In Magnetotellurics (MT) natural electromagnetic field variations are recorded to study the electrical conductivity structure of the subsurface. Thereby long time-series of electromagnetic data are subdivided into smaller segments, which are Fourier transformed and typically averaged in a statistically robust manner to obtain MT transfer functions. Unfortunately, nowadays the presence of man-made electromagnetic noise sources often deteriorates a significant fraction of the recorded time-series by overprinting the desired natural field variations. Available approaches to obtain undisturbed and high quality MT results include, for example robust statistics, remote reference or multi-station analyses which aim at the removal of outliers or uncorrelated noise. However, we have observed that intermittent noise often affects a certain time span resulting in a second cluster of transfer functions in addition to the expected true MT distribution. In this paper, we present a novel criterion for the detection and pre-selection of EM noise in form of outliers or additional clusters based on a distance measure of each data segment with regard to the centre of the data distribution. For this purpose, we utilize the Mahalanobis distance (MD) which computes the distance between two multivariate points considering the covariance matrix of the data that quantifies the shape and the size of multivariate data distributions. As the MD considers the covariance matrix, it corrects not only for different variances but also for any correlation between the data. The computation of both, the mean value and covariance matrix, is susceptible to ouliers (e.g. noise) and requires a statistically robust estimation. We tested several robust estimators, for example median absolute deviation or minimum covariance determinant algorithm and finally implemented an automatic criterion using a deterministic minimum covariance determinant algorithm. We will present results using MT data from various field experiments all over the world, which illustrate successfull data improvement. This approach is able to remove scattered data points as well as to reject complete data cluster originating from noise sources. However, like all purely statistical algorithms the criterion is limited to cases where the majority of the recorded data is well-behaved, that is noise content is below 50 per cent. If the majority of data points originates from noise sources, the new criterion will fail if used in an automatic way. In these cases, additional input by the user either manually or in an automated fashion can be utilized. We therefore suggest to use an add-on criterion to back the MD selection and subsequent robust stacking in form of a physically motivated constraint based on the magnetic incidence direction. This property indicates whether the magnetic field originates from various sources in the far field or from a strong and well defined source in the near field.

No abstract

<p>The Radio-Magnetotelluric (RMT) method is a geophysical near-surface imaging technique with a broad range of possible applications. In 2020, the GFZ Potsdam has acquired a newly developed horizontal magnetic dipole transmitter that allows the application of the RMT method even in regions with an insufficient coverage of radio transmitters which normally serve as source signal. First controlled-source RMT measurements were conducted at three different locations in Chile in 2020. Further measurements were recently conducted in Ireland.  As we are able to store the raw time series, we have full control over the subsequent data processing. The processing tools at GFZ include the modular processing suite EMERALD, which was originally designed for MT processing, but has recently been adapted to process RMT data. One main difference is that in RMT the transmitter data is considered as signal, while in natural source MT this would be regarded as electromagnetic noise that needs to be removed using automated robust statistical approaches. However, processing the entire time series in an automated manner has a large drawback: The different emitted frequencies are transmitted in a sweep implying that only a smaller fraction of the time series contains the required signal for a particular target frequency and leading to an unfavourable signal-to-noise ratio. Since it is technically impossible to have the same time base for the data logger and the transmitter with an accuracy of a few nanoseconds, an automated detection scheme is required to find time segments that contain the transmitter signal. Usually, several Gigabytes of raw time series are collected during field measurements, making manual editing and supervision of the time series virtually impossible. However, a careful selection of appropriate time segments is essential for the success of the data processing. To address the challenge, machine learning algorithms have a high potential to solve both problems. Initial experience was gained with a recurrent neural network approach in order to identify suitable time segments (Patzer & Weckmann, EMTF 2021 – conference contribution and personal communication). However, many questions remained open, e.g. if other machine learning algorithms can result in better performances, which machine learning algorithms are in principle suitable for the characteristics and properties of RMT time series and which parameters should be used as input variables (features) for the algorithms. A large number of machine learning algorithms exist, which can be divided into different groups according to their operating principle and their activity fields. We will test unsupervised methods, especially for clustering the data, to identify a set of suitable input variables. Subsequently, we will use these features to train supervised algorithms as logistic regression, support vector machine and different kinds of neural networks to find the best performing algorithm. We will mainly use the RMT data from Chile within the training process. Furthermore, we will test if the trained algorithm is applicable to other new data sets measured at different locations (e.g. Ireland) and/or with different equipment.</p>

<p>The Bohemian Massif is part of the geodynamically active European Cenozoic Rift System and represents its easternmost termination. The study area is situated at the junction of three different Variscan tectonic units and hosts beside the Eger Rift a series of different fault systems. The entire region is characterized by ongoing magmatic processes in the intra-continental lithospheric mantle expressed by, e.g., the occurrence of repeated earthquake swarms, the presence of Quaternary volcanoes, and massive degassing of mantle-derived CO<sub>2</sub> in mineral springs, mofettes as well as. Several geoscientific studies suggest that fluid circulation along the deep-reaching faults seems to play an important role in explaining the underlying geodynamic processes. As part of an ICDP drilling programme, we applied the Magnetotelluric (MT) method with the goal to contribute to the understanding of the physical and chemical processes and interaction that led to the magma and fluid transport by mapping potential fluid pathways from the crust-mantle boundary up to the surface. Here, we present 3D inversion models of two different overlapping regional and one local MT experiments located in the Cheb basin close to the Hartoušov mofette field. The most prominent large-scale conductivity features of the regional models are several channels from the lower crust to the surface, possibly representing pathways for fluids into the earthquake swarm region, mofette fields, and known spas. However, such a conductive channel is absent in the local model beneath the surface expression of the mofettes. We will test two different hypotheses, namely a vertical ascending channel versus lateral fluid migration. Results from synthetic modelling studies and available geoscientific constraints hint that such a channel might exist directly beneath the mofette field, but due to the given data quality, station distribution, and the subsurface conductivity structure within a conductive sediment basin, it might be challenging to resolve.</p>

<p>The West Bohemian Massif represents the easternmost part of the geo-dynamically active European Cenozoic Rift System. This region hosts different tectonic units, the NE-SW trending Eger Rift, the Cheb Basin and a multitude of different faults systems. Furthermore, the entire region is characterised by ongoing magmatic processes in the intra-continental lithospheric mantle. These processes take place in absence of active volcanism at surface, but are expressed by a series of phenomena, including e.g. the occurrence of repeated earthquake swarms and massive degassing of CO<sub>2</sub> in the form of mineral springs and mofettes. Active tectonics is mainly manifested by Cenozoic volcanism represented by different Quaternary volcanic structures e.g. the Eisenbühl, the Kammerbühl and different maars. All these phenomena make the Eger Rift a unique target area for European intra-continental geo-scientific research. Therefore, an interdisciplinary drilling programme advancing the field of earthquake-fluid-rock-biosphere interaction was funded within the scope of the ICDP. Magnetotelluric (MT) measurements are applied to image the subsurface distribution of the electrical conductivity from shallow surface down to depths of several tens of kilometres. The electrical conductivity is a physical parameter that is particularly sensitive to the presence of high-conductive phases such as aqueous fluids, partial melts or metallic compounds. First MT measurements within this ICDP project were carried out in winter 2015/2016 along two 50 km long perpendicular profiles with 30 stations each and a denser grid of 97 stations close to the mofettes with an extension of 10 x 5 km<sup>2</sup>. Muñoz et al. (2018) presented 2D images along the NS profile of one regional profile. They reveal a conductive channel at the earthquake swarm region that extends from the lower crust to the surface forming a pathway for fluids into the region of the mofettes. A second conductive channel is present in the south of the model. Due to the given station setup, the resulting 2D inversion allows ambiguous interpretations of this feature. 3D MT data and inversions are required to distinguish between different scenarios and to fully describe the 3D structure of the subsurface. Therefore, we conducted a large MT field experiment in autumn 2018 by extending the study area towards the south. Broad-band MT data were measured at 83 stations along three 50-75 km long profiles and some additional stations across the region of the maars, the Tachov fault and the suture zone allowing for 2D as well as 3D inversion on a crustal scale. To improve the data quality, advanced data processing techniques were applied leading to good quality transfer functions. Furthermore, the previously collected MT data were reprocessed using the new approaches. This entire MT data set across the Eger Rift environment together with old MT data collected within the framework of the site characterisation in the surrounding of the KTB drilling are used to compute 3D resistivity models of the subsurface, with combining different transfer functions. These 3D inversion results will be introduced and discussed with regard to existing geological hypotheses.</p><p> </p>