GNGTS 2017 - 36° Convegno Nazionale

712 GNGTS 2017 S essione 3.3 The MT forward modelling used followed Candansayar (2008) and references therein. It is based on the finite-difference technique, which solves the complex system of magnetotelluric equations for TE and TM polarizations. First, the electric and magnetic fields are derived for each mesh node and finally the apparent resistivity and impedance phase are calculated. The PSO technique implies forward modelling to be run several hundred times, thus making the computation time-consuming, especially for 2D problems that have many unknowns. To manage this heavy computation effort, the parallel computing tool of the PSO algorithm was successfully tested on a computer cluster, with the result of large run-time savings. Simulations ran on the High Performance Computing at Politecnico di Torino (HPC@Polito). When 24 cores were adopted, a computation time reduction of about 80% was observed with respect to the non-parallelized release of the code (one single processor or worker). PSO application to 1D field data. Long period MT data were taken from the COPROD dataset, which is publicly available. The Occam-like 1D optimization (Santilano, 2016) was Fig. 3 - a) The solution of the 2D inversion after 400 iterations with no a priori given to the model. b) Comparison between synthetic data (black dots) and predicted model response (red curves) for apparent resistivity and impedance phase for both TE and TM polarizations.