A modelled multi-decadal hailday time series for Switzerland

Wilhelm, Lena; Schwierz, Cornelia; Schröer, Katharina; Taszarek, Mateusz; Martius, Olivia

doi:https://doi.org/10.5194/egusphere-2024-371

Preprints

https://doi.org/10.5194/egusphere-2024-371

Preprints

23 Feb 2024

| 23 Feb 2024

A modelled multi-decadal hailday time series for Switzerland

Lena Wilhelm, Cornelia Schwierz, Katharina Schröer, Mateusz Taszarek, and Olivia Martius

Abstract. In Switzerland, hail is one of the costliest natural hazards, causing extensive damage to agriculture, cars, and infrastructure each year. In a warming climate, hail frequency and its patterns of occurrence are expected to change, which is why understanding the long-term variability and its drivers is essential. Therefore, this study presents new multidecadal daily hail time series for Northern and Southern Switzerland from 1959 to 2022. Daily radar hail proxies and environmental predictor variables from ERA-5 reanalysis are used to build an ensemble statistical model for predicting past hail occurrence. Haildays are identified from operational radar-derived "Probability of Hail" (POH) data for two study regions, namely the north and south of the Swiss Alps. We use data from 2002–2022 during the convective season from April to September. The decision hailday YES / NO is based on surpassing a POH ≥ 80 % for a certain minimum footprint area of the domains. Separate logistic regression models and GAM´s are built for each domain and combined in an ensemble model to reconstruct the final time series. Overall, the models are able to describe the observed time series well. Historical hail reports are used for comparing years with the most and least haildays. For the northern and southern domains, the time series both show a significant positive trend in yearly aggregated haildays from 1959 to 2022. The trend is still positive and significant when looking at the period 1979–2022. In all models, the trends are driven by moisture and instability predictors. In the last two decades, we can see an increase in haildays at the beginning of the hail season and an earlier and longer peak, however, there is no systematic shift in the seasonal cycle. With this time series, we can now study the local and remote drivers of the interannual variability and seasonality of Swiss hail occurrence.

Received: 07 Feb 2024 – Discussion started: 23 Feb 2024

Download & links

Lena Wilhelm, Cornelia Schwierz, Katharina Schröer, Mateusz Taszarek, and Olivia Martius

Status: final response (author comments only)

RC1: 'Comment on egusphere-2024-371', Anonymous Referee #1, 01 Mar 2024

See the attached file.

Citation: https://doi.org/10.5194/egusphere-2024-371-RC1
RC2: 'Comment on egusphere-2024-371', Anonymous Referee #2, 14 Mar 2024

General comments:
This study uses a radar-based hail day time-series for Switzerland to train an ensemble of logistic models. The model is shown to fit historic distributions of hail days reasonably well and is then used to estimate a trend. This trend is positive and significant in both South and North Switzerland.

I am no expert on logistic models. That being acknowledged, I think the article is well written, the methods are appropriate, and the Figures are good. In some places, I disagree slightly with the authors interpretation of the results (see comments below). None of these comments are major and I don’t expect the authors will have trouble addressing them. However, given there are several “deeper” comments and I think the article could be improved by having another round of exchange on some of them, I recommend (very minor) major revisions.

Specific comments:
line 28: You could add Manzato et al. (2022) here in addition to the Augenstein presentation. Their trend (Fig. 3) is not significant (but rather negative than positive) and at least it is published.
100: The intro is nicely structured and compact.
91-94, 130-145, 162-169: I think there are some minor inconsistencies in your approach or how you explain it. You mention POH specifically includes small hail but then you use thresholds for POH and affected area which were tested for car insurance. To my knowledge, damage to structures and cars is dominated by severe hail (>2cm). Hence, to the reader it remains unclear what hail sizes the trend you find is representative of. It might be worth looking into the trend for more or less severe events (e.g., higher values of POH or larger areas?). For instance, it's good that you mention that the sensitivity to the area threshold was tested, but you could elaborate on this more and test other POH thresholds if feasible. I think such tests and discussions would improve the robustness of your results but I will leave it to you.
235 and 299-307: I was wondering the whole time why no kinematic information is included. I would recommend adding a brief sentence here saying something like: „The lack of a kinematic predictor in the northern model will be discussed further later on.“
290-296: Convective updrafts are not resolved in ERA5, so large OMEGA cannot be caused by this. In other words, I don’t think this can be linked to the results of Lin and Kumjian (2022).
Also, while I agree that the role of synoptic lift is likely included in this predictor, the inclusion of omega is surprising because I would have expected orographic lift to dominate over Switzerland? Perhaps because the larger scale lift favors move widespread convection and hence more POH area it is still a useful predictor?
362-364: Lin and Kumjian (2022) saw increasing hail potential until around 2500 J/kg. CAPE doesn't even reach such values in your Fig. 6 and the curve flattens at 500 J/kg already, so I don't think these results should be linked, at least not with further explanation.
376-395 Some of your interpretations here were a bit confusing to me. I think this is a very active research topic but strong storm-relative winds have been shown to promote wider updrafts (Dennis and Kumjian 2017, Peters et al. 2020) and are hence important for hail. Also, weaker low-level winds have been suggested to be better, especially in north-south direction (Dennis and Kumjian 2017, Nixon et al. 2023). So your results or their interpretation are counter-intuitive, which should be clarified (you write that strong storm-realive winds are bad for hail but low-level shear good, at least that’s how I understand your text, maybe you got them mixed up?). One explanation could be that the typical environments in Switzerland are different compared to these studies in the US. So the sensitivities to kinematic variables might be different. Which is worth to be discussed.
397: What do you mean by „circulate“? Most large hail seems to follow a single up-down trajectory while curving around the updraft (e.g., Kumjian and Lombardo (2017), Pounds et al (2023). No re-circulation with repeated ingestion into the updraft seems to happen. How this is in non-supercell storms is still unclear, but I don’t see a reason to assume differently.

Also, I’d suggest rephrasing to „deeper hail growth zone“ or „longer residence time in the hail growth zone“.
398-403: Punge et al. (2023) also found that excluding freezing lvls<2400m helped reduce false hailstorm detections in higher elevation in South Africa. Might be relevant here.
470: I liked that you followed closely and compared your results to Raupach et al. (2023).
Section 5.3: April and September don't have a good sample size and I don't see any increase for either month, but in the text you say "This leads to more events specifically at the beginning of the hail season (April-June)". Perhaps May-June would be more accurate?
525-549: I agree with these explanations, but are your hail day thresholds appropriate then, since they are trained with vehicle damages and hence larger hail ?
600: Raupach et al. had to cover a much larger region and different climate zones. Perhaps it is fair to mention that the skill could be linked to such differences?
647 and abstract: I don’t see a longer peak in Fig. 13. (see also comment on September above)
Overall, I liked your thorough discussion and conclusions.
Acknowledgements: You mention that parameters from thundeR were tested. Maybe I missed this in the text, but some discussion of it might be insightful, no?

Also, since M. Taszarek is in the author list, I’m not sure if it’s necessary to acknowledge his contribution. Your choice though.

technical corrections:
line 27: I’m not sure if this is the standard formatting in egusphere but with some references you use brackets around the years (when the reference starts wiht something line „e.g.“) and in some not. Also, it is common to put a comma after „e.g.“ I think.
43: Just a minor thing, but I think „ERA5“ is more commonly used?
48: I suggest removing „a“ before „sufficient“
68: I suggest writing either "both thermodynamic" in the brackets or "...can be grouped into thermodynamics (instability and moisture) and kinematic conditions." to make clearer that „thermodynamic“ refers to both instability and moisture
195: Suggest removing extra brackets in the exponent.
Fig. 3: Units are missing here in some of the following figures.
455: Add a space before „respectively“.
564: „choose“
The references to Augenstein et al. (2023) and Mohr et al. (2015 a,b) need some fixing.

Manzato, A., S. Serafin, M. M. Miglietta, D. Kirshbaum, and W. Schulz, 2022: A Pan-Alpine Climatology of Lightning and Convective Initiation. Mon. Wea. Rev., 150, 2213–2230, https://doi.org/10.1175/MWR-D-21-0149.1.
Nixon, C. J., J. T. Allen, and M. Taszarek, 2023: Hodographs and Skew Ts of Hail-Producing Storms. Wea. Forecasting, 38, 2217–2236, https://doi.org/10.1175/WAF-D-23-0031.1.
Peters, J. M., C. J. Nowotarski, J. P. Mulholland, and R. L. Thompson, 2020: The Influences of Effective Inflow Layer Streamwise Vorticity and Storm-Relative Flow on Supercell Updraft Properties. J. Atmos. Sci., 77, 3033–3057, https://doi.org/10.1175/JAS-D-19-0355.1.
Punge, H. J., Bedka, K. M., Kunz, M., Bang, S. D., and Itterly, K. F.: Characteristics of hail hazard in South Africa based on satellite detection of convective storms, Nat. Hazards Earth Syst. Sci., 23, 1549–1576, https://doi.org/10.5194/nhess-23-1549-2023, 2023.

Citation: https://doi.org/10.5194/egusphere-2024-371-RC2
RC3:
'Comment on egusphere-2024-371', Julian C. Brimelow, 04 Apr 2024
Review of “A Modelled multi-decadal hailday time series for Switzerland” by Wilhelm et al.

General Comments
The authors apply existing (and at times new) techniques to develop a long-term time series for the occurrence of hail in the vicinity of the European Alps. This work build on previous research and furthers our understanding not only of how hail activity has likely changed on a decadal scale in the study area, but also what is driving those changes. In fact, the latter is a major strength of the paper. Specifically, the authors do an admirable job of trying to understand why certain predictors were chosen (or not) and how they are physically relevant for hail occurrence. This is important, because some of the selected predictors (e.g., the Total Totals index) are, on the face of it, surprising and not necessarily consistent with previous research. Explaining all of this means the paper is long and requires the reader to play close attention through some very detailed/dense text. Suggestions to try and mitigate this, while still retaining valuable information, are provided below.
The content of the manuscript is relevant to NHESS and, in my opinion, meets the scientific rigour expected for an EGU journal paper. The writing and figures are of a high quality, although some minor improvements to the grammar are necessary. The authors clearly describe their methodology and how the various datasets are used, but again, some tweaks are needed in places as indicated below.
Considering all of the above, I am happy to recommend the manuscript for publication.

Specific Comments
The use of “Switzerland” in the title is not consistent with the study area discussed in the text. The radar data used in the analysis cover Switzerland, and parts of France, Italy, Germany and Austria. I would suggest saying “in the vicinity of the European Alps” or something similar.

Along the same lines as (1), it is not clear how the central Alps region, which is excluded from the study area, is defined. Some more information and motivation are required here on how the boundary was determined.

Page 2, lines 32-34: Can the authors please briefly elaborate on how the interannual variability differs between areas north and south of the central Alps.

Figure 1: I would suggest adding borders for surrounding countries for clarity and using either a different colour or bolder line for the Swiss border.

Pages 2-3, lines 56-60. Please briefly discuss what is meant by “local thermodynamic conditions”. Are the authors referring to the mountain-plain circulation. What thermodynamic conditions are unique to the Po valley and not to areas north of the central Alps?

Page 4, lines 123-124: Please provide a reference (or references) to support the assertion about data quality from ERA-5 declining before 1959.

Section 2.1, “ERA-5 environmental parameters”. It is not clear to me over what areas the ERA-5 data were extracted for each day. For example, were all the parameters calculated using ERA-5 profiles at grid points that were within or close to those areas where the POH was at least 80% selected, and then the average of those values used? Or where the values calculated using profiles at grid points over the entire northern or southern domains on days when the POH spatial criteria were met?

Page 8, line 188: Did the authors mean to say “hail potential” and not “convective potential” here?

Section 4.1: I may have missed this, but I found myself asking “why not just use previous logistic regression models” for this? Or, “why was it necessary to build your own logistic regression model”? Please elaborate on your reasoning here. Also, if there are other logistic regression models out there for predicting the occurrence of hail, how does you model compare?

Page 11, line291-296. The thinking here regarding OMEGA and updraft size and strength is not correct. By extension, reference to the work by Lin and Kumjian is not justified. OMEGA refers to large-scale synoptic/dynamic lift and does not provide information on the properties of thunderstorm updrafts. OMEGA does, however, provide information where large-scale ascent favours thunderstorm formation and maintenance. Please check for this throughout the paper, incl. Section 3.

Page 16, line 347: Did the authors mean to say “variance” and not “deviance”? I’m thinking the authors are referring to the variance explained or the coefficient of determination (i.e., R²) here? Check elsewhere in the text for this.

Page 17, lines 399-403: There may be a logical explanation for this seemingly counterintuitive result. One of the conditions for thunderstorm formation is instability and sufficient static energy (referring to the equation for moist static energy = CpT + Lq +gz). One of the reasons thunderstorms tend to be more prevalent during daylight hours and warmer months is because of the higher temperatures. Given that surface temperatures and the height of the freezing level are positively correlated (see Tables A1 and A2), what we could be seeing is that the model is identifying the freezing level as a proxy for surface temperatures. In that framework, the negative linear correlation between deg0l and hail threat makes sense. That said, it is not clear why the model selected deg0l over t2m.

Section 4.3: Can the authors please provide some quantitative information that supports the choice of the ensemble model over the other models in the manuscript. Are the differences sufficiently significant to warrant the use of the ensemble model?

Section 4.3, lines 410-413: The authors only very briefly touch on how the ensemble model was produced. Unfortunately, the text on lines 412-413 was not adequate to provide a clear indication of exactly how the model was constructed. Please elaborate.

It is interesting that the SWISS index was not selected despite being developed for this region and having been found previously for correlating well with the occurrence of large hail. Do the authors have any thought son why this might be?

Technical Corrections
Page 1, line 20: Say, “ Addressing the hail hazard…”.

Page 1, line 22: Are you referring to spatiotemporal dimensions?

Page 2, line 35: Suggest saying, “…is essential for adopting potential adaptation…”

Page 2, lines 40-41: Say, “…, we require a hail time series that is longer than what is currently available.”.

Page 2, line 45: Say, “…ERA-5 is considered one of the most reliable…”.

Page 2, line 48: Say, “…moisture, sufficient vertical wind shear..”.

Page 2, line 55: Suggest saying, “…frontal systems approaching from the west (or north), because unlike over the southern region there is no mountain barrier.

Page 3, line 79: Suggest saying, “…, but the increases are statistically significant only in the northwest and…”.

Page 3, line 92: POH has not yet been explained. Spell out and provide reference.

Page 4, line 106: Replace “dBz” with “dBZ”.

Page 4, line 112: Replace “shielding” with “blocking”.

Page 4, lines 116-117: Suggest rewording to, “Comparing POH data with car insurance loss data, Nisi et al. (2016) found that a threshold of POH ≥ 80% was best associated with the occurrence of hail on the ground”.

Page 5, line 126: Replace “incident” with “event”.

Page 5, line 148: Replace “reaching” with “extending”.

Page 5, lines 150-151: Suggest saying, “Radar-based measurements compliment the archive for more recent periods (i.e., since 2002).”

Page 6, “POH time series”. For context, please provide the areas of the northern and southern regions.

Page 6, line 174: Say, “…, hail over our domain is…”.

Page 9, line 243: Suggest saying, “…model performance is undertaken…”.

Page 10, line 267: Say, “…is connected to environments favouring hail follows”.

Page 11, line 315: suggest saying, “…variables of the model for the northern region on the southern region and vice versa.”.

Page 33, line 577: Say, “…did not present as a skillful predictor”.

Page 33, line 579: Suggest saying, “To address this problem…”.

Page 34, line 585: Say, “Comparison with other studies”.

Page 34, line 614: say, “…, that this study’s modelled trends…”.

Page 36, line 653: Suggest replacing “strong” with “severe”?
Citation: https://doi.org/10.5194/egusphere-2024-371-RC3

Lena Wilhelm, Cornelia Schwierz, Katharina Schröer, Mateusz Taszarek, and Olivia Martius

Viewed

Total article views: 412 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	BibTeX	EndNote
308	89	15	412	7	7

HTML: 308
PDF: 89
XML: 15
Total: 412
BibTeX: 7
EndNote: 7

Views and downloads (calculated since 23 Feb 2024)

Month	HTML	PDF	XML	Total
Feb 2024	94	36	3	133
Mar 2024	138	40	6	184
Apr 2024	76	13	6	95

Cumulative views and downloads (calculated since 23 Feb 2024)

Month	HTML	PDF	XML	Total
Feb 2024	94	36	3	133
Mar 2024	138	40	6	184
Apr 2024	76	13	6	95

Viewed (geographical distribution)

Total article views: 410 (including HTML, PDF, and XML) Thereof 410 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 26 Apr 2024

Short summary

In our study we used statistical models to reconstruct past haildays in Switzerland from 1959–2022. This new timeseries reveals a significant increase in hail occurrences over the last seven decades. We link this trend to climate factors, showcasing the impact of increasing moisture and instability in the atmosphere. The last two decades have seen a surge in early hailseason events. This time series can now be used to study what drives the strong year-to-year variability of Swiss hailstorms.


Total:	0
HTML:	0
PDF:	0
XML:	0