You must have a javascript-enabled browser and javacript and stylesheets must be enabled to use some of the functions on this site.


Neural Networks to Rretrieve Tectonic Acticity Parameters by SAR Interferometry

Fabio Del Frate(1) and Fabrizio Rossi(1)

(1) University of Tor Vergata, Via del Politecnico, 1, 00133 Roma, Italy


SAR Interferometry (InSAR) technique has demonstrated to be a powerful tool for surface movements detection. From its first well known application in 1992 (Landers Earthquake) InSAR has been used in a number of strong (Kobe, 1997; Izmit, 1999; Hector Mine, 2000; Denali, 2002) and moderate earthquakes (Colfiorito, 1997). InSAR allows to reconstruct the co-seismic surface displacement field with a centimetric accuracy using pre- and post- event SAR data. It can also show pre- and post- event movements, whenever present. In the recent years InSAR potentialities, together with classic seismological and geophysical data such as strong motion and GPS, have been also used by geophysicists for the assessment of forward fault models, generally stemming from the Okada formulation. InSAR measurement contains information useful to define the fault geometry (dip and strike angle; width and length), the extension of the rupture, the distribution of slip on the fault plain. In particular, for modelling radar interferometric data the RNGCHN software calculates displacement components, expression at the surface of the seismic source in an elastic half-space. Of great interest and usefulness in this context is the solution of the inverse problem, that means to recover the source parameters from the knowledge of InSAR surface displacement field. To this aim, some significant results have been achieved by means of the simulated annealing technique. However, due to the intrinsic ill-posedeness of the problem, some issues remain still open and more investigation is needed. Neural networks have already been recognized as being a powerful tool for inversion procedure in remote sensing applications. They are composed of many nonlinear computational elements (called neurons) operating in parallel and connected by the so called synapses. The use of neural networks is often effective because they can simultaneously address nonlinear dependencies and complex physical behaviour. In this study we propose an alternative approach for the retrieval of fault parameters from interferometric data based on neural networks. The network is trained by using a data set generated by the RNGCHN software and then is tested on real measured data. The input of the net consists of a set of features calculated from the interferometric image while the output vector contains the parameters characterizing the fault. We started focusing on a restricted number of parameters progressively augmenting the number of components to be retrieved. The results that have been obtained are encouraging and show the validity of such an approach. It has to be reminded that another advantage of neural networks is their flexibility and portability. Such a property could be exploited by adding in the input vector other pieces of independent information derived from several types of ancillary data.


Workshop presentation

Full paper

Keywords: ESA European Space Agency - Agence spatiale europeenne, observation de la terre, earth observation, satellite remote sensing, teledetection, geophysique, altimetrie, radar, chimique atmospherique, geophysics, altimetry, radar, atmospheric chemistry