The effects of noise on Occam's inversion of resistivity tomography data

An Occam's inversion algorithm for crosshole resistivity data that uses a finite-element method forward solution is discussed. For the inverse algorithm? the earth is discretized into a series of parameter blocks. each containing one or more elements, The Occam's inversion finds the smoothest 2-D model for which the Chi-squared statistic equals an a priori value. Synthetic model data are used to show the effects of noise and noise estimates on the resulting 2-D resistivity images. Resolution of the images decreases with increasing noise. The reconstructions are underdetermined so that at low noise levels the images converge to an asymptotic image, not the true geoelectrical section, If the estimated standard deviation is too low, the algorithm cannot achieve an adequate data fit, the resulting image becomes rough, and irregular artifacts start to appear. When the estimated standard deviation is larger than the correct value, the resolution decreases substantially (the image is too smooth), The same effects are demonstrated for field data from a site near Livermore, California, However, when the correct noise values are known, the Occam's results are independent of the discretization used. A case history of monitoring at an enhanced oil recovery site is used to illustrate problems in comparing successive images over time from a site where the noise level changes. In this case, changes in image resolution can be misinterpreted as actual geoelectrical changes. One solution to this problem is to perform smoothest, but non-Occam's, inversion on later data sets using parameters found from the background data set.