Retrieving CO₂ Column Variations
Feasibility and Initial Results

Howard E. Motteler
L. Larrabee Strow
Scott Hannon

Aug. 18, 1998

1  Introduction

• We investigate the feasibility of retrieving CO₂ column variation from upwelling infrared radiation, and present some preliminary simulation results

• Motivation:
  • CO₂ variation is of considerable climatological interest

  • Retrieval of CO₂ variations could be used to refine atmospheric temperature retrievals

• Accurate retrieval of CO₂ variation is difficult, because the CO₂ component of upwelling infrared radiation is very sensitive to temperature variation

2  Methods

• The CO₂ component of upwelling infrared radiation is very sensitive to temperature variation; in fact CO₂ lines are the basis for most temperature sounding

• This temperature sensitivity makes it hard to detect small variations in CO₂ amount

• There is an interval at 792 cm^-1 with significant dB/d(CO₂), compared to dB/dT

• For this study we used our latest AIRS (Atmospheric InfraRed Sounder) fast radiative transfer model, calculating 2373 AIRS channels from approximately 600 cm^-1 to 2700 cm^-1

• We took a large set of atmospheric profiles, varied the CO₂ amount by a few percent, and tried to retrieve this variation

• We used the 1761 TIGR profiles with
  • Surface temperatures simulated by taking a normal distribution with 4K variance, centered at the surface air temperature

  • Emissivity was varied uniformly from 0.8 to 1.0, or from 0.97 to 0.99, for some tests

  • CO₂ varied from the reference profile by a normal distribution with variance of 2% (or 3%, in one test)

• The TIGR set was divided into dependent and independent sets of 880 and 881 profiles, respectively

• For most tests, we used the full set of 2373 AIRS channels, as calculated by our radiative transfer model

• For these initial tests, we used regularized regression, with a 35-element basis for the simulated spectra

• Tests were done with both single-spot AIRS noise, and noise reduced by factors of 4 and 10, to represent averaging over a larger area

• In addition to CO₂, temperature, surface temperature, and emissivity retrievals were also done, mainly to validate the code

3  Results

• Some CO₂ variation was detected; RMS average retrieval error was typically 1.2 to 1.3 percent absolute, 60 to 65 percent relative, varying with details of the particular test

• CO₂ retrievals were not too sensitive to variations of instrument noise, at least with the full set of 2373 channels

• CO₂ retrievals were not too sensitive to variations of emissivity or surface temperature

• CO₂ sensitivity (from regression coefficients) was found across most of the AIRS spectra

• Increasing the CO₂ variation in the simulation increased absolute error but decreased relative error

   Tests 3, 4, 5: Varying Noise
   TIGR surface temps
   CO2 variability: normal distribution, variance 2%
   Emissivity: uniform distribution from 0.8 to 1.0

      Test Number          3           4          5
   AIRS Noise Scaling     1.0         0.25       0.1
   absolute pct error     1.28        1.24       1.24
   relative pct error      64          62         62
   

   Tests 6, 7: Varying Surface Temperature and Emissivity
   Surface temp w/ 4K variance from lowest air
   CO2 variability: normal distribution, variance 2%
   AIRS noise scaling: 0.25

      Test Number            6              7
   Emissivity Range     0.80 to 1.0    0.97 to 0.99 
   absolute pct error       1.35           1.28
   relative pct error        67             64

   Test 9: Increasing C02 Variation
   Surface temp w/ 4K variance from lowest air
   CO2 variability: normal distribution, variance 3%
   emissivity: uniform distribution from 0.97 to 0.99
   AIRS noise scaling: 0.25

      Test Number            9
   absolute pct error       1.42
   relative pct error        48

4  Jacobians and Regression Covariance

• While some of the retrieval covariance matrices showed a significant sensitivity at 792 cm^-1, this was by no means a predominant effect

• Much of the CO₂ sensitivity appears to be in ``water regions''

• These regions also have some temperature sensitivity

• This sensitivity is independent of CO₂ amount, and so in combination with the CO₂ regions can give some indication of CO₂ amount

5  Conclusions and Further Work

• Some information about CO₂ variation can be retrieved with statistical methods

• The sensitivity to CO₂ variation was not where we'd initially expected to find it, occurring over a much wider spectral region

• It is likely that this approach works (as well as it does) due to temperature sensitvity of non-CO₂ components, and in particular, water

• Retrieval accuracy might be improved with further independent temperature information, e.g., from microwave or other soundings, and perhaps with physical iterative methods