Bayesian monitoring design (BayMoDe)
Budsjett
10,07 millionerClimit-finansiering
.5 MNOK from the Research Council. The rest is own financingProsjektnummer
254711
Partnere
University of Bergen, UNI Researc, Heriot-Watt University, Plymouth Marine LaboratoryProsjektperiode
2016 – 2019
The goal of the project is to build a framework based on Bayes theorem that will treat data streams from monitoring time series and quantify suspicion that a leak of geological stored CO2 to marine waters is on going. Quantification of belied further reduces the chance of false alarms that will accelerate the cost significantly. This framework will also be used to optimize monitoring infrastructure design.
This approach will automatically filter out any outliers in a time series; a single leak indication will not automatically sound the alarm but rather increase our awareness by increasing our belief that a leak is on going. Subsequent measurements might reduce or increase our awareness, only when the number of indications reaches a threshold will the extra resources be mobilized.
To test the ability and usefulness of Bayes theorem in the context of environmental monitoring we aim to design a data analysis framework, including monitoring design capabilities, in which the Bayesian approach is the core data treatment. The three main building blocks in the framework
will be a probabilistic map of potential leak locations, environmental baseline statistics, and predictions of leak footprint characteristics. The former two will be part of a site characterization, while the latter will in addition depend on characteristics of seeps.
Even though the focus here is on seafloor monitoring, the approach has the potential to simplify documentation of uncertainty in all monitoring methods. As such the method might accelerate implementation of large-scale storage projects through better procedures for designing and maintaining monitoring programs.
The first nine months of the project period have been used to perform preliminary studies on how to utilize and implement baseline statistics and footprint predictions into the theorem using a limited set of existing model results. Larger datasets from already performed simulations have been identified and are being collected. Additional simulations, to supplement the existing ones, are being planned together with the collaborating HORIZON2020 STEMM-CCS project. In-situ data from Goldeneye will also be made available from STEMM-CCS in the coming years.
Preliminary results have been presented at the joint IEAGHG Monitoring and Modelling combined networks meeting in Edinburgh Scotland in July 2016, and a poster presentation has been accepted at GHGT13 in Lausanne in November.