Using Simulated Atmospheric Motion Vector Wind Retrievals from NWP - - PowerPoint PPT Presentation

Using Simulated Atmospheric Motion Vector Wind Retrievals from NWP Radiances to Characterize Height Assignment Errors

Peter Lean1* Stefano Migliorini1 and Graeme Kelly2

* EUMETSAT Research Fellow, 1 University of Reading, UK

2 Met Office, UK

Background: Atmospheric Motion Vectors (AMVs)

• Wind retrievals from

satellite imagery

– actually observations of apparent cloud motion

• Method:

– Feature detection and tracking between consecutive images – Height Assignment performed (usually based on cloud top temperature and background model temperature profile)

• Errors:

– Height assignment errors significant – Typically assessed against co-located sonde observations

Simulation framework to quantify errors

• Building upon previous AMV simulation studies:

– Wanzong et al (2006) – von Bremen et al (2008) – Stewart and Eyre (2012) – Hernandez-Carrascal et al (2012)

• Met Office UKV model

– 1.5km grid length NWP model

• RTTOV radiative transfer

– produces simulated brightness temperatures from model prognostic fields.

• Nowcasting SAF (NWCSAF) cloud and AMV products

– produces AMVs from the simulated satellite imagery.

• Model winds known at every grid point allowing detailed

quantification of errors for all AMV retrievals.

• Simulation framework requires model to be a reasonable

representation of reality.

NWCSAF AMV package workflow

Cloud Products Feature detection Feature tracking

(which previous features persist?)

Height Assignment

NWP background Cloud Mask Cloud Type Cloud Top Height Standard setup SEVIRI

• bservations

Cloud Products Feature detection Feature tracking

(which previous features persist?)

Height Assignment

NWP background Cloud Mask Cloud Type Cloud Top Height SEVIRI

• bservations

RTTOV

Simulated Radiances

Perfect model framework

NWCSAF AMV package workflow

6 week suite: Feb – March 2013 UKV ran daily from 03z to t+22h (PS31 components) RTTOV9 run on model data at 23:45, 00:00 and 00:15 each day

Wide range of meteorological situations sampled:

• maritime convection, frontal cloud, thin cirrus, stratocumulus over

inversion etc

• 6 weeks: long period of study compensates for relatively small

domain

Quantifying error in NWCSAF Cloud Top Height product

Diagnosing cloud top height from model cloud fraction

Look at model profile of cloud condensate:

• What is the height of the highest

model level with model cloud fraction above some defined threshold?

Results using model CTH defined by cloud fraction threshold

Cloud fraction threshold = 0.5 Low + Medium height opaque cloud categories Semi-transparent high cloud categories

CTH product too low CTH product too high

How realistic are the simulations?

50.0% of pixels are cloudy (excluding zero-height cloud) 69.1% of pixels are cloudy (excluding zero-height cloud)

Distribution of Cloud Top Heights (NWCSAF product)

6 weeks of data

Very low low medium High

• paque

land sea fractional Semi-transparent thin Semi-transparent mainly thick Semi-transparent thick

Model cloud product has:

• too much ‘very low cloud’
• too little ‘semi-transparent cirrus’
• too little ‘fractional cloud’

Distribution of Cloud Types (NWCSAF product)

6 weeks of data

Model cirrus too extensive and optically thin Model cirrus appears too thin/diffuse

• being misclassified

as low cloud Low cloud

Semi-trans high

Comparison of simulated v real brightness temperatures by channel

3.9µm IR 6.3µm WV 7.4µm WV 8.7µm IR 9.7µm IR 10.8µm IR 12.0µm IR 13.3µm IR Differences in cloudy regions much larger than known biases in clear sky regions

Sensitivity to NWP model formulation

Recent upgrades to UKV model affecting cirrus

• Met Office Parallel Suite 28:

– not enough cirrus in model.

• Parallel Suite 31:

– model upgraded to reduce ice fall speeds – more cirrus in model but worse precip

• Parallel Suite 32:

– different ice fall speeds used for aggregates and pristine crystals – more cirrus with no detrimental impact on precip

PS32 v PS31

• Previous results were from PS31
• Reran suite for PS32

Observed ch5 PS31 UKV / RTTOV9 ch5 Slightly more/thicker cirrus in PS32 UKV PS32 UKV / RTTOV9 ch5

Sensitivity to radiative transfer formulation

PS31 UKV / RTTOV9 ch5 PS32 UKV / RTTOV9 ch5 PS32 UKV / RTTOV9 ch5 PS32 UKV / RTTOV11 ch5

RTTOV9 v RTTOV11 (Baran)

Observed ch5 RTTOV9 RTTOV11

Summary

• Simulation framework used to quantify height assignment errors in

AMV wind retrievals.

• Height assignment errors quantified by comparing AMV assigned

height against model truth cloud top.

• Simulation framework provides a valuable methodology to

understand AMV retrieval errors for opaque clouds.

• Issues with thin cirrus:

– Simulated brightness temperatures are very sensitive to:

• model formulation
• radiative transfer formulation

– Caution required applying thin cirrus results from simulation framework to real world.

Thanks for listening! Any questions?