A quality control framework for marine sensing using statistical, causal inference

The manual Quality Control (QC) process under­ taken by experts in marine sensor deployments is becoming increasingly impractical as the size of these deployments increase and streaming applications become the norm. Automated quality control procedures have been developed for real-time sensor deployments proposing a range of tests and checks. Although these tests are often easy to implement, they are often tied to the phenomena being measured in the particular application. In this paper, we develop a new QC approach which is more flexible than previous work by proposing a general QC framework based upon Bayesian Networks (BN). The framework models the causal structure of the quality control process and the relationships between any set of tests, checks and expert knowledge associated with a particular deployment. Furthermore, unlike the hard QC assessments of previous work, we provide probabilistic assessments of quality that enable measurement uncertainty to be estimated. We implement this BN approach upon a deployment of near real-time conductivity and temperature sensor nodes located in Sullivan's Cove, Hobart, Australia. Test results show that our Bayesian Network produce relevant assessments that are similar to those generated by domain experts.