For each profile, the inversion was stopped when the improvement in the RMS misfit was not significant. The combined scheme yields intermediate results. Usually, the L2-norm yields smoothly varying resistivity "blobs" while the L1 yields blocky profiles with horizontal and vertical gradients. It seems to give better results in resolving compact structures where the width and thickness are slightly smaller than depth, such as a cave or an ore-body and RES2DINV GUIs. The Combined Marquardt and Occam inversion method combines the Marquardt (or damped least square method) with the smoothness-constrained method. If there are sharp boundaries, one should choose the robust model inversion constraint. In practice: if the subsurface resistivity changes in a smooth manner, one should use the standard least-squares constraint.
The standard least-square constraint (L2) attempts to minimize the square of the differences between the observed and calculated apparent resistivity values, the robust constraint (L1) is less sensitive to very noisy data points but might give higher apparent resistivity RMS error.
The three outputs are presented with this article. The inversion is done using three different schemes, namely a "normal" scheme (L2 norm, standard least-square constraint), a "robust" scheme (L1 norm) and a "combined" scheme (Combined Marquardt and Occam inversion method). Data points with relative error larger than 2.5% are discarded. ERT data has been processed using RES2DINV.
0 kommentar(er)
