The algorithm to derive the concentrations of water constituents from MERIS (Medium Resolution Imaging Spectrometer, aboard the ESA satellite ENVISAT launched March 2002) is based on Neural Net (NN) technology. The NN not only transforms water leaving radiance reflectances pixel by pixel with high efficiency into concentrations of water constituents but also checks if its input is in the domain which was covered during the training of the NN.

The construction of the NN is based on a large table generated by radiative transfer code. Two NN's are trained with this table: (1) NNinv to emulate the inverse model (reflectance, geometry) -> concentrations, and (2) NNforw to emulate the forward model (concentrations, geometry) -> reflectance.

In the retrieval as it is installed now the two NN's together with a comparison network are combined to give a new NN which first uses the NNinv to obtain an estimate of the concentrations. These are fed into the NNforw and the derived reflectances are compared with the measured reflectances by a third NN component. Large deviations signal a violation of the necessary condition for a successful inversion; corresponding pixels are then flagged.

In this talk we demonstrate an improvement of the above procedure which uses the result obtained by NNinv as a first guess for a minimization procedure. In the minimization procedure the concentrations - input to NNforw - are changed to minimize the distance from the measured reflectances. After minimization the covariance matrix of the concentrations can be derived.

A few examples of such resulting concentration and error maps from the North Sea are discussed.