1. EARWiG: SST retrieval issues for TWP Andy Harris Jonathan Mittaz Prabhat Koner NOAA-CICS University of Maryland Eileen Maturi NOAA/NESDIS, Camp Springs, MD, United States

2. Issues for IR SST Water vapour… –Makes cloud detection more difficult –Makes SST retrieval more difficult Suppresses signal from surface… Two types of retrieval: SST = a 0 + a T yΔSST = Gδy

3. Sensitivity to SST 1:1 is desirable –especially for diurnal warming studies

4. Regression retrieval For daytime, usually 11 & 12 micron Nighttime adds 3.9 micron

5. Sensitivity to Water Vapour 0:1 is desirable –Requires orthogonal relationship

6. Physical retrieval vs regression for Geo-SST Localized linear retrieval using CRTM and NWP profiles If system is “linear”, i.e.: δy = Kδx Then δx can be retrieved K is matrix of partial derivatives of y w.r.t. x Some variant of normal least-squares equation δx = (K T K) -1 K T δy

13. 2-channel bias

14. Summary IR issues in TWP mostly concern water vapour –Signal can be dominated by atmosphere Reduces contrast Makes detection of low cloud more difficult –Regression is “saved” by the close coupling of Air & sea temperature Water vapour and SST –Geophysical correlations mean that sensitivity to SST change is not always 1:1 Responses to SST & water vapour not orthogonal N.B. have ignored changes in water vapour distribution & air temperature

15. Summary cont’d Physical retrieval offers a solution –Can ensure SST response is ~1:1 May introduce noise –Important to remove sources of bias Calibration/characterization Unmodeled effects (e.g. aerosol) –Get better results with 3 channels in daytime than just 11/12 micron –N.B. can derive G empirically – incremental regression MW SST –Rain flagging –All those islands & atolls

16. 15S 35S 150W 120W 15 20 25 30

17. Backup Slides Condition number vs transmittance

18. Early theory required SST – T i = k i F(atm) This allowed SST = k 2 T 1 – k 1 T 2 ——————————— (k 2 – k 1 ) And hence the “split- window” equation, mystique about channel differences, etc. Only need to assume SST – T i  SST – T j to get SST = a 0 +  a i T i Some refinements to account for non-linearity, scan angle: SST = a 0 + a 1 T 11 + SST bg a 2 (T 11 – T 12 ) + (sec(ZA)-1)a 3 (T 11 – T 12 ) “Gamma”

19. The “Gamma” Parameter NLSST Gamma is very smooth – mirrors climatological SST “True” Gamma has more detailed structure N.B. Difference in Gammas must be multiplied by T 11 – T 12

20. How sensitive is NLSST to true SST? If SST changes by 1 K, does retrieved NLSST change by 1 K? CRTM provides tangent-linear derivatives Response of NLSST algorithm to a change in true SST is… Merchant, C.J., A.R. Harris, H. Roquet and P. Le Borgne, Retrieval characteristics of non- linear sea surface temperature from the Advanced Very High Resolution Radiometer, Geophys. Res. Lett., 36, L17604, 2009

21. NLSST Sensitivity to true SSTAir – Sea Temperature Difference

22. Physical retrieval MAP (Maximum A Posteriori) x is reduced state vector [SST(x), TCWV(w)] T y o is observation vector, y b from CRTM+NWP K is Jacobian matrix (∂y i /∂x, ∂y i /∂w) ← CRTM S ε = cov (RTM+instr err), S b cov bkgnd (x, w err) δx= [K T S ε -1 K + S b -1 ] -1 K T S ε -1 (y o - y b ) →δx= a 1 b δy 3.9 + a 2 b δy 11

24. “Flagship” heritage satellite SST product derived from AVHRR data Retrieval coefficients a i are derived empirically by regression of AVHRR channel brightness temperatures matched to in situ [buoy] data Coefficients are time-dependent (5-month rolling average) The AVHRR Pathfinder SST Pathfinder NLSST retrieval algorithm: NLSST = a 0 + a 1 T 11 + SST bg a 2 (T 11 – T 12 ) + (sec(ZA)-1)a 3 (T 11 – T 12 ) Direct regression minimizes effects of mis-calibration & sensor characterization for the training data “Gamma”

25. Apply to GOES data

26. Radiance biases

27. δT = a 0 + a 1 ΔT a + a 2 T b + a 3 S + a 4 ΔT a T b + a 5 ΔT a S ΔT a = SST b - T b

28. Post bias correction

29. c.f. Regression SST Less noisy, less biased –How much is due to bias correction? Top priority: fix bias correction at source