This manuscript investigates an important and practically relevant error source in passive satellite retrievals of greenhouse-gas column abundances. Biases in XCO₂ caused by correlated sub-pixel variability in surface reflectance and surface altitude. The authors argue that the conventional use of an unweighted mean surface altitude within a satellite ground pixel can misrepresent the effective light path when brighter sub-pixel areas are preferentially located at higher or lower elevations. To address this issue, they derive an albedo-weighted effective surface altitude and test the concept using Copernicus Sentinel-2/HLS surface reflectance, Copernicus DEM data, and the FOCAL-CO2M retrieval framework. The manuscript applies the method to idealized two-part pixels and to several real-world examples. The study also proposes an empirical post-processing correction and discusses implications for surface-roughness filtering and spatial-sample size.

I recommend minor revision before the manuscript can be considered for publication.

Major comments

1. The manuscript suggests that the albedo-weighted surface altitude should be used in trace-gas retrievals and that it can reduce biases below future mission requirements. This is a potentially important conclusion, but the evidence presented is not yet sufficient to support it in an operational sense.

The study uses FOCAL-CO2M to generate simulated spectra and then retrieves XCO₂ with FOCAL-CO2M using either unweighted or albedo-weighted altitude input. The authors also state that no radiance noise is added, although a realistic uncertainty based on signal-to-noise ratio is used in the inversion.

I recommend that the authors revise the abstract, conclusions, and relevant parts of Section 4 to make clear that the demonstrated improvement applies to self-consistent FOCAL-CO2M simulations under controlled assumptions. Statements such as “below the requirements for future satellite missions” should be qualified, particularly because the Mont Blanc example still shows substantial residual errors even after applying the albedo-weighted altitude. A more appropriate conclusion would be that the method substantially reduces one specific component of terrain-related XCO₂ bias and may be valuable for future retrieval development, but that its operational benefit remains to be demonstrated under realistic measurement and forward-model uncertainties.

2.  In the theoretical derivation in Section 2, the authors derive an exact effective altitude expression under a simplified Lambert-Beer framework and then apply a first-order Taylor expansion around the mean surface altitude. This leads to the albedo-weighted surface altitude. The derivation is elegant and physically intuitive, but the validity of the first-order approximation is not sufficiently quantified.

The manuscript states that the expansion assumes the deviations of individual sub-pixel altitudes from the mean altitude are small enough. However, later applications include cases with large sub-pixel altitude variations, including mountainous terrain and surface-roughness bins extending to several hundred meters. In such cases, second-order and higher-order terms may not be negligible. The Mont Blanc case appears to be an example where the albedo-weighted approach still leaves residual XCO₂ errors of order ppm, suggesting that the first-order approximation or other simplified assumptions break down for strongly heterogeneous high-mountain scenes.

I suggest the authors add a dedicated analysis comparing the exact effective altitude from their non-linear expression with the first-order albedo-weighted approximation. This should be done across a representative range of altitude variability, albedo contrast, background altitude, solar/viewing geometry, and optical depth. Such an analysis would allow the authors to define a validity domain, for example in terms of surface roughness, maximum sub-pixel altitude range, albedo contrast, or expected XCO₂ residual.

3. The paper discusses scattering using a small number of prescribed scattering scenarios. This is useful, but the range appears limited. In the appendix, the authors test no scattering and two scattering cases with different optical thickness and scattering-layer pressure. The resulting correlations suggest that the empirical correction remains broadly useful, but one of the scattering cases has much lower explanatory power than the others.

I suggest the authors avoid statements implying that the proposed correction is valid over a wide range of scattering conditions unless a broader sensitivity analysis is performed. A more complete test would vary aerosol optical depth, scattering-layer height, Ångström exponent, surface albedo level, solar zenith angle, and viewing geometry. At minimum, the authors should explicitly state that the scattering sensitivity is limited to the few prescribed scenarios tested in this study.

4. The manuscript presents several real-world examples with useful maps and qualitative discussion. A summary table would greatly improve readability and allow readers to compare the performance of the unweighted, albedo-weighted, and post-processed approaches.

Adding a table for each test region is recommended, which can include the number of valid spatial samples, mean bias, median bias, standard deviation, RMSE, etc. This table should include at least Bełchatów, the Black Forest, Mont Blanc, and Germany.

Such a table would also clarify where the method works very well and where it still leaves large residual errors.

Minor comments 

1. The abstract should explicitly state that the study is based on self-consistent simulations, not on real CO2M observations or independent satellite retrieval validation.
2. Line 75: The phrase “this examples shows” should be corrected to “this example shows”.
3. Line 81: The phrase “the first of three satellite” should be corrected to “the first of three satellites”.
4. In Eq. (13), the summation index appears to be written as (i=a). Please check whether this should be (i=1).