I found this paper to be carefully constructed and well written.  The authors provide a clear motivation for the work, and did a good job explaining details that might not be obvious to all readers. For the most part the reader need not be familiar with previous work to understand and follow the discussion in this paper.

Section 1

The authors fail to discuss the version of the TropOMI Level 1B product used in this work until Section 6.3.  The version should be cited early in the paper along with the doi of the data.  As the authors note, the TOA reflectance changes between product versions so it's important to state key facts early.  

A similar criticism can be made regarding accuracy requirements.  The requirements for this work best belong in the introduction where their origin can also be described.

   

Use of a DLER product from TropOMI is pretty much limited to satellite observations in a 1330 orbit, possibly a few others if reciprocity is assumed.  While there are a significant number of instruments orbiting at this time of day it still limits the application of these results.  If the authors were to assume surface BRDF models (e.g. from MODIS) it should be possible to derive a total hemispheric reflectance (THR) from these measurements.  A THR product can broaden the reach of these data, allowing a larger pool of potential comparisons including instruments in morning polar orbits and geostationary instruments.  The authors have effectively already performed such a comparison between afternoon orbit TropOMI data and the morning orbit MODIS data.  Just a thought for a future paper.

Section 4.5

It would be useful to know how sensitive LER is to the AI screening level as a way of evaluating the chosen threshold.  Have the authors performed such a study?  Can they provide some more justification for the screening threshold of AI = 2?

Sections 4.6 & 4.7

In these two sections the authors are perhaps a bit too reliant on the reader having read and remembered their previous publication on this topic.  The reader is left wondering about the general approach.  For example, an elaborate method of selecting a representative LER for each grid is described in Section 4.6.  No mention is given to view angles, so one must assume that all angles are included in the final distributions.  However, a selection of values is then made based on the lowest 10% (or the mode of the distribution in the case of snow/ice).  Such a selection is necessarily biased toward view angles where the BRDF is at a minimum, meaning the Section 4.6 LER is dependent on viewing conditions.  In Eqn. 8 a quantity ALER is introduced for the first time with no explanation of where it comes from.  The reader can deduce from Lines 237-239 that ALER is the reflectivity of water, which is not a Lambertian quantity except in ideal conditions. How does the LER of Section 4.6 relate to Eqn. 8?  Section 4.7 stands out in its need for clearer explanation when compared with the others parts of this paper.

Section 5.4

The authors use a visual example to demonstrate the need for and the effectiveness of their cloud screening method.  In the example shown in Figure 6 it is not immediately obvious that every feature in black is a cloud shadow that should be removed.  Shadows at this location should appear to the north north-east of the actual cloud, and it's rather difficult to imagine clouds that could produce some of the shadows seen to the north and north-east of Iceland in Figure 6a.  It would be helpful if the authors could include a VIIRS RGB image or a TropOMI reflectivity image for the same time period to show the actual cloud field.  There may be clever ways of including this as a transparency overlayed on the minimum reflectivity maps.

Section 6.1

The authors should cite the product versions used in all the comparisons described in this section.

Section 6.2.1, Lines 408-411

The authors cite Rayleigh scattering effects as a reason for not comparing DLER in the UV to the MODIS BRDF.  To first order Rayleigh scattering should already be taken into account via the derivation of surface reflectivity given by Eqn. 2.  The more important reason for not comparing in the UV is that MODIS makes no measurements at these wavelengths and few, if any, estimates of BRDF exist at these wavelengths.  It's worth noting that a UV comparison in conjunction with a long-wave VIS comparison to the MODIS BRDF could provide a useful assessment of how BRDF might change between VIS and the UV.

Section 6.3, Lines 481, 482

The authors state that TropOMI calibration is much improved in the v2.1 Level 1 product compared to v1.0.  Can they provide a reference to substantiate this claim?  There exists evidence that the TOA reflectance in the latter version is actually less accurate than in the earlier version.  Figure 10d in particular is suggestive of a calibration difference of ~2%.

The paper describes an important and timely step that should substantially improve characterization of the Earth’s surface reflectivity derived from the TROPOMI observations. Following Vasilkov et al. (2017 - OMI), Loyola et al. (2020 - TROPOMI) and Tilstra et al. (2021  - GOME-2), the paper re-introduces the directionally dependent Lambertian-equivalent reflectivity (DLER) for TROPOMI and compares the results with various non-directional LER data sets. Such comparisons are indicative, though not conclusive, considering the profound differences in the LER and DLER approaches (e.g., the viewing-angle DLER dependencies from Fig. 4, cf. the angle-independent LER). Hence, the authors validate the results by comparing TROPOMI DLERs to the MODIS surface BRDFs, however only in the near-IR region. The λ< 500 nm domain is critically important for retrievals of many trace-gas species. Surprisingly, the authors completely avoid any comparisons with the publicly available database that is built on similar (directionally-dependent LER, known as GLER: Vasilkov et al. 2017) physical principles, samples the Earth’s surface at similar local solar times, and belongs to the wavelength region of interest: 

https://disc.gsfc.nasa.gov/datasets/OMGLER_003/summary?keywords=AURAMLS

This can be perceived as a serious deficiency of the reviewed study unless there are sound reasons for excluding these data from consideration. If any, such reasons should be explicitly stated and commented on.

More comments:

In Section 2 (Description of TROPOMI) the authors should mention what L1B version is used in the study. This should be tied to the information on how the instrument degradation is accounted for, if applicable to the case. Moreover, the very important and relevant topic of the quality of absolute TROPOMI calibration should be described in some detail. If pertinent, the latter may explain some systematic differences shown in Section 6. Can the introduced c_0 term (Eq. 8) be explored/exploited as a potential link to the absolute calibration uncertainties?

l.173 – briefly comment on the ‘undisputed cases.’ Is/are there any threshold value(s) bounding such cases?

l.199 – mention the source of the AAI data used for screening.

l.217 – the mode of a distribution could be a robust metric for the scenes with permanent ice/full snow coverage. However, this may not be applicable to the partial/thin-snow landscapes. Is there any recipe for addressing such cases that, if mis-represented, may profoundly bias the trace-gas retrievals? Can (should) the ‘clear’ and ‘snice’ fields be mixed in the cases of partial coverages? Please share your experiences (know-how) with this sensitive scenario.

l.239 (also applicable to Section 4.8, i.e., the post-processing routines) – please describe how the solar-glint regions are incorporated (filtered? outright rejected?) under the approach.

l.242 – how does the proposed thresholding work over the solar-glint areas? Please provide more details.

Fig. 5 – the adopted false-color scheme does not serve the purpose. With the exceptions of some cloud/aerosol contamination over the open-water areas, no effects (the claimed cloud-shadow contamination inclusive) are perceivable over the continental landmasses, since one may not be able to distinguish between the seasonal and the contamination-induced trends: e.g., Brazil/Amazonia in (a) vs. (c). Moreover, thus presented, the potential contamination cannot be quantified in any meaningful way. Instead, one may consider providing global maps showing estimates of the contamination magnitudes for some key (trace-gas retrievals, clouds) wavelengths, preferably expressed as percentages of the underlying reflectivities.

l.466, 469, etc. – please clarify the meaning of ‘0.03+10%’.

l.464 and on – please comment, one more time, on the relatively large differences seen in the TROPOMI-GOME-2 comparisons (Table S4).

l.118 - …course…

The Supplement:

Is there any reason behind the very different seasonal trends seen in the Asian (sub)-tropical forests (Figures S1-S3)? The difference between 463 nm and the adjacent 670 and 380 nm samples is rather striking.  One may call the May 380 nm outlier a suspect (instrument – unmitigated straylight? clouds?). This difference is even more striking considering the close match between the average Asian-forest spectra plotted in Figure S4. The same may apply to the Evergreen woodland at 380 nm (August) as seen in Figures S1-S3.

Figure S13 may help to estimate the magnitude of cloud contamination in the continental North America DLER data (April-September biases). This important estimate could be mentioned in the main text.