Neural NetworksThis theme is coordinated by Filipe Aires ()
Artificial Neural Network (NN) techniques have emerged this last deceny as
a practical tool with many successful applications in meteorology and
climatology fields. The NN theory deals with a general set of statistical
non-linear models (Multi-Layered Perceptron, Kohonen Maps, Support Vector
Machines, etc.) used to analyze or simulate complex phenomena.
They can be used as a function approximation tool, for non-linear regression, to solve inverse problems, for classification tasks or for clustering.
In meteorology, they are used in satellite remote sensing to retrieve atmospheric (temperature, water vapor) or surface variables (skin temperature, emissivities) from the radiances measured by the satellite. They are also used to classify observations (vegetation index, wetlands detection, ocean color).
Another branch of applications deals with the analysis of the climate system: NN can be used to analyze feedback process in the climate, define a statistical modeling of dynamical systems such as the El Nino phenomenon, to extract independent climate modes of variability in big datasets of observations or model simulations.