A Bayesian-based Methodology to Detect Suspicious Electrodes for Long-term Permanent ERT Monitoring Experiments

During long-term ERT monitoring experiments, the data quality may reveal time-dependent mainly due to changes in galvanic contact resistance between buried electrodes and soil. Identifying suspicious electrodes in a permanent spread is of major importance as a faulty electrode may affect the quality of tens to hundreds of measurements on each time-slice. An automated methodology was developed to detect suspicious electrodes based on a Bayesian approach. This methodology allows pointing out faulty electrodes based on the analyses of temporal sets of measurements, each one containing multiple electrode arrays. Standard and studentized estimators of the influence of each electrode in the global data quality are computed for each time-slice based on the measurement quality factor Q given as a coefficient of variation of repeated measures. The automated detection of faulty electrodes is obtained by comparing the computed studentized estimators to values expected when every electrode can be considered as good for the given data set. These expected values are computed by Monte Carlo simulations using a distribution of Q factor values of quadripoles selected as good based on reciprocal errors. The efficiency of the proposed methodology is assessed on a field experiment.