Application of maximum entropy to statistical inference for inversion of data from a single track segment

Abstract

An approach is presented for statistical inference, based on maximum entropy (ME) with inversion data from a single source track segment, to account for model mismatch. A previous approach requires data from multiple track segments to set the, otherwise undetermined, ME constraint value specifying the posterior probability density (PPD). One effect of mismatch is that the lowest cost inversion solutions for some parameter values, e.g., source track parameter values obtained from GPS measurements or source levels from towed sources, may be well outside a relatively well known, narrow, uncertainty interval. The basis for the new approach is that the ME constraint value is determined by requiring for such a parameter value that the inferred uncertainty interval encompass the entire uncertainty interval comprising its prior information. Motivating this approach is the hypothesis that the proposed constraint determination, applied to the PPD for a model space broader than that parameter value’s prior, could account for the effect of mismatch on the inferred, but a priori less well defined, values for other parameters.