Mcmillin, Larry | Search NASA

NASA NTRS1994

An improved cloud retrieval algorithim using HIRS2-MSU radiance measurements

Cloud-top heights and cloud amounts are produced as part of the operational processing of polar-satellite data at the National Environmental Satellite Data and Information Service (NESDIS). These products were compared with similar products from the air force's real-time nephanalysis (RTNEPH), from the International Satellite Cloud Climatology Project, and from NASA Goddard's processing of satellite data. It was found that the amount of high-level cloud was too small in the NESDIS results, while the amount of low-level cloud was too large. An examination of the NESDIS algorithm revealed that the differences in cloud distributions were caused by the selection of channels used for the cloud retrievals. Cloud retrievals are most accurate at the levels at which the channels that are used are most sensitive. In addition, it was found that no one pair of channels was best at all levels. A new procedure was developed that varied the channels as a function of an initial estimate of the cloud height. This procedure produced improved cloud retrievals that were then compared with the RTNEPH results. The comparison showed that the two methods provide similar retrievals of cloud height and amount.

Mcmillin, Larry↗

NASA NTRS1993

Neural networks for atmospheric retrievals

We use neural networks to perform retrievals of temperature and water fractions from simulated clear air radiances for the Atmospheric Infrared Sounder (AIRS). Neural networks allow us to make effective use of the large AIRS channel set, and give good performance with noisy input. We retrieve surface temperature, air temperature at 64 distinct pressure levels, and water fractions at 50 distinct pressure levels. Using 728 temperature and surface sensitive channels, the RMS error for temperature retrievals with 0.2K input noise is 1.2K. Using 586 water and temperature sensitive channels, the mean error with 0.2K input noise is 16 percent. Our implementation of backpropagation training for neural networks on the 16,000-processor MasPar MP-1 runs at a rate of 90 million weight updates per second, and allows us to train large networks in a reasonable amount of time. Once trained, the network can be used to perform retrievals quickly on a workstation of moderate power.

Motteler, Howard E.↗