the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 02 Oct 2024
Quantifying the local predictability of the 2021 sudden stratospheric warming event using a novel nonlinear method
Abstract. Sudden stratospheric warming (SSW) is identified as a key role in improving the winter subseasonal-to-seasonal prediction, for its surface impacts up to two weeks through stratosphere troposphere coupling. A better understanding of the predictability of the SSW itself, thus, is fundamental. Most of the previous studies investigate the predictability of SSW events using linear approaches and give the approximate predictability.
In the study, we quantify the local predictability limit the 2021 SSW event, which caused cold extremes across East Asia and North America, by applying a nonlinear method, Backward Searching for the Initial Condition (BaSIC), within ERA5 reanalysis data and the S2S reforecasts. The nonlinear method BaSIC is advanced because the nature of SSW is a chaotic system with intrinsic properties, making it difficult to measure its predictability with traditional linear methods. The local predictability limit of this 2021 SSW event is estimated to be 17 days using BaSIC method, exceeding previous estimations by one to two weeks using linear methods.
To gain further insight into where the errors may originate and propagate, we trace the sources of forecast errors of this SSW to the area of the fastest error growth. At the beginning of the SSW forecast, the overall forecast errors are relatively small over the whole polar stratosphere; the errors grow slowly in the first 2 weeks, but increase rapidly in the mid-high latitudes over central Eurasia (30°–60° E) and propagate into the rest of Eurasia. This indicates that the forecast errors in the 2021 SSW event mainly originate from the high altitude over central Eurasia.
- Preprint
(2618 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on egusphere-2024-2574', Anonymous Referee #1, 05 Nov 2024
General comments:
The authors apply a non-linear analytical assessment of the stratospheric state to asses the local predictability leading up to the 2012 Sudden Stratospheric Warming. The study provides an additional metric for assessing predictability alongside more conventional (linear) approaches and is therefore of merit for publication. However, I feel the manuscript needs substantial modification to enable readers to understand the applied method, the analysis carried out, and the comparison with other dynamical studies (see comments below).
specific comments:
1: The methodology needs a clearer explanation – the authors have referenced previous papers, however more detail on the actual calculations and steps performed on the datasets within this study are needed for a reader to follow. As a minor example, line 181 refers to equations 9 and 10 however these are not found in the manuscript
2: Apologies if I have misunderstood, but it appears the analysis performed here relates only to the zonal wind field at 10hPa, and is therefore missing the influence (and error growth) within the upper troposphere and lower stratosphere (which Cho et al 2021 suggest are important for this event)? The authors discuss the regional error growth within the forecast models in the stratosphere, however this is likely to be driven by the tropospheric wave activity, and will thus affect the resulting predictability limits?
3: Once again, apologies if I have misunderstood, but a key assumption within this methodology is that any state preceding the identified initial condition cannot, by definition, predict the event state? Given the importance of tropospheric conditions driving the stratospheric flow, by only focusing on the 10hPa winds, surely multiple tropospheric states could precede the event (at lead times earlier than the identified initial condition) and would not be captured within this analysis? A practical example of this is that you may find one member predicts the event at a very long lead time (earlier than those utilised here), possibly by chance, but this still represents a physical state leading to the SSW.
Minor comments:
- Suggest adding “practical” to the article title (i.e. “quantifying the practical local predictability…”)
- Please can sub-headings be used to separate out the different sections of the results?
- How does this approach compare or with other statistical approaches e.g. Finkel et al 2023 (Revealing the Statistics of Extreme Events Hidden in Short Weather Forecast Data)?
- If understood correctly, the analysis depends upon the ensemble mean RMSE; can the authors add discussion regarding the regional growth of the ensemble spread, which is also important for understanding predictability barriers (e.g. Sanchez 2020: Linking rapid forecast error growth to diabatic processes)?
- It is also useful to note that new dynamical methodologies demonstrating causality are now available and have been used for SSWs (Kent et al 2023: Identifying Perturbations That Tipped the Stratosphere Into a Sudden Warming During January 2013)
- There are several acronyms which need defining (e.g. WN1, ES, CIS)
- In the abstract the authors state that the surface response is around two weeks following SSWs (Line 11), however the surface impact can be 30-60 days for the downward coupling to influence the troposphere.
- A few spelling errors also need addressing (e.g. "middel")
Citation: https://doi.org/10.5194/egusphere-2024-2574-RC1 -
RC2: 'Comment on egusphere-2024-2574', Anonymous Referee #2, 20 Nov 2024
Summary
The manuscript investigates the predictability limit of the 2021 Northern Hemisphere Sudden Stratospheric Warming (SSW) event using a fairly novel technique known as BaSIC (BAckward Searching for the Initial Condition). Using ERA5 as verification, the authors choose two subseasonal-to-seasonal (S2S) model forecasts in which to show how far in advance the dynamical models could predict the SSW event. As expected, the ensemble-mean forecasts have the best skill (space and time) under 2 weeks before the start date of the SSW event, agreeing with previous works. However, using the BaSIC method on zonal wind conditions, the authors find that predictability may be extended to 14-17 days before the event. Then, the authors examine error growth in stratospheric fields to show that the largest errors in the stratosphere for this event exist over the middle to high latitudes of Eurasia and North America, suggesting that these regions may be the biggest contributors to the errors in the forecast models' performance for the 2021 SSW.
Overall Opinion
The manuscript presents an interesting way of trying to understand how to examine predictability limits for extreme events like major SSWs. However, I have several concerns about the paper that make it not ready for publication at this time. These concerns are mainly twofold: (1) the BaSIC methodology is inadequately described and, with all of the acronyms used, is very confusing to follow, and (2) the paper lacks actual scientific advancement in identifying the source of errors, particularly when it only focuses on stratospheric fields. The writing is generally good but could also use some more proofreading and refinement. At this time, I think that the paper requires significantly more work before it can be published. Therefore, I am unfortunately recommending that it be rejected with the opportunity for resubmission later.Major Concerns
1) Poor Description of BaSIC. While the authors have referenced other papers that use the BaSIC technique, there is a lack of clarity on several terms, equations, and their use throughout the paper. For example, I am very uncertain what CIS and ES refer to (Lines 161 and 162 and elsewhere). I am also unclear what it means that a random state vector x0 "loses its predictability" (Line 167) - e.g., what is the measure for "predictability?" Much more clarity in the methods section needs to be applied in order for just basic understanding of the method, which can then be used for further critiquing its use in the paper.2) Does this method really show longer predictability windows? The authors make a key point of highlighting that, while the dynamical models in the aggregate (key term) have poor prediction skill at leads longer than 2 weeks, the BaSIC technique highlights that the 2021 major SSW event had a predictability limit of 14-17 days. I have a couple of issues with this statement. First, the authors are comparing the skill for ONE event against the skill of models for MANY events. For any extreme event phenomena (heavy rainfall, heatwaves, etc.), one can find evidence that even a model can perform better than expected for MULTI-MODEL means for ONE event versus a collection of them. This is why we have aggregate statistics for model performance. Scientific interest certainly lies in outlier events (e.g., events in which the model did exceptionally well or poorly forecasting at long leads), but the comparison here of an aggregate of 14 days predictability limit vs. 14-17 days for the BaSIC method isn't the right one, in my opinion. Moreover, is a gain of maybe 3 days really useful or even significant for this event? Overall, I find this to be a significant weakness of the authors' main argument for this manuscript.
3) Source of Errors is too limited and does not offer significant scientific advancement. In using their novel technique, the authors also discuss how they are able to find sources of errors for the forecasts which subsequently can result in poorer predictability. However, the authors only look at stratospheric fields, motivated by some literature cited in the Introduction. While certainly there can errors in the evolution of these fields (or maybe even initialization errors), another major source for changes in the polar stratospheric flow fields is wave driving from the troposphere. Models can have significant errors (and/or biases) in wave driving, which subsequently cascade into stratospheric circulation errors. For example, Schwartz et al. (2022) showed that biases in stationary waves within several subseasonal prediction models significantly impact the upward propagation of waves into the stratosphere. This paper does not even consider these errors for this major SSW. I also do not see suggestions for ways to improve even the errors shown in the paper within the Discussion section. These types of analyses and suggestions are where the scientific advancements could be made with this paper and thus make it a useful publication. As such, I recommend that the authors rework their "sources of error" section of the paper to account for tropospheric fields and wave driving and make suggestions for improvement.
More Minor/Specific Comments
1) Line 10. Change the start of the sentence to "Sudden stratospheric warmings (SSWs) are..." Same with Line 32.
2) Line 48. Change "tropospheric fluctuations" to "tropospheric waves."
3) Figure 2 (and others). The pressure level at which these fields are plotted is not indicated in the caption or accompanying text. This is very important to indicate and should be labeled.
4) Figure 3. There is no indication the months/season to which these errors refer. Please be more specific in the figure caption and text.
5) Figure 4 and accompanying text. There is no discussion of how different ensemble members perform (note, for example, there are some ECMWF ensemble members show a major SSW at 2 week leads). This is important, as the authors only compare their method to ensemble-mean statistics.
6) Lines 269-270. I don't know what "lost their forecast skills" means exactly. Is there a quantitative measure for loss of skill?
7) Figure 5. There is no indication of what field is plotted or what the units are.
8) Appendix. I do not know what the script "A" represents in Equations A1, A2, and A3.
9) Data Availability. There is no indication that the authors are making their code and/or datasets publicly accessible. Please consider adding a Zenodo and/or Github repository for this.Reference
Schwartz, C., C. I. Garfinkel, P. Yadav, W. Chen, and D. I. V. Domeisen, 2022: Stationary wave biases and their effect on upward troposphere-stratosphere coupling in sub-seasonal prediction models. Wea. Climate Dyn., 3, 679–692, https://doi.org/10.5194/wcd-3-679-2022.Citation: https://doi.org/10.5194/egusphere-2024-2574-RC2
Data sets
The ERA5 reanalysis dataset Hersbach et al. https://cds.climate.copernicus.eu/datasets/reanalysis-era5-pressure-levels
the S2S prediction database Vitart et al. https://apps.ecmwf.int/datasets/data/s2s-reforecasts-daily-averaged-ecmf/levtype=sfc/type=cf/
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 150 | 50 | 104 | 304 | 4 | 5 |
- HTML: 150
- PDF: 50
- XML: 104
- Total: 304
- BibTeX: 4
- EndNote: 5
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1