CO2 Emissions Estimate From Mexico City Using Ground‐ and Space‐Based Remote Sensing

Ke Che, Thomas Lauvaux, Noémie Taquet, Wolfgang Stremme, Yang Xu, Carlos Alberti, Morgan Lopez, José Agustín García-Reynoso, Philippe Ciais, Yi Liu, Michel Ramonet, Michel GrutterJournal of Geophysical Research Atmospheres 2024 · DOI

TL;DR

Researchers used a combination of ground-based sensors (FTIR) and NASA’s OCO-3 satellite to create high-resolution (1-km, 1-hr) CO2 emission maps for Mexico City. The study found that existing emission inventories overestimate fossil fuel CO2 by up to 18.7%, and while ground and space sensors agree on human-caused emissions, they still struggle to reconcile the city's natural "biogenic" carbon cycle.

What Problem Does This Paper Address?

Urban areas are the primary engines of global CO2 emissions, yet quantifying exactly how much a city breathes out is surprisingly difficult. Traditionally, we use "bottom-up" inventories—estimates based on fuel consumption stats, traffic data, and population density. However, these are often outdated or lack the granularity needed for local policy.

"Top-down" methods use atmospheric observations to work backward to the source (inversion). While ground-based networks provide continuous data, they are expensive to deploy. Satellites offer global coverage but face challenges with precision and temporal gaps. This paper tackles the challenge of reconciling these two data sources to validate emission inventories for the Mexico City Metropolitan Area (MCMA), one of the world's most densely populated regions.

Methods

The authors employed a Lagrangian Inversion Framework to bridge the gap between observed concentrations and surface fluxes.

  • Transport Modeling: They used the X-STILT (X-Stochastic Time-Inverted Lagrangian Transport) model, driven by WRF (Weather Research and Forecasting) at a 1-km resolution. This simulates how air parcels move backward in time from a sensor to identify which surface areas contributed to the CO2 "plume."
  • Analytical Bayesian Inversion: To handle the massive computational load of 1-km, 1-hr resolution across a large metropolitan area, the team used a multi-node parallel strategy (leveraging 120 nodes). This allowed them to solve the "inverse problem"—adjusting prior emission guesses to match observed data—while maintaining a high degree of spatial detail.
  • Multi-Resolution Gridding: To save on memory without losing accuracy, they used a fine 1-km grid over the city and a coarser grid (up to 50-km) for the surrounding rural areas.

Datasets

The study integrated several high-quality data streams:

  • Ground-Based: A dense network of seven Fourier Transform Infrared (FTIR) spectrometers (including portable EM27/SUN units) deployed during the MERCI-CO2 project.
  • Space-Based: NASA’s OCO-3 Snapshot Area Maps (SAMs), which provide 80x80 km² "images" of CO2 concentrations over cities.
  • Prior Inventories: The "starting guesses" included UNAM_EMI (a local Mexico-specific inventory), ODIAC (a global fossil fuel dataset), and CASA (for biogenic/natural fluxes).
  • Background Data: CO2 values from the CAMS global inversion were used to account for air entering the Mexico City basin from elsewhere.

Key Insights and Ideas

  • The "Overestimation" Gap: The local inventory (UNAM_EMI) was found to overestimate fossil fuel emissions by 18.73%. Interestingly, the global ODIAC inventory was much closer, with only a 5.88% overestimation. This suggests that local inventories might struggle with the "point source" intensity of industrial sites like the Tula power plant.
  • Convergence on Fossil Fuels: When the ground-based FTIR data was filtered to match only the exact times the OCO-3 satellite passed overhead, the two independent systems yielded nearly identical corrections for fossil fuel emissions. This is a major win for satellite validation.
  • The Biogenic Divergence: While the sensors agreed on human emissions, they disagreed on the natural carbon cycle. FTIR suggested the city's surrounding vegetation was a stronger "sink" (absorbing CO2), while OCO-3 data suggested a "source." This highlights the sensitivity of biogenic estimates to rural sampling locations and the lower temporal density of satellite passes.
  • The Power of Aggregation: The study found that OCO-3’s raw 1-km data is noisy. However, by aggregating the data to coarser resolutions (up to 30-km), the signal-to-noise ratio improved significantly, leading to much more robust inversion results.

Key Results

  • Fossil Fuel Optimization: Initial estimates of 5.08 and 6.77 GgCO₂/hr were optimized down to a range of 4.85 to 5.51 GgCO₂/hr.
  • Biogenic Flux: The natural sink was refined from an initial -0.14 GgCO₂/hr to between -0.27 and -0.33 GgCO₂/hr based on ground observations.
  • Model Performance: The inversion significantly improved the correlation ($R$) between simulated and observed CO2 gradients, jumping from ~0.6 to >0.9 for the ground-based network.

Limitations and Future Work

The primary limitation remains transport error. If the wind speed or the height of the planetary boundary layer (PBL) in the WRF model is off, the entire inversion shifts. The authors noted that higher transport errors lead to "minimal adjustments," meaning the system essentially gives up on correcting the prior inventory because it doesn't trust the wind data.

Future work aims to incorporate "co-emitted" tracers like NOx or CO. Because these gases are often emitted alongside CO2 during combustion but have different atmospheric lifetimes, they can help researchers distinguish between fossil fuel plumes and natural biological activity more accurately. Additionally, the authors suggest that direct comparison with eddy flux towers (which measure CO2 exchange at the surface) will be the next step in resolving the biogenic discrepancy.

Summary updated

Abstract

Abstract The Mexico City Metropolitan Area (MCMA) stands as one of the most densely populated urban regions globally. To quantify the urban emissions in the MCMA, we independently assimilated observations from a dense column‐integrated Fourier transform infrared (FTIR) network and OCO‐3 Snapshot Area Map observations between October 2020 and May 2021. Applying a computationally efficient analytical Bayesian inversion technique, we inverted for surface fluxes at high spatio‐temporal resolutions (1‐km and 1‐hr). The fossil fuel (FF) emission estimates of 5.08 and 6.77 Gg/hr reported by the global and local emission inventories were optimized to 4.85 and 5.51 Gg/hr based on FTIR observations over this 7 month period, highlighting a convergence of posterior estimates. The modeled biogenic flux estimate of −0.14 Gg/hr was improved to −0.33 to −0.27 Gg/hr, respectively. It is worth noting that utilizing observations from three primary sites significantly enhanced the accuracy of estimates (13.6 29.2%) around the other four. Using FTIR posterior estimates can improve simulation with the OCO‐3 data set. OCO‐3 shows a similar decreasing trend in FF emissions (from 6.37 Gg/hr to 6.36 and 5.04 Gg/hr) as FTIR, but its correction trends for biogenic sources differ, changing from 0.37 to 0.48 Gg/hr. The primary reason is OCO‐3's lower temporal sampling density. Aligning the FTIR inversion timing with that of OCO‐3 yielded comparable corrections for FF emissions, yet discrepancies in biogenic emissions persisted, which can be attributed to their different sampling locations in the rural region and discrepancy in X observations. Our findings mark a significant step toward validating OCO‐3 and FTIR inversion results in metropolitan region.
BibTeX
@article{che2024co,
  author = {Ke Che and Thomas Lauvaux and Noémie Taquet and Wolfgang Stremme and Yang Xu and Carlos Alberti and Morgan Lopez and José Agustín García-Reynoso and Philippe Ciais and Yi Liu and Michel Ramonet and Michel Grutter},
  title = {{CO<sub>2</sub> Emissions Estimate From Mexico City Using Ground‐ and Space‐Based Remote Sensing}},
  journal = {Journal of Geophysical Research Atmospheres},
  year = {2024},
  doi = {10.1029/2024jd041297},
}
Figure