SROCC sea level projections, expert judgement and deep uncertainty

Last update: 11/10/2019

SROCC compared to AR5

This month I would like to give my opinion about the sea level projections of the Special Report on Ocean and Cryosphere in a Changing climate (SROCC). Sea level is just one chapter of the report and future projections of mean sea level is just a small part of that chapter. This post is my own opinion of the SROCC sea level projections, it is not the view of KNMI.

Making sea level projections is a task for which IPCC is generally uncomfortable. It is not just about reviewing the literature and assessing the strength and weaknesses of different ideas, IPCC reports produce projections which have tremendous authority and act as an anchor for following sea level projections. IPCC does not just review published sea level projections, it develops its own projections with for example the analysis of the new CMIP5 global climate model simulations and a new method to add sea level contributors together while propagating the uncertainty in AR5.

In this SROCC report, as in previous IPCC reports, the flagship projections that end up in the Summary for Policy Makers and feed the sea level adaptation section of the report are developed with the bottom-up method. This is based on quantifying the main sea level contributors: thermal expansion, glaciers, the Antarctic and Greenland ice sheets, and ground water, and adding them up to give a total sea level projection.

The decision of the SROCC report was to use the same quantification of the contributors as in AR5 except for the Antarctic ice sheet dynamics. While for the AR5 projections the Antarctic dynamics contribution was estimated using a linear extrapolation of the observed ice discharge (Little et al. 2013) this report decided that now some process-based numerical ice sheet models are adequate for the purpose of making sea level projections. This leads to results that are very similar to the AR5 in 2100 except for the RCP8.5 scenario for which the contribution of Antarctica is tripled from a median of 4 cm to 12 cm leading to an increase of total sea level by around 10 cm with a median of 84 cm and a likely range (17 to 83rd percentile) of 61 to 110 cm.

The difference between SROCC and AR5 is larger for longer time scales. While AR5 authors didn’t know if Antarctica would loose or accumulate mass in 2300 now the likely range of Antarctic mass loss in SROCC is 7 to 37 cm for RCP2.6 and 60 to 289 cm for RCP8.5. That is a big change and it results in much higher long term sea level projections.

A bold decision

One important decision that I was waiting for was which weight SROCC would give to the model that includes Marine Ice Cliff Instability (MICI)? In scientific discussions that took place between AR5 and SROCC, the decision was to discard DeConto and Pollard (2016) but to include in the model mix new simulations that DeConto and Pollard have performed with delayed hydrofracturing and faster ice cliff retreat which lead to less sea level rise in the 21st century. With that approach the upper bound of the likely range of RCP8.5 was 133 cm, much more than the 110 that ended up being published. Unfortunately the lead author, Rob DeConto, was so busy working on the SROCC report that he didn’t have time to finish the study so it couldn’t be used. In the absence of that study SROCC authors decided to completely discarded MICI. The projections rely on the assumption that MICI will not happen before 2100. The text says that it might happen after 2100 but it is also not included in the long term projections because of “deep uncertainty”.

Making sea level projections assuming that no ice shelf is going to break neither because of hydrofracturing nor because of ocean-driven break up is a bold decision. Especially since it already happened on the Antarctic Peninsula. Here is the justification for discarding hydrofracturing:

“The climate forcing used by DeConto and Pollard (2016) simulates the appearance of extensive surface meltwater several decades earlier than indicated by other CMIP5 climate simulations (Trusel et al., 2015). Because their model physics are sensitive to melt water through hydrofracturing, this makes the timing and magnitude of their simulated ice loss too uncertain to include in SROCC sea-level projections. However, their results do demonstrate the potential for brittle ice-sheet processes not considered by AR5 to exert a strong influence on future rates of GMSL rise and the possibility that GMSL beyond 2100 could be considerably higher than the likely range projected by models that do not include these processes. ” (SROCC

So the decision not to include MICI in the projections relies on one paper by Trusel et al. (2015). As a quick reminder Trussel et al. (2015) is a model study in which climate model temperatures are used to project the future surface melt of the ice shelves. There is a step between surface temperature and surface melt, for that the regional atmospheric/snow/firn model RACMO2 is used. That study is very good, I got very inspired by it, however it relies on a few weak hypothesis:

  1. The RACMO2 model present day surface temperature is the truth, it has no bias compared to the real word.
  2. The relation between temperature and surface melt in RACMO also applies to the real world.
  3. The divergence between five global climate models is enough to cover the uncertainty in the future temperature at the ice shelves.

You see my point, I don’t see why one would reject a model study on the basis that it doesn’t agree with another model study.

As to the decision not to include ocean-driven break up, it is simple, no numerical model has so far managed to include this process so once the decision is made to base the projections on models that process is out and it is not even discussed in the report.

An argument I have read to support the choice of the SROCC authors, the argument is not used in the report itself, is that since SROCC only makes likely range projections the fact that they do not include hydrofracturing nor ocean break-up in their projections only means that the probability of these to happen is less than 17%, not that it is 0. But that argument is mathematically incorrect. What happens outside the likely range of Antarctic future contribution has an impact on the likely range of total sea level contribution. It would only be otherwise if the uncertainties of the contributors were assumed to have a rank correlation of 1. Which is not physical and not done in the report.

So what could have been done better? I think there are two ways to avoid the usual flaw of IPCC projections that give likely range conditional on some processes not being important (whole dynamics in AR4, MISI in AR5, MICI in SROCC). First, if one really needs to quantify a probability then do not use models but structured expert judgment. Ice sheet models are not ready to make projections. Second, if a probability is not necessary then embrace the deep uncertainty and provide a few scenarios, which are representing plausible futures, as it is done for the RCPs.

Structured expert judgment

SROCC’s authors were very lucky because a structured expert judgement study from Bamber et al. (2019) was published just on time to be used in the report. It shows that experts are expecting more melt from the ice sheets than SROCC’s projections and with more uncertainty. Using these ice sheet mass loss from structured expert judgment they end up with a median sea level of 110 cm and a likely range of 79 to 174 cm, compared to 84 cm (61-110) for SROCC RCP8.5, for a temperature scenario that is a little warmer than the RCP8.5 (2081–2100 global mean warming of +4.5 °C compared with a median of +4.3 °C for RCP8.5).

Why wasn’t the structured expert judgment used for the SROCC sea level projections? I don’t know, it is not justified in the report. Just this sentence:

“The expert elicitation approach (Bamber et al., 2018) suggests considerably higher values for total SLR for RCP2.6, RCP4.5 and RCP8.5 than provided in Table 4.3.” (SROCC,

Also interesting to note that the authors of AR5 decided to also disregard the previous structured expert judgement from Bamber and Aspinall (2012). Different authors, same decision.

Deep uncertainty

If the authors of the SROCC report were not comfortable with quantifying the uncertainty using expert judgment they could have restrained from providing a single likely range and instead provide a few plausible futures. For example, a scenario without MICI and one with MICI as was done by Edwards et al. (2019). Then comment that in their opinion the MICI scenario has a small but unquantifiable probability to significantly influence sea level projections this century.

Discerning experts

The best discussion I have seen of the repeated reluctancy of the IPCC to engage with ice sheet uncertainties can be found in Discerning Experts, a new book from Michael Oppenheimer, Naomie Oreskes et al. Here is an interesting quote:

“The combination of these three factors—the push for univocality, the belief that conservatism is socially and politically protective, and the reluctance to make estimates at all when the available data are contradictory—can lead to “least common denominator'' results—minimalist conclusions that are weak or incomplete.

Moreover, if consensus is viewed as a requirement, scientists may avoid discussing tricky issues that engender controversy (but might still be important), or exclude certain experts whose opinions are known to be “controversial” (but may nevertheless have pertinent expertise). They may also consciously or unconsciously pull back from reporting on extreme outcomes. (Elsewhere we have labeled this tendency "erring on the side of least drama.”) In short, the push for agreement and caution may undermine other important goals, including inclusivity, accuracy and comprehension.”

I will talk more about sea level projections and deep uncertainty in a following post since I am now involved in a European project to use storylines to investigate the REmote Climate Effects and their Impact on European sustainability, Policy and Trade (RECEIPT) and I have a talk on “Deep uncertainty in the KNMI climate projections” at the Decision Making Under Deep Uncertainty (DMDU) conference next month in Delft.


Bamber, J. L., & Aspinall, W. P. (2013). An expert judgement assessment of future sea level rise from the ice sheets. Nature Climate Change, 2(12), 1–4.

Bamber, J. L., Oppenheimer, M., Kopp, R. E., Aspinall, W. P., & Cooke, R. M. (2019). Ice sheet contributions to future sea-level rise from structured expert judgment. Proceedings of the National Academy of Sciences, 116(23), 11195–11200.

Deconto, R. M., & Pollard, D. (2016). Contribution of Antarctica to past and future sea-level rise. Nature, 531(7596), 591–597.

Edwards, T. L., Brandon, M. A., Durand, G., Edwards, N. R., Golledge, N. R., Holden, P. B., … Wernecke, A. (2019). Revisiting Antarctic ice loss due to marine ice-cliff instability. Nature, 566(7742), 58–64.

Little, C. M., Urban, N. M., & Oppenheimer, M. (2013). Probabilistic framework for assessing the ice sheet contribution to sea level change. Proceedings of the National Academy of Science, 110, 3264–3269.

Trusel, L. D., Frey, K. E., Das, S. B., Karnauskas, K. B., Kuipers Munneke, P., van Meijgaard, E., & van den Broeke, M. R. (2015). Divergent trajectories of Antarctic surface melt under two twenty-first-century climate scenarios. Nature Geoscience, 8(12), 927–932.


  • 11-10-2019: Found a mistake in the numbers for Antarctic contribution in 2300, I used the total sea level instead of Antarctic only. Corrected in this new version.