Benefit of high resolution data assimilation and observing systems in the Met Office UK NWP model

© Crown copyright Met Office

Benefit of high resolution data assimilation and

observing systems in theMet Office UK NWP model

G.T. Dow and B. Macpherson

13th EMS Annual Meeting & 11th European Conference on Applications of Meteorology (ECAM)09 – 13 September 2013, Reading, United Kingdom


Outline

Purpose

UK4 Model and Trial Configurations

Results

General

Case Studies

Observation Network Denials

Summary


Purpose

Local hi-res DA

- why bother ??!

Want to demonstrate added benefit relative to cheaper initialisation options

eg downscaling


Outline

Purpose


Results

General

Case Studies


Summary


UK4 Vital StatisticsUK4 Vital Statistics Resolution: 4km, 70 levels LBCs: from 25km Global

model

3DVAR (with FGAT) + IAU for all observations except Latent Heat Nudging for

radar-derived surface rain rate

Uniform 4km analysis grid

8 three-hourly cycles per day


UK4 Trials to measure and apportion marginal benefitUK4 Trials to measure and apportion marginal benefit

A.A. Full local DA (CONTROL)Full local DA (CONTROL)

B.B. Partial local DA Partial local DA - omitting UK4 only Obs types

C.C. Downscaler Downscaler - run from reconfigured 25km → 4km Global Analysis

Total Benefit of high-resolution

DA

AA vs CC

Benefit from observations

only used in the high-resolution

DA system

AA vs BB

Benefit from higher-

resolution alone

B B vs CC

1. MOPS cloud fraction profiles

2. radar-derived surface rain rate

3. visibility from SYNOPs

4. T2m & RH2m from Highways Agency roadside sensors

5. Doppler radial windsThese observations were denied to the ‘Partial Local DA’ trial


Trial PeriodsTrial Periods

No. of Forecasts

DatesPeriod

4x21=84Mar 10th → Mar 31stSpring

4x38=152Jan 3rd → Feb 10thWinter

4x44=176Nov 1st → Dec 14thAutumn

4x40=160Jul 1st → Aug 10thSummer

Forecasts to T+24 at 00Z, 06Z, 12Z & 18Z

Period picked at RandomPeriod picked due to specific (SCu) event


UK Index MetricUK Index Metric

•Weighted Basket of Indices•6 elements•Combo of ETS & RMS scores

•UK4 Impact trials verified to T+24 at 00, 06, 12, 18 UTC

Element ETS Threshold

1.5m Visibility 200m, 1000m,

5000m

6 hour ppn accumulation

0.5mm, 1.0mm, 4.0mm

Cloud Cover 0.3, 0.6, 0.8

Cloud Base Height

100m, 500m, 1000m

1.5m Temperature

1-RMS2

RMS2(pst)

10m wind 1-RMS2

RMS2(pst)


Outline

Purpose


Results

General

Case Studies


Summary


Verification by PeriodVerification by Period

UK Index Benefit

(-4.83 -- 4.47) = -0.36%-4.47%-4.83%21Spring

(0.78--0.18) = +0.96%-0.18%+0.78%38 Winter

(1.17-0.28) = +0.89%+0.28%+1.17%44Autumn

(2.54--0.96) = +3.50%-0.96%+2.54%40 Summer

of higher resolution aloneof extra obs types used in the UK4 model

of full local DA system

DaysPeriod

Cf. typical annual UK Index progression of 2% per annum


Verification by ElementVerification by ElementColors indicate best setup for element/period

Full DA(9) Full DA(9) Partial DA(11)Partial DA(11) DownScaler(4)DownScaler(4)

Mar 2012

Jan 2012

Nov 2011

Jul 2011

OverallWindTempCloud Base Height

Cloud Amount

PrecipVisPeriod

Note: Some boxes are more significant than

others

=CaseStudy


Threshold

Summer PrecipitationSummer Precipitation

ETS

P(D

etec

tion)

4.0mm/6hr

1.0mm/6hr

8.0mm/6hr

0.5mm/6hrP

( )F

alse D

etectionF

requency B

ias

Full DA Full DA Partial DA Partial DA DownScaler DownScaler Full DA – LHNFull DA – LHN

6-hour accumulations 6-hour accumulations July 2011July 2011


Stratocumulus PeriodStratocumulus PeriodMar 10Mar 10thth -15 -15thth 2012 2012

• Blocked episode• Cloud not breaking soon enough

• Significant T2m errors

• Suspect analysed cloud depth too large

Full DA Partial DA

DownScaler

T+7T+7


Cloud Cover March 2012Cloud Cover March 201216-31st (post-SCu)10-15th (SCu)

Full DA Full DA Partial DA Partial DA DownScaler DownScaler Full DA – MOPS CloudFull DA – MOPS Cloud

Bias Bias

RMSRMS

Similar Bias at T+0Large Bias in Control at T+12

Partial DA ~ Control – MOPS Cloud


TT2m2m impact from SCu errors impact from SCu errors

T+1T+1

Full DA Full DA Partial DA Partial DA DownScaler DownScaler Full DA – MOPS CloudFull DA – MOPS Cloud

T+12T+12

Timeseries 10th-31st10-15th (SCu)

Bias

RMS

SCu

10th-15th

Forecast Range →


Outline

Purpose


Results

General

Case Studies


Summary


UK4 Observation Network UK4 Observation Network denial experimentsdenial experiments

(Autumn period)

Surface +2.9%

Upper Air (excluding aircraft)

+2.1%

Aircraft +2.0%

Radar +2.0%

Satellite +1.7%

“Extra” (all obs networks not in global model)

+0.5%


SummarySummary Consistent Benefit for all elements from full higher-resolution analysis (except perhaps for wind) relative to downscaled analysis

Mixed performance from the extra observations

Sometimes detrimental to the UK Index scores

Consistent summer precipitation benefit up to T+6 from Radar RainRate (LHN) and for some thresholds to T+12

3D MOPS cloud analysis (subsequently replaced) shows overall benefit for cloud cover, but not so good for blocked SCu

Visibility (not presented here) – higher thresholds benefit from vis assimilation, lower thresholds sensitive to RH bias

Beware – largest DA (cloud) impacts can be later than T+1 !

Greatest contribution to UK Index from surface observation network


Questions ?

With thanks to: Bruce Macpherson, Mark Weeks, Dale Barker, Jorge Bornemann, Richard Renshaw

And others...

Thank you for listening


Extra slides...


UK4 Downscaler UK4 Downscaler Configuration & AnalysisConfiguration & Analysis

As UK4 except• Runs 4x daily

• 2stage 4D-Var assimilation at 120km/60km resolutions, resulting 25km Global Analysis then reconfigured to 4km/UK4 70 levels

• Prognostic ‘murk aerosol’ value for visibility (constant value in operational)

• Global Obs Cut Off ~ hh+160 mins


PrecipitationPrecipitation

Full DA

Partial DA Downscaler

55thth DecDec20112011T+1T+1


2.9

2.1 2.0 2.01.7

0.5

0

0.5

1

1.5

2

2.5

3

UK

In

de

x I

mp

ac

t (%

)

Network

Surface

Upper Air (exclaircraft)

Aircraft

Radar

Satellite

All obs networks notin global model

UK4 Observation Network denial experiments

(November 2011 period)


UK4 Observation Network denial experiments

variable Primary Beneficial Network

Secondary

Beneficial

Network

Visibility Surface

Precipitation Radar Upper Air, Aircraft,

Satellite, Surface

Cloud Cover “Extra”

Cloud Base Height

Upper Air

Temperature Surface Radar

Wind Surface Satellite, Upper Air, Aircraft


Stratocumulus PeriodStratocumulus PeriodMar 10 -15 2012Mar 10 -15 2012

• Cloud not breaking soon enough

• Significant T2m errors

• Analysed cloud depth too large• Suspect Cloud Top Height too high

Full DA Partial DA

DownScaler

T+7T+7Full DA Partial DA

DownScaler

T+1T+1

Arctic Ice Syndrome….. the thickness also matters


Cloud Cover March 201216-31st 10-15th (SCu)

Mean ErrorMean Error Mean Difference from Mean Difference from ‘Full DA’ trial‘Full DA’ trial

Full DAFull DAPartial DAPartial DADownScalerDownScalerFull DA – MOPS CloudFull DA – MOPS Cloud

RMS ErrorRMS Error RMS Difference from RMS Difference from ‘Full DA’ trial‘Full DA’ trial


TT2m2m impact from SCu errors impact from SCu errorsMean ErrorMean Error Mean Difference from Mean Difference from

‘Full DA’ trial‘Full DA’ trialFull DAFull DA

Partial DAPartial DA

DownScalerDownScaler

Full DA – MOPS CloudFull DA – MOPS CloudRMS ErrorRMS Error RMS Difference from RMS Difference from

‘Full DA’ trial‘Full DA’ trial

T+1T+1

T+12T+1210-1410-14thth

16-3116-31stst


Verification by ElementVerification by Element

Full DA vs Downscaler

Period

Trial

vis precip cloud

amount

cloud base

height temp wind

Overall

Jul 2011

DS +0.268 +0.151 +0.019 +0.010 +0.078 +0.006 +2.54%

Nov 2011

DS +0.083 +0.123 +0.017 +0.063 +0.019 -0.011 +1.17%

Jan 2012

DS +0.073 +0.077 -0.034 +0.080 +0.026 -0.024 +0.78%

Mar 2012

DS

-0.333 +0.007 +0.082 -0.101 -0.525 -0.091 -4.47%

Full DA vs Partial DA

Period

Trial

vis precip cloud

amount

cloud base

height temp wind

Overall

Jul 2011

NHR +0.036 +0.078 -0.046 -0.000 -0.252 -0.018 -0.96%

Nov 2011

NHR -0.011 +0.051 +0.033 +0.016 +0.002 -0.023 +0.28%

Jan 2012

NHR +0.056 -0.064 +0.029 +0.017 -0.066 -0.017 -0.18%

Mar 2012

NHR

-0.321 -0.095 +0.008 -0.086 -0.349 -0.197 -4.83%

Colors indicate best setup for element/periodFull Full DA(9) DA(9) Partial DA(11)Partial DA(11) DownScaler(4)DownScaler(4)

Period Vis Precip Cloud Amount

Cloud Base Height

Temp Wind Overall

Jul 2011

Nov 2011

Jan 2012

Mar 2012

Awkward Period(see later)


Roadside sensor networkRoadside sensor network OpenRoad – full network SYNOP


Roadside sensor network impactRoadside sensor network impact

Mean T2m error

(2nd half of Dec 2010)

RMS T2m error

control

test


Further WorkFurther Work

• Do additional trial configurations to decipher signals from individual obs types

• Assess benefit from staggered Data Times

• Look at sensitivity of DA signal to synoptic conditions

• Adjoint-based impact studies in the UK Model (see Richard Marriott’s talk tomorrow)

Benefit of high resolution data assimilation and observing systems in the Met Office UK NWP model

Documents