Athias, V. LEGOS, Veronique.Athias@cnes.fr
Mazzega, P. LEGOS, firstname.lastname@example.org
Jeandel, C. LEGOS, Catherine.Jeandel@cnes.fr
TESTING NONLINEAR GLOBAL OPTIMIZATION ALGORITHMS FOR THE INVERSION OF IN SITU DATA FROM DISSOLVED-PARTICULATE EXCHANGES
The dynamics of the dissolved-particulate exchanges in the oceanic water column is ruled by a set of Nonlinear Coupled Partial Differential Equations. The inverse problem we are concerned with consists in recovering the exchange rates of chemical elements between the water column reservoirs from field measurements.
First, we developed a method to solve this nonlinear inverse problem by using a synthetic data set computed by a direct integration of our model. It is based on the computation of a quadratic cost function J, that measures the departure between the observed fields and their model counterparts predicted with trial control parameters. Because of the nonlinearity, the multidimensional cost function displays several minima, saddle points and limited degeneracies over the parameter space. We show that a direct cartography of J allows to solve inverse problems of low dimension. Otherwise, one has to resort to nonlinear global optimization algorithms. We compared the efficiency of the simulated annealing, the genetic and the TRUST algorithms.
Second, we applied this method to the inversion of in situ data, using two-years' time series of trapped, suspended and dissolved material collected in the tropical Atlantic Ocean (EUMELI sites).
Day: Monday, Feb. 1
Time: 02:45 - 03:00pm
Location: Eldorado Hotel