1. 1 GOES-R AWG Ocean Dynamics Team: Ocean Currents and Offshore Ocean Currents Algorithm June 14, 2011 Presented By: Eileen Maturi 1, Igor Appel 2, Andy Harris 3 1 NOAA/NESDIS/STAR 2 NOAA/NESDIS/STAR/IMSG 3 University of Maryland/CICS GOES-R AWG Annual Meeting, Fort Collins, Colorado, June 14-16, 2011

2. 2 Outline  Executive Summary (1 slide)  Algorithm Description (2-4 slides)  ADEB and IV&V Response Summary (1-2 slides)  Requirements Specification Evolution (2 slides)  Validation Strategy (3-5 slides)  Validation Results (3-5 slides)  Summary (1-2 slides) GOES-R AWG Annual Meeting, Fort Collins, Colorado, June 14-16, 2011

3. 3 Executive Summary  The Ocean Dynamic Algorithm (ODA) generates the Option 2 products of Ocean Currents and Ocean Currents: Offshore.  Algorithm Version 4 was delivered in March. ATBD (100%) is on track for a June 30 delivery  A 1-channel IR-only approach has been developed that utilizes improved ABI resolution capabilities over GOES-IM and GOES-NOP.  Validation tools have been developed and applied to 4 weeks of SEVIRI data.  Comparisons with output from the global version of the Navy Coastal Ocean Model (NCOM) indicate 80% spec compliance for all products. GOES-R AWG Annual Meeting, Fort Collins, Colorado, June 14-16, 2011

4. 4 Algorithm Description GOES-R AWG Annual Meeting, Fort Collins, Colorado, June 14-16, 2011

5. 5 Algorithm Summary  This Ocean Dynamic Algorithm (ODA) generates the Option 2 products Ocean Currents and Ocean Currents: Offshore  We use an IR-only approach that provides day/night consistency and takes advantage of the ABI’s higher resolution information  We use three images to monitor motion of the ocean features characterized by the maximum gradient in sea surface brightness temperature  SEVIRI 10.8 micron channel is a proxy for ABI channel 14 (11.2 micron)  A cloud mask is applied to identify cloud free areas GOES-R AWG Annual Meeting, Fort Collins, Colorado, June 14-16, 2011

6. 6 Ocean Currents Mathematical Description: Feature Tracking I 1 =Temperature at pixel (x 1,y 1 ) of the target scene I 2 =Temperature at pixel (x 2,y 2 ) of the search scene Sum-of-Squared Differences (SSD) Summation is carried out for all possible target scene positions within the search region Summation is carried out for all possible target scene positions within the search region The matching scene corresponds to the scene where the function takes on the smallest value The matching scene corresponds to the scene where the function takes on the smallest value Search Region Original Target Scene Matching Scene GOES-R AWG Annual Meeting, Fort Collins, Colorado, June 14-16, 2011

7. 7 EXAMPLE of SUM SQUARED DIFFERENCE RESULTS 1.Example SSD for 5×5 pixel target applied to t 0 &  3 hr, 2.Note displacements are ~symmetrical c.f. zero displacement 3.This is a good consistency check GOES-R AWG Annual Meeting, Fort Collins, Colorado, June 14-16, 2011

8. 8 Ocean Dynamics Algorithm Processing Schematic Read GOES-R IR & SST data Apply Cloud Mask & Land Mask Ocean Currents Algorithm (feature detection, pattern matching – SSD*, vector derivation, QC) Output Ocean Current Vectors *SSD = Sum of Squared Differences GOES-R AWG Annual Meeting, Fort Collins, Colorado, June 14-16, 2011

9. 9 Qualifiers Theses are made with the following qualifiers  Sensor zenith angle < 67 degrees  Region must be cloud-free in image triplet  Cloud & land masks available  Dependent on navigation accuracy  Errors depend on strength of SST gradient in feature being tracked GOES-R AWG Annual Meeting, Fort Collins, Colorado, June 14-16, 2011

10. 10 Expected ABI Performance Relative to Other Sensors Current GOES Imager has approximately 4-km resolution so ABI has a 4x better resolution –Critical, especially for Ocean Currents where image-to-image displacements are small –N.B. MSG-SEVIRI has resolution of 3 km Navigation accuracy of ABI expected to be much better (750 m) –Again, critical for Ocean Currents because image-to-image registration errors contribute directly to retrieval error GOES-R AWG Annual Meeting, Madison, Wisconsin, June 14-16, 2011

11. 11 Next Steps to Reach 100%  Refine internal QC metrics based on various parameters (e.g. gradient strength)  Investigate impact of navigation errors on performance using proxy data to ascertain if expected ABI navigation improvement will reduce errors sufficiently to meet 100% specification »Careful co-registration of sample of individual SEVIRI images for triplets which contain notable retrieval errors GOES-R AWG Annual Meeting, Fort Collins, Colorado, June 14-16, 2011

12. Example ODA Output 12 GOES-R AWG Annual Meeting, Fort Collins, Colorado June 14-16, 2011 North Africa July 8, 2005 Example Ocean Current Vectors derived from MSG- SEVIRI image triplet centered at 12Z Vector length indicates derived current strength (1 degree = 1 m.s -1 ) Shows upwelling off N Africa, combined with complex current pattern in the vicinity of the Canary Islands Canary Islands

13. 13 Algorithm Changes from 80% to 100%  Added multi-layer cloud analysis that accounts for height of the lower cloud layer. At the 80% version, the lower cloud height was assumed to be 2 km above the surface.  Metadata output added.  Quality flags standardized. GOES-R AWG Annual Meeting, Fort Collins, Colorado, June 14-16, 2011

14. 14 ADEB and IV&V Response Summary “OK for advance to 100%, with caveat that team pursue further validation with SEVIRI and also explain N-S vs E-W asymmetry”  Validation expanded to use HYCOM 1/12 degree resolution DA Experiment results for surface currents  Also validation using CODAR with vectors derived from GOES-Imager  N-S / E-W variations in geolocation accuracy are suspected and this is being investigated GOES-R AWG Annual Meeting, Fort Collins, Colorado, June 14-16, 2011

15. 15 Requirements ProductAccuracyPrecisionLatencyhorizontal resolution Ocean Current 0.3 m.s -1 (U & V) 3 hrs2 km Ocean Current: Offshore 0.3 m.s -1 (U & V) 3hrs2 km GOES-R AWG Annual Meeting, Fort Collins, Colorado, June 14-16, 2011

16. 16 Qualifiers Theses are made with the following qualifiers.  Sensor zenith angle < 65 degrees  Cloud optical thickness > 1  Cloud mask and cloud type available GOES-R AWG Annual Meeting, Fort Collins, Colorado, June 14-16, 2011

17. 17 Validation Approach GOES-R AWG Annual Meeting, Fort Collins, Colorado, June 14-16, 2011

18. 18 Validation Strategy: Comparison of MSG-derived current vectors vs. Model  Use ocean current vector components from the global version of the Navy Coastal Ocean Model (NCOM) »Advantage of using model field is that vectors are available “everywhere” »Potential for model-related biases (e.g. displaced currents, errors in forcing fields). However, NCOM does assimilate data (e.g. Altimetry) and is therefore constrained by it »“Global” NCOM is high resolution (at least eddy-resolving) GOES-R AWG Annual Meeting, Madison, Wisconsin, June 7-10, 2010

19. 19 Median = 0.13 R.S.D. = 0.348Median = 0.06 R.S.D. = 0.297 Orange = gaussian curve represented by S.D. Green = gaussian represented by R.S.D. Blue = histogram of errors (MSG – NCOM) Note: green curve matches peak & central distribution Validation Results (2) 5x5 window (Meteosat-8 SEVIRI vs NCOM, 2005 days 2-9, 92-99, 183-190, 275-282) GOES-R AWG Annual Meeting, Fort Collins, Colorado, June 14-16, 2011

20. 20 GOES-R AWG Annual Meeting, Fort Collins, Colorado, June 14-16, 2011 Validation Results (3) 5x5 window (Meteosat-8 SEVIRI vs NCOM, 2005 days 2-9, 92-99, 183-190, 275-282) Note errors in U-component are more prevalent in certain geographical areas V-component errors display little geographical preference

21. 21 GOES-R AWG Annual Meeting, Fort Collins, Colorado, June 14-16, 2011 DEPENDENCE OF RETRIEVAL QUALITY ON SST GRADIENTS – 5x5 WINDOW 80% 100%  =0.3  =0.5

22. 22 PERCENTAGE OF REQUIREMENTS MET Mean US.D. UMean VS.D. V%<0.375 WINTER-0.10020.2218-0.03300.212590.4492.43 SPRING-0.24480.3452-0.02860.295376.2783.05 SUMMER-0.05180.43710.03520.278158.9080.82 AUTUMN-0.17950.3270-0.07700.341477.5270.54 YEAR-0.12990.3078-0.03390.270679.4983.98 GOES-R AWG Annual Meeting, Fort Collins, Colorado, June 14-16, 2011 Validation statistics for 5x5 window excluding weak (<0.5 K) gradients GREEN CELLS=100%YELLOW CELLS=80%RED CELLS=< 80% Meteosat-8 SEVIRI vs NCOM, 2005 days 2-9, 92-99, 183-190, 275-282

23. NEW Validation Strategy: Comparison of MSG-derived current vectors vs. HYCOM 23 GOES-R AWG Annual Meeting, Fort Collins, Colorado, June 14-16, 2011 Filtered using QI>30 (retains 72%) Days 182 – 189 (“Summer”) Every 30 minutes (previously just 12Z) tests effect of day & night cloud masking Median = 0.02 R.S.D. = 0.258Median = -0.03 R.S.D. = 0.247

24. 24 Summary  100% specification requirement is met using a 5x5 target when considering robust statistics (vs. HYCOM data) and “mild” filter on Quality Indicator »Retains 72% for QI>30 (range is 0 – 100)  HYCOM validation has been performed for 30 min intervals throughout day c.f. previous 12Z-only »Increased outliers may be due to variations in cloud mask performance – investigations ongoing  Delivery of Version 5 Algorithm (100%) is on track  Various algorithm aspects (e.g. QC) will be refined in order to meet 100% requirement (e.g. outlier rejection) GOES-R AWG Annual Meeting, Fort Collins, Colorado, June 14-16, 2011

25. 25 Summary cont’d Working with CIOSS on GOES-R Risk Reduction  Analyze the representativeness of feature-tracked motions to ocean currents in a 1-km resolution regional ocean model and the findings will be reported in the next version of the ATBD  Also explore different variants of feature-tracking and possible transfer functions from motion to surface current  Also explore data assimilation approach GOES-R AWG Annual Meeting, Fort Collins, Colorado June 14-16, 2011