Satellite Observation of the Long-Term Dynamics of Particulate Organic Carbon in the East China Sea Based on a Hybrid Algorithm

Round 1

Reviewer 1 Report

1 In fig.1 panel (b), two water types can be seen from the remote sensing reflectance. I guess this is due to water samples from difference seasons. Because only panel b contains two water types, is it possible to use two different line types to discriminate them? 

2 In the legend of fig. 2, please list the data source (cruises) for these two panels.

3 The font of equations 4-6 is different from the main text and other equations, please check it. 

4 (1) For figure 3, I assume the panel a and c present the model establishment using training dataset, and the panel b and d present the performance using testing dataset. If so, please display this information clearly in the figure legend. (2) For y axis name of panel b and d, they are "satellite-derived POC" based on two distinct algorithms, but they have been marked with same y axis name and this could mislead readers, please specify them with accurate information. 

5 There is a nice analysis on POC's annual variations for the last two decades. However, the cited figure 4 and 5 cannot support the statements, simply because on figure 4, there is no significant annual increase over the 20 years and figure 5 is a climatological figure which cannot reflect annual variations. Please add the annual discharge data from figure 6 onto figure 4 (maybe using a different color) 

6 For equation 9, this is a different algorithm with equation 7 as the two offsets are different. If so, the chaoshan dataset is less relevant to the whole study. A remedy could be conducted to test the performance of equation 7 with Chaoshan data and see the uncertainties. It is fine and acceptable if a less good result would be get, because type II water fluctuate all the time and the current study has a very limited in situ data.

However, if the purpose of the paper is to propose a global modified CI algorithm as log10POC = Ax CI +B (Here A and B are two local offsets). Please combine all data as a whole and randomly divide them into training and testing datasets then run the model testing from the very beginning. 

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

The presented article has undoubted relevance and significance, since more and more attention is being paid to the global carbon cycle in the world, and satellite methods allow conducting research in a wide range of spatial and temporal scales. A separate value of the work is that an extensive array of POC natural measurements is used to determine the regional calculation algorithm from Rrs data.

The work deserves to be published in the Remote Sensing journal after some improvements:

1. A general remark is that it is necessary to take more account of the specifics of the Remote sensing journal and add more interpretation of the results from the point of view of remote sensing features, and not only analyze the POC distribution itself (in section 3). First of all, it is necessary to take into account what additional factors from the point of view of ocean optics affect the measurements of Rrs spectra and, consequently, the estimates of POC and chlor_a (CDOM, mineral particles, etc), and how these factors with changes in river flow (or other natural factors) could affect the estimates of POC and chlor_a.

2. In the description of Fig. 1, clarify what the different colors of the Rrs spectra mean.

3. Lines 142-143. The appearance of the peak is influenced not only by the “enormous amount of sediment”, but also by the presence of a high content of CDOM, which strongly absorb exponentially in the spectral region of < 500-550 nm.

4. Line 188. This is a standard algorithm for estimating the concentration of chl-a, which cannot take into account variability of the contribution of CDOM to Rrs. It should be noted that in such optically complex waters as the East China Sea, variations of chl-a/CDOM can significantly affect systematic errors in determining the concentration of chl-a from satellite data.

5. Lines 211-213. It is necessary to clarify how the samples were divided into 70% and 30% of points. Was it done randomly or according to some rule: the first 70% - the next 30%, the first 30% - the next 70%, or something else?

6. In Fig. 3a it can be seen that some of the points are separated to the right from the regression line in the range 0.000 – 0.001 by CI_POC and 1.8-2.4 by log10(POC). It is necessary to specify if there is any unifying feature for these points: a certain area or a certain season, or no feature is determined?

7. Figure 3d shows that, on average, satellite estimates gave overestimated values. So, bias calculation must be added to R2, RMSE and MAPE in order to fully understand the results obtained.

8. Lines 245-247. This statement is not clearly proved, since the graphs and the text present the results for two different algorithms, and not for a hybrid model. To show this, it is necessary to present a table in which the calculated statistical metrics for individual models and for the hybrid approach will be compared.

9. The presented work will be of particular value to the ocean optics community if the proposed algorithms are compared with the standard NASA algorithm for determining POC https://oceancolor.gsfc.nasa.gov/atbd/poc /, or with new developments in this matter (Stramski et al., 2022, https://www.sciencedirect.com/science/article/pii/S003442572100496X ) – it is enough to compare metrics in tabular form without graphs. However, this comparative analysis can be quite time-consuming and deserves to be the topic of a separate article, so the 9th remark is only recommendation and the necessity to resolve it remains at the authors decision.

10. Line 399. Here and in some other parts of the article, CIPOC is written without dropping "POC" into the subscript.

11. Line 20 and 445. Here and in the whole article in some other places it is said that chlorophyll-a affects POC. But strictly speaking, chlorophyll-a is a molecule that by itself cannot be a particulate particle, and it is part of phytoplankton cells. Therefore, when we talk about drivers or processes affecting POC, it is better to write about the content (concentration) of chlorophyll-a, or about phytoplankton directly, depending on the context. But not just - chlorophyll-a.

12. Fig. 6 and Fig. 7. Once again emphasize in the description that these are satellite estimates of POC and chlor_a. Moreover, in the case of POC – with regional algorithm, and in the case of chlor_a – a standard satellite bio-optical algorithm.

13. The comparison of the POC distribution with sea currents is too general and descriptive, based on the multiple use of ready-made results of other works, without any numerical comparisons. Here it is necessary to highlight its novelty more clearly, or to deepen the analysis.

14. The purpose of the work is not clearly indicated.

The abstract says the following: “The results show that the performance exhibited by this hybrid retrieval algorithm confirms the ability to monitor the variability of POC in the shelf sea surface using MODIS/Aqua remote sensing data, and those different drivers such as river discharge, chlorophyll, and sea surface current field jointly influence the spatio-temporal distribution of POC concentration in the East China Sea.”

But the Introduction has only the following purpose: “The purpose of this study is to investigate the applicability and transferability of the CIPOC algorithm for the retrieval of surface POC in coastal water with complex optical properties.” Then there is a description of the results, but not in terms of setting the purpose of the work.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf