Quantification of Temperature with Time-lapse electrical resistivity using the prediction-focused approach-A field case

Standard inversion of time-lapse geophysical suffers from spatially and temporally varying resolution due to the regularization procedure used during the inversion process. In this study, we apply the recently developed prediction-focused approach (PFA) to directly estimate temperature with electrical resistance data, without classic tomographic inversions. PFA is based on a set of prior subsurface models coherent with our prior knowledge of the site. From this set of models, we generate a prior set of temperature distribution and resistance data mimicking the field experiment. Then, we use dimension-reduction techniques to derive a direct relationship between the data and the desired prediction. The use of canonical correlation analysis linearize the relationship and allows using Gaussian regression to sample the posterior. In this paper, we demonstrate the ability of PFA to process time-lapse ERT data during a field experiment. We propose an analysis of time-lapse reciprocals to derive an error model and generate the posterior distribution of temperature. We validate the results using direct measurements in the aquifer. This successful application opens new ways to process and integrate geophysical data in hydrogeological model.