the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 10 Oct 2024
Ensemble design for seasonal climate predictions: Studying extreme Arctic sea ice lows with a rare event algorithm
Abstract. Initialized ensemble simulations can help identify the physical drivers and assess the probabilities of weather and climate extremes based on a given initial state. However, the significant computational burden of complex climate models makes it challenging to quantitatively investigate extreme events with probabilities below a few percent. A possible solution to overcome this problem is to use rare event algorithms, i.e., computational techniques originally developed in statistical physics that increase the sampling efficiency of rare events in numerical simulations. Here, we apply a rare event algorithm to ensemble simulations with the intermediate complexity coupled climate model PlaSim-LSG to study extremes of pan-Arctic sea ice area reduction under pre-industrial greenhouse gas conditions. We construct seven pairs of control and rare event algorithm ensemble simulations each starting from seven different initial winter sea ice states. The rare event simulations produce sea ice lows with probabilities of at least two orders of magnitude smaller than feasible with the control ensembles, and drastically increase the number of extremes compared to direct sampling. We find that for a given probability level, the amplitude of negative late summer sea ice area anomalies strongly depends on the baseline winter sea ice thickness, but hardly on the baseline winter sea ice area. The experiments furthermore indicate a quasi-zero probability to internally generate a seasonally sea ice-free Arctic in this set-up. Finally, we investigate the physical processes in two trajectories leading to sea ice lows with conditional probabilities of less than 0.001 %. In both cases, negative late summer pan-Arctic sea ice area anomalies are preceded by negative spring sea ice thickness anomalies. These are related to enhanced surface downward longwave radiative and sensible heat fluxes in an anomalously moist, cloudy and warm atmosphere. During summer, extreme sea ice area reduction is favoured by enhanced open-water-formation efficiency, anomalously strong downward solar radiation and the sea ice-albedo feedback. This work highlights that the most extreme summer sea ice conditions result from the combined effects of preconditioning and weather variability, emphasizing the need for thoughtful ensemble design when turning to real applications.
- Preprint
(8042 KB) - Metadata XML
-
Supplement
(1506 KB) - BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on egusphere-2024-3082', Anonymous Referee #1, 31 Oct 2024
This study presents an interesting next step from the Sauer et al 2024 paper. It focuses on using a rare event sampling algorithm to diagnose how initial conditions influence the extremely low summer sea ice conditions in the Arctic in an intermediate complexity model. This paper is well written, clear and relatively concise, and would be a good fit for the journal. It is an interesting progression from the original work and different enough to merit publication.
The authors find that initial conditions are very important for determining the extent of extreme low sea ice conditions, and also present plausible mechanisms through which these low conditions form.
I have some points the authors should consider before publication.
Major comments:
One concern is in balancing the needs to learn things from more idealised setups (such as this intermediate complexity model) so that the learnings can be transferred to the real world and comprehensive models, versus the question of how applicable these learnings will be in a much more complex model. Throughout my reading of this manuscript, I kept thinking to myself how many of these results will be applicable in state of the art models, and this is something the authors inherently cannot answer because they have restricted themselves to the intermediate complexity model from the previous study. I think this is ok for this study but would suggest that for further progress to be made, a step will need to be taken towards implementing this in a comprehensive model.
Related to this, the authors barely discuss the implications of the simplifications of the intermediate complexity model on the results (only alluded to in the last paragraph). To my mind there are a number of idealisations which could make a large impact:
- Lack of sea ice dynamics,
- Atmosphere only influences sea ice through thermodynamics,
- Idealised ocean dynamics (I find it hard to tell how idealised),
- Extremely coarse atmospheric resolution,
- Lack of snow on sea ice (I think?)
- Binary sea ice concentration (i.e no sea ice fraction >0 or <1).
- Realism of Arctic cloud cover and how it responds to sea ice changes
- Unsure who well coupled modes of variability are represented in this model
I think that each of these could have important implications for how you interpret your results. And as such I think looking at 0.001% events in this model is challenging to interpret for the real world.
Another major point which needs to be addressed is that you spent a substantial portion of the introduction motivating the question of anthropogenic versus natural to the loss of sea ice in 2007 and 2012. However, you didn’t link your work back up to this question in the conclusions. How has what you have learnt made progress with this question? If the results from the intermediate complexity model are difficult to translate into this question then I suggest you reduce the role of this question in the introduction. However, I think you could have a good paragraph in your conclusions addressing this question.
Minor comments
L25: Potentially useful reference: England et al 2019 https://journals.ametsoc.org/view/journals/clim/32/13/jcli-d-18-0864.1.xml
L30: In this paragraph it would be useful to give the context that summer sea ice conditions are still above the 2012 minimum, which suggests that climate change is not the entire reason, because global warming has obviously continued considerably since 2012.
L64: You also have the need to rely on multiple models rather than a single model which would increase the computational expense further. However ,wouldn’t the existing large ensembles of CMIP5 and CMIP6 models be a useful resource here? https://www.cesm.ucar.edu/community-projects/mmlea There are multiple models with 100 ensemble members and if you are interested in a 2012-like event relative to the forced signal, rather than an event in 2012, you could search through the timeseries for each member which would give a much larger sample size.
I am not convinced by the statement in L155, given that the mean state could be readily tuned to give a reasonable state. This doesn’t necessarily tell us about how suitable the variability or extremes in the model are.
L211: How do you know that M=7 is sufficient to explore the space?
L355: Are the coupled modes of internal variability sufficiently realistic to make this conclusion for the real world or more complex models? I would suggest comparing this with CMIP6 preindustrial simulations with reasonable sea ice mean states. I would assume that there are no ice free summers in these simulations. Also on this point, is there anyone suggesting that this is possible?
How does this fit in the context of these studies: https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2019GL082947 and https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2020GL088335
Citation: https://doi.org/10.5194/egusphere-2024-3082-RC1 -
RC2: 'Comment on egusphere-2024-3082', Anonymous Referee #2, 18 Nov 2024
The authors present a novel simulation ensemble design to analyze statistics of sea ice area lows in the Arctic using a rare event algorithm. The work is an extension of Sauer 2024 and demonstrates the ability to look at seasonal sea ice predictions conditioned on particular initial winter sea ice states. Seven sets of ensemble simulations are generated from seven initial states taken from a long pre-industrial control simulation of the PlaSim-LSG coupled climate model. For each ensemble, 600 trajectories from February 1 to Sepember 30 are run with perturbed initial conditions for both a control and rare event algorithm case with resampling time of 5 days. The rare event ensembles increase the number of extreme lows compared to the control simulations and the analysis demonstrates that the initial state of the ice has a strong influence on the distribution of minimum area. For two particular trajectories, atmosphere and ocean conditions are investigated for contributions to the extreme ice area loss.
Overall, the manuscript is well-written and introduces a novel approach for estimating the probablity of extreme sea ice loss over a season given the winter ice state. The method has the potential to provide new insight into physical drivers of Arctic sea ice loss. The main limitation of the study is the relatively simple sea ice model configuration in combination with the coarse resolution of the coupled climate model. Given that the sea ice model is purely thermodynamic, neglecting drift and deformation, with only a single layer, it is not clear how broadly the conclusions would hold for more sophisticated models. The authors do address these limitations in the manuscript and state their plans to apply the methodology to a more complex model.
A few areas where additional clarification would be helpful are listed below.
Comments/questions:
1. The introduction focuses on the historical sea ice loss events in 2007 and 2012, which are clearly a motivation for the research in this manuscript. However, the analysis in the paper cannot provide insight into questions of how anthopogenic greenhouse gas forcing or sea ice dynamical drivers may have contributed to these events. The introduction also includes a good bit of general descriptive text that matches (at times word-for-word) parts of the introduction in Sauer 2024. It would be worthwhile to edit the introduction to provide a more specific motivation of the analysis in the paper and clarify what questions relating to sea ice loss are being addressed.
2. Lines 145-155, there is a discussion of the model's ability to capture the average seasonal cycle of sea ice area with enough fidelity for the analysis. Given that the winter volume of ice is more predictive of the seasonal minimum sea ice area, it would be good to include information on how well the model matches expected seasonal ice volume and spatially resolved ice thickness.
3. The table in Figure 1a could use additional explanation. What is the composite analysis (Line 243) and what is the significance of the thickness bins (Line 245)?
4. Figure 2 c-f. It is notable that the rare event ensembles have a bimodal probability distribution. However, this is not mentioned in the discussion in the text. Is there a physical reason for this behavior?
Typo:
Line 460: cloduiness -> cloudiness
Citation: https://doi.org/10.5194/egusphere-2024-3082-RC2 -
RC3: 'Comment on egusphere-2024-3082', Anonymous Referee #3, 22 Nov 2024
This paper takes an important next step in the sampling of extreme low Arctic sea ice extent. Rare event simulation techniques are used to study the influence of initial conditions on the rare event of interest. A similar investigation on the influence of initial conditions on an extreme event, albeit using a very different method (action minimization), is performed in Plotkin et al. 2019, which should be mentioned more explicitly.
My main concern about the paper is that the RES runs are performed only once for every initial condition. Variance estimators are also not applied, so the sampling uncertainty in RES cannot be quantified. Nevertheless, statements about the tails of the RES ensembles are made that could be the result of statistical fluctuations. For example, reference is made to "probabilities of less than 0.001%" and "the trajectory with the most extremely negative sea ice area anomaly obtained with the rare event algorithm". It is apparent in Fig.2 that the most extreme realisations in the RES runs originate from just a few ancestors, so there may be a lot of sampling variability involved here.
This uncertainty should be investigated, or the statements should be weakened, perhaps as an investigation of specific storylines that may not be general.
As the aim of the paper is to investigate the influence of climatic drivers, I wonder why RES sampling is performed from a single initial condition (IC) and not from a set of extreme states, for example all those that are the exceeding +/- 2 sigma levels in Fig. 1 right. This would make statements about the influence of drivers more robust against the choice of one particular IC.
line 41: do not managed -> do not manage
line 110: less than 0.001% -> a probability of less than 0.001%
Fig. 1: the spacing in the x-axis label of "delta pan-Arctic sea ice area" makes the hyphen look like a minus sign.
line 313: in what unit is the computational cost expressed?
Fig. 2: The text says simulations are performed until September, but plots continue until October.
Table 2: "control" is used to refer to two different types of runs. Could these be differentiated?
Line 496: The most extreme negative sea ice anomaly in the RES ensemble is dependent on the RES ensemble size, hence quantitative statements about the % of impact are not very meaningful. A high percentile would be more informative here.
Line 635: minimizing -> minimizationCitation: https://doi.org/10.5194/egusphere-2024-3082-RC3
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 129 | 51 | 27 | 207 | 40 | 7 | 3 |
- HTML: 129
- PDF: 51
- XML: 27
- Total: 207
- Supplement: 40
- BibTeX: 7
- EndNote: 3
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1