outline why use atsr? why variational analysis? forward model examples validation level 3 products

Caroline PoulsenATSR-2 Group

Cloud parameters estimated by variational analysis of visible and

infrared measurements from ATSR-2Caroline Poulsen, Richard Siddans, Barry Latter and

Brian Kerridge, Chris Mutlow, Sam Dean2, Don Grainger2, Gareth Thomas2, Graham Ewen2 and

Phil Watts1

Space Science and Technology DepartmentRutherford Appleton Laboratory

UK1. Now at EUMETSAT2. Oxford University

OutlineWhy use ATSR?

Why Variational Analysis?Forward Model

ExamplesValidation

Level 3 productsFuture

ATSR Channels

ATSR2/AATSR • 0.55um• 0.67um• 0.87um• 1.6um• 3.7um• 11um• 12um

Cloud Parameters Retrieved• Cloud top pressure/height• Cloud fraction• Cloud optical depth• Cloud effective radius• Cloud phase

Auxillary information• ECMWF T and q profiles• MODIS surface albedo

Aerosol Parameters Retrieved• Aerosol optical depth• Aerosol effective radius

Comparing measurements with calculations: Ice, water and mixed

phase

waterice

Why use Optimal Estimation?

• Basic principle is to maximise the accuracy the retrieved cloud parameters based on the measurements and any ‘apriori’

• Allows us to characterise the error in each cloud parameter under the assumption of a reasonably plane parallel cloud model

• It’s a very flexible approach that enables us to utilise any prior information, for example on cloud fraction. All the clear sky atmospheric effects can be derived from NWP profiles.

• Allows us to utilise ALL the information in the measurements for each channel contributes to a greater or lesser extent to the retrieval of individual cloud parameters.

Forward Model

Ice clouds: complex particlesCurrently uses a combination of geometric optics (ray tracing); for large ice crystals and a T-matrix (ray tracing); method for small crystals.

Plates

Columns

Rosettes

Aggregates

Water clouds: spherical drops

Mie theory: solution of electromagnetic equations on dielectric sphere

Size distribution

10 m drop, 0.87 m wavelength

Since real time calculations of cloud radiative properties are too slow calculations are made once DISORT (plane-parallel) model and incorporating rayleigh scattering and stored in easily accessible Look up Tables.

Look up Tables

Tbc

Tac(e.g. MODTRAN)

Cloud + Atmosphere/surface

• Separate solar and ‘thermal’ models• Both embed cloud with precalculated radiative

properties (LUTs) in clear atmosphere

re pc (f)Solar model

Rs

TacFrom e.g. RTTOV

re pc (f)Thermal model

Transmitted

Rbc

Cloud emitted

B(T(pc))

Reflected

Rdown

Atmosphereemitted

Rup

Inversion: Optimal estimation

Guess xo

Calculate measurements y(xn)

Adjust (minimise J) x = - J’/J’’ (Newton’s Method)

Stop! J < 0.1 or n>10

Compare J = [ym-y(xn)] Sy-1 [ym-y(xn)]T

a priori xb

+ [xn-xb] Sx-1 [xn-xb]T

= 1D-Variational analysis. Same principles > 3D, 4D Var (assimilation)

Cost Function

Compare J = [ym-y(xn)] Sy-1 [ym-y(xn)]T

+ [xn-xb] Sx-1 [xn-xb]T

J = [ym-y(xn)] Sy-1 [ym-y(xn)]T

Where ym are the radiances, Sy the measurement error covariance and y(xn) the cloud parameters modelled into radiance space. + [xn-xb] Sx

-1 [xn-xb]T

Where Xb is the apriori and Sx the apriori covariance.

Inversion: Optimal estimation

Guess xo

Calculate measurements y(xn)

Adjust (minimise J) x = - J’/J’’ (Newton’s Method)

Stop! J < 0.1 or n>10

Compare J = [ym-y(xn)] Sy-1 [ym-y(xn)]T

a priori xb

+ [xn-xb] Sx-1 [xn-xb]T

= 1D-Variational analysis. Same principles > 3D, 4D Var (assimilation)

Minimising J: optically thick cloud

xoxsolution

-No a priori,-0.55, 1.6 m channels

- , Re only

Retrieved Cloud Parameters

Optical depth

Effective radiusFraction

Cloud top pressure

False colour

Error Analysis and Quality Control

Cost

Ssolution = J’’ solution = (Sx-1 + KT.Sy

-1K)-1

Error Cloud top pressure

False colour

Validation Activities

Re validation against MRF FSSP probe

Optical depth (scaled to fit)

Effective radius

Hercules - ERS-2Coincidence

FSSP

ATSR

Validation at SGP 20th Oct. 1997AATSR overpass17:26Microwave radiometer

SGP ARM data courtesy of Roger Marchand.

Case study 20th October 1997

Parameter ATSR-2 SGP

Optical depth 37.3 35.8

Effective radius 8.8 8.9

Liquid water path 244.0 209.8

Effective radius LWPOptical Depth

SGP validation

Mean: -0.08Stdev: 1.21

Liquid water path is calculated using the technique of Frisch et al, J. Atmos Sci. 1995, the technique is only valid for non-raining, water clouds.

Optical depth calculated using Han et alJ. Atmos Sci.,1995. Errors shown are the standard deviation of the matches used.

Validation of CTH

Chilbolton 94GHz Galileo Radar

Comparison with ISCCP data

ATSR-2 May 1999 Optical depth ISCCP Optical depth May 1999

Level 3 products

Cloud top pressure

Cloud optical depth

Cloud effective radius

Cloud fraction

Summary and plans

• 6 years of ATSR-2 data processed at 3x3km resolution and a variety of level 3 products

• Version 2 to begin soon with many improvements

• Potential is there to use information from other satellites

• Dual view tomographic cloud retrieval

• Extension to AATSR- long time series

• More validation, comparison with met. Office models

The ATSR cloud and aerosol algorithm was developed under funding from the following projects

The end

QC: Summary

• Model adequate (J<1)– Expected errors, S

• parameter dependent• state dependent• Information for

assimilation

• (Discussed today• Not discussed)

• Model inadequate (J>1)– A priori out of range

• rogue values– Measurements out of range

• calibration errors• rogue values

– Model out of range• multi-layer cloud• shadows• incorrect ice crystals• incorrect surface reflectance• incorrect statistical

constraints

Retrieval (inversion): required steps

• “Forward modelling”:

– Optical properties of average particle in ‘single scattering’ event

– Optical properties of a cloud of particles: multiple scattering

– Interaction of cloud radiative processes with atmosphere and surface

– y = y(x)

• “Inverse modelling”:– x = ? (y)

– Guess cloud conditions (x)

– Calculate radiances y(x)

– Compare to measurements

– Change cloud conditionsStop!

Re validation against MRF FSSP probe

Optical depth (scaled to fit)

Effective radius

Hercules - ERS-2Coincidence

FSSP

ATSR

Monthly Averaged Results

May 1999 log10Optical depth May 1999 effective radius

Water clouds: spherical drops

Single particle

Mie theory: solution of electromagnetic equations on dielectric sphere

Size distribution

10 m drop, 0.87 m wavelength

Cloud top pressure