Neural Network Applications in High Resolution Atmospheric Remote Sensing

VOLUME 15, NUMBER 2, 2005 LINCOLN LABORATORY JOURNAL 299 M odern spaceborne atmospheric sounders measure radiance with unprecedented resolution and accuracy in spatial, spectral, and temporal dimensions. For example, the Atmospheric Infrared Sounder (AIRS), operational on the NASA Earth Observing System (EOS) Aqua satellite since 2002, provides a spatial resolution of 15 km, a spectral resolution of ν ν ∆ ≈ 1200 (with 2378 channels from 650 to 2675 cm), and a radiometric accuracy on the order of ±‐0.2 K. Typical polar-orbiting atmospheric sounders measure approximately 90% of the earth’s atmosphere (in the horizontal dimension) approximately every twelve hours. Retrieval algorithms estimate the geophysical state of the atmosphere as a function of space and time from upwelling spectral radiances measured by the sensor. In this article, we present two examples of neural-network-based atmospheric retrieval algorithms being developed and implemented at Lincoln Laboratory. In the first example, we consider the retrieval of atmospheric temperature and moisture profiles (quantity as a function of altitude) from hyperspectral radiance measurements in the thermal infrared. A projected principal component (PPC) transform is used to reduce the dimensionality of the spectral-radiance data and optimally extract geophysical information. A multilayer feed-forward neural network (NN) is subsequently used to estimate the desired geophysical profiles. This algorithm is known as the PPC/NN algorithm. The PPC/NN algorithm offers the numerical stability and efficiency of statistical methods while achieving accuracies comparable to those of physical, model-based methods. In the second example, we consider the retrieval of precipitation rates from passive microwave radiance measurements at frequencies near the oxygen and Neural Network Applications in High-Resolution Atmospheric Remote Sensing