Seismological geothermometer. Part II: Neural network modeling of the temperature prediction from seismic velocity data in the upper crust

Abstract

The temperature prediction accuracy in the upper crust from the seismic velocity data is estimated by neural network modeling. To this end, we used a 2-D temperature model built earlier along the northern Tien Shan latitudinal profile up to a depth of 27 km. We modeled the temperature prediction at depths beneath boreholes and at large distances from the locations where the initial data were available. The estimation of the temperature prediction from the V_P and V_S at depths below the boreholes indicated that the relative accuracy decreases when the ratio (N) between the target and borehole depths increases. As a whole, at any ratio N, the accuracy of the temperature prediction from V_S is higher than that from V_P. In particular, the average relative errors increase when N growths from 2 to 10 in the ranges 6.8–28.4 % and 5.1–20.1 % when using V_P or V_S, respectively. Assessment of the temperature prediction from V_P and V_S in other locations indicated that at distances up to 16 km, the prediction errors are 7.4 % and 5.7 %, respectively. When this distance is increased 4 times, relative errors are increased 2–3 times. We conclude that the neural network temperature prediction from the seismic velocity data (especially, V_S) could be carried out with sufficient accuracy both beneath boreholes and at large distances and therefore could be used as a seismological geothermometer.