Unsupervised data selection for focused time-lapse inversion in electrical resistivity tomography monitoring

Time-lapse electrical resistivity tomography has become a popular technique to monitor many subsurface processes. Inversions and interpretation often remain challenging because of the presence of noisy data and the superposition of several processes influencing the results. In this contribution, we apply for the first time clustering of data time series prior to the inversion process. We then invert only for a subset of data points displaying some interesting temporal variations. This approach allows to reduce data misfit within the data subset and successfully isolate the main processes from other taking place along the measuring profile. This opens new perspectives for the processing of time-lapse ERT data and the monitoring of subsurface processes.