Beyond the Single Score: Decision Making Under Deep Uncertainty

Introduction

Three things are true about climate and societal risk at the same time, and each makes the others harder to manage. Climate change makes future conditions inherently uncertain, not merely imprecisely known — we are not looking at a fuzzy version of a knowable number, we are looking at a genuinely open range of possible futures. The underlying science is still actively evolving, with growing evidence for worst-case dynamics such as tipping points that classical, smooth projection methods were never built to capture. And there is a well-documented, widening gap between the spatial and temporal resolution decision-makers want — “what will happen to this coastline, this city, this asset, by this date” — and the resolution that prediction data can actually deliver.

Institutions that need to give guidance regardless have often reached for a familiar shortcut: the single score. A risk index between one and ten. A probability of exceedance. A star rating. The appeal is obvious — a single number is easy to communicate, easy to put in a spreadsheet, easy to compare across assets or regions. The problem is structural, not cosmetic, and it doesn’t go away with a better model. A single number describing a genuinely uncertain future is either dishonest — projecting false precision onto something that cannot honestly be reduced to one figure — or, once it is appropriately hedged with error bars wide enough to be honest, so broad that it becomes useless for the decision it was meant to support. Either way, a point estimate cannot represent a range of futures, and critically, it cannot tell a decision-maker which futures a given plan survives and which ones break it.

There is an alternative — Decision Making under Deep Uncertainty (DMDU) — with two concrete applications to Swiss climate adaptation problems to show what it does differently. It also deserves real scrutiny rather than uncritical adoption, on two counts: its complexity, and the risk that more rigorous modelling quietly becomes less democratic decision-making.

What “Deep Uncertainty” Actually Means

It’s worth being precise about the term, because “uncertainty” is used loosely in a lot of climate communication. Ordinary statistical uncertainty means you know the shape of the distribution of possible outcomes — you might not know exactly what will happen, but you know the odds. Deep uncertainty is different: it describes situations where the decision-makers involved don’t know, or don’t agree on, the probability distributions over future states of the world, the full set of possible futures, or even how different outcomes should be ranked against each other. Climate adaptation planning sits squarely in this second category. Emissions pathways depend on political choices that haven’t been made yet. Tipping-point dynamics are, by definition, poorly characterized by historical data. And what counts as an acceptable trade-off between cost and risk is a value judgment, not a technical fact a model can resolve on its own.

Traditional risk analysis tools — cost-benefit analysis built around a single expected-value forecast, or a risk score built around a central estimate — implicitly assume the first kind of uncertainty. Applied to the second kind, they don’t fail loudly; they fail quietly, by producing a confident-looking answer to a question that doesn’t actually have one.

What DMDU Does Differently

Decision Making under Deep Uncertainty — often implemented in practice through Exploratory Modeling and Analysis (EMA) — starts from the opposite assumption. Instead of trying to predict the single most likely future and plan around it, EMA generates a large ensemble of plausible futures, spanning the genuine range of disagreement about emissions, climate sensitivity, and socioeconomic development, and asks a different question of each candidate policy: does it hold up, and under what specific conditions does it fail?

That reframing changes the whole shape of the output. Instead of “what will happen,” the analysis answers “what should we do, given that we don’t reliably know what will happen, and how would we know if we needed to change course.” The result is not a single recommendation but a set of adaptation pathways — sequenced options with explicit signposts, so that a decision-maker knows in advance which observable conditions (a certain sea-level rise threshold crossed, a certain frequency of extreme events reached) should trigger a switch from one plan to the next. This is sometimes formalized as Dynamic Adaptive Policy Pathways (DAPP), a specific technique for mapping those triggers and transitions.

Two Applications, Same Logic

Two recent applications from a University of Zurich-linked research group make the abstract idea concrete, and both matter to us because they’re grounded, published, peer-reviewed work rather than a hypothetical case study.

Urban Heat Stress in Basel and Zurich

Federer et al. (2025), published in Urban Climate, apply EMA together with the Wet-Bulb Globe Temperature (WBGT) indicator — a measure that captures heat stress on the human body more directly than air temperature alone — to Basel and Zurich. Rather than producing a single most-likely heat projection for each city, the exploratory approach deliberately captures a wide range of uncertainty across emissions pathways, individual physiological vulnerability, and the thresholds at which different adaptation measures kick in. The paper’s authors are explicit that this integrated, human-system-plus-climate approach is meant to support what they call transformative adaptation — decisions large and structural enough that getting the underlying assumptions wrong has real consequences — not simply to describe risk more precisely. They also flag, honestly, that the approach could be strengthened further by incorporating Dynamic Adaptive Policy Pathways directly and by developing better vulnerability-specific thresholds tailored to local conditions, rather than treating “heat stress” as a single uniform category.

Swiss Ski Resorts and Economic Tipping Points

Vaghefi et al. (2021), in Environmental Science & Policy, apply the same underlying logic to a very different sector: winter tourism. The study models six Swiss ski resorts at low, medium, and high elevation, using high-resolution CH2018 climate projections and exploratory modeling to test two adaptation options — snowmaking and diversifying resort activities beyond skiing — against a wide range of future climate realizations. Rather than reporting a single “resorts will lose X% of revenue” figure, the paper identifies the specific physical and economic tipping points at which each adaptation option stops being sufficient, and uses Dynamic Adaptive Policy Pathways to map when a resort would need to switch from one strategy to the next. One of the paper’s more striking findings is structural rather than numerical: elevation turns out to matter more for a resort’s long-term survival than any single adaptation measure — a conclusion that a single-scenario forecast would have been much less likely to surface clearly, because it emerges from comparing outcomes across the ensemble of futures, not from any one of them.

Read together, the two papers share the same underlying structure: don’t rank scenarios by which one is most likely, map which decisions are robust across as many of them as possible, and be explicit about where that robustness breaks down.

What We’re Not Assuming

It would be easy to present DMDU as a clean solution to the single-score problem and stop there. We don’t think that’s honest, and it isn’t how we want to approach this topic. DMDU has a real, recurring critique that deserves equal weight to its technical strengths: the methods are computationally heavy, methodologically dense, and can be genuinely opaque to the people who ultimately have to live with the resulting policy. A sophisticated pathway map that only a modeling team can interpret hasn’t solved the “black box” problem that plagued the single score — it has just relocated it, from a single misleading number to a large, technically inaccessible dataset. Complexity that only specialists can audit is not obviously more democratic than a simple number everyone can misunderstand in the same way; in some respects it may be less democratic, because it makes the gatekeeping less visible.

That critique is part of what we want to actively research, not a footnote to work around later. Two open questions matter to us as much as the modeling technique itself:

Democratization of method. If DMDU-style approaches prove genuinely useful for a given decision, how do we make the underlying scenario logic — not just the final headline recommendation — accessible and legible to the communities and institutions who are affected by the decision, rather than legible only to the specialists who built the model? This is a design question as much as a communication one: it likely requires rethinking what the output of a DMDU analysis looks like, not just adding a plain-language summary on top of an unchanged technical report.
UX for decision paths. Exploratory modeling can easily produce hundreds or thousands of individual scenario runs. The genuinely unsolved problem, in our view, is representation: how do you show a decision pathway — where it holds, where it breaks, and which observable signal should prompt a change — in a way a non-specialist decision-maker can actually use under time pressure, rather than defaulting back to a single reassuring number because that’s the only format they can act on quickly?

We don’t have settled answers to either question yet. We think that’s the honest position to hold, and it’s the reason this is a research topic for us rather than a service we’re claiming to already deliver.

Where This Connects

This sits directly next to our work on climate risk data quality: both start from the same underlying observation, that a single confident-looking number often hides more than it reveals, and that transparent, use-case-appropriate methods matter more than a headline score. The GARP vendor-benchmarking findings — wide dispersion between vendors assessing the same asset — are, in a sense, deep uncertainty showing up empirically in a commercial data product, even before anyone runs a formal DMDU analysis on top of it. Our Climate Risk Assessment Services work from the same premise: that decision support should make its assumptions and its range of plausible outcomes visible, rather than collapsing them into a single conclusion before the client ever sees the range.

Conclusion

The case against the single climate risk score is not that it’s imprecise — some imprecision is unavoidable and even honest. The case against it is that it answers the wrong question, treating a deeply uncertain future as if it were a fuzzy version of a knowable one. Decision Making under Deep Uncertainty, and the Exploratory Modeling and Analysis techniques that implement it, offer a genuinely different structure: instead of picking the most likely future and optimizing for it, test a policy against the full range of plausible futures and report where it holds and where it breaks. The Federer et al. and Vaghefi et al. papers show this is not just an academic proposal — it’s already being applied to real Swiss adaptation problems, from urban heat to alpine tourism, and producing findings (like the outsized importance of elevation for ski resort survival) that a single-scenario approach would likely have missed.

But we don’t think the story ends with “DMDU is better, adopt it.” The same rigor that lets DMDU escape the single-score trap can also make it harder for the people affected by a decision to actually participate in it. Making these methods more democratic — both in how the scenario logic is communicated and in how the resulting decision paths are represented so ordinary decision-makers can use them — is, for us, just as important a research question as the modeling technique itself.

References

Federer, F., Weibel, F., Huggel, C., Vaghefi, A. S., & Muccione, V. (2025). “Exploratory modeling and analysis of adaptation to urban heat stress under climate change in Switzerland.” Urban Climate. DOI: 10.1016/j.uclim.2025.102729
Vaghefi, S. A., Muccione, V., van Ginkel, K., & Haasnoot, M. (2021). “Using Decision Making under Deep Uncertainty (DMDU) approaches to support climate change adaptation of Swiss Ski Resorts.” Environmental Science & Policy. DOI: 10.1016/j.envsci.2021.09.005