Satellite remote sensing of ozone using a Full-Physics Inverse Learning Machine

Abstract

The new generation of environmental satellites with increased spatial and spectral resolutions imposes critical challenges for the processing of the Big Data. This work employs the newly-developed full-physics inverse learning machine (FP-ILM) to estimate vertical distributions of ozone from Global Ozone Monitoring Experiment - 2 (GOME-2) measurements and analyzed its performance. The obtained ozone profile shapes are further used to derive the vertical column density of ozone. The main advantage of FP-ILM is that, unlike classical retrieval algorithms, the ozone profile retrieval is formulated as a classification problem, producing a significant speed-up and reliable accuracy. The time-consuming radiative transfer computations and neural network training are performed off-line and do not introduce additional performance bottlenecks in the whole processing chain. Therefore FP-ILMs are suitable for processing remote sensing Big Data.