Review of “Two-decade satellite monitoring of surface phytoplankton functional types in the Atlantic Ocean” by Hongyan Xi et al.   

General comment:

The paper submitted by Hongyan Xi et al. focuses on PFTs in the Atlantic Ocean as seen by satellite in the last two decades. The paper is well written and easy to be read and represents a useful analysis and report to understand what is the Atlantic Ocean ecosystem status.

Overall, I have some specific comments that are preferable be included in the revised version of the manuscript. Especially, some sections need to be extended with more information: for instance, some more details about the EOFs technique in the section of time-series analysis, or the definition of the statistical indexes used to evaluate the satellite products in respect to the in-situ dataset. Two key points are: extend the discussion of the results found in the validation test of satellite vs. in situ data, specifically for what concern the statistics; and the inclusion xof a significance test (e.g. p-value) to better interpret the trends found in the different regions.

However, on my opinion, the paper needs only minor corrections before the publication. This is a useful work for the ocean colour community from both scientific (the understanding of the PFTs changes in the Atlantic Ocean ecosystem) and technical (e.g. products validation, long-term time-series) perspectives.

Specific comment:

Section 2.1

I would include a table with the wavelengths/bands of the different sensor here used and that have been controlled. I would add a Figure that pointed out the life cycle of each sensor and the overlap of the different mission. Such a information give to the reader a clear, easy details about the last 20 years of satellite sensors and mission.

Section 2.3

Please extend the section with more details about the method you have applied (i.e. EOFs). Currently, it is not mentioned in the section about the time-series analysis.

Section 3.1

Lines 134: you mentioned that the same correction approach is then applied on Sentinel-3A OLCI derived PFTs. Please add some details about how it works for OLCI.

Lines 139: Please add some more details about the type of errors reported in the table 1. How is defined the MDPD? and the RMSD? What means a MDPD of 89.6%? It indicates a good agreement between satellite and in-situ data or viceversa. Which is the unit of median uncertainty, it is expressed in %.

The authors need to better describe the results found here, since this is the first step behind the successive important and impactful time-series analysis: for instance, an R coefficient of 0.6 is equal to an R2 of 0.36 that implies not a strong correlation between in-situ and satellite-derived PFT. This is consistent or a better results in respect to the previous works and literature?

Section 3.3.

Discussion of trend (increasing/decreasing) needs to be also coupled to a statistical significance test as the p-value. This coefficient could give to the reader a tool to understand the inter-annual variability and if it has a statistical significance in the last 20 years.

Section 3.4

How did you compute the abundance maxima time? How it is defined? Please add more details about the computation in the Section 2.3 (time-series analysis).

Technical Corrections (some examples):

Lines 35-38: please rephrase the sentence.

Line 42: I’d remove “as a whole” à you can leave the sentence as: “phytoplankton biomass does not provide a full description of the complex nature of phytoplankton community and function”.

Line 44: please rephrase the sentence “Phytoplankton composition structure varies across ocean biomes and different phytoplankton groups influence marine ecosystem and biogeochemical processes differently (Bracher et al., 2017) à I would change in “Phytoplankton composition varies across ocean biomes and the different phytoplankton groups influence marine ecosystem and biogeochemical processes differently (Bracher et al., 2017)”

Line 51: I would remove “as well”, it is not necessary here.

Lines 72-74: I do not understand to what referred such sentence, from “Previously” to “measured pigments”.

Line 78: chlorophyll a can be modified in Chla since you have defined in the first part of the paragraph.

Line 80: maybe a refuse of something, maybe you can change in in this work.

Review of “Two-decade satellite monitoring of surface phytoplankton functional types in the Atlantic Ocean”

Hongyan Xi, Marine Bretagnon, Svetlana N. Losa, Vanda Brotas, Mara Gomes, Ilka Peeken, Antoine Mangin, Astrid Bracher

General comments

Using long-time series of satellite-derived PFT products, Xi et al. investigate the two-decade trends, climatology, phenology, and anomaly of PFT in the whole Atlantic Ocean and its different biogeochemical provinces. Firstly, based on their previous studies (Xi et al., 2020, 2021), the authors obtain PFT Chla products (mainly four phytoplankton groups) using three different sets of ocean color data, covering from 2002 to 2021. Through the independent validation and the inter-comparison of among three sets of ocean color data within overlapping time periods, they then identify the systematic differences caused by different data sources, set up the SeaWiFS/MODIS/MERIS-derived products as reference, and correct the other two PFT datasets, to generate a consistent long-time series PFT products over the last two decades. Finally, the trends and variations of PFTs in the Atlantic Ocean and its biogeochemical provinces are analyzed.

In general, the manuscript is well written and logically organized, with clear purposes and conclusions. The findings of this work provide a preliminary distribution and variation of four PFTs in the Atlantic Ocean, and potential contribution of understanding how these phytoplankton groups respond to climate changes. However, there are some issues to be addressed before it can be considered for publication. Please see below for some major and specific comments.

Major comments

1. In this study, when generating the consistent PFT products from different OC data sources, systematic differences among three sources are adjusted based on the relationships between Chla of PFTs derived from each OC data, as shown in section 2.1 and Figure 1. Since the Chla of PFTs are derived from Rrs, i.e., secondary products, why not consider correcting inter-mission bias of the Rrs first, then derive PFTs using the corrected Rrs? In this circumstance, for each of the other two OC datasets, only one correction with respect to Rrs is required, rather than four corrections for different phytoplankton groups. Have you compared the differences of PFT products between these two procedures?
2. As mentioned in the discussion, 20-year observation may not be enough for a robust trend analysis. I was wondering why PFT results derived from SeaWiFS between 1997 and 2002 are not included in this study? If they are included, the length could be extended to ~25 years.
3. Some parts of section Data and Method are lack of details, including: (1) the quality control of pigment data from Aiken et al. (2009); (2) diagnostic pigment analysis; (3) the correct functions among different OC data sources; (4) the sources/website of Longhurst’s geographic classification system; (5) per-pixel uncertainties of PFT products derived from Rrs data; and (6) the calculation of the anomaly of PFTs for year 2021. Please consider including more details here, such as the description, equations, or detailed information from the references cited here (e.g., the number of equations/figures in previous works).

Specific comments

1. Line 89. Swich the order of Table 1 and Table 2. It seems that the Table 2 comes out firstly in the manuscript.
2. Lines 142-144. This sentence is not clear. Please rewrite it.
3. Lines 174 and 251. A decline of prokaryotes from 2013 onwards are observed in the study. Is there any possibility that the decline is related to the removal of MERIS data at this time? As argued in van Oostende et al. (2022), MERIS is able to observe more pixels near the coast and at high latitude, where Chla is higher. It may out of the scope of this study, but the coverage variability among different satellite missions should be taken into consideration in analyzing long-time series studies.  van Oostende, M., Hieronymi, M., Krasemann, H., Baschek, B., & Röttgers, R. (2022). Correction of inter-mission inconsistencies in merged ocean colour satellite data. Frontiers in Remote Sensing, 3(July), 1–17. https://doi.org/10.3389/frsen.2022.882418

4. Line 177. Slight increasing trend of haptophytes on the…
5. Figure 1. Since the SeaWiFS/MODIS/MERIS mission is used as the reference and the MODIS/VIIRS (or OLCI) is corrected to SeaWiFS/MODIS/MERIS, the x-axis should be MODIS/VIIRS and the y-axis should be SeaWiFS/MODIS/MERIS? Also, the equations should be changed accordingly.
6. Figure 3. Please change the colormap of the figures (c)-(f) here. It is not clear whether the white color in the figures represents the slope very close to 0 but significant (p<0.05) or the slope not significant (p>0.05).
7. Figure 4. Consider changing the limits of y-axis for some provinces, such as the NATR, WTRA. Maybe it is worth adding the trend line in these time-series plots if there is a significant trend?

As general comments, I would say that this study is well written, tackles an important subject and would definitely be of interest to a large audience. I have only minor suggestions that will hopefully help the authors to clarify some details, especially in the methodology section.

Line 48: “Bracher et al., 2022” cannot be found in the final reference list. Either it’s missing from there, or it should be 2020 instead of 2022 in the text.

L89: Why Table 2 and not Table 1, since it’s the first one to be mentioned? Also, the CMEMS product mentioned here (OCEANCOLOUR_GLO_CHL_L4_REP_OBSERVATIONS_009_082) seems to be no longer available in the catalogue. I assume there were some changes and the new product containing these datasets is OCEANCOLOUR_GLO_BGC_L4_MY_009_104. Can you please verify?

L123: “Time series analysis is done both, per-pixel […]” – not sure if the comma is needed here.

Section 3.1 – since you are showing scatterplots of monthly PFTs derived from SeaWiFS/MODIS/MERIS merged and MODIS/VIIRS merged Rrs data (figure 1), I think it might be useful (and more consistent, I would add) to show also such scatterplots between corrected MODIS/VIIRS and Sentinel 3A (as you mention in lines 133-134).

L137: I think it can be useful if you briefly explain what this “match-up criterion” is about. For more details readers can always go to the cited publication, but couple of words would make things clearer. Also, can you specify the time range for these match-ups (e.g. “match-ups corresponding to years 2015 and 2016”)?

L139-140: “Slope is always below one indicating that satellite retrievals show overestimation in low concentrations but underestimation in high concentrations.” - A slope below one does not, by itself, infer information on both under and overestimation. The value of the intercept is required, as well. So, it might be useful if you insert these values into current Table 1 or you show the distribution of points as scatterplots. Also, it is not completely clear if the slopes are computed taking into consideration the in situ derived PFT as a function of satellite PFT and not vice-versa. I assume the first option is true, but is not completely clear.

L143: “[…] are consistent with, the median satellite […]” – probably the comma is not required.

L180: “Per-pixel time series in Fig. 3c shows that significant increase is found only in the west coast of Africa (CNRY)” – shouldn’t it be decrease?

L222: “Dinoflagellates show a stable state in 2021 among the four PFTs with only a very slight increase of Chla in the north Atlantic Ocean.” - To what extent this observed stable state is due to the overall low Chla concentration of dinoflagellates? The PFT anomalies shown in figure 6 are given using mg m-3 as units. Would the same stable state be observed if the anomalies are shown as percentages, for example?

Fig. 3 c-f: did you try a different colour palette? One that would allow to spot even the smaller changes? Maybe you could try a version where the interval is set to [-3; 3] instead of [-5; 5]. Just an idea.