Plans for inves-ga-ng -me‐dependent biases in AIRS temperature profiles Bill Irion Jet Propulsion Laboratory, California Ins-tute of Technology With thanks to Chris Barnet, Murty Divakarla, Eric Maddy NOAA John Blaisdell, Thomas Hearty, Gyula Molnar GSFC George Aumann, Eric Fetzer, Evan Fishbein, Steve Friedman, Bjorn Lambrigtsen, Sung‐Yung Lee, Evan Manning, Ed Olsen, Joao Teixeira Jet Propulsion Laboratory, California Ins-tute of Technology
Comparisons of trends in AIRS V5 (Final and MW products), AVN and ECMWF to sonde temperatures Courtesy of Eric Maddy
V5.0.14 + AVN Tq a priori + CarbonTracker CO 2 + no AMSU + CO 2 noise covariance matrix ON NO CO 2 NCV Eric Maddy
V5.0.14 + AVN Tq a priori + CarbonTracker CO 2 + no AMSU + CO 2 noise covariance matrix OFF Eric Maddy
What do we know? • V5.0.14 shows a -me dependent bias w.r.t. sondes ~ ‐100mK yr ‐1 ; similar results against ECMWF • Microwave only retrievals show a lesser, but s-ll significant bias ~ ‐50mK yr ‐1 • Switching to Carbon Tracker CO 2 plus AVN Tq a priori with CO 2 NCV helps at 200 mb and boundary layer, but liale difference in between – Turning CO 2 NCV off brings bias rate down to ~ ‐50mK yr ‐1
AIRS Retrieval System Cloud. Regression Cloudy Parameter sol’n L0 MW LO IR Regression Retrieval #3 products AMSU Cloud‐ AMSU L1A MW L1A IR Retrieval #1 clearing #3 Retrieval #2 MW IR Physical Cloud. Cloud. Calibra-on Calibra-on Retrieval #2 Parameter Parameter Retrieval #2 Retrieval #1 L1B MW L1B IR Error Est. & Cloud‐ Cloud‐ Quality clearing #1 clearing #2 Control MW only Retrieval Regression L2 product Physical Retrieval #1 MW Only Product
AIRS Retrieval System Cloud. Regression Cloudy Parameter sol’n L0 MW LO IR Regression Retrieval #3 products Next Steps: AMSU Cloud‐ AMSU L1A MW L1A IR Retrieval #1 clearing #3 Retrieval #2 1. A test bed for the AIRS PGE with hooks and rou-nes to write out results (incl. Jacobians MW IR Physical Cloud. Cloud. Calibra-on etc.) from each processing step. Calibra-on Retrieval #2 Parameter Parameter Retrieval #2 Needed to evaluate trends introduced by each Retrieval #1 • process. This problem appears to have more than L1B MW L1B IR Error Est. & one contributor. Cloud‐ Cloud‐ Quality clearing #1 clearing #2 A turnkey system will be useful for later algorithm • Control tes-ng and evalua-on. MW only Retrieval Regression L2 product Physical Retrieval #1 MW Only Product
AIRS Retrieval System Cloud. Regression Cloudy Parameter sol’n L0 MW LO IR Regression Retrieval #3 products Next Steps: AMSU Cloud‐ AMSU L1A MW L1A IR Retrieval #1 clearing #3 Retrieval #2 2. A consistent set of radiosonde observa-ons and ECMWF days (i.e. focus days) to compare MW IR Physical Cloud. Cloud. Calibra-on AIRS results against each other. Calibra-on Retrieval #2 Parameter Parameter Retrieval #2 • We’re looking for small but robust effects, so a Retrieval #1 consistent valida-on set across different tests will be L1B MW L1B IR Error Est. & important. Cloud‐ Cloud‐ Quality clearing #1 clearing #2 • Radiosondes remain the best “truth,” but ECMWF Control comparison allows tests under more atmospheric MW only states (different regions, seasons, cloud condi-ons, Retrieval Regression L2 product Physical etc.) Retrieval #1 MW Only Product
AIRS Retrieval System Cloud. Regression Cloudy Parameter sol’n L0 MW LO IR Regression Retrieval #3 products Next Steps: AMSU Cloud‐ AMSU L1A MW L1A IR Retrieval #1 clearing #3 Retrieval #2 3. A consistent method of ver-cally averaging and calcula-ng biases and trends. MW IR Physical Cloud. Cloud. Calibra-on Calibra-on This must be chosen carefully. Too fine a ver-cal • Retrieval #2 Parameter Parameter region (say under the AIRS resolu-on) may produce Retrieval #2 Retrieval #1 an erroneous trend while too thick a region may L1B MW L1B IR Error Est. & mask it. Cloud‐ Cloud‐ Quality Different hypotheses will be tested by different • clearing #1 clearing #2 Control groups, so we need assurance that results can be MW only properly compared to each other. Retrieval Regression L2 product Physical Retrieval #1 MW Only Product
Discussion
Recommend
More recommend
