Update on NOAA CO 2 Retrievals: Validation and Future Directions Eric Maddy, Chris Barnet, Mitch Goldberg, Xingpin Liu, Lihang Zhou and Walter Wolf. NOAA/NESDIS/STAR AIRS Science Team Meeting October 10, 2007 E. Maddy (PSGS, Inc.) Update on NOAA CO 2 retrievals: October 10, 2007 1 / 22
Outline • Description of NOAA AIRS CO 2 retrieval methodologies • How well can we do with a simple climatology? CO 2 ( t ) = a 0 + a 1 ∗ t • Development of error estimates. • Paper on averaging kernels (related to these error estimates) accepted (with revisions) to IEEE TGARS. • Validation with full resolution data vs. NOAA ESRL/GMD Aircraft (2005) 1 and Global Gridded data vs. JAL Matsueda (August 2003 - 2006.) • Comparison of AIRS and models – What new information can AIRS provide to modeling community? 1 Submitted to JGR in review E. Maddy (PSGS, Inc.) Update on NOAA CO 2 retrievals: October 10, 2007 2 / 22
NOAA AIRS CO 2 Retrievals • Use AIRS Science Team Methodology. • Version 4.7. • before cloudy regression introduced. • NOAA O 3 regression on. • 70 channels (mostly 15 micron). • Derive CO 2 in 4 layers in troposphere, 1 stratospheric. • Use Optimal Estimation w/ SVD • Runs within offline science code (consistent RTA/channel set). • Derive 6 - 10 CO 2 basis functions. • Runs very fast No appreciable difference in run-time compared to AIRS Science Team methodology • Validation with full resolution data vs. NOAA ESRL/GMD Aircraft (2005) and JAL Matsueda between August 2003 - 2006 (only AIRS Science Team approach). Each retrieval methodology has the ability to calculate averaging kernels and related diagnostics (d.o.f., etc.) and propagate error estimates. E. Maddy (PSGS, Inc.) Update on NOAA CO 2 retrievals: October 10, 2007 3 / 22
NOAA AIRS CO 2 Retrievals • Use AIRS Science Team Methodology. • Version 4.7. • before cloudy regression introduced. • NOAA O 3 regression on. • 70 channels (mostly 15 micron). • Derive CO 2 in 4 layers in troposphere, 1 stratospheric. • Use Optimal Estimation w/ SVD • Runs within offline science code (consistent RTA/channel set). • Derive 6 - 10 CO 2 basis functions. • Runs very fast No appreciable difference in run-time compared to AIRS Science Team methodology • Validation with full resolution data vs. NOAA ESRL/GMD Aircraft (2005) and JAL Matsueda between August 2003 - 2006 (only AIRS Science Team approach). Each retrieval methodology has the ability to calculate averaging kernels and related diagnostics (d.o.f., etc.) and propagate error estimates. E. Maddy (PSGS, Inc.) Update on NOAA CO 2 retrievals: October 10, 2007 3 / 22
NOAA AIRS CO 2 Retrievals • Use AIRS Science Team Methodology. • Version 4.7. • before cloudy regression introduced. • NOAA O 3 regression on. • 70 channels (mostly 15 micron). • Derive CO 2 in 4 layers in troposphere, 1 stratospheric. • Use Optimal Estimation w/ SVD • Runs within offline science code (consistent RTA/channel set). • Derive 6 - 10 CO 2 basis functions. • Runs very fast No appreciable difference in run-time compared to AIRS Science Team methodology • Validation with full resolution data vs. NOAA ESRL/GMD Aircraft (2005) and JAL Matsueda between August 2003 - 2006 (only AIRS Science Team approach). Each retrieval methodology has the ability to calculate averaging kernels and related diagnostics (d.o.f., etc.) and propagate error estimates. E. Maddy (PSGS, Inc.) Update on NOAA CO 2 retrievals: October 10, 2007 3 / 22
NOAA AIRS CO 2 Retrievals • Use AIRS Science Team Methodology. • Version 4.7. • before cloudy regression introduced. • NOAA O 3 regression on. • 70 channels (mostly 15 micron). • Derive CO 2 in 4 layers in troposphere, 1 stratospheric. • Use Optimal Estimation w/ SVD • Runs within offline science code (consistent RTA/channel set). • Derive 6 - 10 CO 2 basis functions. • Runs very fast No appreciable difference in run-time compared to AIRS Science Team methodology • Validation with full resolution data vs. NOAA ESRL/GMD Aircraft (2005) and JAL Matsueda between August 2003 - 2006 (only AIRS Science Team approach). Each retrieval methodology has the ability to calculate averaging kernels and related diagnostics (d.o.f., etc.) and propagate error estimates. E. Maddy (PSGS, Inc.) Update on NOAA CO 2 retrievals: October 10, 2007 3 / 22
Improvement Over Simple Climatology • Theoretical error analysis for our Version 5 Climatology Polar Midlatitude Tropical 25 25 25 H 2 O T(p) 20 20 20 O 3 T surf Smoothing Total a priori 15 15 15 Altitude, [km] 10 10 10 5 5 5 0 0 0 0 2 4 6 8 10 0 2 4 6 8 10 0 2 4 6 8 10 CO 2 error, [ppmv] CO 2 error, [ppmv] CO 2 error, [ppmv] • Calculation uses a priori covariance calculated as the dif ference between ESRL aircraft and our simple Version 5 climatology. • Ability to partition error sources and their effect on the retrieval. Effect minimized as we have assumed a perfect knowledge of the error covariation of intefering species (assumed ad-hoc : S ( z, z ′ ) = σ ( z ) σ ( z ′ ) · exp( −| z − z ′ | /L ) ). E. Maddy (PSGS, Inc.) Update on NOAA CO 2 retrievals: October 10, 2007 4 / 22
ESRL/GMD Aircraft Validation Approach • Use full resolution AIRS retrievals (previously validated w/ 3 ◦ x 3 ◦ grids) • Average AIRS CO 2 between 6-10 km (nominally where jacobian has maximum sensitivity). • Use nominal jacobians (wrt. latitude) to weight ESRL aircraft. • Enables comparison of scalar measurements • Removes variability in lowest 2.5 km • Average all retrievals within 200km with temporal matchup window between 1 day - 1 month. • Profile statistics will also be shown. NOAA/ESRL CarbonTracker 2 model used to extend profiles above 8 km. 2 http://www.cmdl.noaa.gov/ccgg/carbontracker E. Maddy (PSGS, Inc.) Update on NOAA CO 2 retrievals: October 10, 2007 5 / 22
Example Comparison: Estevan Point, British Columbia Estevan Point, British Columbia NOAA ESRL/GMD Aircraft and AIRS Retrieval Timeseries 395 AIRS CO 2 6km to 8km AIRS CO 2 + / - σ (CO 2 ) 390 Aircraft above 2.5km 385 CO 2 , [ppmv] 380 375 370 365 Dec 03 Jul 04 Jan 05 Aug 05 Mar 06 Date, [1] E. Maddy (PSGS, Inc.) Update on NOAA CO 2 retrievals: October 10, 2007 6 / 22
AIRS Science Team Algorithm vs. ESRL/GMD Aircraft | ∆ t| < 7.00 days 390 Bradgate, Iowa Beaver Crossing, Nebraska Briggsdale, Colorado Dahlen, North Dakota Estevan Point, British Columbia 385 Fairchild, Wisconsin Molokai Island, Hawaii Harvard Forest, Massachusetts Homer, Illinois Park Falls, Wisconsin Worcester, Massachusetts ESRL CO 2 , [ppmv] Oglesby, Illinois 380 Poker Flat, Alaska Rowley, Iowa Rarotonga, Cook Islands Trinidad Head, California N MATCHUP : 495 375 RMS(PPMV) : 2.06 SDV(PPMV) : 1.80 IQR(PPMV) : 2.23 BIAS(PPMV) :-0.99 370 CORREL : 0.77 RANKCORREL: 0.79 365 365 370 375 380 385 390 AIRS CO 2 , [ppmv] • Total magnitude of drawdown at LEF not captured possible over-regularization wrt. characteristic variability. E. Maddy (PSGS, Inc.) Update on NOAA CO 2 retrievals: October 10, 2007 7 / 22
AIRS Science Team Algorithm vs. ESRL/GMD Aircraft | ∆ t| < 7.00 days 390 Bradgate, Iowa Beaver Crossing, Nebraska Briggsdale, Colorado Dahlen, North Dakota Estevan Point, British Columbia 385 Fairchild, Wisconsin Molokai Island, Hawaii Harvard Forest, Massachusetts Homer, Illinois Park Falls, Wisconsin Worcester, Massachusetts ESRL CO 2 , [ppmv] Oglesby, Illinois 380 Poker Flat, Alaska Rowley, Iowa Rarotonga, Cook Islands Trinidad Head, California N MATCHUP : 495 375 RMS(PPMV) : 2.06 SDV(PPMV) : 1.80 IQR(PPMV) : 2.23 BIAS(PPMV) :-0.99 370 CORREL : 0.77 RANKCORREL: 0.79 365 365 370 375 380 385 390 AIRS CO 2 , [ppmv] • 0.5% uncertainty from space! E. Maddy (PSGS, Inc.) Update on NOAA CO 2 retrievals: October 10, 2007 8 / 22
Calculated a priori and Retrieval Error Covariances Tropical Mid-Latitude/Polar A Priori Error Correlation Matrix A Priori Error Patterns A Priori Error Correlation Matrix A Priori Error Patterns 10 10 19.83 (90.4%) σ ( x a ) 30.96 (76.0%) σ ( x a ) Relative Altitude, [km] 5.76 (7.6%) Relative Altitude, [km] 15.27 (18.5%) 8 ^ ) 8 ^ ) σ ( x σ ( x 2.04 (1.0%) 5.78 (2.6%) 1.41 (0.5%) 3.90 (1.2%) 6 6 4 4 2 2 0 0 Retrieval Error Correlation Matrix Retrieval Error Patterns Retrieval Error Correlation Matrix Retrieval Error Patterns 10 10 16.49 (83.1%) 23.49 (63.3%) 8 6.66 (13.6%) 8 15.67 (28.1%) Relative Altitude, [km] Relative Altitude, [km] 2.56 (2.0%) 6.20 (4.4%) 1.40 (0.6%) 3.93 (1.8%) 6 6 4 4 2 2 0 0 0 2 4 6 8 10 -10 -5 0 5 10 0 2 4 6 8 10 -10 -5 0 5 10 Relative Altitude, [km] Scaled Eigenvector, [ppmv] Relative Altitude, [km] Scaled Eigenvector, [ppmv] -1.0 -0.5 0.0 0.5 1.0 • Total variance of the retrieval is less than the a priori indicating a gain in information. • First eigenfunction variance (and percent of total variance) of the retrieval is less than a priori . • Retrieval tends to redistribute variance among higher order eigenfunctions, which are similar in shape to the a priori , indicating we have only 1 piece of information, albeit well constrained, in the vertical. Vertical resolution ≈ 6-8 km. E. Maddy (PSGS, Inc.) Update on NOAA CO 2 retrievals: October 10, 2007 9 / 22
Recommend
More recommend
