High-accuracy temperature prediction at large depth based on electromagnetic sounding data and borehole temperature records

We suggest a method for accurately estimating subsurface temperature by using data from electromagnetic sounding and borehole temperature measurements. This is an updated version of our earlier proposed “electromagnetic geothermometer,” which uses an artificial neural network (ANN) to directly predict temperature at the target depth, based on existing temperature data. Within the framework of a new algorithm, the forecast is carried out step-by-step, so that at each subsequent step, the forecast is fulfilled by an ANN trained not only on initial temperature data and resistivity values determined from magnetotelluric (MT) sounding data but also on the outcomes of all previous forecasts at shallower depths. The proposed recursive algorithm is validated comparing the forecasted temperature data with the actual measurements from deep borehole GPK2 located in the Soultz-sous-Forêts (France) geothermal area. Computer modelling results show that, unlike the direct ANN algorithm, the recursive method allows for temperature forecasting with a level of accuracy that is largely unaffected by the depth being targeted. By means of the proposed approach, temperature forecasts are carried out to a depth of 5 km based on the MT sounding data collected in the vicinity of abandoned oil boreholes. A 2-D temperature model of the study area is created using the resistivity model of the study area built earlier. Its analysis revealed a 200 °C isotherm at a depth of approximately 2 km beneath the abandoned oil boreholes, which could justify economically beneficial exploratory drilling for the development of enhanced geothermal resources at relatively shallow depth.