Logo: to the web site of Uppsala University

We advance a previously established method for 2D inversion of electromagnetic data, in which the smoothness constraints are locally reweighted according to the seismic envelope fractional gradients (SEFGs). As the first step of our modifications, seis-mic envelope values are edited and normalized. This results in the rejection of noise-contaminated parts, clipping the envelope outliers and increasing the reflectivity power of weak seismic signals. Second, we introduce a weighting matrix in the normali-zation process to incorporate prior information in the con-strained inversion regarding the relative contrasts of electrical resistivity in the given parts of the model. Third, due to normal-izing the SEFG, there is no need to search for an optimum sta-bilization factor to modify local smoothing weights, and hence, we set it to a constant value. Finally, an Occam inversion with additional Levenberg-Marquardt damping is used to mitigate possible artifacts in the resistivity model generated by reflection seismic constraints. In applying the proposed scheme to a syn-thetic example, interfaces of various geologic units are restored as sharp boundaries. In addition, artifacts generated in the origi-nal approach are effectively mitigated thanks to the applied nor-malization process. For the first time, we apply the method to radio-magnetotelluric (RMT) and controlled-source audio-mag-netotelluric (CSAMT) field data acquired along a profile across a known aquifer in Heby, Sweden. Our inversion models com-pare favorably to previously presented results from seismic and geoelectric data. By modifying the smoothness weights based on SEFG, the depth to the bedrock is recovered well, being con-strained by the corresponding interfaces in the seismic image. The resistivity models from seismically constrained inversions of RMT and CSAMT data reveal steeply dipping and possibly fractured bedrock underneath the valley-shaped aquifer in the area. This interpretation is verified by borehole logs.