The recent analysis by the Global Association of Risk Professionals (GARP) showed how much climate risk forecasts vary between providers. To some, that looks like failure; if models cannot agree, how can boards, investors or regulators rely on them? Yet the variation is not surprising once you consider what these models are being asked to do. The problem is not that the models are inconsistent, but that they are often expected to do the wrong job. Climate risk is inherently complex, and climate risk models are simplifications of that complexity. Their value lies not in their uniformity, but in their ability to inform judgement and strengthen decision making.

What do you mean by climate risk?

Many people think climate risk modelling is about prediction. In truth, it is about supporting better business decisions by turning complex climate risk data into usable guidance. Would you trust a weather forecast that predicted rain at 3pm in 54 days? Me neither. But would you value an analysis that ranked your 500 international supplier locations by their long-term exposure to heavy rain? Quite possibly. The first is false precision; the second is practical risk management. Good climate risk analysis helps decision makers navigate uncertainty by showing relative exposure and probable outcomes, not certainties.

Climate risk modelling is not about predicting exactly when or where a storm, flood or drought will occur. Nor is it only about measurement; listing metrics without context tells us little. What it is about is decision support. And prioritisation is the most practical expression of that: identifying which assets, locations or operations are most exposed to the risks that matter most for the business. The aim is not to predict next year’s flood losses, but to understand where the greatest sensitivities lie and how they could be reduced.

People often think climate models can predict specific events. They imagine a tool that can tell them when the next flood will hit or which regions will be drought-stricken. That is not what they do. Hazard models estimate likelihoods; they explore possible futures under different assumptions. The science behind them is sound, but they remain scenario-based representations of uncertainty, not forecasts of specific outcomes.

It is like betting on horses. If you asked me to tell you which horse will win tomorrow’s race, I could not. But if you asked me to study their performance histories, the course conditions and their training records, I could rank their chances. That is what hazard modelling does; it ranks relative likelihoods, it does not forecast the winner. Expecting a model to tell you the winner is like expecting the bookmaker to know the outcome before the race begins.

Climate risks depend crucially on location

The GARP study also highlights another source of variation that is often overlooked: geolocation. Many of the climate risk analyses compared in the study may be scientifically robust, but they sometimes refer to locations hundreds of kilometres apart. Climate models are built on grids that vary in resolution One model might represent a 100-kilometre area, another 10. If those grids are loosely mapped to real assets, a risk score for a logistics hub in northern Italy might actually describe conditions centred halfway across the Alps. The model itself is not wrong, it is simply being applied at the wrong scale. A coarse global model can show regional trends very well, but it cannot reliably describe the risk to a specific site. This matters because the precision of a model should match its purpose. The question is not only what is the model predicting, but for where and for what kind of decision.

When users expect a climate risk model to deliver exact outcomes, they are seeking a level of precision the science does not offer. The danger is not just disappointment, it is maladaptation. Acting on a false sense of certainty instead of investing where the evidence is strongest. And when providers encourage that belief, when they imply that hazard models can identify the race winners and losers of future climate events, they risk breaking the trust that the entire field depends on. Clients quickly learn the limits of prediction, and once trust is lost, the value of everyone’s work is diminished. We have a responsibility, as scientists and practitioners, to educate rather than over-promise if we are to serve the market responsibly and maintain the credibility of the discipline.

To map climate risk properly, we need to combine two kinds of input. First, the hazards themselves: the future likelihood and intensity of floods, heatwaves and storms. Second, the vulnerabilities: how exposed and sensitive assets or systems are to those events. Effective climate risk assessment overlays these, but only one of them can be measured directly. Vulnerability can be inspected and tested. You can analyse an asset today, assess its materials, its elevation, its dependence on water or cooling, and its local ecosystem context. Hazards, by contrast, are projections; they depend on emissions scenarios, model ensembles and assumptions about adaptation measures that may never happen.

This means the most reliable route to understanding physical climate risk starts with the world as it is, not the world as we imagine it will be. Reliable data on vulnerabilities enables sound prioritisation and adaptation planning. By describing how assets, supply chains and ecosystems are configured today, we can build a much firmer foundation for understanding how they might respond to future stress. That makes the resulting climate risk assessments both more transparent and more defensible.

refinq leads with vulnerability measurement

At refinq, we employ the best available hazard models with strong academic foundations. But our focus is on mapping vulnerabilities at hyper-granular scale. We quantify how infrastructure, land use and ecosystems interact, location by location. This allows us to show not only where hazards may strike, but how severely each asset could be affected and how this could translate into operational or financial loss.

That is what makes our physical climate risk product industry leading. We translate the science of hazard and vulnerability into asset-level exposure, financial outcomes and adaptation options that can be costed and prioritised. This helps clients see where interventions will have the greatest effect. Resilience begins with understanding your starting point, not with trying to forecast an unknowable future.

So yes, climate risk modelling is about prioritisation, but it is a deeper kind of prioritisation. It is about understanding where exposure is highest and why, and about enabling decision makers to focus effort and capital where it will have the greatest long-term effect. Climate risk models cannot eliminate uncertainty, but they can make uncertainty intelligible and actionable.

A good model does not tell you what will happen next year. It tells you which assets or locations are most likely to struggle under plausible future stress, and how you can strengthen them now. That is how science supports strategy.

The GARP study reminds us that variation between models is not failure. It reflects different questions being asked and different aspects of uncertainty being explored. The most useful models are not those that claim precision about the future, but those that make today’s decisions clearer.

The more we focus on understanding vulnerabilities, and how assets interact with their surroundings, the better we can prioritise meaningful action on resilience. The task for all of us is to help organisations see climate risk modelling for what it truly is: not an exercise in prediction, but a discipline in understanding.

References

Global Association of Risk Professionals (2025). Comparing Climate Risk Forecasts: Why Results Vary Across Providers.

Auld, S. (2025). Sustainability in Plain English: Climate Risk.