Stan E. Dosso

Quantifying uncertainty in geoacoustic inversion. I. A fast Gibbs sampler approach

This paper develops a new approach to estimating seabed geoacoustic properties and their uncertainties based on a Bayesian formulation of matched-field inversion. In Bayesian inversion, the solution is characterized by its posterior probability density (PPD), which combines prior information about the model with information from an observed data set. To interpret the multi-dimensional PPD requires calculation of its moments, such as the mean, covariance, and marginal distributions, which provide parameter estimates and uncertainties. Computation of these moments involves estimating multi-dimensional integrals of the PPD, which is typically carried out using a sampling procedure. Important goals for an effective Bayesian algorithm are to obtain efficient, unbiased sampling of these moments, and to verify convergence of the sample. This is accomplished here using a Gibbs sampler (GS) approach based on the Metropolis algorithm, which also forms the basis for simulated annealing (SA). Although GS can be computationally slow in its basic form, just as modifications to SA have produced much faster optimization algorithms, the GS is modified here to produce an efficient algorithm referred to as the fast Gibbs sampler (FGS). An automated convergence criterion is employed based on monitoring the difference between two independent FGS samples collected in parallel. Comparison of FGS, GS, and Monte Carlo integration for noisy synthetic benchmark test cases indicates that FGS provides rigorous estimates of PPD moments while requiring orders of magnitude less computation time.

An adaptive-hybrid algorithm for geoacoustic inversion

This paper presents an adaptive hybrid algorithm to invert ocean acoustic field measurements for seabed geoacoustic parameters. The inversion combines a global search (simulated annealing) and a local method (downhill simplex), employing an adaptive approach to control the trade off between random variation and gradient-based information in the inversion. The result is an efficient and effective algorithm that successfully navigates challenging parameter spaces including large numbers of local minima, strongly correlated parameters, and a wide range of parameter sensitivities. The algorithm is applied to a set of benchmark test cases, which includes inversion of simulated measurements with and without noise, and cases where the model parameterization is known and where the parameterization most be determined as part of the inversion. For accurate data, the adaptive inversion often produces a model with a Bartlett mismatch lower than the numerical error of the propagation model used to compute the replica fields. For noisy synthetic data, the inversion produces a model with a mismatch that is lower than that for the true parameters. Comparison with previous inversions indicates that the adaptive hybrid method provides the best results to date for the benchmark cases.

Estimation of ocean-bottom properties by matched-field inversion of acoustic field data

A method for estimating properties of the ocean bottom such as bathymetry and geoacoustic parameters such as sound speed, density and attenuation, using matched-field inversion is considered. The inversion can be formulated as an optimization problem by assuming a discrete model of unknown parameters and a bounded search space for each parameter. The optimization then involves finding the set of parameter values which minimizes the mismatch between the measured acoustic field and modeled replica fields. Since the number of possible models can be extremely large, the method of simulated annealing, which provides an efficient optimization that avoids becoming trapped in suboptimal solutions, has been used. The matching fields are computed using a normal mode model. In inversions for range-dependent parameters, the adiabatic approximation is employed. This allows mode values to be precomputed for a grid of parameter values and stored in look-up tables for fast reference, which greatly improves computational efficiency. Synthetic inversion examples are presented for realistic range-independent and range-dependent environments. >

Quantifying uncertainty in geoacoustic inversion. II. Application to broadband, shallow-water data.

This paper applies the new method of fast Gibbs sampling (FGS) to estimate the uncertainties of seabed geoacoustic parameters in a broadband, shallow-water acoustic survey, with the goal of interpreting the survey results and validating the method for experimental data. FGS applies a Bayesian approach to geoacoustic inversion based on sampling the posterior probability density to estimate marginal probability distributions and parameter covariances. This requires knowledge of the statistical distribution of the data errors, including both measurement and theory errors, which is generally not available. Invoking the simplifying assumption of independent, identically distributed Gaussian errors allows a maximum-likelihood estimate of the data variance and leads to a practical inversion algorithm. However, it is necessary to validate these assumptions, i.e., to verify that the parameter uncertainties obtained represent meaningful estimates. To this end, FGS is applied to a geoacoustic experiment carried out at a site off the west coast of Italy where previous acoustic and geophysical studies have been performed. The parameter uncertainties estimated via FGS are validated by comparison with: (i) the variability in the results of inverting multiple independent data sets collected during the experiment; (ii) the results of FGS inversion of synthetic test cases designed to simulate the experiment and data errors; and (iii) the available geophysical ground truth. Comparisons are carried out for a number of different source bandwidths, ranges, and levels of prior information, and indicate that FGS provides reliable and stable uncertainty estimates for the geoacoustic inverse problem.

GEOACOUSTIC INVERSION VIA LOCAL, GLOBAL, AND HYBRID ALGORITHMS

In this paper, local, global, and hybrid inversion algorithms are developed and applied to the problem of determining geoacoustic properties by minimizing the mismatch between measured and modeled acoustic fields. Local inversion methods are sensitive to gradients in the mismatch and move effectively downhill, but generally become trapped in local minima and must be initiated from a large number of starting points. Global inversion methods use a directed random process to search the parameter space for the optimal solution. They include the ability to escape from local minima, but as no gradient information is used, the search can be relatively inefficient. Hybrid inversion methods combine local and global approaches to produce a more efficient and effective algorithm. Here, downhill simplex (local) and simulated annealing (global) methods are developed individually and combined to produce a hybrid simplex simulated annealing algorithm. The hybrid inversion is found to be faster by more than an order of magnitude for a benchmark testcase in which the form of the geoacoustic model is known. The hybrid inversion algorithm is also applied to a testcase consisting of an unknown number of layers representing a general geoacoustic profile. Since the form of the model is not known, an underparameterized approach is employed to determine a minimum-structure solution.

Trans-dimensional geoacoustic inversion

This paper develops a general trans-dimensional Bayesian methodology for geoacoustic inversion. Trans-dimensional inverse problems are a generalization of fixed-dimensional inversion that includes the number and type of model parameters as unknowns in the problem. By extending the inversion state space to multiple subspaces of different dimensions, the posterior probability density quantifies the state of knowledge regarding inversion parameters, including effects due to limited knowledge about appropriate parametrization of the environment and error processes. The inversion is implemented here using a reversible-jump Markov chain Monte Carlo algorithm and the seabed is parametrized using a partition model. Unknown data errors are addressed by including a data-error model. Jumps between dimensions are implemented with a birth-death methodology that allows transitions between dimensions by adding or removing interfaces while maintaining detailed balance in the Markov chain. Trans-dimensional inversion results in an inherently parsimonious solution while partition modeling provides a naturally self-regularizing algorithm based on data information content, not on subjective regularization functions. Together, this results in environmental estimates that quantify appropriate seabed structure as supported by the data, allowing sharp discontinuities while approximating smooth transitions where needed. This approach applies generally to geoacoustic inversion and is illustrated here with seabed reflection-coefficient data.

Data error covariance in matched-field geoacoustic inversion

Many approaches to geoacoustic inversion are based implicitly on the assumptions that data errors are Gaussian-distributed and spatially uncorrelated (i.e., have a diagonal covariance matrix). However, the latter assumption is often not valid due to theory errors, and can lead to reduced accuracy for geoacoustic parameter estimates and underestimation of parameter uncertainties. This paper examines the effects of data error (residual) covariance in matched-field geoacoustic inversion. An inversion approach is developed based on a nonparametric method of estimating the full covariance matrix (including off-diagonal terms) from the data residuals and explicitly including this covariance in the misfit function. Qualitative and quantitative statistical tests for Gaussianity and for correlations in complex residuals are considered to validate the inversion results. The approach is illustrated for Bayesian geoacoustic inversion of broadband, vertical-array acoustic data measured in the Mediterranean Sea.

Array element localization for horizontal arrays via Occam’s inversion

Accurate locations for the individual elements of an acoustic sensor array are required for the application of advanced array processing methods. This paper develops a general method of localizing horizontal line array (HLA) elements which overcomes bandwidth constraints of low-frequency arrays and uncertainty in the experimental configuration. Array elements are localized for two HLA’s associated with the Spinnaker Array, a three-dimensional sensor array located in the high Arctic. Recordings were made of imploding glass light bulbs deployed at a series of locations surrounding the array site. Implosion instants were not measured; hence, the data consist of relative travel times. In addition, the source locations were measured only approximately in the field, and are treated as unknown parameters. The inverse problem of determining hydrophone and source locations is nonunique and ill-conditioned. To determine the most physically meaningful solution, an iterative linearized inversion is developed which ap...

Model selection and Bayesian inference for high-resolution seabed reflection inversion

This paper applies Bayesian inference, including model selection and posterior parameter inference, to inversion of seabed reflection data to resolve sediment structure at a spatial scale below the pulse length of the acoustic source. A practical approach to model selection is used, employing the Bayesian information criterion to decide on the number of sediment layers needed to sufficiently fit the data while satisfying parsimony to avoid overparametrization. Posterior parameter inference is carried out using an efficient Metropolis-Hastings algorithm for high-dimensional models, and results are presented as marginal-probability depth distributions for sound velocity, density, and attenuation. The approach is applied to plane-wave reflection-coefficient inversion of single-bounce data collected on the Malta Plateau, Mediterranean Sea, which indicate complex fine structure close to the water-sediment interface. This fine structure is resolved in the geoacoustic inversion results in terms of four layers within the upper meter of sediments. The inversion results are in good agreement with parameter estimates from a gravity core taken at the experiment site.

Trans-dimensional matched-field geoacoustic inversion with hierarchical error models and interacting Markov chains

This paper develops a trans-dimensional approach to matched-field geoacoustic inversion, including interacting Markov chains to improve efficiency and an autoregressive model to account for correlated errors. The trans-dimensional approach and hierarchical seabed model allows inversion without assuming any particular parametrization by relaxing model specification to a range of plausible seabed models (e.g., in this case, the number of sediment layers is an unknown parameter). Data errors are addressed by sampling statistical error-distribution parameters, including correlated errors (covariance), by applying a hierarchical autoregressive error model. The well-known difficulty of low acceptance rates for trans-dimensional jumps is addressed with interacting Markov chains, resulting in a substantial increase in efficiency. The trans-dimensional seabed model and the hierarchical error model relax the degree of prior assumptions required in the inversion, resulting in substantially improved (more realistic) uncertainty estimates and a more automated algorithm. In particular, the approach gives seabed parameter uncertainty estimates that account for uncertainty due to prior model choice (layering and data error statistics). The approach is applied to data measured on a vertical array in the Mediterranean Sea.