Modeling cumulative sound exposure around marine seismic surveys.

This article presents a method for reducing the computation time required for estimating cumulative sound exposure levels. Sound propagation has to be computed from every source position to every desired receiver location; so if there are many source positions, then the problem can quickly become computationally expensive. The authors' solution to this problem is to extract all possible source-receiver pathways and to cluster these with a self-organizing neural net. Sound propagation is modeled only for the cluster centroids and extrapolated for the entire geographic region. The tool is illustrated for the example of a marine seismic survey over a tropical coral reef. Resident fish species were expected not to flee the reef, but to stay among the corals for the entire duration of the survey. In such cases, the modeling of cumulative sound exposure levels is sometimes requested as part of environmental impact assessments. The tool developed combines a seismic source model, a near-field sound propagation model, and a far-field sound propagation model. The neural network reduces the computation time by a factor of 55. The cost is an error in modeled received levels of less than -1+/-3 dB re 1 microPa(2) s.