Using Double Differences in MICROS for Cross-Sensor Consistency Checks http: //www. star. nesdis. noaa. gov/sod/sst/micros/ Xingming Liang 1, 2, Sasha Ignatov 1 and Korak Saha 1, 2 1 NOAA/NESDIS/STAR 2 CSU/CIRA GSICS Annual Meeting, Beijing, 5 -8 March 2012 Slide 1 of 27
Acknowledgments Advanced Clear-Sky Processor for Oceans (ACSPO; NESDIS Sea Surface Temperature System): Sensor Radiances over Oceans with Clear-Sky Mask and QC Ø J. Sapper, Y. Kihai, B. Petrenko, J. Stroup, P. Dash, F. Xu, M. Bouali – NESDIS SST Team Sensor Characterization & Cross-platform Consistency, including Double Differences (DD) Ø F. Wu, F. Yu, C. Cao, L. Wang, F. Weng, M. Goldberg, T. Hewison, J. Xiong, X. Hu, T. Chang – GSICS Community Radiative Transfer Model (CRTM) Ø F. Weng, Y. Han, Q. Liu, P. Van Delst, Y. Chen, D. Groff – CRTM Ø N. Nalli - Surface emissivity model GSICS Annual Meeting, Beijing, 5 -8 March 2012 Slide 2 of 27
Outline MICROS overview MICROS Double-Differences (DD) Relative Merits: SNO, Hyper-Spectral DDs Conclusion Future plans GSICS Annual Meeting, Beijing, 5 -8 March 2012 Slide 3 of 27
MICROS Overview Objectives Sensors monitored System set-up & Processing time MICROS Hightlights Ways to present M-O bias GSICS Annual Meeting, Beijing, 5 -8 March 2012 Slide 4 of 27
MICROS Objectives Monitor clear-sky sensor radiances (BTs) over global ocean in NRT (“OBS”) , against CRTM with first-guess input fields (“Model”) Understand & minimize M-O biases in BT & SST Ø minimize need for empirical “bias correction” Evaluate sensor radiances for stability Check for cross-platform consistency GSICS Annual Meeting, Beijing, 5 -8 March 2012 Slide 5 of 27
Platforms/Sensors monitored in MICROS Routinely processing 5 AVHRRs Jul’ 2008 -on – – – Metop-A (GAC and FRAC) - Good NOAA 19 - Good NOAA 18 – Good NOAA 17 - stopped processing 2/10; sensor issues NOAA 16 - out of family Under testing / In pipeline – – NPP/VIIRS Terra/MODIS Aqua/MODIS MSG/SEVIRI GSICS Annual Meeting, Beijing, 5 -8 March 2012 Slide 6 of 27
System Set-Up & Processing Time Fully automated - Scripted/Cronned - Back-up processing ACSPO: Identify clearsky pixels (Fortran 95) MICROS: Generate stats (IDL) Post to web: Html/JS/JQuery/JQplot Processing Time (ACSPO/MICROS): Process 24 hrs of Day (N-2) 5 GAC AVHRRs (NOAA 16 -19, Metop. A) ACSPO/MICROS 1 FRAC AVHRR (Metop-A) MODIS/Terra & Aqua VIIRS/NPP 3/0. 5 4/1 10/3 (under opt. ) GSICS Annual Meeting, Beijing, 5 -8 March 2012 Slide 7 of 27
Model - Obs Advanced Clear-Sky Processor for Oceans (ACSPO) Ø “M” = MODEL Clear-Sky BT (currently, in SST bands only) - Ø Calculated using CRTM, with first guess SST (daily 0. 25º Reynolds) and upper air fields (NCEP GFS 6 hr 1º) as input Fast CRTM allows for real-time processing “O”= OBS: Clear-Sky Ocean Sensor BTs - Clear-Sky Ocean pixels identified using ACSPO Cloud Mask and QC Monitoring IR Clear-sky Radiances over Oceans for SST Ø Calculates M-O bias & Runs global daily statistics on it - Ø Processing fully automated, performed in NRT Also, Double-Differences calculated w. r. t. a Reference sensor Graphic summaries reported on the web - http: //www. star. nesdis. noaa. gov/sod/sst/micros/ GSICS Annual Meeting, Beijing, 5 -8 March 2012 Slide 8 of 27
MICROS Highlights End-to-end system Web-Based Statistical analyses performed in global clear-sky ocean domain Near-Real Time MICROS Analyses stratified by Day/Night Only Night data used for sensor analyses Presently, daytime data not used due to sub-optimal treatment of solar reflectance & diurnal cycle Both conventional & robust statistics used Double-differences used to evaluate sensor radiances for cross-platform consistency GSICS Annual Meeting, Beijing, 5 -8 March 2012 Slide 9 of 27
Ways to present M-O Bias Maps Histograms Time series Four ways to present M-O Biases in MICROS Dependencies GSICS Annual Meeting, Beijing, 5 -8 March 2012 Slide 10 of 27
Maps The M-O biases: Close to zero; Uniformly distributed GSICS Annual Meeting, Beijing, 5 -8 March 2012 Slide 11 of 27
Dependencies View Angle dependencies of M-O bias: Close to zero GSICS Annual Meeting, Beijing, 5 -8 March 2012 Slide 13 of 27
Histograms Near-Gaussian # clear-sky oceans pixels ~3 Million/night (global GAC) M-O bias is close to zero In fact, it’s slightly warm: Expected (Discussed next slide) Cross-platform biases close to zero (~0. 2 K) Overpass times from 9: 30 pm -5 am (Diurnal effects) Errors in sensor SRFs (CRTM coefficients) & CAL GSICS Annual Meeting, Beijing, 5 -8 March 2012 Slide 12 of 27
Time Series for GAC M-O bias in Ch 3 B M-O Biases in Ch 3 B BT are in sync with SST oscillations ACSPO version V 1. 00 V 1. 02 V 1. 10 Warm M-O biases result from: (1) Missing aerosols; (2) Using bulk SST (instead of skin); (3) Using daily mean Reynolds SST (to represent nighttime SST); (4) Residual cloud V 1. 30 V 1. 40 SST Biases (Regression-Reynolds) Temporal variability: Due to unstable Reynolds SST (input into CRTM) N 16: Out of family/Unstable (CAL problems) N 17: Scan motor spiked in Feb’ 2010 GSICS Annual Meeting, Beijing, 5 -8 March 2012 Slide 14 of 27
Double Differences (DD) Cross-platform consistency in MICROS Day-to-day noise and spurious variability hinder accurate measurement of cross-platform bias Double-differences (DD) employed to differentiate the “cross-platform bias” signal from “noise” Metop-A used as a Reference Satellite Ø Stable; Overpass time close to N 17/Terra CRTM (Reynolds SST) is used as a ‘Transfer Standard’ DDs cancel out/minimize effect of systematic errors & instabilities in BTs and SSTs arising from e. g. : Ø Errors/Instabilities in reference SST & GFS Ø Missing aerosol Ø Possible systemic biases in CRTM Ø Updates to ACSPO algorithm GSICS Annual Meeting, Beijing, 5 -8 March 2012 Slide 15 of 27
DDs (Ref = Metop-A) Cross-platform consistency in IR 37 Double Differences (DDs) in IR 37 V 1. 02 V 1. 10 V 1. 30 V 1. 40 NOAA 16 V 1. 00 DDs cancel out most errors/noise in M-O biases Relative to Metop-A , biases are - N 16: unstable - N 17: +0. 01 ± 0. 02 K (scan motor failed Feb’ 10) - N 18: +0. 04 ± 0. 05 K (not very stable) - N 19: -0. 06 ± 0. 02 K GSICS Annual Meeting, Beijing, 5 -8 March 2012 Slide 16 of 27
DDs in other bands Cross-platform consistency in IR 11&12 Double Differences in IR 11 Double Differences in IR 12 N 16: Unstable in all 3 bands N 17: biased +0. 05 K high in IR 11; -0. 03 K low in IR 12 N 18: biased -0. 02 K low in IR 11; +0. 06 K high in IR 12 N 19: biased -0. 07 K low in IR 11; -0. 09 K low in IR 12 N 18: Similar pattern in IR 11 and IR 12 with IR 37. Cross-platform biases are due to - CAL errors - SRFs deviation from those used in CRTM - Local time differences (diurnal cycle in SST/GFS) Work is underway to attribute the causes & reconcile platforms GSICS Annual Meeting, Beijing, 5 -8 March 2012 Slide 17 of 27
GSICS Inter-Calibration Methodologies Simultaneous Nadir Overpasses Hyper-Spectral Double-Differences (integrate HS radiances with wideband spectral response) MICROS Double Differences (integrate RTM simulations with wide -band spectral response) – Fit? GSICS Annual Meeting, Beijing, 5 -8 March 2012 Slide 18 of 27
Simultaneous Nadir Overpasses (SNO) Cross-platform consistency in GSICS: 1 of 2 SNO matches two satellites in space and time at nadir Objectives Eliminate uncertainties associated with Atmospheric path View geometry Time difference And estimate cross sensor inconsistency SNO in Ch 3 (NOAA 16~ NOAA 17) SNO in Ch 4 (NOAA 16~ NOAA 17) From SNO web: www. star. nesdis. noaa. gov/smcd/spb/calibration/sno/ GSICS Annual Meeting, Beijing, 5 -8 March 2012 Slide 19 of 27
Hyper-Spectral (HS) DDs Cross-platform consistency in GSICS: 2 of 2 HS DD use GOES as the transfer standard to match up each pair of satellites in space and time at nadir (Wang and Cao, 2008; Hewison and Konig, 2008) (from GSICS Quarterly v 2. 2008) GSICS Annual Meeting, Beijing, 5 -8 March 2012 Slide 20 of 27
MICROS vs SNO & HS DDs MICROS DDs Real-time Online Domain NOBS QC Data Distribution Transfer Standard Effect of Solar reflectance Spectral Response Cross-platform bias precision SNO Hyper-Spectral (HS) DDs NRT Jul’ 2008 -present Online 2002 -08 Online No No Global ocean Clear-Sky only Full sensor swath Polar areas Ice/Ocean/Land All-Sky Nadir only Global match-ups Ice/Ocean/Land All-Sky Nadir Only ~3 Mln/Day (AVHRR GAC) Several match-ups/Day Sensors-specific ACSPO Clear-Sky Mask No QC Gaussian Asymmetric CRTM; No match-up in space/time required Direct Comparisons; Matchup in space/time required GOES; Match-up in space/time required Currently, only used during nighttime (no daytime) Renders data in mid-IR (Ch 3 B) unusable Shortwave bands not always covered by HS measurements Considered Not considered Considered ~0. 01 K ~1 K ~0. 1 K MICROS supplements GSICS Inter-calibration Techniques GSICS Annual Meeting, Beijing, 5 -8 March 2012 Slide 21 of 27
Conclusions M-O biases & Double-Differences (DD) in MICROS - Functional with 5 AVHRR; Terra/Aqua MODIS & NPP/VIIRS being tested DDs Cancel out most errors in M-O biases AVHRR Cross-sensor biases (DDs): ~10 -2 K to ~10 -1 K Cross-sensor biases (DDs) are due to errors in - Sensor Calibration Sensor Spectral Response Functions CRTM Coefficients Need to unscramble cross-platform biases seen in MICROS DDs - With GSICS Colleagues (T. Chang, F. Wu, F. Yu) – Sensor Cal and SRFs CRTM Colleagues – Verify CRTM coefficients MICROS DDs supplement GSICS Hyper-Spectral DDs and SNO - Global clear-sky night ocean domain Difference between M and O: Narrow Gaussian distribution, centered at ~0 Large data volume (GAC: 3 M pixels / 24 hr): Instrumental to beat down noise No collocation with other sensors required GSICS Annual Meeting, Beijing, 5 -8 March 2012 Slide 22 of 27
Future Plans 1 Extend MICROS to more polar sensors and GOES Work with GSICS Colleagues to - Reconcile cross-platform differences - Evaluate MICROS DDs for consistency w/Hyper-Spectral DDs & SNO AVHRRs - Continue monitoring all sensors in orbit; Add Metop-B (May 2012) - Extend back in time to include all AVHRR (1978 -pr) Add Terra/Aqua MODIS & NPP VIIRS in MICROS - Fine tune CRTM and ACSPO Cloud Mask - Evaluate for stability & cross-consistency with AVHRRs - Reprocess MODIS historical data back to 2000. (where is L 1 B data ? ) Add new sensors - GEO: MSG SEVIRI in progress; and GOES-R/ABI (~2015) - ATSR (NRA joint proposal w/JPL/S. Hook and U. Leicester/G. Corlett) - Discussions underway w/CMA on FY 1/VIRR & FY 3/MERIS GSICS Annual Meeting, Beijing, 5 -8 March 2012 Slide 23 of 27
MICROS ver 5 PROXY MODIS & VIIRS streams q MODIS & VIIRS proxy have been trended in MICROS 5 since mid-2011 q Quantitative analyses pending fine-tuning ACSPO Processor q Will report MODIS-VIIRSAVHRR consistency in 2013 GSICS Meeting MICROS Version 5: In preparation for launch of NPP/VIIRS in Oct’ 2011, MICROS 5 was set up - Added proxy NPP VIIRS & Proxy Terra/Aqua MODIS - Added interactive plots for flexible display of multiple platforms GSICS Annual Meeting, Beijing, 5 -8 March 2012 Slide 24 of 27
Future Plans 2 Extend MICROS to Include Reflectance Bands Aerosol Quality Monitor (AQUAM) was set up to prepare for adding aerosol in CRTM ü GOCART and NAAPS identified as sources of 3 D aerosol fields inputs into CRTM ü Initially, use solar reflectance bands to evaluate DDs and CRTM/GOCART&NAAPS - First-guess reflectances will improve ACSPO clear-sky mask ü Subsequently, extend aerosol analyses into thermal IR bands - M-O bias in emission bands will become closer to zero & STD reduced - DDs in emission bands are expected to be improved GSICS Annual Meeting, Beijing, 5 -8 March 2012 Slide 25 of 27
Future Plans 3 Improve Accuracy of MICROS DDs largely cancel out uncertain/unknown factors However, they can be further improved by Using more accurate first guess fields (SST, GFS) Using Improved CRTM (especially daytime) Modeling diurnal variation in first-guess SST So far, checked sensitivity of DDs to first-guess SST (major factor in SST bands) • Overall, very small MICROS DDs are reliable • However, temporal noise may be reduced and DDs estimated more accurately From MICROS paper: http: //www. star. nesdis. noaa. gov/sod/sst/micros_v 5/pdf/JTECH-D-10 -05023. pdf GSICS Annual Meeting, Beijing, 5 -8 March 2012 Slide 26 of 27
Thank you! GSICS Annual Meeting, Beijing, 5 -8 March 2012 Slide 27 of 27
Скачать презентацию Using Double Differences in MICROS for Cross-Sensor Consistency
2c479a456a2537a7b08785ca54da6db1.ppt