In English

invlib: A generic implementation of Bayesian methods for inverse problems in remote sensing

Simon Pfreundschuh
Göteborg : Chalmers tekniska högskola, 2016. 79 s.
[Examensarbete på avancerad nivå]

Spaceborne earth observation missions constitute an essential source of information for climate science. In order to extract physically relevant quantities from the electromagnetic signals recorded by the detector onboard the satellite several data processing steps are required. This process, the so called data retrieval, requires a thorough mathematical formulation as well as an efficient implementation to ensure both the correctness of the retrieved data and the ability to handle large retrievals at a sufficiently high bandwidth. As part of this thesis project the invlib C++ template library has been developed, which implements Bayesian methods for inverse problems arising from the retrieval of remote sensing earth observation data. The library provides functionality for the memory-efficient calculation of maximum a posteriori estimators of Bayesian inverse problems with Gaussian priors and measurement error. This method, known in the field of remote sensing as the optimal estimation method (OEM), has been implemented using generic programming techniques in order to provide a maximum of flexibility and performance to the user of the library. The invlib library has been integrated into the Atmospheric Radiative Transfer Simulator (ARTS), which is a publicly available software package for the simulation of the propagation of electromagnetic radiation through the atmosphere. The integration of the retrieval functionality considerably simplifies the data retrieval workflow with ARTS, which previously required the use of a separate, external software package. In this thesis report the use of the invlib library for tomographic retrievals from the Odin SMR mission is demonstrated. Using the new implementation based on the invlib library, it is now possible to perform the tomographic retrievals of data from a complete half-orbit in a single computation. Furthermore, it is demonstrated how the computations can be parallelized using invlibs generic parallel matrix and vector types. Finally, also the use of the invlib library for the three-dimensional retrieval of simulated data from the MATS satellite mission is illustrated. The principal result of this thesis is a free software C++ template library for remote sensing data retrieval, that emphasizes generality and performance. Furthermore, the invlib library has been integrated into the ARTS software package and the retrieval functionality made available to a large community of existing users. The capabilities of the library to perform and accelerate the solution of real world retrieval problems has been demonstrated by applying it to data from two different earth observation missions.

Nyckelord: bayesian inversions, remote sensing



Publikationen registrerades 2016-10-26. Den ändrades senast 2016-10-26

CPL ID: 244226

Detta är en tjänst från Chalmers bibliotek