Single-site magnetotelluric response functions using b-robust w-estimators, with an application to earthquake prediction research

Résumé

The accumulation of stress and strain is known to induce changes in the electrical properties of rocks, which can be monitored for signs of earthquake preparation processes. To this effect, the Magnetotelluric sounding method presents some unique advantages. However, single-site MT data are notoriously susceptible to natural or anthropogenic time-varying coherent noise, which may severely bias the response function estimators and degrade their repeatability, unless treated with dedicated processing techniques. Such a technique is presented herein, involving the W-estimator with random error weighting, followed by an iterative robustification scheme based on an influence function approach. The algorithm is demonstrated on a set of severely distorted data exhibiting a marginal distribution of outliers, and is shown to effectively reduce the bias errors and the variance. It is also applied to the long-term monitoring of crustal resistivity with MT response functions at a noisy site located near Aerino village, SE Thessaly, Greece, achieving a sustainable repeatability threshold of 10-20% and faring very well with respect to the data quoted from the international literature.