The manuscript analyses decorrelation in time and space from both a Lagrangian and Eulerian perspective with the ultimate aim to estimate how well Argo float act as Lagrangian platforms. The motivation for the paper is sound and it addresses some very important questions. I’m excited to use the results from a published version of the MS in future studies and believe it to have a wide potential utility. There are, however, a couple of major questions/concerns I need resolved before recommending publication.

1. The first equation suggests, to my understanding, that the Chl field is fixed in in space. This is a bold assumption that needs to be carefully motivated. I would have expected the advection decorrelation term to be applied to the Eulerian observer since Chl is advected with the velocity field. One could possibly argue that biomass might originate from stationary processes at for example seam mounds, but this is rather the exception than the rule. As a consequence, I expect that a Lagrangian sampling platform in general, with some specific exceptions, experiences longer temporal decorrelation time scales compared to the Eulerian observer. I’m willing to admit that I might have misunderstood Eq1 and the reasoning around it, but I don’t think I’m the only one if so. This need to either be explained better or changed.

2. The use of Chl fields with a 0.25° spatial resolution and the removal of sub- and mesoscale variability weakens the study significantly. It is abundantly clear that sub-mesoscale processes are of first-order importance in controlling the variability of Chl, as mentioned in the MS and cited publications by Amala Mahadevan or Marina Levy. A general analysis of decorrelation time- and length scales can get away with using coarser grids by defining the domain of interest carefully but this study doesn’t have that luxury. One specific aim, as I understand, is to evaluate the utility of float which requires the use of the highest resolution possible. I would have preferred that a 1km product had been used (OC-CCI at 1km is for example available from Plymouth Marine Laboratory), but I understand if a 4km product is used out of necessity. Aren’t the results quite dependent of the rather arbitrarily chosen 0.25° pixel size? How much would the results differ if 0.125°, 0.5, or 1° pixels were used instead?

3. The use of geostrophic velocities to estimate QPI is problematic. There are many processes that attribute to Lagrangian decorrelation missing from these fields- I’m not even sure if Ekman drift is included? Many of these forces are also likely to affect the upper ocean to larger extent, creating an even further biasing when being omitted. One easy test is to calculate QPI for the drifters the same way as the floats to see how representative the geostrophic velocity fields are. Another option is to conduct the excercise in a high resolution ocean model using virtual drifters and floats.

4. I’m not happy with how the Chl data for the floats is handled. The mean Chl concentration in the mixed layer is not what is observed by satellite. This is of particular importance in regions with deep Chl maxima where most Chl is close to the base of the mixed layer and not visible from space. This issue can easily be amplified in this study if there is MLD variability over short timescales or if the isolines are sloping. Each case could lead to spurious variability in Chl observed by the float, compared to the drifters. The correct approach would be to use attenuation or PAR from the float (or Kd490 from satellite in if not available on the float) and average the Chl data down to the first optical depth. An even better approach would be to match satellite Chl to the floats the same way as to the drifters. I don’t see any benefits in using In-situ observations for one platform and satellite-derived data for the other when comparing the two. 5. Finally, while the formalism in the MS is thorough and impressive, I think it might scare many potentially readers away. Cleaning up the text by explaining the reasoning in a way that can be understood by a wide audience and move a portion of the analytical description to an appendix would probably increase the readership statistics and potential of citations.

This manuscript presents extensive work evaluating Eulerian and Lagrangian time and length scales of velocity and chlorophyll, as well as discussion about how they correlate.  The proper interpretation of drifting phytoplankton observed in a Eulerian fashion is a long-standing paradigm in ocean ecology. However, estimates of Lagrangian phytoplankton statistics and comparisons with Eulerian counterparts are rare. This study represents a significant contribution towards best understanding how to interpret phytoplankton/chlorophyll measured in both Eulerian and Lagrangian platforms.  The authors are very thorough in their analysis and description of the results. Nonetheless, I have a few comments to be addressed prior to recommending publication.

Major comment: There are several data limitations that guide methodological decisions in an analysis of this type (e.g., the broad spatial averaging, chlorophyll averaging in the MLD). While some of the issues arising from these are mentioned briefly throughout the text, I would prefer to see a dedicated discussion section with the limitations and caveats.

Specific comments:

1. I find that, while technically correct, talking about Lagrangian-Eulerian “statistics” in the title and throughout the text can be misleading. Why not refer to the specific statistics that are included in the analysis? i.e., Lagrangian-Eulerian time and length scales.
2. The notation of upper case L for both Lagrangian and length-scale can be a bit confusing. I suggest using upper and lower case or a different notation to improve readability.
3. Equation 5. Terminology becomes confusing here too when calling the nominators Lagrangian, Eulerian and Spatial (chlorophyll) scales. Is there a different name that could be more appropriate and less confusing? This is essentially a change in chlorophyll, correct?
4. Table 1 is also confusing. Why are ACF bins different? Why are time windows for Eulerian and Lagrangian different? Does that have any effect on the comparison? (I think it would if you were calculating other statistics). Where does the 27.8km ACF bin for Eulerian length scale come from? I probably missed it.
5. Figure 1d. orange profiles: QPI<5km; blue: all others (i.e., not total)
6. Line 180. Please specify the convention for flagging. It is my understanding that BGC-Argo flagging may have changed through the years and between institutions. (I’ve used Sprofs where 3 means bad).
7. How does the GlobColour product compare to other products? Why is this one selected over others? (OCCCI, for instance). I suggest including a brief sentence.
8. Section 3.3 could be simplified. Two sets of chlorophyll anomalies are estimated: 1. Anomalies with respect to a 31-day smoothing filter, and 2. Anomalies with respect to the climatology. I would suggest stating something like that to start, and then continue with the details.  
9. The climatology is based on the same 31-day filter + a boxcar function? This is not exactly what comes to my mind when “climatology” or “repeating annual cycle” is mentioned.
10. I don’t like the use of the satellite-based “subtrahend” to estimate chlorophyll anomalies from the MLD-averaged chlorophyll from the float. How does the MLD average compare to the satellite? I think some type of bias correction may be needed. You mention that the subtrahend is regressed against float data. Do you mean you corrected a bias? That should be included in a supplement.
11. Line 240. Why aren’t Eulerian and Lagrangian segments equal?
12. Section 3.4 could be simplified as well. If I understand correctly, you tested two approaches to estimate spatially averaged scales. In lines 272-275 you mention you use one or the other. When and why you use each one should be clearer.
13. Line 292. Picks?
14. Lines 319-320: “If we take …” this sentence is confusing.
15. I probably missed this. Are the results in figures 5 to 9 based on all profiles or only QPI<5km?