RETRIEVAL OF ATMOSPHERIC TEMPERATURE PROFILES FROM SATELLITE SOUNDING DATA WITH A NEURAL NETWORK

THE PROPOSED STUDY WILL DEMONSTRATE THE FEASIBILITY OF USING A NEURAL NETWORK TO RETRIEVE ENVIRONMENTAL PARAMETERS SUCH AS THOSE DISCUSSED IN THE 1986 JOINT CHIEFS OF STAFF MEMORANDUM MJCS 154-86. BECAUSE NEURAL NETWORKS ARE INHERENTLY PATTERN PROCESSORS, THEY ARE EXCELLENT CANDIDATES TO PROCESS THE OUTPUTS OF SATELLITE-BORNE SOUNDERS. A NETWORK IMPLEMENTED ON A DESKTOP COMPUTER WILL BE TRAINED TO MATCH SOUNDING DATA TO MEASURED TEMPERATURE PROFILES TAKEN OVER LAND AND OCEAN, IN TROPICAL AND MID-LATITUDE REGIONS, AND IN DIFFERENT SEASONS. IT WILL, THEREAFTER BE ABLE TO QUICKLY SYNTHESIZE TEMPERATURE PROFILES FROM NEW SOUNDING DATA, EVEN IF THE DATA WERE COLLECTED UNDER CONDITIONS NOT EXACTLY MATCHING THOSE REPRESENTED IN THE TRAINING SET. PHASE I OF THE PROJECT WILL CHOOSE THE BEST NETWORK FOR THE APPLICATION AND ESTABLISH THE FEASIBILITY AND ACCURACY OF THE METHOD. TESTS WILL BE RUN USING PROPRIETARY NETWORK DESIGNS TO DEMONSTRATE THE EFFICACY OF THE METHOD. THE ULTIMATE GOAL OF LATER PHASES OF THE RESEARCH IS TO INCREASE THE ACCURACY AND UTILITY OF ATMOSPHERIC PARAMETER RETRIEVAL USING DESKTOP COMPUTERS AND A PROPRIETARY APPROACH DESCRIBED IN THE PROPOSAL.