the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 03 Sep 2026
Measurement report: Multi-platform greenhouse gas monitoring of a major petrochemical industrial complex in South Korea
Abstract. Quantifying greenhouse gas (GHG) emissions from large industrial complexes remains challenging due to the coexistence of diverse source types, highly variable operational conditions, fugitive releases, and limited representativeness of bottom-up inventories. These complexities require observational approaches resolving emissions across multiple spatial and temporal scales. To address this gap, this study conducted a coordinated multi-platform observation campaign from 6–14 January 2024 at the Daesan Petrochemical Industrial Complex (DPIC), one of South Korea’s largest petrochemical hubs, which integrated in situ and ground-based column observations, mobile surveys, aircraft measurements, and satellite data across complementary spatial and temporal scales. Ground-based measurements confirmed sustained CO2 enhancements of 27.72 ppm above background concentrations within the DPIC and observed frequent downwind transport of emissions beyond the complex boundary. Mobile measurements identified combined signatures of combustion-linked emissions and fugitive methane releases at the two largest-emitting facilities, while aircraft emission ratio characteristics corroborated these findings, revealing methane leakage concentrated in the DPIC center alongside increased combustion activity. An annual GHG emission rate of 2,595.45 ± 333.09 GgCO2eqyr–1 derived from the two largest-emitting facility hotspots alone accounted for approximately 37.1 % of the nationally reported inventory values, suggesting systematic underestimation of bottom-up accounting frameworks, especially of fugitive methane. Satellite observations further identified methane enhancements in access-restricted areas consistent with ground-based measurements. This study demonstrates that coordinated multi-platform observation strategies provide critical constraints for atmospheric emission verification and a scalable framework for improving the accuracy of the quantification of greenhouse gas emissions in complex industrial environments.
- Preprint
(21934 KB) - Metadata XML
-
Supplement
(1369 KB) - BibTeX
- EndNote
Status: open (until 17 Oct 2026)
- RC1: 'Comment on egusphere-2026-3755', Anonymous Referee #1, 10 Oct 2026 reply
-
RC2: 'Comment on egusphere-2026-3755', Anonymous Referee #2, 12 Oct 2026
reply
This manuscript presents results from a short-term campaign designed to characterize CO2 and CH4 emissions from a major petrochemical industrial complex in South Korea using coordinated ground-based, mobile, aircraft, and satellite observations. Although the campaign covered only a relatively short period, the measurements were comprehensive, and the different observational platforms consistently captured clear emission signals from the industrial complex. I consider this a useful measurement study and potentially suitable for publication in ACP as a Measurement Report, provided that the issues below are adequately addressed.
Before publication, however, several aspects of the observational data and their interpretation need to be clarified. In particular, the manuscript should more clearly describe what types of emission sources are represented by each measurement and quantification approach. In addition, the interpretation needs to be more cautious given the limited duration of the campaign. My major and specific comments are given below.
### Major commentMy main concern is the conversion of emission estimates obtained during a campaign lasting only several days into annual emission rates. Such an extrapolation implicitly assumes that the emission rates and operating conditions observed during the campaign are representative of those throughout the year. This assumption requires much stronger justification. In particular, extrapolating a single or a limited number of 30-min emission estimates to an annual emission rate is not justified without evidence that these measurements are representative of typical operating conditions. If the authors wish to retain annualized emission estimates, they should provide independent evidence, supported by appropriate references or operational information, showing how representative the observed industrial activities are of annual conditions. The potential temporal variability in facility operation and emissions should be discussed and, as far as possible, quantitatively incorporated into the uncertainty of the annual estimate. The representativeness of the meteorological conditions during the campaign and their implications for the applied plume/inversion model should also be considered. Importantly, this uncertainty associated with temporal extrapolation should be evaluated separately from uncertainties intrinsic to the emission estimation method, such as the posterior uncertainty derived from MCMC or other components of the inversion framework. Even if such uncertainty is quantified, an annual emission estimate based on only several days of observations will inevitably remain poorly constrained. I therefore recommend presenting the emission rates representative of the campaign period as the primary quantitative result. The annualized emission values, the percentage comparison with the annual inventory, and the associated claim of systematic underestimation should be removed from the Abstract and Conclusions, or at least substantially qualified unless their annual representativeness can be convincingly demonstrated.
### Specific commentsLines 129-140: Please provide a clearer description of the land cover, terrain, and major surface structures within and surrounding the study area. In particular, please clarify whether large buildings, stacks, complex terrain, or other heterogeneous surface features could influence atmospheric transport and introduce uncertainty into the plume model or inversion. This information is important for evaluating the applicability of the modeling assumptions to this industrial environment.
Line 177: Please specify the height at which wind speed was measured. Please also explain whether the wind measurement at this height can be considered representative of the wind field relevant to plume transport and the plume model. If wind speed was adjusted to the effective plume or source height, the procedure should be described.
Line 257: Please see my major comment regarding the annualized emission estimates. In addition, mobile measurements conducted near the surface may not fully capture emissions released from elevated sources such as stacks or flares. Please clearly describe the source/release heights represented in the plume model and clarify whether emissions from elevated point sources are included in the estimated emissions. If some elevated sources may not be adequately captured, the potential resulting bias should be discussed.
Figure 4 caption: The caption refers to "CH4 measurements." If the values plotted in the figure are background-subtracted values, should this instead be "CH4 enhancements"?
Figure 5: Please indicate the locations of A1 and A2 in panels (a) and (b). The caption should also state that these measurements were conducted as 30-min stationary observations and specify the date and time of each observation. In addition, please clarify the temporal resolution represented by each data point in the figure (e.g., whether each point represents a 1-s instantaneous measurement or an averaged value).
Lines 322-325: Please see my major comment regarding the annualized emission estimates. Figure 8 appears to show several types of industrial infrastructure, including stacks. Please clearly specify which types of emission sources are actually represented in the quantitative emission estimates (e.g., near-surface fugitive emissions, stacks, flares, or other sources) and which source types may not be adequately captured by the applied approach.
Line 354: Please identify the location of "the nearest automated weather station" more explicitly, including its location relative to the study area. The captions of Figs. 9c, 10c, and 11c should also clearly state where and at what height the wind direction and wind speed shown in these panels were measured.
Lines 549-551: Although a relationship is apparent, the correlation is relatively weak. This may indicate contributions from multiple emission sources or processes with different emission ratios, although differences in atmospheric transport, mixing, and sampling may also contribute to the scatter. Please discuss these possibilities in greater detail and, where possible, identify plausible source types within the facility.
The manuscript discusses fugitive releases as a possible source of the observed CH4 enhancements. If fugitive natural-gas emissions are specifically invoked, please provide a locally relevant C2H6:CH4 ratio for the natural gas supplied to or handled within this region, with an appropriate reference, and compare it quantitatively with the observed C2H6:CH4 relationship. Such a comparison would provide stronger support for the proposed source attribution.Citation: https://doi.org/10.5194/egusphere-2026-3755-RC2 -
RC3: 'Comment on egusphere-2026-3755', Anonymous Referee #3, 13 Oct 2026
reply
This paper presents results from a 9-day multi-platform observation campaign conducted at the Daesan Petrochemical Industrial Complex (DPIC). The study analyzed data from in-situ, ground-based column, mobile, and aircraft observations, as well as satellite observations from three months after the campaign. By combining five independent platforms, the authors identify emission hotspots and sources in a complex industrial environment. The authors estimate annual emission rates using in-situ and mobile measurements and compare them to the nationally reported inventory. The measurements were comprehensive, encompassing a range of distinct observations, and together they constitute a valuable dataset suitable for publication as a measurement report. However, the presented analysis does not support the authors’ main claim that their study provides a robust constraint on emission magnitudes and indicates that the reported inventory values substantially underestimate the annual emission rate of the DPIC. The authors should substantially strengthen their quantitative analysis before the manuscript can be considered for publication in ACP. I will point them out in major comments.
Major Comments:
1.This study claims that the national inventory systematically underestimates emissions from the DPIC. However, this conclusion rests on emission estimates from only two hotspots, each quantified from roughly 30 minutes of mobile measurements. Given this limited spatial and temporal coverage, the data cannot support a claim of systematic underestimation, either for the complex as a whole or on an annual basis.
Furthermore, although the authors emphasize their multi-platform, cross-validating framework (for example, line 572 and 580), the manuscript lacks quantitative analysis across all platforms and does not demonstrate any actual cross-validation between them.
I suggest providing emission estimates using aircraft and ground-based column obversions as well. For example, emission rates could be quantified using measurements from the circular flight loop (Conley et al, 2017) and EM27/SUN (Makarova et al. 2021) using mass balance approach. Crosswind transects from the airborne survey could also be used for quantification (Conley et al, 2016) or combine EM27/SUN with X-STIMT for quantification. Linking and integrating emission estimates from individual platforms would strongly demonstrate the added value of the multi-platform approach, increasing confidence in both the derived emission rates and the attribution of emissions to specific sources.
2. The authors should clarify how the OTM-33A method was applied. This method assumes a Gaussian plume and relies on concentrations measured near the plume centerline. However, the manuscript does not describe how the data were processed and filtered to meet these requirements. In addition, the determination of background concentrations requires further clarification. For example, the authors state that background concentrations were determined as the 5th percentile within a 5-minute moving window, but it is not specified whether this was applied to the entire mobile survey dataset for the day or only to the 30-minute fixed-point observations.
3. The manuscript does not state whether a side-by-side intercomparison was performed for the three EM27/SUN instruments. Please clarify this point. If no side-by-side calibration was conducted, this would be a critical issue, as instrument-specific biases could not be ruled out and could directly affect the column differences used in the analysis.
For example, I don’t see a reason why CO2 and CH4 had different background sit for January 8. If no side-by-side calibration was conducted, it is possible that this indicates the instrument-specific biases. Different pattern between CO2 and CH4 on Figure 10 and 11 is also suspicious.
Minor Comments:
- Line 29-32: Please clarify that the estimated emission rate accounts for 37.1% of the emission rate reported for DPIC.
- Line 173-175: Please provide a more detailed description of the OTM-33A method, including the requirements that must be satisfied given the assumptions underlying the derivation of Eq. (1).
- Line 174-175: Correct σx to σz.
- Line 177: Please specify the empirical model that contains the I, J, and K coefficients, and provide the corresponding reference.
- Section 3.1: Please provide the exact location where the 30-minute fixed-point measurements was carried.
- Figure 5: Please provide wind rose plot that would more clearly show the prevailing wind condition.
- Line 266: Do you mean 30-minute stationary observations?
- Line 265-290: Please provide a more detailed explanation of how the processes associated with each operation could lead to stronger or weaker correlations and to larger or smaller emission ratios.
- Figure 7: Please provide zoomed map around DPIC to better show where SNUCO2M was location. It would be also good to compare the location with 30-minute fixed-point measurements by the mobile vehicle.
- Figure 15: It is difficult to identify where concentrations are high or low relative to the DPIC layout. Please revise this figure to improve readability, for example by adding facility boundaries or key source locations, or by adjusting the color scale or transparency.
- Line 587-597: Please use unified units discussing estimated emission rates.
Citation: https://doi.org/10.5194/egusphere-2026-3755-RC3
Data sets
Dataset for multi-platform greenhouse gas monitoring of a major petrochemical industrial complex in South Korea H. Park et al. https://doi.org/10.5281/zenodo.21201694
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 362 | 157 | 58 | 577 | 70 | 57 | 59 |
- HTML: 362
- PDF: 157
- XML: 58
- Total: 577
- Supplement: 70
- BibTeX: 57
- EndNote: 59
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Park et al. report a nine-day winter campaign (6–14 January 2024) at the Daesan Petrochemical Industrial Complex (DPIC). The campaign combined a low-cost in situ CO2 sensor (SNUCO2M), three EM27/SUN spectrometers, a vehicle-based mobile platform, a KMA research aircraft, and one EMIT scene acquired three months after the campaign. The mobile platform identifies two hotspots (A1 and A2). Applying OTM-33A to 30-minute stationary measurements at these sites gives an annualized emission of about 2.5 Tg CO2eq yr–1. The authors compare this with the 2022 national inventory value for the DPIC (37.1 %) and interpret it as evidence that bottom-up inventories systematically underestimate emissions, especially fugitive methane. Coordinated greenhouse gas observations at a large Asian petrochemical complex are rare, and a well-documented dataset of this kind would be a useful measurement report for ACP. In its current form, however, the analysis does not support the manuscript’s central quantitative claim. The headline number is the sum of the single highest 30-minute OTM-33A estimates for each site and gas, taken from different days. It is then annualized and compared with a complex-wide inventory total whose scope is never defined. The OTM-33A implementation departs from the method in ways that can change the result several-fold. The uncertainty budget is a fixed ~13 % that does not respond to sampling conditions. None of the other four platforms is quantitatively linked to the mobile estimates, so the repeated claims of “five independent” and “recursive, cross-validating” confirmation are asserted rather than demonstrated. There are also many internal inconsistencies (headline totals, flight altitudes and loop counts, source directions, units) that need to be resolved. I recommend major revision. The authors should either substantially strengthen the quantification (concerns below) or reframe the paper as a descriptive measurement report and remove the conclusions about inventory underestimation.
Main concerns
The comparison with the national inventory does not support the conclusion of systematic underestimation. The abstract and conclusions rest on the statement that two hotspots account for 37.1 % of the DPIC inventory. Several problems compound here. The annual total is built from maxima. The combined CO2 (2343.72 GgCO2 yr–1) is exactly the highest A1 rate (10 January) plus the highest A2 rate (6 January). The combined CH4 (179.90 GgCO2eq yr–1) is likewise the A1 maximum (6 January) plus the A2 maximum (7 January) (L252–257). Annualizing the largest single 30-minute snapshot at each site, from different days, gives a near-worst-case number, not an annual mean. The authors should report day-by-day estimates for every fixed-point deployment (Table S1 already lists the schedule). The annual value should be based on the mean or median and its day-to-day spread, together with a discussion of whether daytime January operations represent the annual mean.
The OTM-33A implementation and its uncertainty budget issues. OTM-33A was designed for near-field quantification (typically ~20–200 m) of ground-level point sources from ~20-minute stationary deployments with a 3-D sonic anemometer. The present implementation in the paper departs from default use in several ways that matter. C is set to the “maximum observed enhancement” (L180). The instantaneous maximum within 30 minutes is systematically larger than the fitted Gaussian peak of the wind-direction-binned means. It will bias Q high, possibly by a large factor, and makes Q depend on sampling duration and analyzer response. The stability class is also assumed to be A or B because “daytime conditions in South Korea generally correspond to” these classes (L176–177). The stability class should be derived from the on-site wind measurements for each deployment. The source geometry is not constrained. The authors acknowledge “ambiguity in locating emission sources within the DPIC” (L184), yet the method effectively assumes a single ground-level point source at a known distance. If the CO2 at A1/A2 comes from combustion stacks, the plume is elevated and buoyant. A vehicle at ground level a short distance away will at best sample a fragment of the plume (or building downwash), while most of the exhaust is carried away aloft, so the method may just as well underestimate.
The application of OTM-33A to the SNUCO2M record (L321–323) is not justified. SNUCO2M is a fixed sensor in the middle of the complex, not a short deployment placed deliberately downwind of a single known source. The “sources” are broad facility areas (red ellipses in Fig. 8a) a few hundred metres across and a few hundred metres away, in a dense industrial area where many sources share each wind sector. A pollution rose can identify a sector, not a source, and attributing the signal to one facility and computing its emission rate is not supported. Moreover, the estimate for the northern/northwestern source uses 00:00–05:59 LT on 12 January (Fig. 8b). That is nighttime, when the stable boundary layer is inconsistent with the A/B daytime stability classes assumed in Sect. 2.3. The background used for the SNUCO2M enhancements is not stated.
The WRF configuration is not described at all. Please give the domains, horizontal resolution, vertical levels, PBL scheme and any nudging, and evaluate the simulated winds against the wind measurements made at the three sites. The 1 km footprint resolution (L147–148) is coarse relative to the distances between hotspots and receptors.
For the aircraft enhancement-ratio analysis, the statistics also need revision. The samples within a bin are strongly autocorrelated and come from only a few plume crossings. Resampling individual points 500 times treats them as independent, so the bootstrap intervals (e.g. 12.16–15.67 ppb ppm–1 with R² = 0.40) reflect the number of data points rather than the number of independent plume samples. With so few laps, a point-level bootstrap is not meaningful. The authors should resample whole plume crossings (block bootstrap) or report ratios per crossing. Ordinary least squares with errors in both variables biases slopes low, so York or orthogonal regression should be used. Many bins in Fig. 17 have R² of 0.01–0.2, where the slope is not meaningful as an enhancement ratio and should not be interpreted. The very wide boxes in some bins (e.g. ΔCH4/ΔCO near 36.96–36.97° N on 6 January, with R² = 0.04 and whiskers from about –0.3 to 1.2) more likely reflect where each pass sampled the plume (core, edge, or outside it) than real differences in source composition. Finally, taking the background as the 10th percentile of the whole 500–900 m subset includes plume-influenced data on days with a near-surface plume.
Several interpretations are not physically sound. L484–486 attributes a low ratio to “the boundary of the flare plume where CH4 has been oxidized”. Methane is not oxidized in a plume on these timescales; destruction happens in the flame and is reflected in the flare’s combustion efficiency. A low ΔCH4/ΔCO2 therefore indicates efficient combustion at the source, not plume-edge chemistry. The aircraft data also leave a clear opportunity unused. The loops flown around the complex at several altitudes are the classic design for a closed-path mass-balance (Gauss divergence) flux (e.g. Conley et al., 2017; Fiehn et al., 2020). Such a flux would give a complex-wide CO2 and CH4 emission rate, which is the most direct test of the inventory claim. The authors should either compute it, explaining how the unsampled layer below ~600 m is handled, or explain why it is not possible.
The multi-platform integration is asserted rather than demonstrated. The paper’s central framing is that the five platforms yield “mutually consistent signals converging across five independent observational approaches” (L579–581) within a “recursive, cross-validating architecture” (L572–576, L631–632). In practice, only the mobile platform produces an emission estimate for the hotspots. SNUCO2M and the EM27/SUNs show that enhancements exist, the aircraft shows latitude-dependent ratios, and EMIT observed in a different season. No quantitative cross-check between platforms is performed. For example, the paper never tests whether the A1/A2 emission rates, transported with WRF/X-STILT, reproduce the EM27/SUN or aircraft enhancements. The Conclusions themselves acknowledge that the agreement “may reflect shared sensitivity to the same source regions rather than fully independent confirmation” (L628–629), which contradicts the earlier claims.
Other comments
The platforms carried tracers that are not used. The mobile platform had a NOx analyzer (Table 1), and ΔNOx/ΔCO2 would help separate combustion CO2 from vented process CO2. This matters most at A1, which is next to a CO2 recovery and dry-ice plant fed by refinery and ethanol off-gas (L285–287). The CO2 there could plausibly be vented process CO2 rather than combustion CO2, which would also affect how it is treated in the inventory. Likewise, the CH4 from the excluded pipeline construction site (L230–236) is natural gas, so its C2H6/CH4 ratio would provide a useful reference signature for interpreting A1 and A2.
Units and GWP need care throughout. CO2-only emissions are reported in GgCO2eq yr–1 (e.g. L253, L322). The GWP used to convert CH4 should be stated and should match the one used in the inventory being compared.
Many quantities are reported with far more precision than the measurements support (e.g. 29,833.35 ppb, 1053.37 ± 134.30, 67.873 ± 8.657, 16,800.81). Please round to a precision consistent with the uncertainties.
Several figures are difficult to read. Legends, scale bars and colour bars in Figs. 4 and 5 are illegible at print size, and the fonts in Fig. 16 are too small.
Regarding data and code availability, please confirm that the Zenodo repository is public and contains the data at native resolution. It should also include the analysis code (background definitions, OTM-33A implementation, regression/bootstrap, matched-filter configuration), the EMIT granule ID, and the WRF/X-STILT configuration, and the paper should give the sources of the AMY data and of the DPIC inventory value.
L30–32: The value of 2,595.45 ± 333.09 is inconsistent with L257 (2523.62 ± 251.23) and L557 (2,595.49 ± 333.09). “Two largest-emitting facility hotspots” is not demonstrated: A1 and A2 are the largest enhancements observed along the accessible route, which depend on distance from the road and on wind, not necessarily the largest emitting facilities. “Systematic underestimation … especially of fugitive methane” is not supported.
L40–41: Friedlingstein et al. (2025) is cited, but the reference list gives ESSD vol. 18 with a 2026 DOI. Please make these consistent.
L42: “approximately 86 times more potent than CO2 on decadal timescales” is the AR5 GWP20 value. Since IPCC (2023) is cited, please use the AR6 values (GWP20 ≈ 81 for CH4, ≈ 83 for fossil CH4) and state the time horizon explicitly. “Global warming potential … more potent” is also redundant.
L50–52: The motivation is generic. It would be stronger if it pointed to the specific gap, namely facility-level verification of reported emissions in petrochemical complexes, where fugitive and vented sources rely on emission factors.
L60–62: The long lifetime of CO2 is not what limits local attribution. The limitation is the small local enhancement relative to a large and variable background. The citation of Åström et al. (2025) seems tangential.
L80–82: Please name the major facilities and operators in the DPIC and give their approximate reported emissions. This context is necessary to evaluate A1/A2 and the inventory comparison.
L83: The text cites “Choi et al., 2007”, but the reference list gives a single author (Choi, 2007).
L84 and L86: “Ethylene” and “ethene” are used interchangeably; please choose one. The 2 % share of global ethene production is cited to an atmospheric chemistry paper; an industry source would be more appropriate.
L86–89: Please clarify whether the 11,135.98 GgCO2eq refers to process (IPPU) emissions only or to all emissions from the chemical industry including fuel combustion, and explain how it relates to the 6796.59 GgCO2eq used for the DPIC in L258. The precision of these values is excessive.
L129–131: The institute’s name is “National Institute for Environmental Studies”, here and in affiliation 6. “Were calibrated against the KIT standard and demonstrated consistent reliable measurement performance” should be supported with side-by-side results and correction factors from this campaign.
L141–148: Please give the WRF configuration (domains, horizontal resolution, vertical levels, PBL scheme, nudging) and an evaluation against the measured winds. Please also state the number of X-STILT particles and how the slant path and averaging kernels are treated. Is the 1-hour back-trajectory length long enough to capture influence from the whole complex on the days with weaker wind?
L172: Please cite OTM-33A properly (US EPA; Thoma and Squier, 2014) together with Brantley et al. (2014). The reference-list entry for Thoma and Squier (2014) is garbled and contains unrelated text about permits.
L228–230: Please define “hotspot” (threshold, and whether it recurred across days).
L276–278: This attribution to non-combustion sources rests only on facility type. The weak CO2–CH4 relationship at A2 is consistent with CH4 not being co-emitted with CO2, but it does not identify the process. Please use C2H6/CH4 signatures and, if available, NOx or CO.
L332: Five days of EM27/SUN measurements are mentioned, but only three are shown.
L477–478: Elevated ΔCO/ΔCO2 has other possible sources (FCC regenerators, process heaters, ships), and ratios of up to ~31 ppb ppm–1 are high for well-controlled combustion. Please discuss.
L501–502: These are emission or enhancement ratios, not “plume chemistry”. Two flights cannot establish variation “with facility operation and meteorological conditions”.