Methane Emission Aggregation at LNG Facilities
Investigating aerial data pre-analysis schemes and site-level methane emission aggregation methods at LNG facilities
In this work, we are interested in providing a comprehensive analysis of site-level methane emissions at liquefied natural gas (LNG) facilities, which is a current and critical gap in understanding.
First, a bit of context. The Russian invasion of Ukraine has shifted geopolitical dynamics, prompting the European energy market to seek alternative energy sources to Russian pipeline gas while maintaining its strong commitment to reducing greenhouse gas emissions. Concurrently, the United States has substantially increased its LNG exports, becoming the world’s largest exporter in 2023. The liquefaction process involves cooling and condensing natural gas into a liquid form for efficient storage and transportation over long distances. This surge highlights the critical need for accurate methane emissions reporting at LNG facilities, which is essential for effective emission reduction strategies.
So, back to the problem.
Aerial remote sensing technologies capture a momentary snapshot of a specific site’s emissions profile, potentially revealing disparities between instantaneous site emissions and inventory-style emission estimates that are based on temporal averaging. The concept is demonstrated in the following figure.
We evaluate three data pre-analysis schemes, summarized in the following table.
Our solution.
To address this problem, we use two aggregation methods to calculate emissions at the site level using data from the QMRV project collected on two LNG facilities, thereby addressing the problem directly.
Here’s how we compute site emissions:
- Site-level Inventory
- Spatial grouping of measured emissions is conducted at functional element level (e.g., LNG train).
- Bootstrap 10K average emission rates to determine the distribution of each functional element. Assuming Normal distribution for each element, compute mean and standard deviation.
- Generate 10K emission rates for each functional element by sampling data from a Normal distribution using parameters obtained in the previous step.
- Draw and sum individual average emission rates from each functional element to determine the site total rate.
- Instantaneous Emissions
- Randomly draw one emission rate at a time from each point source where measurements were conducted. Sum them to find site total emission rate. Repeat 10K times.
Key Take-Aways
In this work, we’ve learned that data pre-analysis significantly influences the emission estimates. Specifically, adjusting for persistence compared to assuming 100% persistence has a statistically significant impact (p-value < 2e-16) on resulting emission estimates. Excluding detected but not quantified emissions compared to using the 90% POD or campaign averages did not result in a significant difference (p-value = 0.14).
Also, similar LNG facilities operated by the same company can have different emission characteristics. Although the standardized emission rates for Sites A and B were similar in shape and magnitude when calculated using the average-based method, the standardized emissions at Site A have a much wider spread than those at Site B when assessed with the instantaneous method. Furthermore, the flares were the largest contributors to emissions on Site A, while trains were the largest contributors on Site B as demonstrated in the following figure. This figure shows the percent contribution of each functional element’s average emission rate to the scan-level total emissions at each facility prior to scan-level zero emission rates being imputed.
And finally, from our work, we were able to develop a method for estimating the distribution of instantaneous site-level emissions. We find that the distribution of instantaneous emissions is much wider than the distribution of average emissions for both sites. This implies that snapshot measurements, which capture the instantaneous emission rate of an LNG facility, can fall outside of the distribution describing the annual inventory (i.e., the annual mean) and the uncertainty surrounding it. The instantaneous approach should not be used for MII computations, because it describes the variability of the emissions at specific moments in time, rather than providing an accurate representation of the annual inventory. See the following figure, which shows the site-level emissions distributions using both the averaging and instantaneous methods for Sites A and B. The 95% confidence intervals are identified using solid lines for average site-level emissions and dashed lines for instantaneous site-level emissions. Emission rates have been standardized for anonymity.
