Statistical Methods for Integrating Multiple CO<sub>2</sub> Leak Detection Techniques at Geologic Sequestration Sites

2018-07-01T05:10:14Z (GMT) by Ya-Mei Yang
<p>Near-surface monitoring is an essential component of leak detection at geologic CO2 sequestration sites. With different strengths and weaknesses for every monitoring technique, an integrated system of leak detection monitoring methods is needed to combine the information provided by different techniques deployed at a site, and no current methodology exists that allows one to quantitatively combine the results from different monitoring technologies and optimize their design. More importantly, an evaluation that is able to provide the assessment of possible size of a leak based on the multiple monitoring results further helps the managers and decision makers to know whether the unexpected leakage event is smaller than the required annual seepage rate for effective long-term storage. The proposed methodology for this application is the development and use of a Bayesian belief network (BBN) for combining measurements from multiple leak detection technologies at a site.</p> <p>The Bayesian Belief Network for CO<sub>2</sub> leak detection is built through an integrated application of a subsurface model for CO<sub>2</sub> migration under different site conditions; field-generated background information on several monitoring techniques; and statistical methods for processing the field background data to infer the leak detection threshold for each monitoring technique and the conditional probability values used in the BBN. Several statistical methods are applied to estimate the detection thresholds and the conditional probabilities, including (1) Bayesian methods for characterizing the natural background (pre-injection) conditions of the techniques for leak detection, (2) the combination of the characterization of the background monitoring results and the simulated CO<sub>2</sub> migration for estimating the probability of leak detection for each monitoring technique given the size of leak, (3) a probabilistic design of CO<sub>2</sub> leak detection for estimating the detection probability of a monitoring technique under different site conditions and monitoring densities, (4) a Bayesian belief network for combining measurements from multiple leak detection technologies at an actual test site, with the site conditions and the probability distributions of leak detection and leakage rate estimated for the site.</p> <p><br>The BBN model is built for the Zero Emissions Research and Technology (ZERT) test site in Montana. The monitoring techniques considered in this dissertation include soil CO<sub>2</sub> flux measurement and PFC tracer monitoring. The possible near-surface CO<sub>2</sub> and PFC tracer flux rates as a function of distance from a leakage point are simulated by TOUGH2, given different leakage rates and permeabilities. The natural near-surface CO<sub>2</sub> flux and background PFC tracer concentration measured at the ZERT site are used to determine critical values for leak inference and to calculate the probabilities of leak detection given a monitoring network. A BBN of leak detection is established by combing the TOUGH2 simulations and the background characterization of near-surface CO<sub>2</sub> flux and PFC tracer at the sequestration site. The BBN model can be used as an integrated leak detection tool at a geologic sequestration site, increasing the predicted precision and inferring the possible leak distribution by combining the information from multiple leak detection techniques. Moreover, the BBN model can also be used for evaluating each monitoring technique deployed at a site and for determining the performance of a proposed monitoring network design by a single or multiple techniques.</p>