Smooth MT Transfer Function Estimation by an Inverse Scheme

The geophysical exploration of geothermal areas strongly relies upon Magnetotellurics (MT) method, that exploits the measurement of the low-power natural EM field at the Earth's surface. Through the estimation of the MT transfer functions (TF), relating the electric to the magnetic field, the subsurface resistivity distribution can be inferred. Due to the diffusive nature of the low-frequency EM field, the MT TF's are inherently smooth, and smoothness is the main criterion adopted by EM community to assess the estimation quality. In the presence of Gaussian noise, the frequency-domain the least squares (LSQ) method provides the best possible estimate; natural MT data, however, contains a significant amount of non-stationary data that constitute outliers for the LSQ procedure. These outliers make the TF’s sharp at several frequencies, according to the nature of the noise. In order to circumvent this problem, robust methods were introduced, and provide a smooth TF, if non-stationarity is a minor fraction of the record. However, as geothermal development proceeds on a global scale, investigations can involve densely populated and industrialised areas. In these zones, high-power artificial is more likely present. Since these disturbance signals can by persistent, robust methods can fail; moreover, preliminary filtering can be ineffective. In these conditions, the applicability of MT is severely hampered. In some cases, smoothness is a-posteriori introduced by splines or smoothing procedures, but this approach lacks physical consistency. It can also be part of estimation methods but implying the adoption of some arbitrary assumptions. Here we propose a new heuristic algorithm to reach the maximum TF's smoothness through an inverse scheme applied to event rejection in frequencydomain. The algorithm searches for frequency-dependent power thresholds to be applied to the events, in order to achieve the maximum smoothness in the TF's. The smoothness is the objective function to be minimized, and the model space is constituted by the infinite set of threshold vectors. We found that after the completion of the process, the distribution of the event powers is more Gaussian, and then, more suitable for LSQ estimation; the corresponding residuals are consequently closer to a Rayleigh distribution. Physical consistency of the resulting TF has been tested by 1D inversion. The algorithm can, therefore, be combined with the MT remote-reference technique, and we found that it is effective to reduce the effects of artificial strong-power signal that can deteriorate the acquisitions. We present the successful application of this new technique to MT data collected over an East African Rift System geothermal area.