Review of ‘Early twentieth century Southern Hemisphere cooling’ by Bronnimann et al.

 Overview: This paper focuses on improved evidence of the 1908-1911 cooling in the Southern Hemisphere by examining the effect of newly digitized ship logbooks on the 20^th century reanalyses and other models.  Overall, it finds that the inclusion of the new data has an important effect on the solution, especially for pressure, but nonetheless confirms that the cooling still exists in the early 20^th century. 

The paper is well-written and the figures are sound.  I enjoyed reading the manuscript and feel the paper makes a valuable contribution to the field.  I suggest a few possible new points of data – perhaps not necessary to include in the offline assimilation unless it is easy to do - and clarify a few comments in the manuscript below.  I am suggesting a minor revision of this work.

My regards to the authors,

Ryan Fogt

Comments:

L39-46: The wording here is slightly misleading.  The Fogt and Connollly (2021) dataset is a blend of the Fogt et al. (2019) spatial pressure reconstruction poleward of 60°S and the 20CRv3 equatorward of 60°S.  Thus, everywhere north of 60°S the Fogt and Connolly merged data are exactly the 20CRv3.  Poleward of 60°S, their data does agree best with 20CRv3 (it at least shows the least drift in the early 20^th century).  So perhaps the text could be revised to say, starting at line 42 “Poleward of 60°S, the “Twentieth century reanalysis” for this period fits best with their reconstruction, whereas other products showed spurious trends.  However, there are marked differences between all products prior to 1957 south of 60°S due to the sparseness of pressure data...’.  The Fogt and Connolly (2021) paper found more agreement in the various products over Southern Hemisphere landmasses in the early 20^th century, for what it is worth.

L67: Grateful for you to submit these ship logbook data to ICOADS.  Are there additional pressure data in recent releases of ISPDv4 that could also be used in this study, that aren’t in ICOADS or assimilated into 20CRv3? Perhaps some early Antarctic data from British and Norwegian expeditions, as examples?  Perhaps also the daily data (could maybe be close to the 12Z values) from Orcadas used in Zazulie et al. (2010), which from conversations with Gil Compo aren’t included in the early 20^th century in 20CRv3?

Table 1 – need to clarify what 20CRv3+ is in the caption – I assume it is the 20CRv3 with the additional data assimilated.  If that is correct, you may wish to introduce the acronym when you first mention this dataset on L51-52.

L187-188: Technically, the above mentioned ship with a correlation to 20CRv3 of 0.71 (Ruapehu) does increase to 0.94 in 20CRv3+.  It is the Paparoa ship with r=0.77 that actually decreases to r=0.73 in 20CRv3+.

Fig. 4: I’m having trouble interpreting the reduction of spread figure, perhaps overthinking it.  The darker areas are the least reduction of spread based on the legend, which at face value to me would imply that they have had the least impact on the spread as they didn’t reduce it much.  But yet the text focuses on these as areas of improvement (near Cape Town, and along the ship tracks).  Is the figure perhaps showing the ratio of spread, instead of the reduction in spread (which I take as the original spread (variance or standard deviation) minus the new spread (therefore, larger values are more of an improvement)?  Please clarify in text or in caption.

Fig. 6a: can you comment in the text about what percentage the SH temperature anomaly is due to the tropical Pacific / La Niña -like cooling (Am I correct to assume this average is area-weighted by the cosine of latitude)?  It seems like the zonal mean temperature plot shows much of the cold anomaly is in the tropics.  I’m not sure much can be said of the temperature anomalies south of 60°S – what data constrain these over the Antarctic continent (the spread is also huge there!)?  You can perhaps add a few sentences here after lines 272-273.

Fig 8 – note misspelling of Connolly in top row.

Fig 8 – interesting comparison with the Fogt and Connolly (2021) data here – thanks for doing this!  One important note here that is worth mentioning in the discussion following lines 283-284: The Fogt and Connolly (2021) data do include direct observations from Orcadas station, which starts in 1903.  These data strongly control the solution of the Fogt and Connolly (2021) data near the Antarctic Peninsula, Weddell Sea / South Atlantic, in particular.  Thus the positive pressure anomaly in MAM in Fogt and Connolly dataset near the tip of the Antarctic Peninsula is likely real (based on observations at Orcadas), whereas the strong positive pressure anomaly in 20CRv3+ over the northern Antarctic Peninsula may be overstated.  You could quickly make comparisons of the pressure anomalies at the grid point of Orcadas station using the READER archive to verify these results – there is some information! (there are also data at Grytviken in South Georgia, if you want to further verify – these data are NOT used in Fogt and Connolly (2021).)

Fig 9 – can you plot the observed SOI from the Australian BOM here as well?  http://www.bom.gov.au/climate/enso/soi/

General comments

This paper analyzes a cooling period in the Southern Hemisphere in the early 20^th century using observations, existing reanalyses, and an augmented reanalysis in which additional observations were assimilated offline. I found the paper clear, logical, and well-organized. I just have some relatively minor comments.

Specific comments

1. Line 49: maybe clarify “very little pressure data was ingested into 20CRv3 during these years, *particularly in the Southern Hemisphere*”. In addition, the authors may wish to include a link to maps of the observations assimilated in 20CRv3 by year: https://psl.noaa.gov/data/20CRv3_ISPD_obscounts/
2. Line 51: maybe here is a good place to introduce 20CRv3+, since it’s used in the Table caption prior to introduction in the text.
3. Section 2.2: I recommend adding a few more details about 20CRv3 to this section. In particular, note that 20CRv3 does not use HadISST SSTs in the period of study, but rather SODAsi.3 (Giese et. al 2016, https://doi.org/10.1002/2016JC012079); there are also places later in the paper where it is stated that 20CRv3 uses SSTs “similar” to HadISST1.1 (lines 160 & 313). It might also be worth mentioning that 20CRv3 uses a cycling EnKF (unlike the offline approach used for 20CRv3+). Finally, it may be worth emphasizing that no station temperature data went into 20CRv3 at all; although the text in this paper is accurate, readers often miss this detail (in my experience.)
4. Line 189/Figure 4 caption: Clarify whether this is reduction in time-averaged spread over some time period? If so, is there a reason for not (also) showing reduction in spread for the single date plotted in the lower part of the figure? Also, if this is time-averaged spread, then on line 191, you may want to add “since there were only few ships in the vast space *and time period*, the reduction is statistically weak”
5. Line 225: Suggest changing “observed SSTs” to “prescribed SSTs”, since 20CRv3 actually uses an ocean reanalysis for its SSTs in this period (see comment 3 above).
6. Line 307: clarify what “beneficial” means here, beyond the indication of a robust signal. Maybe, that the comparison also shows where uncertainty remains?

Technical corrections

1. Line 102: “spatial resolution is” should be “spatial resolution of”
2. Line 124: maybe replace ybar with y (or define ybar)
3. Line 135: outliers were removed *when* y-Hx…
4. Line 186-187: Clarify that you’re referring to 20CRv3+ here
5. Figure 4 caption: clarify what “results” are here (perhaps “analysis fields”?)
6. Line 226: “as the” should be “as with the”