Full-Physics Inverse Learning Machine for Satellite Remote Sensing Retrievals
Abstract
The satellite remote sensing retrievals are usually ill-posed inverse problems that are typically solved by finding a state vector that minimizes the residual between simulated data and real measurements. The classical inversion methods are very time-consuming as they require iterative calls to complex radiative-transfer forward models to simulate radiances and Jacobians, and subsequent inversion of relatively large matrices. In this work we present a novel and extremely fast algorithm for solving inverse problems called full-physics inverse learning machine (FP-ILM). The FP-ILM algorithm consists of a training phase in which machine learning techniques are used to derive an inversion operator based on synthetic data generated using a radiative transfer model (which expresses the "full-physics" component) and the smart sampling technique, and an operational phase in which the inversion operator is applied to real measurements. FP-ILM has been successfully applied to the retrieval of the SO2 plume height during volcanic eruptions and to the retrieval of ozone profile shapes from UV/VIS satellite sensors. Furthermore, FP-ILM will be used for the near-real-time processing of the upcoming generation of European Sentinel sensors with their unprecedented spectral and spatial resolution and associated large increases in the amount of data.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2017
- Bibcode:
- 2017AGUFM.A52G..07L
- Keywords:
-
- 3315 Data assimilation;
- ATMOSPHERIC PROCESSES;
- 0430 Computational methods and data processing;
- BIOGEOSCIENCES;
- 1816 Estimation and forecasting;
- HYDROLOGY;
- 7212 Earthquake ground motions and engineering seismology;
- SEISMOLOGY