Using neural network ensembles for the operational retrieval of ozone total columns

This paper presents the operational retrieval of ozone total columns from atmospheric spectrometers using an algorithm based on neural network ensembles. Single neural networks are trained to approximate subregions of a complex multidimensional function; the neural networks are then combined using the mixture-of-experts model. The resulting multinetwork is being used as part of the operational processing of the GOME/ERS-2 data, including a near-real-time service.