Receiver Operating Characteristic Curve Analysis-Based Evaluation of GCMs Concerning Atmospheric Teleconnections

Round 1

Reviewer 1 Report

Reviewing: Receiver operating characteristic curve analysis-based evaluation of GCMs concerning atmospheric teleconnections

It is a motivating topic, and the authors tried to propose methods to track the changes in the action centres' geographical positions and apply them to the regional studies. Their results are based on observation and reanalysis datasets and have been used to evaluate the CMIP5 simulations.

The authors have provided detailed information and covered the teleconnection between the North Pacific Ocean, North Atlantic Ocean, Mediterranean Sea, North Africa, and Asia. Finally, they offered frequent errors from the model simulations, indicating further investigation on CMIP6 models are needed and a pronounced tool for model development.

Here I will suggest accepting this paper with a bit of grammar adjustment.

My final comments are if the package is available online or if the authors could provide their code to the development team of the ESMValTools? If so, that will benefit most of the model users.

 

 

 

 

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

This manuscript presents a statistical method to evaluate the performance of GCMs concerning teleconnections patterns with respect to the ERA-20C reanalysis dataset taken as reference. The authors use, in this case, the negative extrema method and the receiver operating characteristic curve analysis. They quantify the performance of GCMs with the Matthew Correlation Coefficient and the geographical distances between the action centres detected in the reanalysis and GCM clusters. They also show an example of the application of the proposed methodology to construct mobile teleconnections indices. This paper continues the study presented in Kristóf et al., Atmosphere, 2020.

I found this paper interesting, and the approach has potential benefits in teleconnection studies. Overall, the manuscript is well written, even if I am not a native English speaker and I am not qualified to provide a reliable judgment about it. The methodology is fully described and well explained (with details that allow reproducibility). The Supplementary Material contains almost all the information to analyse the results deeply.

Nevertheless, in my opinion, several revisions are required. In particular, some pieces of information are missing in the Data and Methods section, and the description of the results could be slightly improved. Details are reported in the following.

- abstract: the authors should specify that the study is limited to NH and 500 hPa to provide the correct context in which the performance of the GCMs was tested

-line 107: authors should specify why they chose the December-February period

-Figure 1: why was a threshold of -0.3 used for correlation? Moreover, in Figure 1 in the second CP map (7.5%), I can identify only 3 clusters instead of 4

-line 235: in the footnote, what is the meaning of "local" MCCmax?

-lines 361-364: the fact that GFDL-CM3 is not able to correctly reproduce reference MED cluster cannot be deduced from Table 3 but only from Table S7. In Table 3, the value 3 is associated also with other GCMs

-Figure 7 is not very clear because of the large number of cases reported. In particular, the increase of MCC and the decrease of the average distance with time are hard to see. I suggest trying to connect the numbers of the same model in panel (b) with continuous lines, but I let the author decide if this change will improve the clearness of this plot

-section 3.3.4: in this section, the authors show an example of the possible advantages of using the detected clusters to construct mobile teleconnections indices. They chose, as an example, the MED cluster, considered linked to MO. They tested the method by selecting the analysed GCMs, shown in Figure 8, based on the average distance and MCC. Nevertheless, only the dependence on the average distance is discussed in the text (lines 511-517), referring to the figure. In fact, the relationship with MCC values is discussed at lines 518-525 but referring to models not shown neither in the manuscript nor in the Supplementary Material. Moreover, in my opinion, in this section, or in the Discussion, a short description of why the authors chose to analyse the MED cluster and why they examined only the correlation with the t2m variable should be included

-Figure 8: I do not understand why the panels are shown in this order, which reflects neither the values of the average distance nor the alphabetic order of the models.  I suggest also adding the MCC value for each panel together with the average distance

Author Response

Please see the attachment.

Author Response File: Author Response.pdf