2-DIMENSIONAL NONLINEAR MAGNETOTELLURIC INVERSION USING A GENETIC ALGORITHM

We have developed a nonlinear magnetotelluric inversion based on a standard finite difference TE/TM mode forward solution, including static distortion effects, and a new genetic algorithm for general functional optimisation and hypothesis testing. We have used this to invert a subset of the COPROD2 data in terms of best-fitting 2-D electrical conductivity distributions. Our optimal electrical conductivity model, defined by 66 electrical conductivity parameters and 20 static shift coefficients, attains an rms misfit of 1.48, for standard errors in the data of at least 10% in apparent resistivity and 3-degrees in phase. This may represent the minimum level of misfit given this coarse parameterisation of the earth. The optimal model contains certain features, including the North American Central Plains conductivity anomaly and a surface layer of 1000 S conductance, that are consistent with previous electromagnetic inversions and the local geology. The global optimisation took approximately 12 days to compute on a approximately 20-40 Mflop (million floating point operations per second) computer. We have chosen not to seek a smooth model consistent with the data, a task well handled by existing, faster regularized inversions, but instead to use the genetic algorithm for the more demanding task of extracting the global best-fitting conductivity model.