3-D inversion techniques have become a widely used tool in magnetotelluric (MT) data interpretation. However, with real data sets, many of the controlling factors for the outcome of 3-D inversion are little explored, such as alignment of the coordinate system, handling and influence of data errors and model regularization. Here we present 3-D inversion results of 169 MT sites from the central San Andreas Fault in California. Previous extensive 2-D inversion and 3-D forward modelling of the data set revealed significant along-strike variation of the electrical conductivity structure. 3-D inversion can recover these features but only if the inversion parameters are tuned in accordance with the particularities of the data set.Based on synthetic 3-D data we explore the model space and test the impacts of a wide range of inversion settings. The tests showed that the recovery of a pronounced regional 2-D structure in inversion of the complete impedance tensor depends on the coordinate system. As interdependencies between data components are not considered in standard 3-D MT inversion codes, 2-D subsurface structures can vanish if data are not aligned with the regional strike direction. A priori models and data weighting, that is, how strongly individual components of the impedance tensor and/or vertical magnetic field transfer functions dominate the solution, are crucial controls for the outcome of 3-D inversion. If deviations from a prior model are heavily penalized, regularization is prone to result in erroneous and misleading 3-D inversion models, particularly in the presence of strong conductivity contrasts. A 'good' overall rms misfit is often meaningless or misleading as a huge range of 3-D inversion results exist, all with similarly 'acceptable' misfits but producing significantly differing images of the conductivity structures. Reliable and meaningful 3-D inversion models can only be recovered if data misfit is assessed systematically in the frequency-space domain.

To advance and optimize secondary and tertiary oil recovery techniques, it is essential to know the areal propagation and distribution of the injected fluids in the subsurface. We investigate the applicability of controlled‐source electromagnetic methods to monitor fluid movements in a German oilfield (Bockstedt, onshore Northwest Germany) as injected brines (highly saline formation water) have much lower electrical resistivity than the oil within the reservoir. The main focus of this study is on controlled‐source electromagnetic simulations to test the sensitivity of various source–receiver configurations. The background model for the simulations is based on two‐dimensional inversion of magnetotelluric data gathered across the oil field and calibrated with resistivity logs. Three‐dimensional modelling results suggest that controlled‐source electromagnetic methods are sensitive to resistivity changes at reservoir depths, but the effect is difficult to resolve with surface measurements only. Resolution increases significantly if sensors or transmitters can be placed in observation wells closer to the reservoir. In particular, observation of the vertical electric field component in shallow boreholes and/or use of source configurations consisting of combinations of vertical and horizontal dipoles are promising. Preliminary results from a borehole‐to‐surface controlled‐source electromagnetic field survey carried out in spring 2014 are in good agreement with the modelling studies.

S U M M A R YWith advancing computational resources, 3-D inversion techniques have become feasible in recent years and are now a more widely used tool for magnetotelluric (MT) data interpretation. Galvanic distortion caused by small-scale near-surface inhomogeneities remains an obstacle for 3-D MT inversion which so far has experienced little attention. If not considered properly, the effect on 3-D inversion can be immense and result in erroneous subsurface models and interpretations. To tackle the problem we implemented inversion of the distortion-free phase tensor into the ModEM inversion package. The dimensionless phase tensor components describe only variations of the conductivity structure. When inverting these data, particular care has to be taken of the conductivity structure in the a priori model, which provides the reference frame when transferring the information from phase tensors into absolute conductivity values. Our results obtained with synthetic data show that phase tensor inversion can recover the regional conductivity structure in presence of galvanic distortion if the a priori model provides a reasonable assumption for the regional resistivity average. Joint inversion of phase tensor data and vertical magnetic transfer functions improves recovery of the absolute resistivity structure and is less dependent on the prior model. We also used phase tensor inversion for a data set of more than 250 MT sites from the central San Andreas fault, California, where a number of sites showed significant galvanic distortion. We find the regional structure of the phase tensor inversion results compatible with previously obtained models from impedance inversion. In the vicinity of distorted sites, phase tensor inversion models exhibit more homogeneous/smoother conductivity structures.

S U M M A R YOver the last decades, electromagnetic methods have become an accepted tool for a wide range of geophysical exploration purposes and nowadays even for monitoring. Application to hydrocarbon monitoring, for example for enhanced oil recovery, is hampered by steelcased wells, which typically exist in large numbers in producing oil fields and which distort electromagnetic fields in the subsurface. Steel casings have complex geometries as they are very thin but vertically extended; moreover, the conductivity contrast of steel to natural materials is in the range of six orders of magnitude. It is therefore computationally prohibitively costly to include such structures directly into the modelling grid, even for finite element methods. To tackle the problem we developed a method to describe steel-cased wells as series of substitute dipole sources, which effectively interact with the primary field. The new approach cannot only handle a single steel-cased well, but also an arbitrary number, and their interaction with each other. We illustrate the metal casing effect with synthetic 3-D modelling of land-based controlled source electromagnetic data. Steel casings distort electromagnetic fields even for large borehole-transmitter distances above 2 km. The effect depends not only on the distance between casing and transmitter, but also on the orientation of the transmitter to the borehole. Finally, we demonstrate how the presence of steel-cased wells can be exploited to increase the sensitivity and enhance resolution in the target region. Our results show that it is at least advisable to consider the distribution of steel-cased wells already at the planning phase of a controlled source electromagnetic field campaign.

Over the last decade, an increasing number of numerical studies have proposed controlledsource electromagnetic (CSEM) techniques for monitoring of fluid flow in reservoirs, for example, in the framework of hydrocarbon production or CO 2 storage scenarios. A fundamental prerequisite for any monitoring application in practice is repeatability of the measurements, particularly in areas with high noise levels.Here, we report on CSEM data acquired across a producing oil field on land in three consecutive surveys between 2014 and 2016. As major conductivity changes in the reservoir structure are not expected for this time frame, the data sets provide an excellent basis to study accuracy and repeatability of such measurements over a time span of 2.5 yr.Our results show that uncertainties of single CSEM measurements lie between 0.1 and 10 per cent with a focus around 1 per cent in all surveys. For source-receiver offsets <2 km uncertainties are in the range of ∼0.1-0.3 per cent, proportional to the transfer function amplitudes, and are dominated by intrinsic noise of the measuring system. At source-receiver distances >4 km external noise resulting from natural electromagnetic field variations and powered installations dominates uncertainties that assume minimum absolute values of 10 −9 -10 −10 V A −1 m −1 with lowest values at frequencies between 0.1 and 10 Hz.Overall, repeatability of CSEM measurements depends on a range of factors, including source-receiver distances, component of the transfer function, source-polarization and relocation errors, in particular at sites close to the source, where the geometry and characteristics of the source fields vary rapidly in space. Best repeatability was observed for receiver stations at 2-4 km distance from the source and frequencies <20 Hz. At these stations, phases and amplitudes of transfer functions usually agreed within ±1 • and ±5 per cent between measurements. Such values are in a range as expected from time-lapse signals due to resistivity changes in target (reservoir) formations. Hence, precise surveying procedures are essential.We also measured the vertical electric field (E z ) with a newly developed receiver chain in a 200 m deep observation borehole. The vertical electric field component shows generally higher sensitivity to resistivity changes in reservoir structures than the horizontal electric fields measured at surface. Although amplitudes of E z are about one to two orders of magnitude smaller than amplitudes of horizontal electric fields, recordings of E z are stable. More importantly, E z transfer functions of three measurements between 2015 and 2016 show excellent quality and repeatability within <±2 • and <±5 per cent, similar as horizontal electric fields indicating that noise conditions at depth improve when compared with sensors at surface.