Project Canary has recently improved our quantification algorithm to predict total site emissions even more accurately. Our high-fidelity sensors look at < 1 ppm level of emissions and can now accurately detect on-pad, off-pad, and intermittent “minor” emissions. Utilizing a new machine learning model, we look at large amounts of data trends to detect and quantify all operational and fugitive emissions, no matter the size. We also have enhanced the emission detection algorithm to confidently exclude offsite emissions sources in most cases, which ensures emission totals are calculated only for that pad.
The second iteration of quantification (Quantification_02) was predicated by our initial version (Quantification_01). After reviewing all emissions from all operators in all basins, well by well, our Technology and Fields Teams noticed that the initial version of quantification was failing to detect small intermittent emissions. The emissions totals were less than what was expected and this is what caused us to shift our focus towards quantifying total site emissions over fugitive emissions.
An estimated 70-90% of all emissions on a well pad originate from pneumatic devices. These pneumatic devices emit intermittently and are often minor emissions that are notoriously hard to quantify. Project Canary Quantification_02 blends trend data and historical data with real-time monitoring so we can detect, localize, and quantify these smaller emissions adding to the functionality as the initial version of quantification. This means that Project Canary can detect and quantify both intermittent and persistent emissions accurately and reliably.
Another improvement area is the localization algorithm. Previously, we triangulated a source from within an area marked as a “potential leak polygon.” Now, we’ve identified individual leak sources utilizing high-resolution drone imagery. This allows us to be more accurate in determining where leaks appear to be occurring. Most importantly, offsite emissions will now be excluded from well pad emission totals.
While Quantification_02 is an improvement over the first iteration, it is not without some limitations. The underlying physical model is still using a Gaussian plume dependent upon air and wind being able to move freely between the source and the pad. Any physical obstruction will cause aberrations in emission totals. It is also designed to quantify the total site emissions by looking at a much larger set of data. What this means is that very short large emissions (<1 hour) may not cause a spike in the emissions but instead the spike will be averaged into the data set. Longer lasting fugitive emissions will cause a spike in the emissions.Quantification_01 had a limiting factor – only detecting and quantifying a single leak source at a time. This assumption is not the reality on an actual well pad, leading to possibly misleading emission totals. These minor errors in the emission totals were due to the algorithm bucketing all emissions for a specific period into the most likely leak location. Now, we can quantify emissions from multiple leak sources simultaneously. This is done by analyzing more data over an extended period and, more specifically, localizing emission sources from multiple wind directions within the larger dataset. Quantification_02 can quantify emissions coming from an individual pad while excluding any background detection which may be detected from nearby pads or midstream equipment.