Sonar uncertainty and sensitivity analysis techniques

Current sonar performance prediction models provide outputs such as probability of detection. It is desirable to give measures of uncertainty in the output due to uncertainty in all the inputs. This process is uncertainty analysis (UA). It is also desirable to indicate the sensitivity of the output to uncertainty in each of the inputs. This process is sensitivity analysis (SA). In this paper, a set of techniques has been incorporated into a sonar uncertainty and prediction tool (SUST) which carries out the main steps of UA/SA: (1) sampling of inputs, (2) propagation of inputs through a model to yield outputs, (3) UA from outputs and (4) SA from outputs. In the following, two enhancements to a previous UA/SA tool are presented: (a) empirical orthogonal functions, a general method of representing uncertainty in the sound speed profile. and (b) Latin hypercube sampling, a more efficient means of sampling the input space. This paper demonstrates the use of the techniques by performing UA/SA for an environment offshore from Sydney, Australia.