1. Prototyping SST Retrievals from GOES-R ABI with MSG SEVIRI Data Nikolay V. Shabanov 1,2, Nikolay.Shabanov@noaa.gov, (301)763-8102 ×154Nikolay.Shabanov@noaa.gov Alexander Ignatov 1, Boris Petrenko 1,2, Yuri Kihai 1,3, XingMing Liang 1,4, Wei Guo 1,2, Feng Xu 1,4, Prasanjit Dash 1,4, Michael Goldberg 1, John Sapper 5 1 NOAA/NESDIS/STAR; 2 IMSG Inc; 3 Perot Systems Government Services; 4 Colorado State University- CIRA; 5 NOAA/NESDIS/OSDPD Geostationary Operational Environmental Satellite-R Series (GOES-R) will carry Advanced Baseline Imager (ABI) onboard. Sea Surface Temperature (SST) algorithm for ABI is being developed by the SST Team which is a part of the GOES-R Algorithm Working Group (AWG). ABI SST production is prototyped with the Meteosat Second Generation (MSG) Spinning Enhanced Visible and IR Imager (SEVIRI) data (Schmetz et al., 2002). The Advanced Clear-Sky Processor for Oceans (ACSPO) developed at NOAA/NESDIS and currently operational with AVHRR data onboard NOAA-18 and MetOp-A has been adopted to SEVIRI. ACSPO-SEVIRI system processes 15-minute full-disk (FD) SEVIRI Level 1 data in near-real time (NRT) and generates Level 2 clear-sky products over ocean, including top-of-atmosphere (TOA) clear-sky brightness temperatures (BTs), SST and aerosols. This poster evaluates initial BT and SST retrievals from Meteosat-9 SEVIRI for one full month of data in June 2008. The views, opinions and findings contained in this report are those of the authors and should not be construed as an official NOAA or U.S. Government position, policy, or decision. This poster does not reflect the views or policies of the GOES-R Program Office or Algorithm Working Group. ACSPO-SEVIRI System CRTM simulations for SEVIRI longwave bands 9 (11 μm) and 10 (12 μm) are consistent with similar results for the AVHRR. More analyses are needed to better quantify these preliminary observations. M-O bias in Ch4 (3.7 μm) is inconsistent with AVHRR and requires further analyses. Comparison with Global Reference SSTs The SST Quality Monitor (SQUAM) currently routinely generates consistency statistics between the retrieved AVHRR SST and multiple reference SST fields (Dash et al., 2009). SQUAM was applied to quickly evaluate SEVIRI SST retrievals. The following reference SST were used: Daily Reynolds: OISST (AVHRR-based) and OISST-A (AVHRR+AMSR-E), OSTIA, RTG HR (High Resolution) and Low Resolution (LR), and Pathfinder SST climatology. SST Diurnal Cycle Introduction CRTM and BT Simulations Fig. 2. Example of MSG-2 SEVIRI 15-min FD data for June 14, 2008, 2:00pm UMT, processed by the ACSPO-SEVIRI system. (a) Visible true-color image constructed from albedo. Black strip expanding from SE corresponds to night-time; (b) Cloud Mask; (c) SST; (d) AOD retrieved from Ch01 using single-channel algorithm (De Paepe et al., 2008). Dash et al., (2009). The SST Quality Monitor (SQUAM), submitted, RSE. De Paepe et al., (2008). Aerosol retrieval over Ocean from SEVIRI for the use in GERB Earth’s radiation budget analysis, RSE, 112, 2455-2468. Liang et al., (2009). Implementation of CRTM in ACSPO and validation against nighttime AVHRR radiances, submitted, JGR. Merchant et al., (2008). Optimal estimation of sea surface temperature from split- window observations, in press, RSE. Schmetz et al., (2002). An Introduction to Meteosat Second Generation, BAMS, 83, 977-992. Walton et al. (1998). The development and operational application of nonlinear algorithms for the measurement of sea surface temperatures with the NOAA polar-orbiting environmental satellites. JGR, vol. 103(C12), 27999- 28012. Disclaimer Literature (a) (b) (d) (c) Input to ACSPO-SEVIRI system is 15-min SEVIRI FD channel data (optical Ch1-3 and thermal Ch4, 9-10). 1°-resolution weekly Reynolds SST and 6-hour National Centers for Environmental Prediction Global Forecast System (NCEP/GFS) data are used as input to the fast Community Radiative Transfer Model (CRTM) to simulate clear-sky BT in Ch4, 9-10 (Liang et al., 2009). Two SST algorithms are implemented. The Regression (split- window, NLSST) algorithm is based on Walton et al. [1985] equation: Currently used regression coefficients a 0 –a 3 were derived by EUMETSAT for Meteosat- 8. The Physical algorithm uses CRTM to invert BT 9 and BT 10 for SST and water-vapor optical depth scaling factor (Merchant et al., 2008). The regression and physical algorithms will be cross-evaluated, to generate a superior quality hybrid SST algorithm. For this study, a simplified cloud-mask was implemented which currently uses only two tests: SST test (comparison of retrieved and Reynolds SST) and BT test (comparison of CRTM simulated and measured BTs). The end-to-end NRT processing system has been set up for data stream processing: (1) downloading SEVIRI data from NOAA operational servers (in collaboration with AWG Land Team), (2) data processing at SST Team servers, (3) data analysis with web-based QC tools. SEVIRI data for 2008 are online, 2006-2007 are on external disks. The product file contains up to 32 data layers, file size of one FD product file is ~450MB. Execution time is ~7min per FD on DELL PowerEdge 2900. Fig. 6. SST bias as function of Number of Ambient Clear-Sky pixels (NAC; calculated within 21×21 GAC pixels). FD data from 1-7 June 2008 are binned at NAC=20 and averaged for 2 hours centered on Midday and Midnight. Fig.3. Solid lines show FD monthly mean diurnal cycles of “Retrieved minus Reynolds SST”. Individual data points correspond to different days in June 2008. SST=a 0 +a 1 BT 9 +a 2 T Reynolds (BT 9 -BT 10 )+a 3 (BT 9 -BT 10 )(sec  - 1), First Guess SST Reference SST SEVIRI Dynamic Data:  Optical bands [1, 2, 3]  Thermal bands [4, 9, 10]  Illumination geometry (SZA, RAA) (HDF, 5x5 km (at nadir), 15 min) SEVIRI Static Data:  Land/Water mask  View geometry (VZA)  Lat/Lon (HDF, 5x5 km (at nadir)) Ancillary GFS Data:  Air and Surface Temperature  Water vapor  Wind speed  Ozone concentration (HDF, 1°x1°, 6-hours) Regression (Split window) Physical (CRTM-based) CRTM simulations of BT and BT Jacobian at coarse (1°x1°) Grid Cloud Mask Algorithm (7 tests) 1) SST (currently on) 2) BT (on) 3) Ch 3 Albedo (currently off) 4) Reflectance Ratio (off) 5) Three-Five (off) 6) Ultra Low Stratus (off) 7) Uniformity (off) Channel Calibration Bi-Linear Interpolation (BT and BT Jacobian) 1 o x1 o -> 5x5 km (at nadir) Single Channel 6S-based LUTs Output (up to 32 layers/SDS):  TOA albedo in VIS bands (1, 2, 3)  TOA BT at Thermal bands (4, 9,10)  SST (Regression & Physical)  Aerosol Optical Depths in VIS bands (1,2,3)  CRT BTs  Geometry and Geolocation  QC (cloud mask, L/W, day/night flag, etc)  OISST (Reynolds)  GFS fields interpolated to SEVIRI (HDF, 5x5 km (at nadir), 15 min) Ancillary OISST Data:  Reynolds SST (HDF, 1 o x1 o, weekly or 0.25 o x 0.25 o, daily ) SST Algorithm AOD Algorithm Web-based QC Tools:  Error characterization as function of Environmental Conditions (SSES)  SST evaluation with reference Fields and Buoy Data (SQUAM)  BT evaluation with CRTM simulations (MICROS) Fig. 1. Flow-chart of the ACSPO-SEVIRI system. Reference SST Bi-Linear Interpolation (Reynolds SST) 1 o x1 o -> 5x5 km (at nadir) Individual Cloud Mask Tests results AOD flag L/W Mask Nay/might flag Glint flag QC Mask Regression Physical Fig.4. Time series of SST anomalies evaluated over clear sky pixels for June 2008. Retrieved SST exhibits the diurnal cycle with an average amplitude of 0.3°C. SST anomaly changes from day to day and as a function of time of day. Fig.5. Physical-Reynolds SST as a function of wind speed. FD data from 1-7 June 2008 are binned at V=1 m/s and averaged for 2 hours centered on Midday and Midnight. At night, SST is a fairly insensitive to wind speed. During daytime, SST changes as a function of wind speed with an amplitude of ~0.7 K. Difference between day and night is largest at low winds, due to forming a strong diurnal thermocline. The higher the wind speed, the smaller the diurnal heating. SST depends upon ambient clear-sky conditions during both day and night (Xu et a., 2009). The diurnal signal is smallest when ambient cloudiness is high (NAC→0) and largest under clear skies (NAC~400). JP2.15 89th AMS Annual Meeting and 5th GOES Users' Conference, Phoenix, AZ, 11-15 January, 2009 Correspondence: Nikolay.Shabanov@noaa.gov, Tel.: 301-763-8102 x154, Fax: 301-763-8572 Day Fig.9. Example FD histogram of daytime (SZA<90°) SEVIRI SST (physical retrievals) referenced to OSTIA SST for 15 June 2008 @ 23:45UTC. Global statistics of SST anomaly are superimposed. The shape of the histogram is close to Gaussian The mean bias averaged over the FD is ~-0.06K. The global standard deviation is ~0.62K. (Robust Stdv~0.59K.) Fig.8. Time Series of M-O biases in SEVIRI Ch4, 9 and 10 in June 2008. (Nighttime data with VZA<70°). The root causes of the two drop-out anomalies on 11 and 15 June 2008 are being investigated. Community Radiative Transfer Model (CRTM) is used in ACSPO in conjunction with Reynolds SST and NCEP Global Forecast System (GFS) upper air fields to simulate clear-sky BTs in SEVIRI Ch4, 9, and 10. CRTM BTs are used for cloud masking, physical SST retrievals, and to monitor quality of SEVIRI radiances. Here, Model (CRTM) minus Observation (SEVIRI) (“M-O”) biases are preliminarily evaluated using nighttime SEVIRI data (Sun Zenith Angle > 90°). Fig.7. Nighttime “M-O” biases in SEVIRI Ch4, 9, and 10 for one FD image taken on 14 June 2008 @ 3:00am. Data shown are for Satellite VZA < 70°. High M-O biases of ~+0.6K in Ch9 (11μm) and ~+0.3K in Ch10 (12 μm) are consistent with AVHRR (Liang et al., 2009). They may be due to CRTM (missing aerosol, using bulk rather than skin SST, not correcting SST for diurnal cycle) or SEVIRI data (residual clouds). The low bias in Ch4 (3.7 μm) is inconsistent with AVHRR analyses. The ACSPO-SEVIRI end-to-end system has been initially set up and preliminarily evaluated. Future work will be aimed at its optimization, fine-tuning and stress- testing with long time series of SEVIRI data. The following tasks are planned: Optimize and validate CRTM in conjunction with Reynolds SST/NCEP GFS Fully implement and optimize cloud mask Optimize Physical/Regression SSTs & combine into more accurate Hybrid SST Calibrate/Validate SEVIRI SSTs against in-situ data Optimize ACSPO processing speed and product volume Analyze diurnal cycle and explore diurnal data compression (using e.g. PCA) Analyze SEVIRI SST long-term wrt multiple reference and in situ SSTs Conclusion and Future Work Fig.10. Time series of bias for daytime (SZA,VZA<90°) SEVIRI SST wrt multiple reference SSTs. Each data point represents FD average over 24hr interval. (Note that physical SST was tuned against Reynolds SST and therefore bias wrt both OISSTs is smallest whereas wrt OSTIA it is highest ~0.2K). Fig.11. Time series of STD for daytime SEVIRI SST wrt multiple reference SSTs. Each data point represents STD around mean over FD×24hr. The STD changes from 0.4K to 0.8K depending upon day and reference state, and there is a significant variability from day to day. Part of noise comes from the unconstrained domain SZA,VZA<90°. The SQUAM methodology will be routinely applied to analyze SEVIRI SSTs.