Statistical aspects of monitoring and verification

Statistical methods will be important tools in the monitoring of CO2 storage sites, because any signals of leakage are likely to be small compared with measurement and modelling error and natural variability. To conclude that there is no leakage at a storage site necessarily involves consideration of leakage models, as proving “no leakage” as an isolated proposition is impossible. It is important in the circumstances of carbon capture and storage (CCS) to have clear and reproducible methods for testing leakage and no-leakage models against each other in the light of data, and some statistical methods are more useful than others in achieving this. This article reviews three broad approaches to the statistical interpretation of data containing a small leakage signal, with the objective of clarifying the concepts involved for practitioners in CCS who are not statisticians. Bayesian methods are given particular emphasis because they most clearly answer the questions that stakeholders ask, and give a natural framework for dealing with a comprehensive suite of models. The nature and importance of baseline data is discussed in the context of statistical interpretation of monitoring data.

► Proving “no leakage” is key to the success of CCS but is logically impossible.
► No leakage may be more likely than any plausible specific leakage models.
► We must choose between such models statistically, because data are imperfect.
► The statistical methods for making these choices are reviewed.
► The case is made for the well-developed Bayesian methods of model choice.