2D MODELING WITH LINEARIZED INTEGRAL-EQUATIONS

A reduced COPROD2 data set with response estimates at twenty sites for four periods (85 s to 683 s) is interpreted, using an iterative model ling scheme on the basis of linearized integral equations. Input data are the anomalous fields, here E(ax) and B(az) for E-polarisation, whi ch are derived from the supplied impedances Z(xy) and the magnetic tra nsfer functions T(zy). Prior to 2D modelling a normal 1D reference mod el is introduced (here a 3-layer model) and an anomalous domain define d (here from zero to 40 km depth and 200 km in width). It is subdivide d into M subdomains of constant anomalous conductivity. The linearisat ion of the non-linear data functional is performed by approximating th e internal field E(x) within the anomalous domain. An iterative proces s is started with the normal field E(nx) of the 1D reference model as internal field, gradually improving this first approximation. The evol ving linear problem is solved by the least-squares method, adapting th e data kernel with each iteration step better to the model which arise s from the application of this kernel to the data. No Frechet derivati ves of the data functional are involved and no starting model is requi red. A first set of models is derived from MT data alone, a second set from combined MT/GDS data, increasing the number of subdomains from M = 1 to M = 64. It is found that with M = 8 (i.e. with 20 x 50 km2 Sub domains) the resolution power of the data is exhausted. The resulting models have an almost uniform top layer and a deep-seated central regi on of reduced resistivity of 10 OMEGAm at 20 to 40 km depth. Further m odelling studies show that a deep origin of the observed anomalies is indeed more likely than a shallow origin and that the modelling result s do not depend significantly on the used periods. The mean residual ( if only MT data are used) is greater than the data error and the indiv idual residuals are not randomly distributed; both indicate that the d ata have not been exploited to their fullest possible extent. Forward modelling shows that the models are not in good agreement with B-polar isation impedances.