Hyperspectral data compression and science algorithms for the NEMO satellite

The Naval EarthMap Observer (NEMO) will carry a hyperspectral sensor to characterize the littoral environment for scientific studies and to meet Joint Naval needs. The sensor will measure 210 spectral channels over the range of 400 to 2500 nm. A 30 or 60 m GSD and a total crosstrack swath of 30 km will be used. The values of GSD and SNR will be sufficient to allow for long term monitoring and real-time characterization of the coastal environment. A unique feature of the spacecraft will be the on-board processing system which will allow for real time processing, feature extraction and data compression using the Optical Real-time Adaptive Spectral Identification System (ORASIS) algorithm. The ORASIS spectral processing system is based on linear spectral unmixing, which decomposes the observed spectrum of each pixel into a linear sum of the products of the spectra of the "pure" substances in the scene, termed the endmembers, multiplied by their respective abundances. It is non-statistical in nature. ORASIS compression is a variation of I,earned Vector Quantization (LVQ) that, while lossy, preferentially preserves the important spectral information. The compression allows for a data collection rates exceeding 60 GB/day. Because of the large data volumes, science algorithms being developed are designed to operate directly on the compressed data. Algorithms include those for atmospheric correction, bathymetry, water clarity, chlorophyll content, water hazards, etc. We demonstrate the effect of compression on the output of the ATREM code as well as general classification algorithms.