A Climatology of Atmospheric Patterns Associated with Red River Valley Blizzards

Round 1

Reviewer 1 Report

MAJOR COMMENTS

 

1.  Section 4:  The paper notes the success of the SOM technique in objectively classifying patterns.  Given the good agreement between the objective and subjective techniques, what benefits does one technique have over the other once the patterns have been recognized?  Could the results from this study be used operationally for blizzard prediction in RRV based on one of the four synoptic types?  Discuss.

 

MINOR COMMENTS/EDITORIAL REVISIONS

 

1.  Section 1, Line 95: Replace “this is considered…” with “Storm Data is considered…” 

 

2.  Section 2, Line 117: Explain here how winter seasons were defined (e.g., DJF) or if all blizzards occurring during the time period were included.

 

3.  Section 2, Lines 118-119:  Were the subjective classifications regarding the event type (i.e., Colorado Low, Alberta Clipper, Hybrid, or Arctic)?  If so, clarify.

 

4.  Section 2.2, Lines 143:  Insert Principal before Component Analysis.

 

5.  Section 2.2, Line 171: Insert “a” before “…goal to classify…”

 

6.  Section 3.1 and Figure 3a:  Any thoughts as to the variability in blizzard frequency?

 

7.  Section 3.2, Line 274: Usually see AVA described as Negative Vorticity Advection (NVA) (and cyclonic motion as Positive Vorticity Advection (PVA)).

 

8.  Section 3.2, Lines 264-279:  Do you find Arctic Fronts often followed after Colorado Low or Alberta Clipper events?  Often synoptic analysis shows a “double cold front” whereby the main low and cold front are then followed by a trailing Arctic front and subsequent strong anticyclone.

 

9.  Section 3.2, Lines 294-296:  Any physical explanation for the lull in late February extratropical cyclones?

 

Author Response

MAJOR COMMENTS

1.  Section 4:  The paper notes the success of the SOM technique in objectively classifying patterns.  Given the good agreement between the objective and subjective techniques, what benefits does one technique have over the other once the patterns have been recognized?  Could the results from this study be used operationally for blizzard prediction in RRV based on one of the four synoptic types? 

This is a great question--  there is no reason why a seasoned forecaster couldn’t do this type of vetting by hand. The true benefit is freeing them of this burden so the forecaster can focus on other duties.  We have expanded on this topic in the discussion:

‘A possible solution is for SOMs to provide real-time identification and classification of forecast atmospheric patterns from deterministic or ensemble NWP systems. While this could be done subjectively, this places more burden on the forecaster to identify patterns in the large range of modeling systems now available. Further, there is always the issue of human bias. A hypothetical system could identify the fractional number of ensemble members with forecasted blizzard patterns. Pattern typing would then inform forecasters on the scope of impacts. For a system such as the Global Ensemble Forecast System (GEFS), doing this subjectively would require the forecaster to individually inspect 21 members, a process that is too arduous in an operational setting.’

MINOR COMMENTS/EDITORIAL REVISIONS

1.  Section 1, Line 95: Replace “this is considered…” with “Storm Data is considered…” 

Fixed

2.  Section 2, Line 117: Explain here how winter seasons were defined (e.g., DJF) or if all blizzards occurring during the time period were included.

Modified to: 'To compare events to NARR data, the time period was limited to the winter seasons (October – April, determined by reported blizzards in Storm Data) of 1979-1980 to 2017-2018.'

3.  Section 2, Lines 118-119:  Were the subjective classifications regarding the event type (i.e., Colorado Low, Alberta Clipper, Hybrid, or Arctic)?  If so, clarify.

Correct. Modified to: Subjective classifications of event types (Alberta Clippers, Arctic Fronts, Colorado Lows, and Hybrids) were made by Grand Forks NWSFO meteorologists using available observations, model, and reanalysis output.

4.  Section 2.2, Lines 143:  Insert Principal before Component Analysis.

Fixed

5.  Section 2.2, Line 171: Insert “a” before “…goal to classify…”

Fixed

6.  Section 3.1 and Figure 3a:  Any thoughts as to the variability in blizzard frequency?

Research in a forthcoming paper has found equivalent variability in the frequency of blizzard patterns. Given ties of cyclone frequency and snowfall anomalies to teleconnections such as ENSO and the NAO in this region (Seager et al. 2010, Eichler et al. 2006), this isn’t too surprising. We have added this information to the paragraph:

'While the source of this variability is beyond the scope of this study, snowfall and cyclone variability in this region have been tied in part to phases of the El Niño Southern Oscillation (ENSO) and the North American Oscillation (NAO) [38,39].'

7.  Section 3.2, Line 274: Usually see AVA described as Negative Vorticity Advection (NVA) (and cyclonic motion as Positive Vorticity Advection (PVA)).

Good point- modified to NVA.

8.  Section 3.2, Lines 264-279:  Do you find Arctic Fronts often followed after Colorado Low or Alberta Clipper events?  Often synoptic analysis shows a “double cold front” whereby the main low and cold front are then followed by a trailing Arctic front and subsequent strong anticyclone.

 This has certainly been noticed in the past, and timing is one source of variability when objectively matching patterns between the different categories. In some of the more extreme (longer duration) cases, one could make the argument that across the CWA, multiple patterns (e.g. Colorado/Clipper -> Arctic Front) could be associated with an event. This will be worth discussing in the next paper when we describe how the objective classification can be used for historical retrospective or forecasting purposes.

9.  Section 3.2, Lines 294-296:  Any physical explanation for the lull in late February extratropical cyclones?

The cyclone climatologies may be tied to the mean position of the polar jet during this time period which is apparently less favorable for lee cyclogenesis. Adding any comments on this topic borders on speculation, so we have not added any text expanding on this topic.

 

Reviewer 2 Report

The paper is interesting and it can be a useful contribution for the study extreme events, such as the blizzards, and the classification of the synoptic conditions that lead to their occurrence, though in its current form it contains some weaknesses that need to be addressed before it can be accepted for publication. Thus, I would recommend a major revision of the manuscript.

I hope the authors will find the following comments useful.

Major Comments

Please see specific comments for details.

1.       The SOM method is not clearly presented. I would suggest to revisit the respective section, providing a comprehensive description of the methodological steps and choices. Moreover, the tests for the evaluation of the clustering results are missing (q-error and t-error).

2.       The paper is not carefully prepared, e.g. there are mistakes in the numbering of the figures, the resolution of the plots is low etc.

Specific Comments

3.       Figure 2: The images are of low quality and the lettering is not readable. Please try to improve their quality.

4.       Page 4, lines 127-129: This statement here is not accurate. The reanalysis datasets can exhibit large differences, thus, a better justification for the selection of NARR is needed.

5.       Page 4, line 143: I would suggest to provide some information regarding the “other advantages” that SOM method offers.

6.       Page 4, lines 143-144: PCA and EOF is the same method. Please see for example Wilks’ Statistical Methods in the Atmospheric Sciences, Chapter 11.

7.       Page 5, line 170: Here, you refer to 93 blizzard cases, but in the rest of the analysis 100 (or 98) cases are used. Please clarify.

8.       Page 5, line 182: The justification of the 4x2 class selection is not adequate. Did you try other SOM arrays apart from 3x5? For example, a 3x3 array could provide better results as the symmetrical rectangular could separate the 4 synoptic types to the corners of the SOM map.

9.       Figure 4:

a.       The isotherms are not visible, nor their labeling.

b.      It is not clear to me what the 12h differences of the MSLPs add to the discussion.

10.   Figure 5: It is not the same as Figure 4, as there are no isotherms. Please correct the caption.

11.   Figure 6&7: The time difference is not clarified here. Please clarify which time steps are used.

12.   Page 10, line 307: The rightmost nodes are 4/8.

13.   Page 11, lines 324-334: In my opinion, the four circulation types are not depicted in the SOM maps (maybe the Colorado Lows). It seems too optimistic to state that there is good agreement subjectively and objectively identified patterns. I would suggest to tests with different SOM parameters selections.

 

Minor comments

14.   Page 5, line 200: Which “paper”? Sorry, I got lost here.

15.   Page 6, line 267: There is no Fig. 3d. Do you mean 4d?

16.   Page 6, line 271: Do you mean 5,7d, respectively?

17.   Figure 9: Please correct spelling in label.

18.   Page 10, lines 294-296: How this lull is explained? I would be helpful to add some information for the not familiar reader.

19.   Page 11, line 323: This should be figure 10.

20.   Appendix A. It would be interesting to add the names of the named blizzards. You should note which two cases are excluded.

Author Response

See .doc for responses.

Author Response File: Author Response.docx

Reviewer 3 Report

Please see the attached file. Thanks. 

Comments for author File: Comments.pdf

Author Response

Reviewer #3

 

General remarks:

 

This study investigates the climatology of blizzards within the Grand Forks NWS Forecast Office County Warning Area (CWA) for about 40 years (1979/1980 - 2018/2019). Then they demonstrated the atmospheric patterns using objective classification method known as a Self- Organizing Map (SOM). The four subjective, but well known categories as the typical blizzard within their CWA, Alberta Clippers, Arctic Fronts, Colorado Lows, and Hybrids, are used for the classified blizzard patterns. Furthermore, these patterns have seasonal variability. The SOM reproduces the general characteristics of these four classified blizzard patterns. In general, this manuscript is well motivated/written, and its main findings are clearly presented. Although some figures are not easily readable and SOM section still has some unclear description, the authors organize their main findings well and present their analysis in a reasonable way. I will accept the

publication of this article after a couple of minor revisions.

 

Minor comments:

Figure 2: This map is a way too RED I understand the snow cover is denoted by pink/red, and there are other things denoted by other colors, but they are not well visible. And it is very hard to tell what the authors are explaining. If you put some locations on the map or put arrows in it to point out what each color means or what locations they are referring (connecting to Fig. 1), that would be helpful.

This figure has been redrafted to improve clarity. Contrast has been improved and the primary region of blowing snow has been identified. Geopolitical boundaries have been added/weighted more heavily so Figures 1 and 2 can be connected together. 

Line 137: Put the reference for K-mean clustering algorithm.

           

Added.

 

Lines 176 -190: The authors stated that the key parameter of the SOM is the number of classes chosen. Two questions are raised.

 

1) What is the difference between SOM and K-mean then?

           

This was briefly mentioned when SOMs were introduced. After rereading this section, our prior sentence could be interpreted as SOMs are similar to K-means algorithms that include a neighborhood function (this is incorrect). We have reworded this section.

‘A competitive neural network, SOMs are most similar to a K-means clustering algorithm [30]. Unlike K-means clustering, SOMs include a neighborhood function during the training process.’

 

2) Why do you choose the 8-class? Why not 4 or 12? How are they related to the total number of blizzards studied here?

 

While we tested a number of different sizes (ranging from 8 to 24-classes), we opted to display the 8-class SOM as it does a reasonable job segregating patterns despite higher quantization errors (e.g. less variability in patterns compared to larger maps). As noted in the response to reviewer #2, blizzard patterns have smaller errors when matched to the presented SOM vs. the composite patterns. In forthcoming work that identifies these patterns in full sets of reanalysis and climate model data, the methodology has been modified to use larger maps as we need to detect these events amongst all possible patterns. 

 

Table1: Very hard to follow.

 

We have added an additional column that provides noes about each parameter. This should provide sufficient information that someone could reproduce our methods with SOM PAK.

 

Figure 9. Put (a) to the top panels, and put (b) to the bottom panels. Currently, they are all

written as Figure 9 . Please change your lines into solid. And the figures (8 panels) are so alike especially top panels of Figure 9. (MSLP and surface temperature), what do they mean?

 

We have kept these figures separate, but have fixed the numbering and turned the lines to solid. Regarding the patterns, we have discussed how the SOM captures a transition in patterns that range from Colorado Low to the Arctic Front. At the surface, this is seen in a progression of patterns from more intense lows underneath a shortwave trough/closed low, to Clippers/Fronts that have smaller amplitude waves embedded within strong NW flow. Although many of these patterns look similar, the composite plots also have many similarities (e.g. connection to a nearby surface low). It is the intensity, direction, and upper-level features that distinguishes the types of blizzards.

 

Figure 10. Each number of blizzards were written in the box. Instead of showing the absolute number of the blizzards, how about showing the percentage of blizzards for each classified blizzards (number of blizzards assigned to each SOM node/ total number of blizzards)? This way, you will show the relative number of blizzards assigned to each category.

 

Good suggestion, Fig. 11 has been modified to show the percentage.

 

 

Table 2. It will be nice if Figure 10 is also presented in the relative percentage of blizzards.

 

We think this is referring to the numbers within Table 2. Relative percentages could be calculated across all blizzards, by month, or by pattern.  We have included this information per pattern to be consistent with Figure 10.

 

This study more focuses on the local, dynamic factors to affect the atmospheric patterns associated with red river valley blizzards. While authors mentioned in Line 83-84 that while blizzards are often thought of a large-scale event associated with the juxtaposition of winds and snowfall associated with mid-latitude cyclones, there is no description of the large-scale, backgound conditions (PV, putting surface low and upper-level trough together etc). It would have also been interesting if the authors show the large-scale condition of these blizzards. Another suggestion is that showing the other thermodynamic factors, such as temperature, potential temperature, diabatic heating, and precipitation/snow pattern over the regions during the winter periods and how they vary with season, and how they affect the development and intensification of each blizzard types. But I think these suggestions are not essential for this current form of a manuscript so take this recommendation for the future study.

 

Noted, and we wholeheartedly agree.

Reviewer 4 Report

Summary:

This study evaluates the frequency of blizzards in the Red River Valley and links the events with 6-hr MSLP and 500hPa patterns. This is a very good study with interesting and valuable results. I only have a few minor questions and comments so I think that the paper only requires minor revision. See my comments/questions below.

 Comments/Questions:

The paper addresses the MSLP and 500hPa patterns through composite analysis and SOMs for the blizzard events but does not comment on how often the same weather patterns occur without blizzards (hit/miss). The authors do elude to this issue for future efforts to apply the results to forecasting applications (L355-356). I think that this issue should either be directly addressed as a shortcoming of this study in the discussion or some additional analysis should be provided.

Figure 2: it is hard to make out the features with the red background. I suggest adjusting the color scheme to better contrast against the blowing snow.

L68: For those not familiar with the USA please identify the “namesake”

L80: Please spell out NWSFO here (it isn’t defined until L90)

There are two Figure 9’s and no Figure 11

Include the node numbers on all of the panels in Figure 9’s

Author Response

Reviewer #4

This study evaluates the frequency of blizzards in the Red River Valley and links the events with 6-hr MSLP and 500hPa patterns. This is a very good study with interesting and valuable results. I only have a few minor questions and comments so I think that the paper only requires minor revision. See my comments/questions below.

Comments/Questions:

The paper addresses the MSLP and 500hPa patterns through composite analysis and SOMs for the blizzard events but does not comment on how often the same weather patterns occur without blizzards (hit/miss). The authors do elude to this issue for future efforts to apply the results to forecasting applications (L355-356). I think that this issue should either be directly addressed as a shortcoming of this study in the discussion or some additional analysis should be provided.

We have expanded the discussion to discuss the potential limitation of the current methodology, and how it can be applied in practice (which is more complicated). Because this issue includes problems such as human observers (whether an event was sampled) and snow conditions (modeling needed, as this information is not provided by the reanalyses), we have not included additional analysis, as this will be more fully covered in the follow-up manuscript.

Figure 2: it is hard to make out the features with the red background. I suggest adjusting the color scheme to better contrast against the blowing snow.

This figure has been redrafted to improve clarity. Contrast has been improved and the primary region of blowing snow has been identified. Geopolitical boundaries have been added/weighted more heavily so Figures 1 and 2 can be connected together. 

L68: For those not familiar with the USA please identify the “namesake”

Added (state of Colorado).

L80: Please spell out NWSFO here (it isn’t defined until L90)

Fixed.

There are two Figure 9’s and no Figure 11

Fixed.

Include the node numbers on all of the panels in Figure 9’s

Numbers are now included for all nodes.

Round 2

Reviewer 2 Report

The authors have taken under considerations the comments of the first review.

The quality and the resolution of the figures are substantially improved, though Figure 9 (one of the SOMs), which is a key figure of the paper is missing.

I would also suggest to add the discussion in the response letter regarding re-analysis datasets in the manuscript as well (and the necessary reference).

Author Response

The quality and the resolution of the figures are substantially improved, though Figure 9 (one of the SOMs), which is a key figure of the paper is missing.

This figure was omitted on accident by the editors and is included in the manuscript.

I would also suggest to add the discussion in the response letter regarding re-analysis datasets in the manuscript as well (and the necessary reference).

This discussion (clarification about validity of our statement) was previously added, but the King and Kennedy citation has also been added.