Permeability of intact and fractured core samples collected in EPS1 borehole at the Soultz-sous-Forêts geothermal site (France) is used for neural network permeability prediction beyond the borehole location using electrical conductivities obtained by 2-D inversion of the magnetotelluric (MT) profile data. The estimates of prediction accuracy as a function of depth indicate that the average relative accuracy of permeability prediction from electromagnetic (EM) data, for example, at double borehole depth ranges from 93% to 97% depending on rock fracturing. Permeability predicting by ANN trained on intact permeability values results in its lower-bound assessment (in contrary to upper-bound values, which could be predicted by ANN trained on “equivalent" permeability values determined using 2-D flow model or depth dependence of fracture apertures. It is shown that the EM forecast of cross-well permeability gives an idea of the overall distribution of this parameter in the section while the comparison of this prediction with the temperature model is fairly conclusive for locating the most promising sites and depths for geothermal exploration. On the other hand, the EM-based permeability prediction could reveal practically impermeable domains in the sedimentary cover and granitic basement where drilling of exploration boreholes is inadvisable. Based on the 2-D MT sounding data, permeability model of the Soultz-sous-Forêts geothermal site is built. The model, in particular, revealed a 500 m thick poorly permeable belt in the sedimentary cover, presumably formed by the precipitation of secondary hydrothermal minerals during cooling. The belt can be considered as a transition zone between conductive (above) and convective (below) heat transfer mechanisms hypothesized earlier based on the analysis of geotherms from different boreholes.
Read full abstract