A Neural Network Retrieval Technique for High-Resolution Profiling of Cloudy Atmospheres

Abstract

The synergistic use of microwave and hyperspectral infrared sounding observations gives rise to a rich array of signal processing challenges. Of particular interest are the following elements which are combined for the first time in the retrieval technique presented here: 1) radiance noise filtering and redundancy removal (compression) using principal components transforms and canonical correlations, 2) data fusion (infrared plus microwave at possibly different spatial and spectral resolutions) and stochastic cloud clearing (SCC), and 3) geophysical product retrieval from spectral radiance measurements using neural networks. In this paper, we describe the algorithm and demonstrate performance using the Atmospheric Infrared Sounder (AIRS) and the Advanced Microwave Sounding Unit (AMSU). We show that performance is improved by approximately 25%-50% using the neural network method relative to other common techniques. Furthermore, we quantify the improvement in the vertical resolution of the retrieved products.