A geostatistical and probabilistic spectral image processing methodology for monitoring potential CO2 leakages on the surface

Remote sensing has demonstrated success in various environmental applications over the past three decades. This is largely attributed to its ability for good areal coverage and continued development in sensor technologies. Carbon dioxide Capture and Storage (CCS) is an emerging climate change mitigation technology where monitoring is vital for its sustainability. This research investigates the use of spectral remote sensing imagery in detecting potential CO₂ occurrences at the surface, should a leakage occur from subsurface reservoirs where CO₂ is stored. Currently, there are no known leakages of CO₂ at industrial storage sites, therefore, this research was carried out at the Latera natural analogue site in Italy, in order to develop the methodology described. This paper describes the use of a popular probabilistic information fusion theory, referred to as the Dempster-Shafer theory of evidence, to analyse outlier pixels (anomalies). Outlier pixels are first determined using a new geostatistical image filtering methodology based on Intrinsic Random Function (IRF), Independent Component Analysis (ICA), and the industry standard parametric Reed-Xiaoli (RX) anomaly detection. Information fusion of detected outlier pixels and indirect surface effects of CO₂ leakage over time, such as stressed vegetation or mineral alterations, assigns a confidence measure per outlier pixel in order to identify potential leakage points. After visual validation using direct field measurements, it was demonstrated that the proposed methodology is able to detect majority of the seepage points at Latera, and holds promise as a new unsupervised CO₂ monitoring methodology.