H.H. Aumann (hha@airs1.jpl.nasa.gov), AIRS Project Scientist, Jet Propulsion Laboratory
M.T. Chahine, Jet Propulsion Laboratory, is the AIRS Science Team Leader. He and the other members of the team have the responsibility of developing the computer program, referred to as the retrieval algorithm, which converts the radiances measured by the AIRS/AMSU/MHS instruments to the desired temperature and moisture profiles. The observational conditions, i.e., typical temperature and moisture profiles under a wide range of climatic, geographical, and day/night conditions, are reasonably well known from more than a decade of experience with the HIRS 2/MSU data (TIROS Operational Vertical Sounder, TOVS). The generation of simulated data (by converting the observational conditions to the radiances typical of those to be observed by AIRS, AMSU, and MHS) is a critical part of the AIRS algorithm development.
The AIRS algorithm development effort involves several independent groups within the AIRS Science Team:
The simulated data are distributed electronically to the teams involved in developing retrieval algorithm concepts and prototype software. Three types of data are distributed to facilitate the task of the algorithm developers:
The algorithm development teams return their results from the test data and the truth data, together with the software used to obtain the results, to the simulation team at JPL. The retrievals are evaluated for accuracy. The software is evaluated for computer resource requirements (CPU and I/O utilization) and compliance with reasonable software engineering standards. Periodic meetings of the AIRS Science Team are used for discussions of simulation procedures, retrieval accuracy, and retrieval resource requirements.
The algorithm development using the separation between Level 2 data simulation, Level 1 data simulation, and Level 2 data retrieval as described above was started in 1992. The initial tests were simple: Night time, cloud free, surface with no elevation (i.e., at 1000 mb pressure) and with known, wavelength-independent emissivity and reflectivity. Since then the simulation has advanced to include daytime, wavelength-dependent and unknown surface emissivity and reflectivity, and realistic topography, but until recently it was still cloud free.
TOVS data from HIRS 2/MSU indicate that 45 percent of the time there are clear conditions, about 35 percent of the data are partly cloudy, but the retrievals are acceptable, while the remaining 20 percent of the data are so cloudy that the HIRS 2/MHS data can not produce usable retrievals. The first test data including clouds was released to the algorithm development teams in August 1994. This test was called the single layer gray cloud test. The statistical distribution and cloud granularity were patterned using the statistics obtained from the TOVS data. For this test the simulation program read Level 2 data from four satellite tracks crossing the model area from south to north (tracks A, B, C, D in Figure 1) and converted them to the spectral radiances as described above. (The curvature of the tracks is an artifact of the mercator map projection). As this was the first simulation of cloudy data, the data set was limited to a single cloud layer and the clouds were simulated as spectrally gray, i.e., the emissivity and reflectivity were unknown, but wavelength independent. Figure 2 shows the fractional cloud cover averaged along track B. The fractional cloud cover in the AIRS FOV ranged from 20 to 90 percent. The cloud top pressure ranged from 850 mb to 100 mb. Figure 3 shows the cloud liquid water content along track B. It averages about 0.01g/cm2, but exceeds 0.03 g/cm2 near latitudes N44 and N52. The onset of precipitation is between 0.02 and 0.04 g/cm2. This data set represented a severe test of the ability of the combined infrared and microwave sounding capability of AIRS/AMSU and MHS
Figure 1. The simulation used temperature and moisture
profiles from a mesoscale
model provided by the National Meteorological Center. Data
from parts of flour satellite
tracks (A, B, D, D) were converted to spectral radiances and
used as input to the
retrieval algorithms.
Figure 2. The fractional cloud cover averaged along track B
is shown. the fractional
cloud cover in the AIRS FOV ranged from 20 to 90
percent.
Figure 3. The cloud liquid water content along track B is
shown. It averaged about
0.01g/cm2, but exceeds
0.03 g/cm2 near latitudes N44 and N52.
The onset of precipitation is between 0.02
and 0.04g.cm2.
The key questions posed at the November 1994 team meeting to the core algorithm development teams were twofold:
Both questions were answered affirmatively.
Mitch Goldberg and Larry McMillin (NOAA/NESDIS) used an extension of the algorithm used operationally for the TOVS data. This is a sequence of four steps: First, the data are cloud-cleared, using the TOVS-tested N-star method (McMillan and Dean 1982). In the second step the cloud-cleared radiances are compared to the radiances produced by one of several thousand temperature/moisture profiles in the NOAA operational matchups library (McMillan 1991). In step three, the closest match from the library search is used as a first guess to a microwave only retrieval. The output from step three is used in the fourth step as the first guess to a physical retrieval using the infrared data. With this method Goldberg achieved 0.9 K rms retrieval accuracy.
Jun Li and Allen Huang (U. Wisconsin), members of Bill Smith's team, presented a new approach to cloudy profile retrievals. Their method first estimates the cloud top pressure and fraction. This is followed by the simul-taneous retrieval of atmospheric profile and cloud parameters from the AIRS and AMSU data. Unlike the N-star method (used in the NOAA TOVS operational retrievals), this method is claimed not to amplify the noise. The rms error in the cloud top height was 37 mb, rms fractional cloud cover error was 5 percent, rms error in the temperature was 1.33 K, and rms error in total water vapor was 9.8 percent. The algorithm reached the required accuracy for water vapor, but not for the temperature retrieval. This was attributed to the need of the algorithm for accurate component transmittances (rather than the combined transmittance of all active gases). Component transmittances will be posted on the network.
Joel Susskind and his team at GSFC use a nine step startup procedure followed by an iteration, which is based on experience with TOVS data (Susskind et al. 1983). AMSU and AIRS data are used first to evaluate a parameter eta, which is related to the fractional cloud cover and cloud contrast (Chahine 1974). A first guess temperature profile is then derived using the AMSU data only. The rms error for the first guess temperature retrieval from the AMSU data alone is typically 2.56 K (for the A- track). The first guess profile is used as the input to the iterative loop to evaluate the final temperature and moisture profile, the surface temperature, and the ozone burden. The iteration uses the combined AIRS/AMSU/MHS data. The rms retrieval accuracy improved (for the A-track data) from 2.56 K to 0.98 K. The performance of this retrieval algorithm meets the AIRS 1 K rms accuracy requirement. Susskind noted that the cloudy data set is good as a test of the retrieval algorithm under cloudy conditions, but the cloud contrast conditions are much more severe than the TOVS data indicate: with TOVS, 46% of the time the field of view is clear of clouds and the average eta for the remaining data is 1.27. As eta becomes larger, the cloud contrast decreases and high quality retrievals become more difficult. The simulated cloudy test data contained no clear fields, with the average eta=2.0. Susskind also felt that the liquid water effects on the AMSU data were stronger than expected.
The science team presentations and discussions showed that the simulated data are suitable to proceed with their use for the core algorithm development. No team was expected to present retrievals from all test data. The fact that retrieval results from the NOAA and GSFC teams already met the 1 K rms retrieval accuracy requirement for part of the test data is very encouraging. The next meeting of the AIRS Science Team will be held from February 21-23, 1995, at UCSB, Santa Barbara, CA. The focus of the meeting will be the final results from the single layer gray cloudy test and a discussion of the simulation approach for the next two data sets: multilayer gray clouds and non-gray clouds. Selection of the core algorithm from the combination of the best algorithm elements generated by the three teams will be based on multilayer non-gray clouds. This selection is scheduled for the end of 1995.
References
Aumann, H. H., and R. J. Pagano, 1994: "Atmospheric Infrared Sounder on the Earth Observing System." Optical Engineering, 33 (3), 776-784.
Chahine, M. T., 1974: "Remote Sounding of Cloudy Atmospheres. I. The Single Layer Cloud." J. Atmos. Sci., 31, 233-243.
JPL, 1991: "AIRS Science and Measurement Requirements." JPL Publication D6665.
McMillan, L.M., 1991: "Evaluation of a Classification Method for Retrieving Atmospheric Temperatures from, 432-446.
McMillan, L.M., and C. Dean, 1982: "Evaluation of a New Operational Technique for Producing Clear Radiances." J. Appl. Meteor., 21,1005-1014.
Susskind J., J. Rosenfield, and D. Reuter, 1983: "An Accurate Radiative Transfer Model for Use in the Direct Physical Inversion of HIRS 2 and MSU Temperature Sounding Data." J. Geophys. Res., 88 C, 8550-8568.