the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 24 Oct 2024
Attributing the occurrence and intensity of extreme events with the flow analogues method
Abstract. Extreme event attribution methodologies have been proposed to estimate the impacts of anthropogenic global warming on observed climatological and meteorological extremes. The classical risk-based approach uses Extreme Value Theory (EVT) to derive changes in the unconditional probabilities of yearly maxima but bears the risk of comparing events with different dynamical mechanisms. The flow analogues method on the other hand is a conditional attribution method which compares events with similar synoptic scale dynamics. Here we propose a procedure for estimating both the intensity change and the probability ratio of observed extreme events with this method. We illustrate the procedure on three recent extreme events in Europe and compare the results obtained to the EVT-based approach. We show that the conditional flow analogues method gives more significant results for these events, which suggests a stronger climate change signal than the one detected with the unconditional approach.
- Preprint
(4780 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
- RC1: 'Comment on egusphere-2024-3167', Anonymous Referee #1, 30 Oct 2024
-
RC2: 'Comment on egusphere-2024-3167', Anonymous Referee #2, 07 Jan 2025
Review: Attributing the occurrence and intensity of extreme events with the flow analogues method
This paper presents a methodology for attributing extreme weather events using flow analogues, using three different event types within Europe (precipitation, heat, and wind) as worked examples. The method is compared to the ‘traditional’ probabilistic attribution method. The paper provides arguments for the methodological choices in the analogues attribution technique, stressing both the advantages compared to GEV and advantages compared to earlier uses of analogues. The approach methodology presented is well, and is a great development for the use of analogues in attribution.
However, I have a few major comments that I think need addressing, followed by some more minor comments.
1. I think more consideration of the results and how/why they initially appear to (in some cases) contradict the GEV method needs to be made. More thought and discussion on the ‘attribution statement’, and importantly the framing of the questions that each attribution methods can be used for, needs to be made. A related point which is not discussed in detail in the paper, is the advantage of the method presented over other conditional methods (e.g. those referenced in line 48), and how using a range of methods to provide multiple lines of evidence could be useful. As is, the paper appears to pit the analogues method against GEV, rather than consider how it is useful to consider both to increase understanding.
This could be a useful reference - Coumou, D., Arias, P.A., Bastos, A., Gonzales, C.K.G., Hegerl, G.C., Hope, P., Jack, C., Otto, F., Saeed, F., Serdeczny, O. and Shepherd, T.G., 2024. How can event attribution science underpin financial decisions on Loss and Damage?. PNAS nexus, 3(8), p.pgae277.
2. For the precipitation event many of the analogues show little rainfall (particularly after detrending). This suggests that the analogues poorly represent the precipitation of the observed event. Can you show that the analogues do represent the precipitation adequately? (This could also be needed for temperature and wind, but it appears less of an issue for these variables – perhaps suggesting the method is only suitable for specified event types).
3. One point that is mentioned in the discussion (line 363), but I feel should be stressed further, is that the method only works when there are good analogues. Further discussion of this would be valuable – are there certain event types better suited to the method? Or some events (perhaps hurricanes) where the event should not be used due to insufficient past analogues (though maybe large model ensemble could be used instead)? What if good analogues only occur in later decades?
Minor comments -
The ‘past’ starts in 1950, restricted by the ERA5 dataset. Do you think this matters? Do you think results would differ significantly if you were able to go right back to 1850?
Line 65 – dates formatting, could remove the ‘of’’s throughtout (e.g. 4th July, not 4th of July)
Line 98 – why do you take Z500 rather than SLP (as used in Faranda et al 2024)?
Line 112 - Some other analogue studies use spatial correlation to identify analogues, why did you choose to use Euclidean distance?
Line 200 – Assessing linear trend per decade – there aren’t many data points, did you test sensitivity shifting the decades (i.e. 1955-1964, 1965-1974 or other starting years)?
Fig.2 caption – the final sentence is a bit misleading as no trends are shown. I think this should be removed, and just referred to in the results (I spent a while trying to spot the trend!)
Fig 2 / 3 / 6 – it would be great to title the columns as you do in Fig4
Fig3/6 could you align the zeros?
Lin 255 - ‘northern France and ‘southern England’ rather than north of etc.
Line 271 – You chose to use the same number of analogues for all events, but some events may have more good analogues (i.e. be less dynamically extreme). Would there be a way to incorporate this into the method, so use a different number of analogues depending how good the analogues are? Maybe by using the information in Fig2a,b,c?
Fig.4 - z500 contours are not very clear to me
Line 279 – precipitations (no need for ‘s’)
Line 345 – stray ‘?’ in references
Line 392 (final sentence) This doesn’t quite read right to me, consider rewording.
Citation: https://doi.org/10.5194/egusphere-2024-3167-RC2
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 169 | 41 | 10 | 220 | 5 | 9 |
- HTML: 169
- PDF: 41
- XML: 10
- Total: 220
- BibTeX: 5
- EndNote: 9
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1