You must have a javascript-enabled browser and javacript and stylesheets must be enabled to use some of the functions on this site.


Non-linear inverse problems in satellite remote sensing

Johanna Tamminen(1), Marko Laine(1), Erkki Kyrölä(1) and Pepijn Veefkind(2)

(1) Finnish Meteorological Institute, POBox 503, 00101 Helsinki, Finland
(2) KNMI, PO Box 201, NL-3730 AE De Bilt, Netherlands


Complicated and challenging non-linear inverse problems can be solved in fully Bayesian way by applying MCMC (Markov chain Monte Carlo) methodology. This technique is nowadays commonly used in various complex modeling problems in a wide range of applications. In satellite remote sensing the non-linear nature of the problem is commonly not studied much. However, thorough understanding of the posterior distribution is needed when the data is combined with other measurements or further assimilated into atmospheric models.

The adaptive variations of the MCMC are flexible and easy to implement. The MCMC technique allows also non-Gaussian measurement noise and prior information. We demonstrate the capabilities of the MCMC methodology by applying it to Envisat/GOMOS and EOS-Aura/OMI retrieval problems. As a special case we show how aerosol model selection can be included in the retrieval problem in a Bayesian way.