Review attached.

Leguijt et al. presented their emission estimates of CO over several African cities based on TROPOMI observations and a computationally efficient cross-sectional flux (CSF) method. They presented methods for identifying the plume areas, calculating CO enhancements and CO fluxes, and for validating the CSF approach. They found discrepancies between space-based estimates and inventory-based estimates and leveraged emission ratios (ER) of CO to CO₂ from sector-specific emission inventories for further clues. An interesting “weekend effect” has also been revealed using CSF and TROPOMI XCO data. 

General comments:

African cities or African land, in general, are somewhat mysterious and challenging to study from both the carbon budget and air pollution perspective. Hence, this study provides some interesting insights into the emissions of those rapidly growing cities, which could now be made possible using satellites monitoring the entire globe. Overall, this paper provided details and figures to support the descriptions of their plume detection and CSF methods. Aware of the potential limitations associated with the simplified CSF method, the authors conducted analyses to quantify emission uncertainties and reveal the flux threshold using the CSF method. The emission results are nicely presented to clearly show regional-specific differences between top-down (TD) and bottom-up (BU) estimates.

Yet, based on the current presentation, it took me longer to comprehend the validation section (Sect. 2.6), e.g., the purposes of using WRF and what model quantities were the authors trying to optimize/evaluate without reading the previous studies in-depth. Although many ingredients of this paper were built on Varon et al. (2018), it would be more accessible to general readers if the authors could show how each obs/modeling component in Sect. 2 is related to one another since several different model products/variables were used for different purposes (e.g., TROPOMI, CSF, WRF, U10, Ueff, U from NECP, GEOS-FP, and ERA5...). For instance, having an overall flow chart may be helpful in showing what was derived from CSF, what was fed into/obtained from WRF simulations, and adjustments from Varon 2018 for urban emission estimates (e.g., a better characterization of wind fields). Detailed confusion about the validation procedure is included in the specific comments.

Three minor comments include 1) the lack of discussion of impacts from wildfire emissions and 2) secondary CO productions and CO sinks, and 3) possible discrepancy in the spatial extents of emissions from BU versus TD estimates (e.g., Fig. 6).

• Wildfire and biofuel combustions play significant contributions to XCO signals and combustion efficiency over the African land. It is likely that influences from wildfire and chemical sources and sinks are minimized by the subtraction of the background XCO. However, the authors should provide some supplementary materials or investigations by, for example, examining wildfire inventories or satellite-based burned areas during the study periods. It helps verify whether TD CO emissions from TROPOMI for certain cities are affected by non-anthropogenic emissions, which may lead to systematic differences from the BU fossil fuel estimates (since pyrogenic emissions usually have relatively higher CO/CO₂ ERs than most FF emissions).
• It is unclear if TD emissions derived from different TROPOMI overpasses represent roughly the same spatial extent of the selected grid cells from the two inventories (which may affect BU estimates). To yield an apple-to-apple comparison between the TD and BU estimates, one needs to ensure the spatial extents represented by the two perspectives are similar OR the TD vs. BU differences (e.g., in the map of Figure 6) are not sensitive to how the authors selected the inventory grid cells.

Specific comments:

L93: “To test and calibrate our emission quantification approach we apply it to simulated data.” – unclear. What do “it” and “simulated data” stand for?

L94: What did the authors mean by “simulate emissions”? Should it be “simulate column CO concentrations/mixing ratios”?

L168: TROPOMI XCO AKs are accounted for in the WRF-based simulations. However, how TROPOMI XCO averaging kernel profiles were accounted for in the CSF method is not super clear. Typical XCO AKs deviate from 1 towards the surface.

L 169 – L171: “...see whether the CSF can reproduce the emissions used as model input.” - So, emissions Q estimated using CSF and TROPOMI (Sect. 2.4) were fed into WRF to produce modeled XCO that can be evaluated against TROPOMI XCO? Or CSF is used to calculate emissions based on pseudo-XCO created by WRF + priors (e.g., EDGAR) like an OSSE experiment? What do “idealized” pressure weighted columns mean?

L174: “Parameters like the width of the transects are tuned to get optimal quantification estimates on the simulated data” – I guess my confusion is still related to the previous comment. I might miss something here, but how could the authors determine when the estimates are “optimal”, especially when both CSF and WRF provide modeled values, not true observations? For example, the WRF 10m wind may not be accurate. Were there any observed wind observations that could be leveraged?

L175-L177: “We then use the simulations to calibrate the CSF method following the procedure by Varon et al. (2018). The wind speed in Eq. 2 is replaced by an effective wind speed...” – Without reading Varon2018, readers may be confused by the sudden introduction of effective wind speed (Ueff not mentioned in Sect. 2.4). Also, did the authors end up using the U(x, y) in Eq.2 or the alternative Ueff? I would suggest providing some context to this Ueff and to the calibration procedure in Varon2018 (e.g., what it was designed for).

L217-218: What does “concentrated emissions” mean? Please reword.

L223-225: Does the regional inventory DACCIWA have a higher CO/CO2 ER than EDGAR, since DACCIWA seems to agree better with TROPOMI-based estimates?

Fig. 7: What are the emission sectors plotted in Fig 7? Only fossil fuel sectors or does it include other non-FF anthropogenic sectors (e.g., biofuel also with typical high CO/CO₂ ERs)? The EDGAR-based ERs appear to be always more variable than DACCIWA-based ratios. Would the choice of spatial extent affect those spatially mean ERs?

L239 – 240: Why modeled winds like NCEP and GEOS-FP become important when discussing results for Nigeria, but not for other regions?

Fig 8: interesting to see the huge inventory-inventory discrepancy in emissions over Lagos.

L279 – 280: “We evaluated the CSF method by applying it to a full-year of WRF simulations over three distinctly different African cities (Cairo, Lagos, and Bamako).” – Again, it would be more helpful to summarize what from CSF has been evaluated? Is it to evaluate CSF’s wind representations like U(x, y) or Ueff, its hyperparameters like # of transects, or its general capability in retrieval emissions (i.e., related to the simplified formula in Eq. 2 vs. full-physical models like WRF)?