AEA Technology Ltd is carrying out research for the European Space Agency to attempt to make progress in weather prediction by training a neural network to do the task. It has been using data on ocean surface wind velocities as the basis of the work. A satellite scatters light radar beams onto the ocean’s surface and the reflections that come back are recorded. The data gleaned from this is fed into a multilayer perceptron type of network to build a model of what the surface looks like, a factor related to the speed and direction of the wind. The project is being hampered by the fundamental ambiguity of the wind data being used as the training set – such as the possibility that local winds may not have anything to do with the waves at any given point because they are generated thousands of miles away and can take two weeks to reach any given point, for example. – Abigail Waraker
