the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 02 Sep 2026
Advancing seasonal storm surge forecasting through weather pattern–informed ensemble subselection
Abstract. This study investigates how large-scale atmospheric circulation regimes can improve seasonal storm surge predictions in the North Sea using the high-resolution Max Planck Institute Earth System Model (MPI-ESM-HR) forecast ensemble. Although the ensemble mean shows limited skill in predicting interannual variations of seasonal storm surge activity, circulation-dependent information within the ensemble can be exploited through weather-pattern-informed subselection to improve predictions beyond the full ensemble mean. The approach provides a physically interpretable framework for extending circulation-based coastal hazard forecasting towards seasonal timescales. An idealised approach using perfect knowledge of weather patterns achieves anomaly correlations of 0.78 and 0.64 for seasonal predictions of surge height and surge-event counts, respectively, substantially exceeding the forecast-based subselection (0.27 and 0.31). This gap highlights the potential to improve seasonal storm surge forecasts through better prediction of surge-relevant atmospheric regimes.
- Preprint
(1747 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 25 Oct 2026)
- RC1: 'Comment on egusphere-2026-5065', Anonymous Referee #1, 23 Sep 2026 reply
-
RC2: 'Comment on egusphere-2026-5065', Anonymous Referee #2, 29 Sep 2026
reply
Review of manuscript „Advancing seasonal storm surge forecasting through weather pattern–informed ensemble subselection” by Anna K. Miesner, Daniel Krieger, and Leonard F. Borchert
Summary: In the framework of model-based seasonal climate prediction, the manuscript presents a subsampling strategy that is intended to increase the storm surge prediction skill of the MPI-ESM-HR seasonal forecast ensemble. The occurrence of storm surges over a defined storm surge season (ONDJFM) is measured in two complementary metrics (cumulative storm surge height and storm surge count). The authors connect the occurrence of storm surges to atmospheric conditions, identifying two storm surge promoting and various storm surge inhibiting weather regimes within the well-known Hess-Brezowsky weather pattern classification. The subsampling strategy consists of a selection rule that depends on the number of forecast members exhibiting storm surge promoting/inhibiting weather regimes. The skill in the two complementary storm surge metrics increases after subsampling from overall insignificant (full ensemble) to minimally significant values in some, but not all skill measures.
General Comment: although the authors go to considerable lengths to defend their subsampling strategy as an effective way to extract predictive information out of the forecast ensemble, the increase in skill measures is inconsistent and, if at all, very limited to the extend that the subsampled forecast has no relevance for decision-making. Still, also failed experiments can represent valuable information that pushes scientific progress. A deeper investigation of the reasons, why this attempt was not successful, could increase the usefulness of this manuscript considerably. Besides the signal-to-noise paradox that is ubiquitously plaguing numerical climate models discussed by the authors, reasons for the disappointing results could also possibly be found in
- The limited suitability of the Hess-Brezowsky weather pattern classification for the promotion/inhibition of storm surges (the differences in the probability of occurrence of storm surges between the various Hess-Brezowsky weather patterns are in the single-digit percentage range—not very indicative for an event of interest). More suitable classifications could be found or even custom-tailored.
- The decision to do without any additional predictive information coming from sources outside the numerical forecast model. Successful subsampling strategies published to date in the scientific literature are in their majority based on additional information often extracted from observed boundary conditions such as for instance ERA5 SST by means of statistical procedures. It has been shown that such statistical procedures can, under certain conditions, achieve greater predictive power than the numerical climate models. It is for instance conceivable to incorporate a statistical estimate of the probability of occurrence of promoting weather patterns into the sampling procedure instead of relying exclusively on the numerical forecast ensemble.
- Lastly, the subsampling strategy to retain only forecast members featuring promoting weather patterns in case there are more than 16 of them, as opposed to retain only members with inhibiting features in the opposite case, might seem on one hand not fully defined (cases not covered by this dichotomy are possible), and on the other hand slightly exotic (this feeling is not alleviated by the procedure’s lacking efficacy).
Specific comment: at various points in the manuscript, the authors make the claim that the repeated existence of a forecast member, which comes close to the observed storm surge metrics, is proof of the ensemble containing predictive information. This claim is disputable. The ensemble covers the observation in most instances by a wide margin and the ensemble mean barely varies, with no correlation to the observed storm surge measures. That means the ensemble reproduces the climatological distribution almost invariably, regardless of the initial conditions. So necessarily, one or another forecast member will almost always come close to the observation. But unless there is some sort of information inherent in the ensemble, which allows its identification without the knowledge of the observation, this does not point to predictive information, but is the result of pure chance. An ensemble consisting of white noise would also have this property, given that its distribution is wide enough to cover the observations.
Recommendation: the manuscript would gain very much relevance if the above-mentioned points would be discussed, and even more, if the authors would include a variety of subsampling strategies in their investigation to get a better grip on the challenges of the problem.
Citation: https://doi.org/10.5194/egusphere-2026-5065-RC2
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 217 | 91 | 42 | 350 | 41 | 39 |
- HTML: 217
- PDF: 91
- XML: 42
- Total: 350
- BibTeX: 41
- EndNote: 39
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Dear Authors,
My detailed comments and revision requests are provided in the attached PDF file.