(638c) Geologic CO2 Storage Leakage Detection with Statistical Analysis of Controlled-Release Field Data

For geologic CCUS to be accepted, leak detection capability must be guaranteed. Furthermore, it is ideal for leak detection algorithms be very practical and applicable at any geologic CCUS site. The primary objective of this analysis is to develop and test algorithms for effective detection of leakage from geologic carbon capture and sequestration (CCUS) sites. Other leak detection techniques such as monitoring eddy flux and atmospheric modeling have proven effective at detecting simulated leaks CCUS sites. These methods require knowledge of turbulent flow patterns and complex boundary layer models. Monitoring eddy flux is difficult due to the significant impacts that variable terrain, buildings, seasonal cycles, diurnal cycles, and biological processes have on the flux measurement. It is evident that we need a more straightforward, probabilistic method to leak detection and identification that could be employed at any monitoring site. The techniques developed in this study have filled that need.

As part of the monitoring, verification, and accounting (MVA) efforts of the Southwest Regional Partnership, this study developed and tested a leak detection methodology using single and multiple greenhouse gas detection towers on the University of Utah campus. Controlled releases of CO2 and CH4 were used to simulate various point sources by changing duration, location, and flow rate of the point source emissions. High frequency background data (on the order of 10 Hz) and the controlled release experiment data were collected. Point source emission concentration thresholds for both CO2 and CH4 were determined by probabilistic analysis of the data. Data was also used to construct a visual â??probability density roseâ? (PDR) diagram. A PDR is a polar diagram that represents the probability of a leak coming from a certain direction with respect to the detection tower. PDRâ??s were constructed using CO2 and CH4 concentration data and meteorological data collected at the detection tower over a specified time span, which was typically at least 24 hours to account for varying wind directions throughout the day. The wind direction was split into a designated bin size, which for this study was 5°. The number of measurements above the concentration threshold of a leak for each bin was divided by the total number of measurements in each bin to give a â??probabilityâ? of the concentration exceeding the leakage concentration threshold for each bin.

The resulting statistical analysis (both descriptive and circular statistical analyses) on the data collected indicated that there were significant differences in wind direction, wind speed and background concentration data throughout the day as well as from day to day. This confirmed the importance of collecting long periods of background data for successful leak detection methodology development. Furthermore, in most instances the direction from which readings representative of a point source emission were significantly different than the dominant wind direction. When compared to traditional wind rose diagrams and pollutant concentration rose diagrams, PDR diagrams were shown to be far more useful in determining the direction from which leakage signals were originating from. This was attributed to the probabilistic nature of the diagram and that by dividing the number of measurements above the concentration threshold by the total number of measurements in each bin, the diagram has accounted for the likelihood of readings coming from that direction. It was concluded that PDR diagrams were not only effective visual tools, but also effective at determining the leakage source location with respect to the detection tower.

Ongoing and planned future work includes conducting an uncertainty analysis on the leak detection methodology developed on the University of Utah campus. This analysis will focus on quantifying uncertainty associated with variable monitoring scenarios such as the number of detection towers, location of detection towers, and various site characteristics associated with the monitoring effort. Furthermore, the data collected during the controlled release experiments described in the above paragraphs will be used to determine whether or not an inverse model can be built and used to effectively locate and quantify leaks based on data collected at the detection tower and machine-learning algorithms.