Satellite SST Radiance Assimilation and SST Data Impacts

James Cummings Naval Research Laboratory

Monterey, CA 93943

2013 Sea Surface Temperature Science Team MeetingSeattle, Washington, USA

Global HYCOM SST 25 Oct 2013 00Z

• incorporate impact of real atmosphere above SST field • remove atmospheric signals in SST radiance data • include variables known to affect satellite SST radiances:

• atmospheric temperature• atmospheric water vapor• SST• aerosols (dust, smoke) - not used

yet

• all variables are available from NWP, aerosol, and ocean model forecasting systems

SST Radiance Assimilation: Objectives

GOES-15 Water Vapor

HYCOM SST Forecast

SST Radiances (BTs)

Calibration, QC, Cloud Clearing

Compute Differences: δBTs

3DVAR Minimization

SST Inverse Model: δSST

NWP Fields

CRTM Jacobians (Tb/x)

CRTM Forward Model: TOA-BTs

SST Radiance Assimilation:

Observation Operator

Other Observations: T,S,U,V

Observation & First Guess Errors

Ocean Model SST

SST Radiance Monitoring

GOES METOP GAC METOP LAC

NOAA GAC NOAA LAC NPP VIIRS

SST Radiance Assimilation: Observing Systems

NAVOCEANO SST Radiance Data:

• GOES-13, GOES-15• METOP-A, METOP-B (GAC and LAC)• NOAA-18, NOAA-19 (GAC and LAC)

• NPP VIIRS• COMS-1

No radiance data available for MSG and MTSAT

Ch3

Ch4

Ch5

NAVO Buoy

Matchup Data

25 Aug to 8 Sep 2013

NWP Priors:90 min,16 km

METOP-A

CRTM Forward Model: Obs vs. Simulated BTs NPP VIIRSNOAA-19

CRTM Forward Model: Bias Correction

• by definition satellite SST radiances are cloud free• but NWP priors may have clouds • TCWV bias correction modeled as quadratic function (red

curves)• calculated for all satellites and all channels using 15-day

sliding time window of NAVO SST radiances and buoy matchups

METOP-A NOAA-19

Liquid Water Path kg/m2 Liquid Water Path kg/m2

Computes SST correction (Tsst ) given TOA BT innovations (BT) and CRTM Jacobians (J):

Requires specification of prior error statistics:• air temperature: εt

• specific humidity: εq

• sea surface temperature: εsst

• satellite BTs + radiometric error: εbt

Partitions BT innovations into Tsst , Ta , Qa

corrections

SST Inverse Model

a

a

sst

qqqtqsstq

qttttsstt

qssttsstsstsstsst

q

t

sst

Q

T

T

JJJJJJ

JJJJJJ

JJJJJJ

JBT

JBT

JBT

=

NAVGEM Ensemble: Specific Humidity

SST Inverse Model: NWP Prior Errors

Provides situation dependent uncertainty of atmospheric forecasts

Specific humidity variability greatest at low latitudes

NAVGEM Ensemble: Air Temperature

SST Inverse Model: NWP Prior Errors

Air temperature variability greatest at high latitudes

SST Inverse Model: Ocean SST Errors

Atlantic Indian Pacific

Computed from time history of model forecast differences at update cycle interval (24-hr)

SST variability greatest in tropics, Antarctic circumpolar, and western boundary currents

HYCOM 3DVAR SST Background Error

5 Sep 2013

SST Inverse Model: NAVO SST Corrections

METOP-A 5 Sep 2013 Day Night

•large positive SST corrections at high latitude dry atmospheres•shows globally defined NAVO SST retrievals biased in some regimes

•some day/night differences in SST corrections (e.g., U.S. west coast)

Tsst

SST Inverse Model: Air Temperature Corrections

METOP-A 5 Sep 2013

Day Night

•air temp corrections generally small given expected range of atmospheric temperatures

•corrections tend to be largest at high latitudes where NWP model air temperature variability is high (some exceptions, e.g. WestPAC)

Ta

SST Inverse Model: Water Vapor Corrections

METOP-A 5 Sep 2013

•water vapor corrections tend to be greater at low latitudes where NWP model water vapor variability is high

Day Night

Qa

SST Inverse Model: Correction Data Density

METOP-A METOP-B

NOAA-18 NOAA-19

NAVO SST Correction vs. Total Column Water Vapor: 5 Sep 2013

Tsst

Observation(y) NCODA

3DVAR HYCOMForecast

(xf)

Forecast ErrorJ: (J/xf)

Background(xb)

Analysis(xa)

Adjoint ofHYCOM

ObservationSensitivity

(J/y)

Initial ConditionSensitivity

(J/xa)

Adjoint of 3DVAR

What is the impact of observations on measures of forecast error (J) ?

Data Impact System

Analysis – Forecast System

𝛿𝑒 𝑓𝑔=⟨ ( 𝑦−𝐻𝑥𝑏) , 𝜕 𝐽𝜕 𝑦 ⟩Observation Impact

Equation (Langland and Baker, 2004)

Observation Impact Equation: Interpretation

< 0.0 the observation is BENEFICIAL - forecast errors decrease

For any observation assimilated, if ...

gfe

gfe > 0.0 the observation is NON-

BENEFICIAL - forecast errors increase

Non-beneficial impacts:

- not expected, assimilation should decrease forecast error

- if it is persistent, may indicate observing system problems

SST Data Impact: Satellite Observing Systems

Global HYCOM - November 2012Per Ob Data Impacts for Reducing HYCOM 48-hr SST Forecast

Error Atlantic Pacific

Data assimilated are NAVOCEANO SST retrievals

SST Data Impact: Non-beneficial Impacts

METOP-A - November 2012(averaged at model grid locations)

PacificAtlantic

SST Data Impact: Non-beneficial Impacts

NOAA-19 - November 2012(averaged at model grid locations)

PacificAtlantic

SST Data Impact: Non-beneficial Impacts

METOP-A vs. NOAA-19 - November 2012

(averaged at model grid locations) METOP-A NOAA-19

More non-beneficial impacts from assimilation of NOAA-19 retrievals than METOP-A Differences in quality of AVHRR instruments on the two

satellites?

SST Data Impact: Non-beneficial Impacts

GOES - November 2012(averaged at model grid locations)

PacificAtlantic

• CRTM forward and SST inverse modeling:• removes atmospheric signals from radiance observations• real atmosphere needed to understand changes in TOA BTs

• Data Impacts:• assimilation of satellite SST data reduces HYCOM forecast

error• non-beneficial impacts show geographic, instrument, and

satellite zenith angle dependencies

• Science Team Objectives:

• data classification: TOA-BT Tsst , Ta , Qa partitions and data impacts provide pixel level information

• data merging and gridding: global, model-based 3DVAR with full error analysis and data impact components

• communication: contribution to MICROS? Other inter-comparison activities along the lines of the ESA CCI?

Assimilation and Data Impact: Conclusions

Questions?