1. 4 June 2009GHRSST-X STM - SQUAM1 The SST Quality Monitor (SQUAM) 10 th GHRSST Science Team Meeting 1-5 June 2009, Santa Rosa, CA Alexander “Sasha” Ignatov*, Prasanjit Dash*, John Sapper**, Yury Kihai* NOAA/NESDIS *Center for Satellite Applications & Research (STAR) **Office of Satellite Data Processing & Distribution (OSDPD)

2. 4 June 2009GHRSST-X STM - SQUAM2 NESDIS Operational AVHRR SST Products  Heritage Main Unit Task (MUT) -1981 - present (McClain et al., 1985; Walton et al., 1998).  New Advanced Clear-Sky Processor for Oceans (ACSPO) -May 2008 – present http://www.star.nesdis.noaa.gov/sod/sst/squam/ Employ L4 SSTs (Reynolds, RTG, OSTIA, ODYSSEA,..) to Evaluate MUT and ACSPO SST products in near-real time for self-, cross-platform and cross-product consistency Identify product anomalies & help diagnose their causes (e.g., sensor malfunction, cloud mask, or SST algorithm) Objective of the SST Quality Monitor (SQUAM)

3. 4 June 2009GHRSST-X STM - SQUAM3 Customarily, satellite SSTs are validated against in situ SSTs However, in situ SSTs have limitations  They are sparse and geographically biased (cover retrieval domain not fully and non-uniformly).  Have non-uniform and suboptimal quality (often comparable to or worse than satellite SSTs).  Not available in near real time in sufficient numbers to cover the full geographical domain and retrieval space.

4. 4 June 2009GHRSST-X STM - SQUAM4 AVHRR SST MetOp-A GAC, 3 January 2008 (Daytime) Heritage MUT SST product ACSPO SST product SST imagery is often inspected visually for quality and artifacts. Large-scale SST background dominates making it not easy to discern “signal” from “noise”.

5. 4 June 2009GHRSST-X STM - SQUAM5 Heritage MUT SST product Mapping deviations from a global reference field constrains the SST “signal” and emphasizes “noise”. This helps reveal artifacts in SST product (cold stripes at swath edges). Removing large-scale SST background (daily 0.25 º Reynolds) emphasizes ‘noise’ ACSPO SST product

6. 4 June 2009GHRSST-X STM - SQUAM6 View angle dependence of ‘MUT - daily Reynolds SST’ (NOAA-17) Such ‘retrieval-space’ dependent biases are difficult to uncover and quantify using customary validation against in situ data, which do not fully cover the retrieval space. The SQUAM diagnostics helped uncover a bug in the MUT SST which was causing across-swath bias >0.7K. After correction, bias reduced to ~0.2K and symmetric with respect to nadir.

7. 4 June 2009GHRSST-X STM - SQUAM7 Use global L4 SST products to quantitatively evaluate satellite SST  Satellite & reference SSTs are subject to near-Gaussian errors T SAT = T TRUE + ε SAT ; ε SAT = N(μ sat,σ sat 2 ) T REF = T TRUE + ε REF ; ε REF = N(μ ref,σ ref 2 ) where μ’s and σ’s are global mean and standard deviations of ε‘s  The residual is distributed near-normally ΔT = T SAT - T REF = ε SAT - ε REF ; ε ΔT = N(μ ΔT,σ ΔT 2 ) where μ ΔT = μ sat - μ ref ; σ ΔT 2 = σ sat 2 + σ ref 2 (if ε SAT and ε REF are independent)  If T REF = T in situ, then it is customary ‘validation’. If (μ ref, σ ref ) are comparable to (μ in situ, σ in situ ), and if ε SAT and ε REF are not too strongly correlated, then T REF can be used to monitor T SAT

8. 4 June 2009GHRSST-X STM - SQUAM8 Global Histograms of T SAT - T REF ( Nighttime MUT)

9. 4 June 2009GHRSST-X STM - SQUAM9 Histogram of SST residual Reference SST: In situ 30 days of data: ~6,500 match-ups with in situ SST Median = -0.04 K; Robust Standard Deviation = 0.27 K

10. 4 June 2009GHRSST-X STM - SQUAM10 8 days of data: ~483,500 match-ups with OSTIA SST Median = 0.00 K; Robust Standard Deviation = 0.30 K Histogram of SST residual Reference SST: OSTIA

11. 4 June 2009GHRSST-X STM - SQUAM11 8 days of data: ~483,700 match-ups with daily Reynolds SST Median = +0.08 K; Robust Standard Deviation = 0.44 K Histogram of SST residual Reference SST: Daily Reynolds

12. 4 June 2009GHRSST-X STM - SQUAM12  Global histograms of T SAT - T REF are close to Gaussian, against all T REF including T in situ  Normal distribution is characterized by location (median) and scale (robust standard deviation, RSD)  Reduced number/magnitude of outliers with respect to L4 T REF compared to T in situ  For some T REF (e.g., OSTIA), VAL statistics is closer to T in situ than for others (e.g., Reynolds). * More histograms (ACSPO/MUT, day/night, other platforms / reference SSTs) are found at SQUAM page Observations from global histograms analyses

13. 4 June 2009GHRSST-X STM - SQUAM13 Time Series Global Median Biases of (T SAT - T REF )

14. 4 June 2009GHRSST-X STM - SQUAM14 Global Median Biases T SAT – T in situ 1 data point = 1 month match-up with in situ Median Bias within ~0.1 K (except for N16 - sensor problems) MetOp-A and N17 fly close orbits but show a cross-platform bias of ~0.1 K

15. 4 June 2009GHRSST-X STM - SQUAM15 1 data point = 1 week match-up with OSTIA SST Patterns reproducible yet crisper (finer temporal resolution) Cross-platform biases: Slightly differ from Val (diurnal cycle) OSTIA artifacts observed in early period (2006-2007) Global Median Biases T SAT – T OSTIA

16. 4 June 2009GHRSST-X STM - SQUAM16 1 data point = 1 week match-up with Reynolds SST Patterns reproducible but noisier than with respect to OSTIA Artifacts also observed but different from OSTIA Global Median Biases T SAT – T Reynolds

17. 4 June 2009GHRSST-X STM - SQUAM17  Number of match-ups is more than two orders of magnitude larger against L4 T REF than against T in situ  Major trends & anomalies in T SAT are captured well against all T REF. More detailed and crisper than against T in situ  Some T REF are “noisier” for VAL purposes than others. Different artifacts are seen in different T REF  Nevertheless, time series of ( T SAT – T REF ) can be used to monitor T SAT for cross-platform & cross-product consistency * More time series (ACSPO/MUT, other reference SSTs) are available from SQUAM page Observations from time series of global biases

18. 4 June 2009GHRSST-X STM - SQUAM18  Cross-platform consistency of T SAT can be evaluated from time series of T SAT - T REF overlaid for different platforms  For more quantitative analyses, one ‘reference’ platform can be selected & subtracted from all other ( T SAT - T REF )  N17 was selected as ‘reference’, because it is available for the full SQUAM period, and its AVHRR is stable  Double-differences (DD) were calculated as DD = ( T SAT - T REF ) - ( T N17 - T REF ) for SAT=N16, N18, and MetOp-A Cross-Platform Consistency Using Double-Differences (T SAT – T SAT_REF )

19. 4 June 2009GHRSST-X STM - SQUAM19 Global Median Biases T SAT – T in situ Same as slide 14

20. 4 June 2009GHRSST-X STM - SQUAM20 In situ Double-Differences (T SAT – T in situ ) - (T N17 – T in situ ) Biases are due to errors in T SAT and T SAT /T in situ skin/bulk differences Before mid-2006, all SSTs agree to within ~0.01 K In 2006, N16 develops a low bias up to ~-0.7 K, and N18 and MetOp- A a warm bias up to ~+0.1 K

21. 4 June 2009GHRSST-X STM - SQUAM21 OSTIA Double-Differences (T SAT – T OSTIA ) - (T N17 – T OSTIA ) DD’s with respect to global reference fields: Errors in T SAT + Missing diurnal signal in T REF (T REF do not resolve diurnal cycle) N16: sensor problems. MetOp-A: suboptimal regression coefficients Diurnal correction to T REF is needed to rectify inconsistencies in T SAT

22. 4 June 2009GHRSST-X STM - SQUAM22 Reynolds Double-Differences (T SAT – T Reynolds ) - (T N17 – T Reynolds ) DD’s are consistent for different T REF (biases/noises in T REF largely cancel out in calculating DD’s)

23. 4 June 2009GHRSST-X STM - SQUAM23  In situ DD’s are close to ‘true’ cross-platform bias in T SAT (bulk T in situ partially accounts for diurnal cycle in skin T SAT )  DD’s with respect to global T REF additionally include diurnal signal (current L4 T REF do not resolve diurnal cycle)  Employing diurnal-cycle resolved T REF in DD’s (or adding diurnal correction on the top of existing T REF ) should rectify the ‘true’ cross-platform inconsistency in T SAT  The DD’s provide quick global ‘validation’ of the diurnal cycle model (e.g., Gentemann et al, 2003; Kennedy et al, 2007; Filipiak and Merchant, 2009) Observations from Satellite-to-Satellite Double Differences

24. 4 June 2009GHRSST-X STM - SQUAM24  Day-Night consistency of T SAT can be evaluated as DD = ( T DAY - T REF ) - ( T NIGHT - T REF ) Day-Night Consistency Using Double-Differences T DAY – T NIGHT

25. 4 June 2009GHRSST-X STM - SQUAM25 In situ Day-Night Double-Differences (T DAY – T in situ ) - (T NIGHT – T in situ ) During daytime, all platforms show a warmer ~+(0.1±0.1) K bias (except for N16 – sensor problem) Seasonal structure seen in DD’s Different capturing of diurnal cycle by skin T SAT and bulk T in situ

26. 4 June 2009GHRSST-X STM - SQUAM26 OSTIA Day-Night Double-Differences (T DAY – T OSTIA ) - (T NIGHT – T OSTIA ) Day-Night DD’s wrt OSTIA show biases due to diurnal warming Seasonal variability seen in all DD’s For N17 and MetOp-A (~10am/pm), diurnal signal is (+0.1±0.1) K For N18 (~2am/pm), diurnal signal is (+0.3±0.1) K

27. 4 June 2009GHRSST-X STM - SQUAM27 Reynolds Day-Night Double-Differences (T DAY – T Reynolds ) - (T NIGHT – T Reynolds ) DD’s are closely reproducible for all T REF (biases/noise in T REF largely cancel out in calculating DD’s)

28. 4 June 2009GHRSST-X STM - SQUAM28  DD’s wrt in situ data more closely represent cross-platform inconsistencies in T SAT, less difference in the diurnal  If global T REF is used, then DD’s additionally include diurnal signal (currently, T REF ‘s do not resolve diurnal cycle)  Employing diurnal-cycle resolved T REF in DD’s is expected to improve cross-platform consistency  The DD’s provide quick global ‘validation’ of the diurnal cycle model (e.g., Gentemann et al, 2003; Kennedy et al, 2007; Filipiak and Merchant, 2009) Observations from Day-Night Double Differences

29. 4 June 2009GHRSST-X STM - SQUAM29  Validation against global reference fields is currently employed in SQUAM to monitor two NESDIS operational AVHRR SST products, in near-real time  It helps quickly uncover SST product anomalies and diagnose their root causes (SST algorithm, cloud mask, or sensor performance), and leads to corrections Summary and Future Work  Work is underway to reconcile AVHRR & reference SSTs -Improve AVHRR sensor calibration -Adjust T REF for diurnal cycle (e.g., Kennedy et al., 2007) -Improve SST product (cloud screening, SST algorithms) -Provide feedback to T REF producers Objective is to have a single “benchmark” SST in NPOESS era  Add NOAA-19 and eventually MetOp-B, -C and VIIRS to SQUAM  We are open to integration with GHRSST and collaboration (to test other satellite & reference SSTs, diurnal correction,..)

30. 4 June 2009GHRSST-X STM - SQUAM30  SQUAM page http://www.star.nesdis.noaa.gov/sod/sst/squam/ Real time maps, histograms, time series (including double differences), dependencies http://www.star.nesdis.noaa.gov/sod/sst/squam/  CALVAL page http://www.star.nesdis.noaa.gov/sod/sst/calval/ Cal/Val of MUT and ACSPO data against in situ SST (currently, password protected but will be open in 2-3 months)http://www.star.nesdis.noaa.gov/sod/sst/calval/  MICROS page http://www.star.nesdis.noaa.gov/sod/sst/micros/ (Monitoring of IR Clear-sky Radiances over Oceans for SST) Validation of SST Radiances against RTM calculations with Reynolds SST and NCEP GFS inputhttp://www.star.nesdis.noaa.gov/sod/sst/micros/ NESDIS NRT SST analyses on the web