. negative LHN increments and cooling where the precipitation rate should be

increased)



possible adaptations III:

• temporal delay effect (generated precipitation reaches the ground with some delay):

an immediate reference information, on how much precipitation the temperature increment has initialised already, is required within each time step

use of a ’reference precipitation’ RRref: refRR

RR

RR

RR obs

mo

obs

• diagnostically calculated precipitation rate (by additional call of diagnostic precipitation scheme without any feedback on other model variables)

• vertically averaged precipitation flux (more consistent, however it does not eliminate the temporal delay completely)

for LHN: temporal delay effect found to be much more important than spatial displacement



verification against German radiosondes, 11-day period (8 – 18 July 2004): dashed: with latent heat nudging / solid: without latent heat nudging

bias

temperature

relativehumidity

+ 0 h + 6 h + 12 h + 18 h

more stable

colder

drier

moister



verification against German radiosondes, 11-day period (8 – 18 July 2004): dashed: with latent heat nudging / solid: without latent heat nudging

r m s e

t e m p e r a t u r e

r e l a t I v e h u m I d I t y

+ 0 h + 6 h + 12 h + 18 h

worse

better



• ‘blacklist’ for radar data: avoids introduction of spurious rain at radar locations

• several adaptations to LHN to cope with prognostic precipitation; most important:use of an ‘undelayed’ reference precipitation (vertically averaged precipitation flux)

• revised LHN, assimilation mode: – simulated rain patterns in good agreement with radar observations, – overestimation of precipitation strongly reduced

• subsequent forecasts, impact on precipitation (10-day summer period):– large positive impact for 4 hours (longer than in simulations with diagnostic precip)– mixed ETS impact beyond + 6 h (interpretation yet unclear,

need verification without ‘double penalty’)

• upper-air verification (11-day summer period): – LHN cools and dries PBL, increases mid-tropospheric stability and upper-

tropospheric moisture– overall neutral impact on rmse of forecasts

• strong gravity waves induced during assimilation LHN forcing too strong

Summary of Results

Klaus Stephan , Stefan Klink and Christoph Schraff and Daniel Leuenberger

Documents

Transcript of Klaus Stephan , Stefan Klink and Christoph Schraff and Daniel Leuenberger