Climate Risk Data Quality Is a Market Problem, Not Just a Methods Problem

Introduction

Most discussion of climate risk data quality treats it as a technical problem: are the models good enough, is the downscaling sound, is the vendor’s methodology documented. Those questions matter — we’ve written about them directly in Why Climate Risk Vendors Disagree. But we think they sit on top of a more basic structural problem that gets skipped over: who can afford to know their climate risk accurately, and who can’t. That’s not a methods question. It’s a market-structure question, and we think it deserves to be treated as one.

This piece lays out the chain of reasoning that gets us there — from public data’s resolution limits, to the market that has emerged to fill the resulting gap, to what we think that market’s current shape does to climate justice, and finally to what we think should actually be challenged, rather than quietly accepted as the cost of getting better data.

Public Data Is Abundant — and Coarse by Design

A large amount of the climate and earth-observation data the world runs on is publicly funded and openly available. NASA and ESA (through the EU-funded Copernicus programme) produce and release enormous volumes of satellite and reanalysis data at no cost to the user. ERA5, the widely used global atmospheric reanalysis produced by the Copernicus Climate Change Service at ECMWF, provides hourly data on a 31 km grid. The CMIP6 global climate models that underpin the IPCC’s assessment reports run at resolutions typically between roughly 0.9° and 2.5° — on the order of 100–250 km per grid cell, depending on latitude. Even purpose-built, peer-reviewed downscaling efforts that refine CMIP6 output against the higher-resolution ERA5-Land reanalysis only get as far as a 0.1° grid, on the order of 10 km.

That is genuinely useful data for understanding regional and global climate dynamics. It is nowhere close to sufficient for the question a bank, insurer, or property owner actually wants answered: what’s the flood or heat risk to this specific building, on this specific street. A single 10 km grid cell can contain an entire small city, with wildly different exposure from one side to the other. Public data, in other words, is not low-quality — it’s simply operating at a different resolution than asset-level decisions require, by design and for good scientific reasons.

The Resolution Gap Creates a Market

That gap between what public data can resolve and what asset-level decision-makers want doesn’t stay empty. It gets filled — and the actors with both the strongest incentive and the deepest pockets to fill it are exactly the ones GARP’s benchmarking study surveyed: well-funded banks, insurers, and portfolio managers who need building-level or even sub-building-level risk estimates to price loans, set premiums, and meet emerging regulatory disclosure requirements. That demand has produced a genuine, active commercial market of physical climate risk data vendors — the same market whose thirteen benchmarked providers, per our earlier piece, can look at the same 100 properties and return meaningfully different answers.

We want to be precise about our hypothesis here, because it’s a hypothesis, not a settled finding: as this market has grown to close the resolution gap — frequently using proprietary downscaling techniques and increasingly AI-driven methods to translate coarse public inputs into fine-grained, asset-level outputs — we think there is a real risk that data quality has not kept pace with resolution. Going from a 31 km reanalysis grid cell to a single-building estimate is a large extrapolation, and the commercial incentive to offer building-level precision doesn’t automatically come with the scientific rigor needed to justify that precision. The vendor disagreement documented in the GARP study — a 1,507 km geocoding error, wildly divergent flood-depth estimates for identical properties — is at least consistent with that hypothesis, even if it doesn’t prove it outright.

The Democratization Problem

Here is where we think the conversation usually stops short. If the organizations capable of producing genuinely high-quality, fine-grained climate risk data are primarily well-resourced commercial vendors — companies that can afford the computational infrastructure, the proprietary datasets, and increasingly the AI/ML expertise needed to do sophisticated downscaling — then high-quality climate risk data becomes, in practice, a product that only well-capitalized buyers can access.

That has a specific and uncomfortable distributional consequence. The regions and institutions most exposed to physical climate risk, and often least responsible for the emissions driving it, are frequently the ones with the least ability to pay for commercial-grade risk data. A city government in a lower-income country trying to plan flood defenses, or a smallholder agricultural cooperative trying to understand drought exposure, is far less likely to be able to afford the same fine-grained commercial dataset that a global reinsurer uses to price its own book. Meanwhile the public data both groups can access equally — ERA5, CMIP6 — is exactly the data that’s too coarse to answer the asset-level question either of them actually has.

This is a climate justice problem in a very direct sense: the ability to know your own physical climate risk with any precision is being distributed by ability to pay, layered on top of a crisis whose harms are already distributed unfairly by history and geography. Good climate risk data increasingly functions like other forms of climate adaptation capacity — available in proportion to existing wealth, not in proportion to need.

When It’s Not Even More Accurate, the Problem Compounds

The situation gets worse, not better, when the commercial data closing this access gap isn’t reliably more accurate than what it’s replacing. If paying for proprietary, fine-grained climate risk data reliably bought genuinely superior accuracy, the justice problem would still exist, but it would at least be a problem of unequal access to something that works. The GARP benchmarking findings complicate even that: vendor disagreement was wide across every peril studied, driven by identifiable methodological inconsistencies (downscaling technique, geocoding practice, inconsistent hazard metrics) rather than an irreducible feature of the underlying physics. In other words, the premium being paid for fine-grained commercial data does not obviously and consistently buy proportionally higher reliability — sometimes it buys false precision instead, wrapped in a more expensive and more exclusive package.

That combination — unequal access and uncertain reliability — is, in our view, the real climate risk data quality problem. Neither half of it is adequately addressed by only talking about methodology, or only talking about access. They need to be treated as the same problem.

What We Think Should Actually Be Challenged

We don’t think the answer is to challenge only the technical methods vendors use, while leaving the underlying funding structures and market model unquestioned. We think all three need to be challenged together:

The funding model. Public investment (NASA, ESA/Copernicus) produces the foundational data everyone builds on, but the value-added, asset-level layer that decision-makers actually need is being built almost entirely by private capital seeking a commercial return. That’s not inherently illegitimate, but it means the priorities of that layer — which regions get modeled first, at what resolution, for which perils — are set by where paying customers are, not by where climate risk is highest or where the need for accurate information is greatest.
The underlying methods. This is where our existing work on vendor evaluation and on decision making under deep uncertainty connects directly: methods that produce a single confident-looking asset-level number from a 31 km public data foundation deserve real scrutiny about what that precision is actually built on, and what its honest uncertainty range looks like.
The market model itself. Even with better methods and more transparent vendors, a market that serves whoever can pay will keep producing the access gap described above. That’s a structural feature of the market, not a bug any individual vendor can fix by improving their own product.

Grounds for Cautious Optimism

We want to be honest that we don’t have a general solution available today, and we’re wary of presenting one as though we did. But there are two developments that genuinely give us reason for optimism, and we think both are worth actively pursuing rather than waiting for.

Moving away from single scores helps, even before the market changes. The same argument we make in Beyond the Single Score applies directly here: if the field moves away from presenting asset-level climate risk as one confident number, and toward presenting the actual range of plausible outcomes and the assumptions behind them, that alone makes the current market’s outputs more honest and more useful, independent of who can afford them. Realistic uncertainty is a prerequisite for realistic pricing and realistic public debate about who needs help.

Cheaper, more efficient models could close the access gap from the technology side. The rapid trend toward smaller, more computationally efficient AI models — capable of doing more with less infrastructure and lower cost — has real potential to lower the barrier to producing fine-grained climate risk data, if that potential is deliberately aimed at broad access rather than purely at commercial differentiation. We’re not claiming this will happen automatically, or that it’s already happening at meaningful scale. It’s a hypothesis about where technical progress could be directed, not a prediction that it will be.

Conclusion

Climate risk data quality looks, at first glance, like a question of whose model is more accurate. We think that framing is incomplete in a way that matters. The deeper story is that public, openly available climate data is coarse by design, asset-level decisions need far more resolution than that, and the market that has grown to close that gap is currently organized around ability to pay rather than around climate need — with data quality problems inside that market compounding, rather than offsetting, the resulting access gap. Challenging vendor methodology alone treats a market-structure problem as a technical one. We think the funding model, the underlying methods, and the market model all need to be challenged together, even though — and we want to be clear about this — no general solution is available today. What we can say with more confidence is where we think progress is possible: moving the field away from false-precision single scores, and directing the current wave of smaller, cheaper AI models toward broad access to good climate risk data rather than only toward another round of proprietary, exclusive products.