Uncertainty analysis for tactical applications within the underwater domain
Publish date: 2020-02-07
Report number: FOI-R--4674--SE
Pages: 22
Written in: Swedish
Keywords:
- LHS
- uncertainty analysis
- POD
Abstract
We have studied the possibility of using stratified sampling for uncertainty analysis concerning underwater applications. The method is called Latin Hypercube Sampling (LHS). The idea behind this approach is to sample only a representative part of the probability distribution, so that the number of runs can be reduced significantly. Usually, the number of runs is in the order of 50-100 compared to traditional Monte Carlo sampling, where millions of runs are required. We illustrate the basic principle of LHS using an electromagnetic example. After that, we discuss the difference between uncertainty and sensitivity analysis using an active acoustic scenario. Thereafter, we show how LHS can be used in an area where you have quite large uncertainties in the geo-acoustic properties. We have identified the need of presenting the uncertainties in a simple way without having to analyze a number of runs. A new visualization parameter is defined, denoted total POD (Probability of Detection), to meet this requirement. This parameter represents the total probability of detection from all individual POD-calculations. Finally, we illustrate a case where the uncertainties of the geo-acoustic properties are small, which has bearing towards ranging station applications. Our studies show that LHS is easy to apply and gives useful insight in complex uncertainty dependencies. However, for relevant output, we need to have good knowledge about the individual parameter probability distributions, both regarding absolute values and representative distributions. This remains to be investigated as part of future work.