Characteristics of STILT footprints driven by KIM model simulated meteorological fields: implication for developing near real-time footprints

This study presents an analysis of the atmospheric footprint sensitivities and CO 2 enhancements measured at three in situ stations in South Korea (Anmyeondo (AMY), Gosan (JGS), Ulleungdo (ULD)) using the KIM-STILT and WRF-STILT atmospheric transport models. Monthly aggregated footprints for each station were compared between the models for July and December 2020. The footprints revealed major source regions and the sensitivity of atmospheric mole fractions at the receptor to upstream surface fluxes. In July, both models showed similar major source regions for the AMY station, including Korea, the Yellow Sea, and Japan. However, a discrepancy was observed in the Eastern Pacific Ocean, with KIM-STILT showing larger sensitivity compared to WRF-STILT. In December, both models indicated strong sensitivity over Northeast and Eastern China, but KIM-STILT exhibited smaller sensitivities towards North-western China and Mongolia compared to WRF-STILT. At station ULD in July, both models exhibited comparable source regions, but a notable difference was found in Southeast China, where KIM-STILT showed stronger sensitivity. For the JGS station, both models agreed on major sources, but WRF-STILT demonstrated stronger sensitivity over North and Northeastern China. Regarding CO 2 enhancements, both models generally underestimated the amplitude of CO 2 enhancements, especially in July. However, in December, there was better agreement with observed data. The models were able to reproduce the phase of measured ΔCO 2 reasonably well despite the underestimation of CO 2 amplitudes. The contribution of biospheric CO 2 to the observed enhancements, along with fossil-fuel emissions, was highlighted. In specific cases with significant CO 2 enhancements, the models provided varying estimates of CO 2 ff values, particularly in the source regions of Eastern China. The differences in sensitivity estimations emphasize the need for further investigation to understand the underlying factors causing disparities. Overall, this study provides valuable insights into the potential advantages of each model in capturing dispersion patterns in specific regions, highlighting the importance of understanding these differences to improve the accuracy of atmospheric transport models. Further work is necessary to address the observed disparities and enhance our understanding of the transport models in the studied regions.


Background
In the framework of carbon mitigation efforts, understanding the detailed characteristics of the source regions for attributing temporally high-resolution greenhouse gas (GHG) enhancements is essential for improving the accuracy of carbon budget estimates.To delineate the characteristics of the emission source regions, we employed the Stochastic Lagrangian Transport model (STILT), forced by meteorological fields obtained from the Korean Integrated Model (KIM) and Weather Research Forecasting (WRF) models (Nehrkorn et al. 2010).The footprint is calculated from the particle density and residence time in the layer that sees surface emissions, defined as half of the PBLH planetary boundary layer height (PBLH) (Gerbig et al. 2003).However, model simulations are profoundly susceptible to an array of uncertainties, such as winds and vertical mixing (Lin and Gerbig, 2005;Pillai et al. 2012).The accuracy of near-surface meteorological fields, particularly wind predictions, is intricately tied to initial conditions and boundary layer (BL) schemes (Zhang and Pu, 2019;Liu et al. 2020).Notably, the choice of meteorological models and their associated parameterization wields a more substantial influence on simulated footprints than the selection of dispersion models (Angevine et al. 2014).For instance, Hegarty et al. (2013) compared multiple dispersion models (STILT, HYSPLIT, FLEXPART) driven by the same WRF simulated meteorological fields to underscore this phenomenon.
The primary objective of this study was to conduct a comprehensive assessment of two distinct transport models, namely, WRF-STILT and KIM-STILT, in the context of footprint simulations.While the WRF model serves as a widely adopted research tool, offering insights into atmospheric processes, its applicability for realtime or near-real-time meteorological data remains constrained due to computational complexity.In contrast, the KIM model, an operational forecasting tool endorsed by the Korea Meteorological Administration (KMA), excels in delivering continuously updated meteorological data.This quasi-real-time capability renders it particularly valuable for conducting footprint simulations with timely and pertinent meteorological inputs.Such capabilities are particularly indispensable in scenarios necessitating accurate and prompt results, such as pollution tracking.Through the utilization of both the KIM-STILT and WRF-STILT models, our intention was to holistically analyze the merits and limitations of each approach.This comparative analysis facilitates the identification of the advantages inherent to KIM-STILT for near-real-time footprint simulations, thereby significantly enhancing the pertinence and efficacy of such simulations in GHG tracking endeavors.This work emphasizes the importance of employing the KIM-STILT model, particularly in contexts mandating real-time or near-real-time meteorological data.
In our endeavor, we introduced a near-real-time system designed to elucidate the source regions contributing to elevated atmospheric greenhouse gas (GHG) mole fractions at three distinct receptor sites in Korea, namely, AMY, JGS, and ULD.Central to this system were STILT footprints harnessed with KIM meteorological data.With the dual objectives of illustration and assessment of KIM-STILT footprints in delineating source areas, we pursued a twofold approach.Firstly, we initiated a comparison of the simulated footprints generated by leveraging both KIM and WRF meteorological data.This comparative analysis offers insights into the nuanced differences arising from the use of these distinct meteorological datasets, thereby enabling a comprehensive understanding of their respective contributions to the modeled footprints.Secondly, our investigation involved the quantification of time-series enhancements in CO 2 mole fractions for the designated stations.This quantification facilitated the elucidation of these enhancements through the KIM-STILT and WRF-STILT simulated CO 2 ff values.The integration of these simulated CO 2 ff values with the time-series data permitted to the assessment of the models' quality.

KIM-STILT model
The Korean Integrated Model (KIM) (Hong et al. 2018) provided simulated meteorological fields (Table 1) at a horizontal resolution of 0.125˚ by 0.125˚ latitude-longitude with a temporal resolution of 3 h.The input data were converted to STILT compatible data format using KIM2ARL conversion code (Fig. 1).The converted data were then used as input for STILT (version 2, Fasoli et al. 2018) model to generate concentration footprints to identify the source-receptor relationship for near-surface measurements of GHGs such as CO 2 at regional stations across Korea: AMY, ULD, and JGS.Footprints give the sensitivity of observation to upwind source regions expressed in unit of mole fraction per unit flux (Lin et al. 2003).Footprints are proportional to the number of the particles in a surface influenced volume (defined as the lower half of the planetary boundary layer) and the time spent in that volume.In each model run, 500 particles were released from an inlet height of 40 m, 10 m, and 12 m, above the ground at AMY, ULD, and JGS, respectively, and traced backward in time for 5 days to spread the particles over East Asia.The average trajectory of the particle cloud was computed using meteorological wind fields, and random velocities were introduced for each particle using a Markov chain process to approximate turbulent motions.A larger footprint coverage indicates higher influence region from upstream locations, and a higher magnitude reflects more particles in contact with the surface.The influence functions in the given time intervals are highly variable across a month due to changes in wind patterns, turbulence effects, PBL dynamics, and radiation changes.Based on KIM meteorological fields, STILT determines the PBL heights using a modified Richardson number method as stated in Vogelezang and Holtslag (1996).The surface fluxes are instantaneously diluted within an effective mixing depth of h equal to half of the PBL height, for which the vertical mixing timescale is comparable to the advection time steps (Gerbig et al. 2003).The footprints were provided on a horizontal resolution of 0.1° × 0.1° for 5 days back in time with a time resolution of 1 h using the following equation: where m air is the molar mass of air, ρ is the average parti- cle density below h, N total is the total number of released particles, and t p,i,j,z is the amount of time a particle spends within the volume (i, j, k ≤ h) t k at each time step (k is the time-step).The chosen 5 days backward in time is a reasonable time to cover the dispersion of particle over East Asia domain (95°E-150°E, 20°N-55°N) (Table 2).
Footprints have been used to indicate the source areas, as well as estimating CO 2 enhancement through combining with anthropogenic emissions.

WRF-STILT model
We also used WRF model version 4.3.3(Skamarock et al. 2021) to simulate meteorological fields to drive STILT model.Initial and boundary conditions were taken from ERA5 reanalysis data (Hersbach et al. 2020), which has a horizontal resolution of 0.25° × 0.25°, and a temporal resolution of 1 h on 37 model vertical levels.
The WRF run was carried out at horizontal resolution of 0.27° × 0.27°.The simulation domain is centered on East Asia, covering 1000 km by 1000 km at a 27-km resolution.We used Yonsei University (YSU) scheme for PBL (Hong et al. 2006) and the physics schemes have Fig. 1 Schematic flow chart for the footprint calculation.Receptor informations refer to geographic latitude, longitude, and height of the measurement locations; parallel simulation settings indicate number of core and node of the resource; model control includes forward/ backward trajectory time and number of particle release; footprint grid settings define horizontal resolution.For further details, please see Section 2.1 been described in Table 3.The WRF model employs the vertical grid levels of 10 below 1.5 km.The differences in meteorological fields can be attributed to differences in parametrizations within the model configurations, and initial and boundary conditions (Angevine et al. 2014).

Resources and simulation time
We run all our simulations in the super computer at the Korean Meteorological Administration (KMA).Before running STILT, there is an interface which is KIM2ARL code that converts KIM output meteorological variable to the STILT compatible data format.In the KIM2ARL conversion process, we generated data at the global scale.The elapse time of conversion processing is approximately 2.5 h to complete one-month data.In STILT, to run a 5-day backward trajectory in hourly basis with a release of 500 particles at the receptor site, it will take approximately 9 min to complete.It was run parallel simulation settings: number of cores = 4 and number of nodes = 1.

Near-surface situ measurements and site descriptions
We utilized near-surface in situ CO 2 mole fraction measurements measured at three regional stations, Anmyeondo (AMY:36.538°N, 126.33°E), Gosan (JGS:33.29382°N, 126.1628°E), and Ulleungdo (ULD:37.48°N, 130.9° E), Korea, for July and December  Notably, the area contains huge thermal power plants fueled by coal and heavy oil, located within 35 km to the northeast and southeast of the station.JGS is located on the western expanse of Jeju Island, the largest volcanic island in Korea, around 90 km from the mainland.The island's economic drivers are its vibrant tourism and flourishing livestock industries.ULD is positioned to the east of Ulleungdo Island, located in the eastern fringes of Korea, at a distance of approximately 155 km from the mainland.Along the southeastern shoreline of the Korean Peninsula, a range of heavy industries including steel, chemical, and petrochemical sectors are found within a distance of about 200-250 km from the island's location.This coastal region also contains two natural gas power plants.Ulleungdo Island, spanning 72 km 2 , has volcanic origins.Detail activities surrounding the sites were briefly presented in Lee et al. (2019).
Regarding the method for ΔCO 2 extraction, we have calculated the CO 2 mole fraction enhancements (ΔCO 2 ) from in situ measurements for all stations by comparing them to the background data.To derive ΔCO 2 , we employed a curve fitting technique to eliminate the longterm trend and seasonal variations in CO 2 mole fractions, as described in Thoning et al. 1989.The resulting fit-curve (CO 2 .fit)was then subtracted from the original measured values to calculate the ΔCO 2 , which was already applied for those stations (e.g., Lee et al. 2019;Kenea et al. 2023).

Emission inventory data
We utilized anthropogenic CO 2 emission data from ODIAC2022 (https:// db.cger.nies.go.jp/ datas et/ ODIAC/ DL_ odiac 20222.html) (data accessed: July, 2023) (Oda et al. 2018) data to determine the contribution of the source regions.The emission data were derived from fossil fuel combustion, cement product, and gas flaring, which are the major anthropogenic sources of CO 2 (Peters et al. 2012).CO 2 enhancements due to fossil fuel emissions (CO 2 ff ) were estimated for each CO 2 sampling at AMY, JGS, and ULD during July and December 2020, by multiplying footprints with the gridded CO 2 surface fluxes from ODICA2022.It is noted that the horizontal resolution of ODIAC data of 1 km by 1 km has been re-gridded into STILT footprint horizontal resolution of 0.1° × 0.1°.When computing CO 2 ff, we used a 2020 monthly CO 2 emission data from ODIAC2022.It should be noted that the ODIAC data does not account for maritime emissions.As a result, the CO 2 contributions stemming from the maritime area emissions have not been included in our analysis.Furthermore, it is important to note that the emissions data does not encompass hourly variations, which can notably affect the accuracy of CO 2 ff estimates.

Characteristic of footprints at Korea in situ stations
In this work, we conducted a thorough analysis of the source regions for different stations (AMY, JGS, ULD) and performed a comparative assessment of the monthly aggregated footprint sensitivities between two models, namely, KIM-STILT and WRF-STILT.Monthly aggregated footprints for AMY, JGS, and ULD sites were shown for July and December 2020 (Figs. 3, 4 and 5).The color bar indicates the footprint's magnitude, reflecting the extent to which the atmospheric mole fraction at the receptor is influenced by upstream surface fluxes.Owing to differences in station locations, noticeable disparities arise concerning their source coverage.Upon examining the results (Figs. 3, 4 and 5), it becomes apparent that during July, both the AMY and ULD stations showcased a notably heightened sensitivity towards the Northeast in contrast to JGS.However, while December, the principal source regions for all stations converge upon eastern and northeastern China.To a certain extent, the influence of South Korea emerged more prominently for ULD compared to the other two stations.
In July 2020, both models exhibited similar major source regions for the AMY station, encompassing areas such as Korea, the Yellow Sea, the East China Sea, the East Sea, Taiwan, and Southern Japan (Fig. 3a  and b).However, a noticeable discrepancy emerged between the models concerning their representation of the Eastern Pacific Ocean as a source region, with KIM-STILT showing a large footprint magnitude and WRF-STILT depicting not well captured.In December, both models indicated strong sensitivity of footprints covering Northeast and Eastern China, as well as the Yellow Sea (Fig. 3c and d).However, as we moved further towards Northwestern China (particularly, Inner Mongolia), and Mongolia, the magnitude of the footprint sensitivity shown in KIM-STILT was smaller than that in WRF-STILT.
Figure 4 delineates the monthly aggregated footprints of the JGS stations.In July, marked variations in footprint sensitivity were observed across northern Japan, North Korea, and Liaoning province in China (Fig. 4a and b).Within these regions, KIM-STILT exhibited higher sensitivity compared to WRF-STILT, while consistency prevailed in other areas.For December, similar to AMY station, substantial differences in aggregated footprint sensitivity manifested away from the receptor station (Fig. 4c and d).These disparities manifested across South Korea, Mongolia, Inner Mongolia, and Liaoning provinces in China, with WRF-STILT demonstrating notably stronger sensitivity in contrast to KIM-STILT.
At ULD, a closer examination of the monthly aggregated footprints from both models was undertaken (Fig. 5) for July and December.In July, similar source regions were identified, encompassing Korea, the East Sea, the Yellow Sea, East China, Taiwan, Northern and Southern Japan, and Northeast China, with comparable footprint magnitudes.Nevertheless, a distinct distinction arose in Southeast China, particularly within Fujian province, where KIM-STILT simulated a heightened footprint sensitivity compared to WRF-STILT.In December, the disparities in footprint sensitives observed at AMY station were also noted by the ULD station.Additionally, we identified discrepancies in the southern part of Korea during our analysis.The findings highlight the potential strengths of each model in capturing dispersion patterns within specific regions, emphasizing the importance of discerning these differences to enhance the accuracy of atmospheric transport models.Further work is needed to delve deeper into the factors contributing to the observed disparities and to improve our understanding of the transport models in the studied regions.

CO2 enhancement, CO2ff contribution, and source region
In the subsequent sections, we present a comprehensive analysis of the time series of CO 2 mole fraction enhancements measured at three in situ stations in Korea (AMY, JGS, ULD).To explain the observed enhancements, we employed two models, i.e., KIM-STILT and WRF-STILT, which simulate CO 2 contributions from fossil fuel emissions.To ensure a thorough examination, we selected a few specific cases for evaluation, taking into account the distinct capabilities of each station in capturing varying levels of enhancement strength due to their locations.This kind of assessment will be helpful in highlighting the consistency and discrepancies between the models.

Anmyeondo
The variability of CO 2 mole fraction enhancements is governed by the spatial distribution of surface fluxes and atmospheric transport.Accurate CO 2 fluxes and transport models with high resolution are essential for the realistic simulation of CO 2 variability (Patra The simulated CO 2 mole fraction enhancement due to surface fluxes was compared with the measured ΔCO 2 to assess how well the models (KIM-STILT and WRF-STILT) reproduce the time series of the observed ΔCO 2 .Figure 6 presents the comparison of time series of the hourly simulated CO 2 ff and measured ΔCO 2 for both July and December 2020.In July, we observed that the fluctuations of the simulated CO 2 ff from both models were insufficient to capture the measured ΔCO 2 , as indicated by a low correlation coefficient of 0.26.Upon examining the amplitude of CO 2 ff, both models underestimated the values.In December, there was a better agreement between the models and observations, with a correlation coefficient of 0.60 for KMI-STILT and 0.51 for WRF-STILT.Despite the underestimation of CO 2 amplitudes compared to the observed values, the models were able to reproduce the phase of measured ΔCO 2 reasonably well.We also examined a few cases with significant measured CO 2 enhancements and analyzed the footprints grid coverages to identify their respective source areas.For instance, on July 9th (daily mean), there was a substantial CO 2 enhancement of 26.3 ppm, which is mainly originated from the night and early morning estimates.The corresponding CO 2 ff values were estimated to 5.8 ppm and 5.6 ppm by KIM-STILT and WRF-STILT, respectively (Fig. 7a and b).However, both models exhibited an underestimations of the observed CO 2 mole fraction enhancements.In another case in December, on the 28th, a high in situ CO 2 enhancement of 19.3 ppm was detected.The estimates provided by KIM-STILT Fig. 7 Spatial distributions of daily averaged surface grid contribution for high CO 2 enhancements (09 July and 28 December 2020) cases at AMY site (left panels for KIM-STILT and right panels for WRF-STILT) Unit for CO 2 ff is given in log10 (ppm) and WRF-STILT were 5.1 ppm and 17.6 ppm, respectively.The estimates from WRF-STILT are closer to the observation.This significant contribution originated from the nearby Anmyeondo station and Seoul metropolitan area, which was not reproduced by KIM-STILT, as shown in Fig. 7a and b.Despite the significant role of fossil-fuel emissions in the observed CO 2 enhancements, biospheric flux is also responsible for CO 2 enhancement and reduction.The contribution of the biospheric CO 2 component varies significantly due to the strong seasonality of biospheric flux.During winter, the increases of CO 2 levels by the biosphere were found to be larger than 3 ppm in East Asia, with variations across regions, and more than 5 ppm in North China and Southeast China.On the other hand, during summertime, the biosphere leads to a substantial reduction (> 7 ppm) in CO 2 levels in most areas in East Asia (Kuo et al. 2015).In addition to that, the discrepancy between the observed and simulated can be caused by uncertainties of bottom-up emissions, the accuracy, and the limitation of horizontal resolution of the transport model.

Gosan
At JGS, we present the comparison of time series between the simulated CO 2 ff and measured Δ CO 2 for July and December 2020 in Fig. 8.In July, we obtained a correlation coefficient of approximately 0.3, which is similar in magnitude to that of AMY.This indicates that the models did not accurately reproduce the timing of CO 2 ff variations.Furthermore, both models underestimated the amplitude of CO 2 ff during this period.In December, the models showed some improvement by fairly capturing the phase of the observed CO 2 , as indicated by a correlation coefficient of 0.57 for KMI-STILT and 0.55 for WRF-STILT.However, the amplitude of CO 2 ff in the model simulations remained underestimated.Overall, the results suggest that while the models performed better in December, there are still some discrepancies in reproducing both the timing and amplitude of CO 2 ff variations between the simulations and observations.
We selected a specific case on July 8 (daily mean) and observed a CO 2 enhancement of 5.1 ppm.Remarkably, the models were able to accurately reproduce this observed CO 2 enhancement.For this particular case, the KIM-STILT and WRF-STILT models estimated corresponding CO 2 ff values of 5.3 ppm and 6.6 ppm, respectively (as shown in Fig. 9).The differences between the models' estimations can be attributed to variations in CO 2 ff originating from the southern part of Korea and eastern China.As illustrated in Fig. 9, the WRF-STILT model estimates a slightly higher spatial distribution of CO 2 ff over the Shandong area in China compared to KIM-STILT.Similarly, we took an additional case on December 7 (daily mean), during which the CO 2 enhancement measured by in situ was 7.1 ppm.The estimates produced by KIM-STILT and WRF-STILT were 1.4 ppm and 3.1 ppm, respectively.Upon examining the spatial distribution of CO 2 ff estimated by both models, we found consistent patterns in identifying the sources of CO 2 ff.However, it is worth noting that WRF-STILT yielded slightly higher estimates than KIM-STILT in the region where major contributions originated, specifically eastern China (Fig. 9c and d).These differences can be attributed to variations in the footprint sensitivity between the two models.This comparison emphasizes the importance of accurately characterizing CO 2 ff from specific source regions, as these variations can significantly impact the overall modeling results.By identifying such discrepancies and understanding their underlying causes, we can enhance the reliability and accuracy of the models, leading to a more comprehensive understanding of CO 2 transport.

Ulleungdo
Figure 10 illustrates the time series of observed CO 2 mole fraction enhancements and CO 2 ff from KIM-STILT and WRF-STILT for July and December 2020 at the ULD site.The correlation coefficients between in situ measurements and model predictions were found to be 0.4 and higher than 0.7 during the respective months, indicating a reasonably good capture of the timing, especially in December, compared to the AMY and JGS stations.However, like the other stations, the models underestimated the amplitude of the CO 2 ff compared to the actual observations.As discussed in Section 3.2.1, when considering the CO 2 ff component alone, a realistic representation of CO 2 enhancements cannot be reproduced, particularly evident in the summer.In December, the consistent underestimation of CO 2 ff amplitude against observations was more evident in KIM-STILT than in WRF-STILT.For instance, on December 7, KIM-STILT and WRF-STILT estimated to 2.5 ppm and 5.8 ppm, respectively, which are lower than the observed value of 10.1 ppm (Fig. 11).This discrepancy within the models was mainly attributed to differences in footprint sensitivity, which were obvious far from the receptor stations.The overall underestimations of the models compared to observations can also be attributed to uncertainties in the spatial distributions of CO 2 emissions from the ODIAC inventory.One factor contributing to the uncertainty of ODIAC emissions is the spatial disaggregation methods that utilize nighttime light data from satellite images to distribute emissions.This approach tends to concentrate hotspots more strongly in areas with high nighttime light, potentially leading to an underestimation of industrial and transportation emissions (Han et al. 2020 and references therein).Han et al. (2020) also indicated that the emissions from ODIAC were not captured over western China, where other emission inventories showed lower emissions.Additionally, the ODIAC emissions provided monthly estimates that do not account for hourly variations, which can impact the time series of hourly CO 2 ff estimates, in addition to the limitations of the transport models.

Conclusion
Our work introduces a near-real-time system that employs KIM meteorological data to drive STILT footprints, enabling precise identification of source regions responsible for elevated atmospheric GHG mole fractions at Korea's in situ stations.The strategic utilization of the KIM model, supported by the Korea Meteorological Administration, ensures continuous access to up-to-date meteorological information critical for accurate assessments.This serves as a pivotal asset, overcoming the limitations faced by the WRF model in offering timely data due to its complexity and computational requirements involved.Our comparison of the WRF-STILT and KIM-STILT transport models underscores the significance of the KIM-STILT model's development.
We analyzed the major source regions and footprint at three different Korea in-situ stations (AMY, JGS, ULD) using KIM-STILT and WRF-STILT.The comparison of monthly aggregated footprints for July and December 2020 revealed similarities and differences in their representation of major source regions.Both models showed similar major source regions for the AMY station in July, except for a discrepancy in the representation of the Eastern Pacific Ocean as a source region, with KIM-STILT showing a larger footprint magnitude.In December, both models indicated strong sensitivity of footprints covering Northeast and Eastern China, as well as the Yellow Sea, but KIM-STILT exhibited smaller sensitivity in Northwestern China (particularly, Inner Mongolia) and Mongolia compared to WRF-STILT.For the ULD station in July, both models indicated similar source regions with comparable footprint magnitudes, but a notable difference was observed in Southeast China, where KIM-STILT simulated a stronger sensitivity compared to WRF-STILT.Similarly, at the JGS station, both models agreed on major source areas but differed significantly over North and Northeastern China, with WRF-STILT showing stronger sensitivity.
Regarding CO 2 mole fraction enhancements, both KIM-STILT and WRF-STILT simulated CO 2 ff, which is generally underestimated the amplitude of ΔCO 2 , and struggled to capture the timing accurately especially in July.The contribution of biospheric CO 2 to the observed enhancements was also highlighted, and it varies significantly due to the strong seasonality of biospheric flux.However, in December, the models showed better agreement with the observed data, especially in reproducing the timing of CO 2 ff variations as evidenced by the correlation coefficient of 0.51 to 0.76.Despite the underestimation of CO 2 amplitudes, the models were able to reasonably reproduce the phase of observed ΔCO 2 .Specific cases with significant CO 2 enhancements were examined, and the models provided varying estimations of CO 2 ff values for these cases.The differences in the estimated CO 2 ff values could be attributed to variations in the footprint sensitivity between the two models, particularly in specific source regions like eastern China.Additionally, uncertainties in the spatial distributions of CO 2 emissions from the ODIAC inventory can contribute to the overall underestimations of the models compared to observations.
Overall, our study provides valuable insights into the potential advantages of each model in capturing dispersion patterns in specific regions.It underscores the importance of understanding these differences to improve the accuracy of atmospheric transport models.The study also revealed the need for further investigation to understand the underlying factors causing discrepancies in the models' performance.

Fig. 11
Fig. 11 Spatial distributions of daily averaged surface grid contribution for large CO 2 enhancements for ULD site (left panels for KIM-STILT and right panel for WRF-STILT)

Table 1
List of meteorological variables derived from KIM model used as input data for STILT are presented

Table 2
Short brief of model is provided

Table 3
Physics parameterizations used in WRF Version 4.3.3 are summarized Physics SchemeMicrophysicsWSM 6-class graupel scheme Short-wave radiation Rapid Radiative Transfer Model (RRTMG) scheme Long-wave radiation rrtmg scheme Land-water surface Noah land-surface model and urban canopy model Cumulus Kain-Fritsch (new Eta) scheme Fig. 2 Map of measurement sites, AMY, JGS, and ULD (marked by star symbol), located over Korea.The satellite map provided by © Google maps is embedded using QuickMapServices plugin