General comments:

The paper is well structured and readable, and addresses a region that is not commonly discussed in atmospheric blocking papers. While I have some clarifying questions about the methodology below, I think that the overall work is quite robust and a useful contribution to the literature.

Specific comments:

Line 93: How does ERA5 improve representation of the Southern Hemisphere, relative to ERA-Interim, in a manner that is relevant to the topic at hand? Better representation of wind, temperature, moisture transport?

Line 127: If you had hourly data, why not use the full hourly resolution rather than 6hr resolution? Was it resource constraints or something else, like trying to maintain consistency with past papers?

Line 128: What is the vertical resolution? 50hPA, something else?

Sections 2.3.2 and 2.3.3: Initially I struggled to understand the difference between the two methods just based on the text/Table 1. It took looking at the Figure 1 schematic for me to finally understand it . They both sounds like they’re anomaly-based from the descriptions, but one is called “Absolute threshold” and the other is called “standard deviation-based anomaly”. Would suggest revising the text of these two sections to more clearly outline the calculation steps, as Figure 1 does a very nice job of explaining it visually.

Section 2.3.4: The data has 6hr resolution, but the thresholds are established on a daily basis? (my assumption based on line 246) Does that mean that the full day has a single threshold rather than one per time step? If so, why that single daily threshold and not per time step?

Figure 4: Would like to see the polygons in 4b) shown in all items, not just b), so that it is clear which areas are being summarized in Table 2.

Line 429 and Section 4.1/4.2 in general: Would we expect to see much blocking over Antarctica at all for the Z500 gradient method given the constraints of the latitudinal range? Trying to compare the frequencies of the Z500 gradient method and the other methods in the higher latitudes is rather misleading given the different latitudinal extents. Discussion should account for this fact.

Figure 5: as with Figure 4, would like to see the polygons shown so that it’s clear which areas are being summarized in the table

Table 2: Perhaps would be more appropriate as a figure? It’s very hard to easily digest this information in its current format.  Suggest something like grouped boxplots.

Figure 9: Would it be possible to provide a background color of a relevant field, such as 500hPA geopotential height, to provide some context to what exactly is being captured by the various methods? It would be interesting to see how “accurate” or “inaccurate” the various methods are or what they are specifically capturing (i.e., a ridge vs a rex block vs an omega block). Perhaps not feasible if it obscures the contours, but maybe worth trying?

Section 4.5: I found this discussion to be rather unsatisfying with respect to the events being shown in Figure 9. It discusses the results in broad strokes rather than referring to what happened during the specific selected events. For example, in 9d, what is going on with those contours off the eastern coast of South America? Did they merge with their respective counterparts in 9e, or was that the detection of a local anomaly rather than a persistent blocking feature?

In the article "A Multi-Method Antarctic Atmospheric Blocking Dataset (1979-2024)", submitted to Earth System Science Data, the authors apply a selection of seven of blocking detection methods developed in the midlatitude meteorology literature, and apply them to the Southern hemisphere excluding the tropics but including Antarctica. The authors then compare the climatological properties of blocking according to each metric (frequency and persistence) in great detail and also demonstrate the differences in instantaneous detection using a case study. Importantly, the dataset is made public and is described in enough details as to let researchers use it for their own work without onboarding pain, which is commendable.

The authors focus their discussion on the polar region where there is still a methodological gap in blocking detection. Their solution to help fill this gap is to gather many different methods for midlatitude blocking detection and to apply them all to this polar region without modification, to the thresholds most crucially, and then compare the results. While this provides a good starting point and the full dataset is a useful tool, I believe a critical assessment of the threshold values in this very different region could have made this work a lot more significant. Compounding with this issue, the authors do not provide the analysis code that created this dataset, such that the users cannot easily reproduce the dataset with the same methodology but with more carefully chosen thresholds in the region of interest. I will let the editor decide whether more work on the thresholds in the polar region should be done before resubmission or if this extra work is beyond the scope of this journal and the dataset as-is is sufficient to justify the publication.

There are other, smaller methodological points I would like to raise.

Is a fixed area threshold of 1million km2 the best choice? This is bound to create biases in number of PV and Z500 events because the fields inherently have different spatial coherence characteristics.

To go back to the threshold problem, the paper the authors cite the most and seem to take most of their methods from is Pinheiro et al. 2019. In that article, the authors justify their choices of thresholds (maximum between 1.5 sigma and 100m, e.g.) by quantifying the standard deviation of the fields in the regions they are interested in. No such justification is provided for the polar region in this present work, and I think it should be there.

Related to the previous point: Between 1.5 std and 100m or 1.1 PV, which threshold is triggered more often and where ?

Was is the most judicious choice to apply the reversal index as is? It seems mostly unable to detect reversals over Antarctica, but with all thresholds kept identical to those used for the midlatitudes how might the reader interpret these results? Would a modified reversal index be more sensitive in this region? Or are there actually no reversals there?

Tracking is beased on grid point overlap, but grid points represent diminishing areas with increased absolute latitude. Is persistence at the pole therefore comparable to persistence at 60°S?

Why even use the gradient reversal method if by definition (without changing the latitudinal bands over which the gradient is calculated which could have been an option) it cannot detect blocks over Antarctica?

Smaller points:

Introduction:

Line 1: Maybe half a sentence to define what is a block? General enough that it agrees with any detection method.

In intro, when citing papers that discuss blocking in Antarctica, maybe briefly mention how they diagnose it; or dedicate another paragraph to this

maybe worth mentioning in the intro; PV framework also allows natural decomposition of conservative and non-conservative (cross-isentropic PV fluxes) processes affecting evolution of blocks

Line 86-88: "[...] its use for multi-method Antarctic blocking has not previously been integrated within a single dataset". Please rephrase

Paragraph starting line 91 should have one sentence stating what you are doing, for those who have not read Pinheiro et al. 2019. I propose something like

"Earlier work by Pinheiro et al. (2019) provided a dataset of blocking characteristics derived from the ERA-Interim reanalysis ...."

Methods

Do the authors use the full spatial resolution of ERA5 (0.25°) or do they regrid to a coarser grid?

(typesetting): cleaner maths typesetting; units and integral's d should be upright, indices as subscripts

Line 159: 3-day smoothing of anomalies: is this standard in blocking detection? How much does this choice affect the persistence requirements?

Line 173: "All methods share a common framework for spatial coherence, ..." please rephrase to a more precise statement. After each method, the the same area threshold is used to filter out too small events.

This might be me but I struggle to understant the difference between method 2 and 3. I see later in the results section that they do not detect the same amount of blocking so there must be a difference, but the description in the methods section makes me think they do the same thing.

I will summarise what I understood: At each grid point and time step, we can define a field x, its climatological value smoothed with five harmonics, x_clim, and the anomaly compared to this climatology x' = x-xclim, itself smoothed using a 3-day window. A smoothed daily standard deviation, sigma, is also computed and smoothed like the climatology. Finally a fixed threshold of 100m or 1.1 PVU is also defined, let's call it x. Method 2 defines a blocked grid point as respecting x > x_clim + max(1.5 sigma, x), while method 3 defines it as respecting x' > max(1.5 sigma, x), i.e. x-x_clim > max(1.5 sigma, x) which is the same inequality as the one of method 2. The only differences I can find are slight changes in smoothing (number of harmonics in lon, lat, time) for the climatology and sigma, and the fact that x in method 2 is not smoothed in time while x' is in method 3. What am I missing? Being able to see the code might make this a lot clearer.

Table 1:

row "threshold" type: use "or" instead of the confusing character "/" in "100 m / 1.1 PVU"

Threshold type for gradient method should read "fixed thresholds of north and south gradients" or something, currently it is the same text as the following row. suggestion: Use one large row spanning multiple columns to emphasize when all columns share the same entry especially for last two rows

Dataset description:

I appreciate the user friendliness of this section. The right amount of details for the users to start using the dataset.

Good also to provide the summary statistics (especially mean length)

3.2: if i understand correctly, in the case of splitting, the child with the most overlap with the parent retains the ID? and similarly for merging, the child retains the ID of the parent with the most overlap? how about the other one(s)? is there a dataset of all such merging and splitting? it would be very valuable for more advanced feature-based analysis.

Results

Figure 4: "Regions shown in white correspond to latitudes where blocking detection is not applied": not quite true, it also corresponds to regions with very low values of blocking frequency, which makes it hard to see at a glance where the reversal index stops applying and is in my opinion very important to visualise. Hatching maybe? or a light gray shading? or no white in the colormap of blocking frequency?

Line 416; "The PV-based methods exhibit weaker spatial contrasts than the Z500-based method": Isn't that purely a matter of choice of thresholds in intensity and area? The authors could compare with the midlatitude literature: what is the expected difference in blocking detection frequency between PV and z500 methods, with these thresholds?

Line 434: "In June-July-August (JJA), the spatial distribution of blocking over high latitudes changes markedly. The Z500-Grad method shows localized enhancements in blocking over the Amundsen and Bellingshausen Sea sectors, where the mid-latitude jet weakens and becomes less zonally continuous over the South Pacific" this last statement deserves a citation

Line 483. In this paragraph, are the authors referring to the results of the previous subsections or to what one can see on figure 6? If the latter, it is important that the figure be referenced early on.

Line 497: "The persistence of some winter blocking signals at high latitudes..." I do not understand this sentence. At this point the authors have not presented any persistence results, and even when they do, they do not separate by season. Is this section out of order?

Line 505: I appreciate this, this is a good idea.

FIgure 6: the computation of sigma_metric and sigma_variable should be in the main text and more detailed. The description in the figure caption is confusingly written.

Line 557: "consistent with the smoother structure of vertically integrated PV fields.". Is that so? This smoothness should be quantified somehow, it is not obvious it should be the case. PV on single levels is very spatially variable, and it is not obvious to me that it should be completely smoothed out by vertical integration.

Code availability:

As a reviewer, I would have liked to be able to inspect the code to make sure the methods applied really correspond to their descriptions, mostly to try and catch potential errors that we all make. For complex methods such as these, the code might provide another angle to the explanation that can clarify some points like the order of operations (smoothing, climatology computing, time-averaging...). I think just pointing out one of the tools used is not enough. Since the authors made a lot of choices not everyone might agree with with respect to threshold values, it would also allow the users re-create a similar dataset with their own choices of thresholds, but otherwise applying the same methodology.