Summary

The authors have created an extensive database of MHW results based on the ESA SST CCI v3.0 product. They have applied a numeric algorithm, similar to the Hobday et al. 2016, 2018 definition, to detect MHWs based on a range of different factors. Specifically they have detected events based on different climatology periods, raw vs de-trended data, different minimum lengths, and different percentile thresholds. All of these outputs are saved in a series of NetCDF files and available via HTTPS.

While the database appears to be technically sound, I am not convinced that it will see much use in its current form. I provide examples for why this may be in the specific comments below. Even if the authors were to address the comments I’ve made, or point out why I may be mistaken, I don’t think a database of these sorts of results would fundamentally see much use. I say this because the Hobday definition has existed for almost a decade now, and databases of global results using this analysis are not new. Many have been published or are otherwise publicly available. To be clear, this is not a reason that this current work should not be published, rather it would have been something to consider before all of this work was performed. Now that the files exist online and are publicly accessible it would be best to publish this work here in ESSD. Though do consider the points I’ve made below as they may improve the likelihood of extended use and valorisation of the work performed by the authors.

Title

- If the authors want to spell marine heat waves as three words, ‘waves’ should be capitalised in the title.

Abstract

- If there are plans to update the database on an annual basis (or more regularly), it would be good to mention that somewhere in the abstract. The other plans for future development, like adding other satellite products, should also be mentioned.

1. Introduction

- 51 : km → km^2

- 64 : ‘MHW events’ Here and throughout the text. A MHW is an event, so writing ‘MHW event’ is tautology. This is a common grammatical error in the MHW literature.

- 91 : v.30 → v3.0

- 94 : ‘​dataset​’ Here and throughout. I think the size of this output, and its standardised structure as a series of NetCDF files, could warrant the title of ‘database’ instead of a dataset. As the authors prefer.

2. SST data

- 113 : extra space after ‘x’

3. MHW definitions

- 161-168 : I’m not certain that would result in a trend-free time series. I could see how de-trending based on each DOY could still allow for an overall long-term trend in the full time series. Did the authors test the difference in output for their de-trending method vs a simple linear model fit to the full time series at a given pixel?

- 175 : Some sort of formatting issue with header sizes

- 179-195 : The choice of the moving smooth window is non-negligible, as the authors noted in the text. Did the authors perform any sensitivity studies for how the choice of the moving window for the smooth chosen here may create different basic results (i.e. 90^th percentile 5-day limit) from the standard python script created for Hobday et al. 2016?

Note that the basic Hobday methodology performs two smooths, not one, as noted in this manuscript. The first smooth is an 11 day window (5 day double sided running mean centred on the DOY) that takes into account all DOY values across all years and is used to calculate the DOY percentiles. This is similar to what the authors have done in their methodology with the 21 day window. Meaning that the base Hobday definition uses a smaller window than the authors have.

However, a second smooth of a 31 day total window is applied to the outputs of the first smooth. This is done so that any outsized effects from one or a few particularly extreme years does not overly bias the percentile thresholds applied to all other years.

Therefore the methodology employed by the authors is arguably a large departure from the initial methodology on which the analysis in this manuscript departed from. It would be preferable to provide at least a basic comparison of how this differs.

I suspect that this is just one of the unintended differences caused to the results by diverging from the base Hobday methodology as there are more specific steps in the process than most researchers realise. These sorts of issues come up often when one incorporates CDO into a methodology instead of performing everything in a more malleable language like R or Python because CDO is more limited in its operations.

Ultimately, this is just a technical issue. The authors have clearly explained what they have done and why, so it is up to readers/users to decide if they want to use the outputs. As the authors noted in the intro, there is no agreed upon methodology to be used. Nor should there be one single numerical approach.

- 218 -221: So does the authors methodology insert an NA value on DOY 60 for non-leap years? If so, how does the algorithm handle this missing day? For reference, the Hobday algorithm creates DOY 60 from DOY 59 and 61 via a linear interpolation BEFORE then taking all DOY values when calculating the seasonal and 90^th percentile signals. This may be another source of varying outputs.

- 248 : So the authors have not provided metrics like mean, maximum, or cumulative intensity? These would be very relevant for a number of applications. That a MHW lasted 10 days with an average anomaly of 0.1°C is very different from an event with an average anomaly of 1.0°C.

4. Results

- No comments.

5. Data

- Why add the 10^th and 50^th percentiles to every 90^th, 95^th, and 99^th percentile climatology file? Wouldn’t it be much more space efficient on the server to host these as separate files?

- 363 : ‘difference above​ ​threshold’ Do the authors mean to say the difference above the seasonal signal?

- As the files are organised now, if a researcher wanted to access all of the years of data for their particular study site (even if it is very small), they would need to download all of the files on the server for their input values of choice (i.e. 95^th percentile, 10 day limit). Then open each one and extract just the portion they want. Combine these all into another file and save it for continued use. This task is generally outside of the competence of most non-physical oceanography marine researchers. Have the authors considered rectangling their data to also provide the same output as a series of files organised along latitude or longitude lines? For example, providing all of the DOY percentile values for all days for all pixels in the latitude range of 0-5°E. The exact method isn’t too important. The point is that, one can create an impressive database of results (as the authors have) but if it is too complex or inconvenient to use, it will not be used. Have the authors consulted with a broad range of stakeholders to see who amongst them would be able to make use of the database as it stands now?

- Additionally, the lack of standard MHW metrics (e.g. duration, cumulative intensity) provided for discrete events will likely prove problematic for most researchers. If I have understood correctly, the author provide the per day results per pixel, but no summary metrics. It is the summary metrics (e.g. total duration) that tend to be of most interest to a researcher. Therefore, any use of this database will inherently require a lengthy post-processing analysis by the user in order to arrive at the final metrics that they will probably need for their work. Have the authors considered this? If so, why not provide files that give these summary metrics? This would not be too terribly difficult to create, and would require quite a lot less work from a user. For example, create a file that contains all of the summary results for all pixels for a given longitude band. This would contain the default Hobday-ish output that lists each discrete event, its duration, max intensity, etc. etc. Nothing new or innovative, just the standard stuff that people are used to.

- Ultimately, who is the target user for this database? Whose life would be made easier by using these data as they exist now? In what circumstance would a researcher with a limited study area not simply calculate the results for exactly what they need? With an output that matches exactly what they need? Because these data are stocked in a NetCDF file format (a choice that I completely agree with, just to be clear) it will generally require that a user (or someone in their lab) be able to access, open, and extract data from them using a programming language. Considering that there are already MHW toolboxes for Python, R, and MATLAB, why would they not just use one of these to perform the tests that they need? While the database that the authors have created is impressive, and would be very useful, I do not immediately see that it will receive much use as is. While true that this database would inherently be more useful to a research question with a global scope, most researchers tackling investigations at this scale will already have the necessary competence, and will most likely already have a particular dataset or other peculiarities that they will need to pursue that would prevent them from exploiting this database.

6. Usage

- 403 : The point about climate model evaluation is a tenuous one. Unless the model is in some way related to the ESA CCI SST product, the modelling of temperature extremes tend to be so incongruous between different models that comparing them directly is rarely a fruitful endeavour.

- 411 : The original python code, written in 2016, already allows a user to make all of the changes detailed in this manuscript. Including the changing of the DOY moving window. Minus of course the de-trending of the raw data.

7. Summary

- 415-420: A problem with this statement is that the authors themselves have deviated from the base Hobday algorithm, meaning that the results in the database may already differ from an analysis that uses the exact same climatological baseline, duration limit, etc. Let alone comparisons for results with different baselines.

- 420 : HMW → MHW

8. Conclusions

- No comments.

Table 1-2

No comments.

Figure 1-7

No comments.

General comments:

This manuscript introduces a dataset of marine heatwaves identified at each location globally using different baselines and thresholds. The authors have applied these various definitions to the ESA SST CCI climate data record and saved NETCDF files of daily MHW climatologies, anomalies, and categories. They compare the MHWs detected on January 1, 2014 under different definitions to understand how changes in baseline and threshold affect MHW detection.

I agree with many comments made by Anonymous Referee #1, so I attempt to not restate those comments here and below. While the data product is novel and could make certain analyses more efficient by pre-calculating MHWs with different definitions, I agree with Anonymous Referee #1 that it is unlikely to see much use. The dataset is complete, but I am not convinced it is unique or useful. The ESSD review criteria that “it should not be possible to replicate the experiment or observation on a routine basis” may be true for the extent of the analysis, but there are several software packages available that make MHW identification a relatively straightforward and customizable process. For usefulness, the goal of presenting multiple different definitions of MHWs is somewhat at odds with the goal of providing a constant foundation to facilitate common analysis. A standard data product should present some clear benefit over other satellite and in situ SST data, like OISST or OSTIA, which many MHW studies use as the observational reference.

The presentation quality of the manuscript is good and the writing is clear. The manuscript could be improved by providing more summary statistics for how the different definitions perform over the whole time period, demonstrating how changes in definition generally affect MHW identification. Further explanation of potential use cases would also improve the manuscript. A demonstrated use case for this specific data product, rather than MHW statistics generally, would warrant the manuscript to be published. Alternatively, this paper may be better presented as a scientific analysis of the effects of different MHW definitions as opposed to an article about this data set.

Specific comments:

L184: Why not also compute the 1^st and 5^th percentiles to facilitate marine cold spell analyses? In section 5, it states that the daily climatology files record the 10^th, 50^th, and 90^th/95^th/99^th percentiles, so why not also include the same levels for MCSs?

L187: I agree with Anonymous Referee #1 that this change from the Hobday definition and methodology should be carefully explained and justified.

L216: Using the median climatology instead of the mean climatology is another slight departure from the Hobday definition.

Section 3.5: Mixing multiple thresholds and severity categories may obfuscate the severity of MHWs by requiring both to be reported. For example, a 95^th percentile category 2 MHW may also be a 90^th percentile category 4 MHW and a 99^th percentile category 1 MHW, as shown in Fig. 6 and Fig. 7. Multiple thresholds could be used as another way to categorize MHW severity instead of using the threshold differences defined here and in Hobday et al. (2018).

Figs. 6, 7: Consider just showing one of these, as Fig. 7 is just an inset of Fig. 6.

Section 4: I appreciate choosing a large and well-studied MHW here, but this specific MHW certainly does not show “typical” summary statistics, as opposed to averaging those statistics over a longer period or the whole record. For example, there is almost no difference between Fig. 7g and Fig. 7h. The Blob is clearly defined across MHW definitions, as you state, but how does that generalize to less severe MHWs?

Section 6: I agree with reviewer #1 that it is unclear who the target of this data product is. The two applications listed here are ecosystem impact studies and model evaluation efforts. Both applications may require specifications and data products not included here. The manuscript states that one could compare models and observations “by applying the same algorithm to SST output from models,” but in that case, one would likely want to apply the algorithm themselves. To that end, there already exist several packages in python, Matlab, and R to identify MHWs and calculate statistics based on user-defined parameters. The manuscript also states that “our dataset offers a common reference” for model-observation comparisons, but a central point of this manuscript is to include many different observational MHW catalogs, so which would serve as a common reference for comparison?

Technical corrections:

L250: Change “Fig. 4” to “Fig. 5” and “section 2.5” to “section 3.5.”

L263: Change “Fig. 4” to “Fig. 5.”

I spotted these two, but there may be more instances where figures and section references have changed since they were originally written.