AGU Highlights

1
AGU Highlights

• Vijay Natraj

2
CO2 Retrieval Simulation from GOSAT Thermal IR
Spectra

• 15 um CO2 band 0.2 cm-1 res, 300 S/N
• 110 layers for forward model, reduced grid for
  retrievals
• MAP retrieval
• If T known perfectly, excellent agreement above
  700 mbar
• If T had random errors, results with 1.5
  precision if channels selected on the basis of
  CO2 IC or CO2 IC T IC as appropriate

3
Impact of Aerosols on CO2 Retrievals using NIR
GOSAT Data

• 1.6 um CO2 band
• Large CO2 errors for aerosols at high latitudes
  even for low aerosol od (gt 0.05)
• CO2 errors also large when surface albedo is
  large
• Simultaneous retrieval of aerosol, CO2 and
  surface albedo reduces bias

4
Cirrus Cloud Characteristics from GLAS
Observations

• Geoscience Laser Altimeter System
• Cirrus clouds located at 13 km in tropics and 8
  km in mid-latitudes, with 2 km thickness
  everywhere
• Optical thickness less than 0.2 in UT and approx.
  constant at 0.25 in mid and lower trop in the
  tropics
• Mean value of optical thickness increases with
  latitude
• In the tropics, 56 of cirrus cloud events occur
  above other cloud layers!

5
Ozone Profile Retrieval from OMI Data

• 270-330 um
• 18-layer atmosphere 6-8 km vertical res
• DOAS technique with optimal estimation
• 6-stream LIDORTpolarization correction LUTRRS
• Results good for levels lt 50 mbar

6
Accounting for Non-uniform Spatial IC of Remotely
Sensed Data

• Spatial characteristics of observations different
  from those of assimilation model
• Typically use point-based interpolation
  techniques such as bilinear interpolation
• Such techniques ignore footprint characteristics
  of observations hence uncertainty inherent in
  resampling
• Geostatistical Inverse Modeling (GIM)
  incorporates spatial scale of observations and
  models uncertainty inherent to making estimates
  at different spatial scales
• Essentially, GIM is a bayesian approach similar
  to traditional inverse modeling
• Treats each pixel as an non-uniform integration
  of footprint depending on sensors point-spread
  function and viewing geometry, and not as a point
  or rectangle with uniform information
• Inverse modeling used to estimate value for
  center pixel using information from both center
  and surrounding measurements