Predictive vs Probabilistic Climate Risk

The architecture capital markets are quietly building

Something interesting is happening in climate risk right now!

AI-driven predictive models are moving incredibly fast, with high-resolution wildfire modelling, dynamic flood signals and asset-level forecasts updating almost in real time, and often forecasting near-term events with remarkable accuracy. It’s technology that genuinely wasn’t available a few years ago.

And yet, when the conversation turns to capital, the basics haven’t shifted nearly as much.

I’m often asked whether predictive AI is replacing probabilistic catastrophe models. It’s a fair question, but what I’m seeing in practice is much more nuanced.

For me, the debate isn’t really predictive versus probabilistic. It’s about matching the right modelling approach to the right financial decision, without losing sight of uncertainty.

When I speak with underwriting or operational teams, the focus is naturally predictive. The questions are practical and immediate. Which assets are most exposed this season? Has the risk profile shifted since last quarter? What’s changed on the ground?

These are questions of visibility and action. Predictive AI is undoubtedly transformative here.

But when the discussion moves to capital committees or risk oversight, the tone shifts. I’ve seen it happen when someone quietly asks, ‘How confident are we in that number?’ From there, the questions change. It’s no longer about the forecast itself. It’s about the distribution of loss over 20 or 30 years, how sensitive the outcome is to different climate pathways, what the tail really looks like, and whether the methodology stands up to governance scrutiny.

At that point, it isn’t a forecasting conversation anymore. It’s a probabilistic one.

Over the past year, across research, investor forums and supervisory commentary, I see one recurring theme -  deep uncertainty isn’t a modelling flaw, it’s a structural feature of climate risk.

Financial risk is about tails, not averages. Capital buffers and stress tests depend on distributions, not single forward views. Governance expectations are rising too. Deterministic outputs that imply precision without quantified uncertainty are increasingly uncomfortable for boards.

And, climate is non-stationary. AI models trained on historical data are powerful, but the future doesn’t behave like the past. Extrapolation without uncertainty bounds can create a dangerous sense of confidence.

None of this diminishes the importance of predictive systems. Actually, quite the opposite. They’re transforming real-time monitoring, event response, asset-level downscaling and the speed at which institutions can act. In fast-moving perils, that shift really matters.

But when it comes to capital allocation, pricing, solvency and long-horizon portfolio exposure, probability distributions remain the currency of decision-making.

What feels most interesting now is that the market is maturing beyond a binary debate. It’s no longer probabilistic versus predictive. It’s more like deterministic predictive outputs versus probabilistic predictive systems.

The most sophisticated institutions I see aren’t choosing sides. They’re designing unified risk stacks. Probabilistic frameworks anchor capital strategy and governance. Predictive intelligence refines underwriting and real-time exposure management. That feels much less like competition and more like architecture.

So, for me the takeaway is simple.

Predictive models improve visibility. Probabilistic models provide rigour. The future of physical climate risk isn’t about picking a methodology. It’s about combining resolution and defensibility, speed and uncertainty.

Because in capital markets, precision without quantified uncertainty doesn’t reduce risk - it redistributes it.

A note to the Women in Climate community

As climate intelligence evolves from probabilistic hindsight to predictive foresight, the real question is not just technical. It is about judgement.

Who decides which models are trusted?
Who challenges assumptions?
Who asks whether we are predicting the right things, in the right way?

Many of the women in this community are working at exactly that intersection, between science, strategy and capital. The shift from probabilistic to predictive thinking is not just a modelling evolution. It is a leadership one.

And leadership in this space has never mattered more.