Ocean Color Quality Control Masks Contain the High Phytoplankton Fraction of Coastal Ocean Observations

Round 1

Reviewer 1 Report

This problem has been persistently challenging since the first ocean color sensors (CZCS) and continues to difficult research in the field.

The document is well organized, with the chapters of introduction, materials and methods detailed with care. The scientific literature included in these sections are appropriately selected and referred.

The results and statistical analyses are correct, they clearly support the discussion and conclusions. Furthermore, the figures and additional references properly highlight the findings.

The examined regions: California and Benguela Current Systems, are suitable representatives of productive ecosystems, thus the outcomes can be easily applied to other biologically reach coastal areas.

Finally, the recommendations and suggestions provided should be welcomed by a large community concerned by similar remotely-sensed complications.

Author Response

We thank the reviewer for the positive comments.

Reviewer 2 Report

The objective of masking setting of ocean color data products is to ensure the high quality of ocean color data products. For the evaluation of masked regions, it should be well organized according to the factors which induce the masking. After several reading of the manuscript, it gives me the following impression, the manuscript should be well reorganized, I cannot get the focus. So it's recommended that the authors could re-organize the manuscript and make a resubmittion.

Author Response

We agree that masks are intended to ensure the high quality of ocean color data products, but we show in our analysis that masking also produces an unintended consequence of removing biologically important observations. The focus of the manuscript is to test whether the loss (due to masking) of the high biomass data fraction imparts bias into ocean color datasets. The focus of this study is explicitly stated in the abstract (lines 18-23), in the introduction (lines 121-125), and repeated in the conclusion (lines 507-509). Our results suggest that the biomass distribution of ocean color datasets may be modified by the removal of masked pixels, altering the satellite perspective of coastal ecosystems. Key findings of this study are described in the discussion section and listed again in the conclusion (lines 509-518). The recommendation that the manuscript be re-organized is not supported by other reviews, and we politely maintain that the manuscript is clear and appropriately organized. If the reviewer provides more specific feedback, then we could better address their recommendations.

Reviewer 3 Report

The manuscript titled “Ocean color quality control masks contain the high phytoplankton fraction of coastal ocean observations” by Houskeeper and Kudela investigates the detrimental effects of data quality control flags in productive coastal ecosystems. The standard quality control flags are applied to NASA’s Level-2 (L2) and Level-3 (L3) satellite imagery to ensure higher quality products for the end-user applications; however, authors suggest the possible downside of such flags, especially in coastal waters, as their application may remove several important pixels with high chlorophyll content which eventually causes errors in the phytoplankton biomass estimation and overall biological viewpoint in productive coastal ecosystems. Using a red band difference (RBD) as a proxy for phytoplankton biomass on L2 images and L3 composites in two eastern boundary current systems, the effect of standard masking was evaluated in providing an unbiased perspective of phytoplankton biomass. The results suggest the satellite retrieval in productive waters were frequently assigned the standard flags on L2 and L3 products and such pixel-masking (i.e., the removal of high biomass pixels) eventually lowered the overall RBD in the study area. This study suggests that screening such important pixels may possibly undermine the importance of biological processes within the region or the importance of a region itself in meso-scale processes. Authors have also suggested the necessary remedy to detect and avoid such issue using methods described in the manuscript to the end-users. Overall, the manuscript is well-written and scientifically sound. Knowing the importance of this work and its usefulness to a large scientific community, I would recommend this manuscript for the publication in Remote Sensing.

 

Minor comments:

Line 231: “at” TOA

Section 3.1: This section describes the efficacy of the RBD algorithm in MB, but the results for SHB are not mentioned anywhere. Please add discussion on the performance of the three algorithms in the context of SHB.

Figure 4: RBD is defined as a difference of surface reflectance between 678 nm and 667 nm. Surface reflectance is unit less while RBD in a figure is in radiance. Please clarify this.

Line 342: “L3 composites (4km, 1day)” – I didn’t understand “1 day”. Figure 5 shows the monthly averaged L3 products, right?

Figure 6: “σ” for Standardized Bias has been already used for standard deviation in equation 3.

Author Response

We thank the reviewer for the positive comments.

Line 231: “at” TOA

Thank you, this has been corrected.

Section 3.1: This section describes the efficacy of the RBD algorithm in MB, but the results for SHB are not mentioned anywhere. Please add discussion on the performance of the three algorithms in the context of SHB. 

Thank you for identifying this omission. We agree, and have added the following text to section 3.1 to improve the description of the matchup results:

The highest Pearson coefficient for each product was derived from SHB matchups, with RBD showing the strongest association with in situ Chla among the evaluated remote products. The SHB matchups were unique from the two MB sites in that the in situ measurements were obtained by ship at various distances from shore, allowing the SHB matchup dataset to include a wider diversity of water types than either the wharf or mooring datasets in MB.

Figure 4: RBD is defined as a difference of surface reflectance between 678 nm and 667 nm. Surface reflectance is unit less while RBD in a figure is in radiance. Please clarify this. 

Thank you for identifying this mistake. We agree that RBD is unit less, and the figures have been corrected.

Line 342: “L3 composites (4km, 1day)” – I didn’t understand “1 day”. Figure 5 shows the monthly averaged L3 products, right?

Figure 5 shows the standardized bias derived from the full satellite record of daily, spatial-composite images, according to equation 3. The software used to generate L3 composites allows a variety of default temporal windows (1 day, 8 day, monthly, etc.) as well as spatial grids (4km, 9km, etc.). We chose to composite the L2 images onto 4km standard grids, but to maintain native temporal resolution because the standardized bias equation derives the mean of the RBD values within each standard (4km) grid, and thus using temporal composites would result in a "mean of means" comparison. To help clarify this, we have revised "L3 composites" to "L3 spatial composites" on line 348.

Figure 6: “σ” for Standardized Bias has been already used for standard deviation in equation 3.

Thank you, we have removed “σ” from the figure.

Round 2

Reviewer 2 Report

This revised manuscript can be published in the present form!