Welcome to the NEUROSAT home-page at Aston.

Introduction

NEUROSAT is a European Union funded project involving research groups throughout Europe and assessing the potential benefits that neural networks can bring to remote sensing. The Aston contribution to the project, led by Ian Nabney working with Dan Cornford, will be largely concerned with investigating the potential of neural networks (and other mathematical techniques) to convert satellite scatterometer measurements of the ocean surface to wind vectors. An overview of the project can be found in NEUROSAT: An Overview [compressed postscript]. Additional funding for the work on ambiguity removal was provided from EPSRC grant GR/L 03088, Combining Spatially-Distributed Predictions from Neural Networks. This web page is intended to give a brief outline of our activities, but those interested in the details should down-load the referenced documents since these contain much more detail.

What is a Scatterometer?

Simply, a scatterometer is an instrument that measures the amount of back-scattered radiation received from the observed area, when illuminated by radiation of a given wavelength. Backscatter is denoted by sigma-0, the ratio of power transmitted to power received. In the context of this work we are investigating backscatter from the scatterometers aboard the ERS-1 and ERS-2 satellites. These active instruments operate at wavelengths at which the atmosphere is largely transparent so the strength of the signal received by the satellite will depend on the properties of the surface that the beam is `bounced off'. Over the ocean, the instantaneous wind causes small ripples of the order of centimeters wavelength, and it is these that affect the measured back-scatterometer. A large spatial average, as well as repeated measurements are taken to ensure representativity. More information on scatterometers and scatterometer links can be found here.

Problems We Are Addressing

The problem of wind retrieval from scatterometer data can be broken down into several components, each of which we hope to address. The majority of the methods are outlined in NEUROSAT: An Overview so we only list them here:

• modelling the unconditional distribution of local satellite measurements (sigma-0) for novelty detection
• modelling the local forward problem (recreation of a sensor model to predict the local satellite measurements given the local wind vector)
• modelling the local inverse problem (retrieval of the local wind vector(s) given a local satellite observation)
• ambiguity removal (selection of the correct wind field)

Modelling the forward problem

Modelling the inverse problem

Ambiguity removal

Software

More Information

We attempt to produce online versions of all relevant publications arising from the NEUROSAT project. You can see a list of all the available technical reports by following the link below:

Online NEUROSAT Publications

This list does not include some of the papers which can be accessed from these pages.

Software

Further information can be obtained by mailing d.cornford@aston.ac.uk.