| dc.description.abstract | We study data from marine magnetotelluric (MT) surveys. In MT surveys the objective is to study the distribution of resistivity in the Earth's subsurface, and the quantity of interest is the impedance Z(ω). The impedance is a multiplicative relationship betweenelectric, e, and magnetic, h, fields in the frequency domain, and hence a convolution inthe time domain, e(n) = z(n) ∗ h(n). MT data is usually analyzed in the frequency domain, but we argue that it is easier to model the observation error process in the time domain. As a first step to explore the assumed convolution model in the time domainwe propose to consider only lags sufficiently far apart so that the convolution model can be approximated by a multiple linear regression model. We explore this strategy forboth simulated and real MT data. For the simulated data we are in most cases able torecover the impedance z(n) used to generate the data, but for the real MT data we are inmost cases obtaining inconsistent estimation results when different sections of the sameobserved e(n) and h(n) series are used to estimate the impedance z(n). Heavy tails inthe observation errors, which is not taken into account in our regression model, may bethe reason for the less promising results for the real data. | nb_NO |
