Towards Community Consensus SSTs and Clear-Sky Radiances from AVHRR

8 November 2010 SST Science Team Meeting 1

Towards Community ConsensusSSTs and Clear-Sky Radiances

from AVHRR

SST Science Team Meeting8-10 November 2010, Seattle WA

Sasha Ignatov

NOAA/NESDIS

Prasanjit Dash, Xingming Liang, Feng Xu

NOAA/NESDIS and CSU/CIRA


Contributions

AVHRR Level 2/3 SST products - J. Sapper, Y. Kihai, B. Petrenko, J. Stroup: ACSPO (GAC: 5 platforms, FRAC: MetOp-A)

- P. LeBorgne: O&SI SAF MetOp-A FRAC

- D. May, B. McKenzie: NAVO SEATEMP

- K. Casey, T. Brandon, R. Evans, J. Vazquez, E. Armstrong: PathFinder v5.0

Level 4 SST products (additional L4 SSTs are being tested)- R. Grumbine, Xu Li, B. Katz: RTG (Low-Res & Hi-Res), GSI

- R. Reynolds: OISST (AVHRR & AVHRR+AMSRE)

- M. Martin: OSTIA foundation, GHRSST Median Product Ensemble

- D. May, B. McKenzie: NAVO K10

- E. Autret, J.-F. Piollé: ODYSSEA

- E. Maturi, A. Harris, W. Meng: POES-GOES blended

- B. Brasnett: Canadian Met. Centre, 0.2 foundation

AVHRR Radiances- C. Cao, X. Wu, J. Mittaz, A. Harris, A. Heidinger, L. Wang: AVHRR Cal

- C. Cao, T. Hewison, M. König: GSICS (Global Space-based Inter-Cal System)

- Y. Han, M. Liu, Y. Chen, P. Van Delst, D. Groff, F. Weng: CRTM


SST data come from various sources, sensors, and processing algorithms

L4 Reynolds (AVHRR; +AMSR-E) RTG (Low, High Resolution), GSI OSTIA (UKMO) ODYSSEA (France) GMPE (GHRSST) NAVO K10 NESDIS POES-GOES Blended JPL G1SST CMC 0.2 (Canada) GAMSSA (Australia) JAXA (Japan) RSS (MW, MW+MODIS)

In situ Sources-GTS, ICOADS, GODAE/FNMOC

Platforms-Drifters, Moorings, Ships, ARGO Floats

Quality Control-May be unavailable or non-uniform

L2/L3 POES-AVHRR (NESDIS, NAVO, O&SI SAF, U. Miami, NODC)-MODIS, ATSR, VIIRS, ..-Microwave

GOES-GOES (NESDIS, NAVO, O&SI SAF)-SEVIRI (NESDIS, O&SI SAF)-MTSAT (NESDIS, JAXA)

Are these products self-consistent?Cross-consistent? Processed using community consensus algorithms?


Initial Focus: AVHRR

Past: Climate Data Record- Pathfinder Ocean: 1981-present

Present: Initial Joint Polar System (IJPS)- NOAA/EUMETSAT Cooperation

- NOAA: GAC (4km), NOAA-18 & -19 (2 pm/am)

(NOAA-16 and -17 are also in orbit but degraded)

- EUMETSAT: FRAC (1km) 3 mid-morning birds (9:30 am/pm)

Metop-A (19 Oct 2006), -B (2012), -C (2016)

Future: Joint Polar Satellite System (JPSS)- Europe: AVHRR/3 onboard Metop-B (~2012) & -C (2016)

- AVHRR will serve the data through at least 2021

- USA: Visible and IR Imagers Suite (VIIRS) on NPP & NPOESS- Need to ensure VIIRS/AVHRR synergy


Operational POES SST at NESDIS

Heritage Main Unit Task (MUT)- 1981 - pr (MCSST - McClain et al., 1985; NLSST - Walton et al., 1998)

- 1993 - pr: Re-hosted to NAVO (“Shared Processing Agreement”).

Robust end-to-end system. No fundamental redesign since 1981: Data sub-sampling (e.g. 2×2); No RTM; No reprocessing capability.

New Advanced Clear-Sky Processor for Oceans (ACSPO)- Development started in late 2005 (IJPS) - Operational in May 2008

Process all AVHRR pixels (GAC, FRAC). RTM & Reprocessing capability.

Joint Polar Satellite System (JPSS) - Generate AVHRR-like ACSPO products from VIIRS radiances.

Fall-back for NPOESS SST. Benchmark to measure VIIRS improvements. Smooth transition for SST users.

- Cross-evaluate against VIIRS SST generated by NPOESS contractor.

Contractor’s cloud mask + VIIRS instrument + Heritage NLSST algorithm


GAC ~4kmGlobal Area Coverage

GAC ~4kmGlobal Area Coverage

NESDIS ACSPO (new); MUT (heritage)

NESDIS ACSPO (new); MUT (heritage)

MetOp FRAC ~1kmFull Resolution Area Coverage

MetOp FRAC ~1kmFull Resolution Area Coverage

EUMETSATO&SI SAF

EUMETSATO&SI SAF

NAVOSEATEMPNAVO

SEATEMPU. Miami + NODC Pathfinder Ocean (L3)

U. Miami + NODC Pathfinder Ocean (L3)

NESDISACSPO (new)NESDIS

ACSPO (new)

Cross-evaluate ACSPO against heritage MUT SST

Current AVHRR SST Products

.. and against other available L2 AVHRR SST products

L2 vs. in situ

Customary Validation

78 November 2010 SST Science Team Meeting


Global Mean “SAT – In Situ” Global Std Dev “SAT – In Situ”

Outliers in in situ data:Critically affect validation statistics

MUT SST versus in situ SST(no QC applied to in situ SST)


Global Mean “SAT – In Situ” Global Std Dev “SAT – In Situ”

MUT SST versus in situ SST(after QC of match-up data)

QC improves both moments (Mean, Std Dev).

Most dramatically, it affects the Std Dev. Absolute values are sensitive to QC.


QC & Monitoring of in situ SST:Need community consensus

GHRSST coordinates efforts towards community consensus - Quality Control of in situ SST

- Match-up procedures and practices

NESDIS contribution- Implement in situ QC consistent with the UK Met Office

- Set up near-real time web-based in situ SST Quality Monitor (iQuam) http://www.star.nesdis.noaa.gov/sod/sst/iquam/

- Monitor QC’ed data against global L4 SST (currently, daily Reynolds)

- Serve QC’ed in situ data to SST community, in near-real time

Work in progress- Validate L2/3/4 products against in situ SSTs, using consistent QC

and match-up procedures

- Evaluate match-up criteria and iQuam QC through sensitivity studies; Adjust as needed; Seek community consensus

- Add ARGO floats to iQuam


iQuam webpage http://www.star.nesdis.noaa.gov/sod/sst/iquam/

L2/3 vs. L4

Validation against global L4



Validation against global L4 SSTswas first considered because…

…in situ SSTs have limitations

Non-uniform / Suboptimal quality (may be worse than satellite

SSTs)

Sparse, geographically biased, and do not cover retrieval domain,

fully and uniformly

Not available in sufficient numbers in near-real time (limited

statistics). Diagnostics delayedUsing L4 products provides global snapshot (map) of TSAT

performance, in near-real time, with uniform global coverage.

Also, QC & smoothing done at L4 production stage. Errors of

L4 SST are more uniform in space & time than for in situ data.


ΔTS=TSAT - TREF are mapped to identify issues in data, in near-real time (e.g., cold biases likely due to residual cloud)

http://www.star.nesdis.noaa.gov/sod/sst/squam/


NIGHT

DAY

NIGHT

DAY

O&SI SAF SST minus OSTIAMetop-A FRAC 16-Oct-2010


NIGHT

DAY

NIGHT

DAY

NESDIS ACSPO SST minus OSTIAMetop-A FRAC 16-Oct-2010


NIGHT NIGHT NIGHT

DAY DAYDAY

O&SI SAF: ~40 M clear pixels/night and ~40 M/day.

At night, ACSPO produces ~15% clear-sky ocean pixels than O&SI SAF. Global median biases and Robust Std Dev (RSD) wrt. OSTIA are comparable.

During daytime, # of clear-sky ocean pixels comparable in ACSPO and O&SI. Global median biases: comparable; RSD: slightly smaller for ACSPO.

O&SI SAF ACSPO

O&SI SAF ACSPON

um

ber

of

Ob

serv

atio

ns

(x 1

.0e7

)

Rob

ust

Sta

nd

ard

Dev

iati

on

Med

ian

Dif

fere

nce

Statistics of ΔTS for O&SI SAF and ACSPOMetop-A FRAC minus OSTIA


Pathfinder v5.0 (Day) - Reynolds SST

http://www.star.nesdis.noaa.gov/sod/sst/squam/PF/

Reynolds OISST uses Pathfinder as input. Still, the two products may differ by several tenths K


“Pathfinder v5.0 (Day) - OISST” vs. Wind SpeedPla

tform

sYear

Wind Speed (ms-1)

NOAA-17 mid-morning platform – Diurnal warming suppressed


L2-L4 statistics for various L2s(L4 = OSTIA)

SST systems

* +Attributes

8 November 2010 20SST Science Team Meeting

Red: DayBlue: Night

Avg. num. of retrievals/24hr

Avg. Median Diff. (K)

Avg. Std Dev (K)

Robust Conv.

NESDIS MUT 73,00054,000

0.25-0.02

0.430.28

0.570.42

NAVO SEATEMP 163,000153,000

0.230.06

0.400.30

0.530.37

ACSPO GAC 2,500,0002,200,000

0.13-0.03

0.500.34

0.570.40

Pathfinder v5(L3P)

2,700,000(N_L2: 6,060,000)

1,843,000(N_L2: 4,100,000)

0.12-0.07

0.450.47

0.540.52

ACSPO FRAC 42,500,00044,500,000

0.140.04

0.450.33

0.590.47

O&SI SAF FRAC 42,900,00037,000,000

0.140.06

0.470.33

0.630.48

+ Average NOAA-18 is shown. Outliers not removed

SST systems

* +Attributes

8 November 2010 21SST Science Team Meeting

Red: DayBlue: Night

Avg. num. of retrievals/24hr


Avg. Std Dev (K)

Robust Conv.

NESDIS MUT 73,00054,000

0.22-0.05

0.430.37

0.530.49

NAVO SEATEMP 163,000153,000

0.220.04

0.380.33

0.47**0.39**

ACSPO GAC 2,500,0002,200,000

0.11-0.06

0.490.42

0.550.47

Pathfinder v5(L3P)

2,700,000(N_L2: 6,060,000)

1,843,000(N_L2: 4,100,000)

0.11-0.10

0.480.48

0.540.54

ACSPO FRAC 42,500,00044,500,000

0.160.05

0.460.40

0.550.52

O&SI SAF FRAC 42,900,00037,000,000

0.190.06

0.520.45

0.670.59

L2-L4 statistics for various L2s(L4 = Reynolds “AVHRR”)

+ Average NOAA-18 is shown. Outliers not removed **NB: 2006-pr OISST uses NAVO SEATEMP as input


Observations from validation against global L4 SSTs

Choice of L4 affects Mean and Std Dev statistics of L2-L4

Absolute values are L4-specific.

However, relative performance of L2 products can be

evaluated using (L2-L4) analyses, with any L4.

Main advantage of L2 vs. L4: Product performance is

monitored in near real time, using a global instantaneous view.

Work in progress

Add MODIS and VIIRS SSTs to SQUAM.

Sensitivity to reference L4 calls for “L4 vs. L4” evaluation.

Although beyond our initial plans, our NCEP RTG

colleagues suggested adding L4 comparisons to SST Quality

Monitor.

L4 vs. L4

Cross-evaluation of various L4s



Contribution to GHRSST Inter-Comparison Technical Advisory Group

http://www.star.nesdis.noaa.gov/sod/sst/squam/L4/

“OSTIA – GMPE” mean zonal difference


Year

Latitude

“L4s – GMPE”: time series statistics


More time-series at: www.star.nesdis.noaa.gov/sod/sst/squam/L4/l4_delsst_timeseries.htm

DOI_AVDOI_AARTG_HRRTG_LRGOESPOESOSTIACMC 0.2NAVO K10ODYSSEAGMPE

Reynolds OISST (AVHRR)Reynolds OISST (+ AMSR-E)Real Time Global high resolutionRTG low resolutionBlended POES and GOESOperational SST and Sea ice analysisCanadian Met. Centre 0.2 degreeNAVOCEANO 1/10 degreeMERSEA IFREMER/CERSATGHRSST Median Ensemble Product

Roughly, L4 products form 3 major groups:

DOI_AV, DOI_AA, RTG_LR, NAVO K10

RTG_HR, GOES-POES blended (with seasonal variation between: RTG_HR, RTG_LR)

OSTIA, CMC, & GMPE*ODYSSEA: production temporarily halted

Mean

Year

Std

Dev

Year

wrt GHRSST GMPE wrt GHRSST GMPE

Match-up statistics in SQUAM for various daily L4 products wrt. GMPE

SST systems

* +Attributes

Type/resolutionInput


Avg. Std Dev(K)

*Robust Conv.

Reynolds (AVHRR) Bulk, 0.25°IR (PF till `05, then NAVO), & in situ

0.02 0.31 0.48

Reynolds (+AMSR-E) Bulk 0.25°AVHRR IR, AMSR-E MW, & in situ

0.04 0.33 0.47

RTG low resolution Bulk, 0.5°AVHRR IR, & in situ

-0.09 0.29 0.57

NAVO K10 Bulk, 0.1°AVHRR, VISSR, AMSR-E, JPL cli.

0.01 0.27 0.43

POESGOES blended Bulk, 0.1°AVHRR, GOES

-0.06 0.27 0.52

RTG high resolution Bulk, 1/12°AVHRR IR physical retr., & in situ

-0.06 0.27 0.57

OSTIAFoundation, 0.05°

IR: AVHRR, AATSR, SEVIRIMW: AMSR-E, TMI, SSMI ice, & in situ

0.0 0.17 0.31

CMCFoundation, 0.2°

IR: AVHRR, AATSRMW: AMSR-E, & in situ & CMC sea/ice

-0.02 0.15 0.31

ODYSSEASubskin, 0.1°

IR: AATSR, AVHRR, VISSR, SEVIRI, MW: AMSR-E, TMI

-0.03 0.29 0.40

8 November 2010 27SST Science Team Meeting*robust parameters are resistant to outliers, may hide local issues. Should be used with additional diagnostics


Conclusion to L4 vs. L4 Comparisons

Currently, nine L4 daily products are continuously monitored in

SQUAM, wrt. each other.

Foundation SST products (OSTIA, CMC) appear more stable in

time, less noisy in space, and more consistent with satellite data.

G1SST is being evaluated (G1SST). RSS MW is in pipeline.

Future plans…

Validate all L4 SSTs against independently QCed in situ data (iQUAM).

Explore DV model. This should reduce cross-platform biases and spurious

noise in “L2-L4”, and will help to globally validate the DV model.

Cooperate with L4 producers to add missing L4s

Radiances

Monitoring Clear-Sky Radiances



Monitoring of IR Clear-sky Radiances over Oceans for SST (MICROS)

Web-Based NRT tool to monitor M-O bias- M (Model) = Community Radiative Transfer Model (CRTM) used in

ACSPO to simulate TOA Brightness Temperatures.

- O (Observation) = AVHRR Clear-Sky BTs in Ch3B, 4 & 5.

Key objectives- Fully understand and reconcile CRTM and AVHRR BTs

- Minimize cross-platform biases -

Users/Applications- Test and improve ACSPO products

- Validate and improve CRTM performance

- Contribute to sensor characterization and inter-calibration within Global Space-based Inter-Calibration System (GSICS)


MICROS Overview

www.star.nesdis.noaa.gov/sod/sst/MICROS/

MICROS is end-to-end system


M-O Biasesand Double-Differences

Warm M-O biases are due to a combined effect of incomplete model (aerosols not included; bulk SST used instead of skin; daily mean Reynolds SST is used to represent nighttime SST) and biased satellite sensor radiances (residual cloud).

Double differences (DDs) cancel out many possible systematic errors in CRTM and its input (SST and GFS, missing aerosol, etc).

Non-zero DDs are mainly due to errors in sensor calibration and spectral response functions. Largest systematic errors are in N18 (Ch4) and N19 (all bands).

NOAA-16 is unstable in whole monitoring period.


The effect of CRTM input on M-O bias (Reynolds SST used as CRTM input)

μ and σ: median and RSD over time

Spurious variations found when Reynolds SST used as CRTM input.


The effect of reference SST on M-O bias (OSTIA SST used as CRTM input)

Spurious time variability reduced when OSTIA used as CRTM input.The Std Deviations are also dramatically reduced.


Conclusion to Monitoring Radiances

SST is an unresolved combination of 2-3 sensor bands. Monitoring

individual bands is needed to unscramble SST anomalies.

Web-based near-real time Monitoring of IR Clear-sky Radiances

over Oceans for SST (MICROS) established

http://www.star.nesdis.noaa.gov/sod/sst/micros/

Currently, three users groups actively use MICROS

- ACSPO SST developers

- CRTM developers

- Sensor calibration scientists

Future plans…

Minimize M-O biases through: Adding aerosol in CRTM; Improving

AVHRR sensor characterization; Improving CRTM accuracy

Conclusion

Community Consensus

SSTs & Radiances



Conclusion

Community consensus methodologies and tools for SST

evaluation and validation are needed

Prototypes have been established at NESDIS

- SST Quality Monitor (SQUAM) for L2 & L4 products

http://www.star.nesdis.noaa.gov/sod/sst/squam/

- In situ SST Quality Monitor (iQuam)

http://www.star.nesdis.noaa.gov/sod/sst/iquam/

Satellite radiances in individual sensor bands should be monitored

- Monitoring of IR Clear-sky Radiances over Oceans for SST

http://www.star.nesdis.noaa.gov/sod/sst/micros/

We are open to any suggestions to make SQUAM, iQuam,

and MICROS the “community tools”

Questions?

Thank you!



Back-Up slides


Std Dev of “In Situ – Daily Reynolds SST” stratified by

in situ data types

All SSTs are strongly contaminated by outliers (drifters affected most).

Effect of QC on NCEP GTS SST(in situ minus daily Reynolds)

Ships

Drifters

Tropical Moorings

Coastal Moorings


Percent of in situ data excluded by QC

Percent of outliers in drifter SSTs is largest–

QC is needed most.

Percent of NCEP GTS dataexcluded by QC

Ships

Drifters

Tropical Moorings

Coastal Moorings


Pathfinder v5.0 (Night) - Reynolds SST


Pathfinder v5 was recently added to SQUAM


Pathfinder v5.0 (Day) - Reynolds SST


Pathfinder v5 was recently added to SQUAM

Documents

Towards Community Consensus SSTs and Clear-Sky Radiances from AVHRR