By Ron Dembo
● Climate Risk 2.0 is a minimum standard for physical climate risk measurement, not a product or vendor claim.
● It requires five things current models mostly don't do: current data, full physics, full probability distributions, compound-risk modelling, and financial translation.
● The gap is already showing up in losses: a 60% global insurance protection gap, and billion-dollar US weather events now occurring roughly every two weeks, up from about three per year in the 1980s.
● The EU, UK, and Canada are moving fastest on mandatory climate stress-testing; the US rescinded federal climate risk guidance in October 2025.
A decade of investment in NGFS pathways, SSP projections, and integrated assessment models has produced tools built on assumptions increasingly detached from observed reality. Three structural failures define the current generation.
Most providers still calibrate on meteorological forcing data from around 2015. 2024 was the first year on record to exceed 1.5°C above pre-industrial levels — a milestone absent from every model built on that baseline.
The January 2025 Los Angeles wildfires caused $40B in insured losses that pre-2015-calibrated models had not priced. H1 2025 global insured catastrophe losses reached $80 billion— almost double the 2015–2024 average. Non-peak perils drove $166B in 2025 total losses, against a 30-year average of $90B.
Most providers apply a scaled "climate change factor" to historical data instead of running full hydrologic models per scenario. It's cheaper, but systematically inaccurate — it smooths out extremes and produces outputs that conflict across scenarios when bias correction draws on different datasets, in ways buyers and regulators are starting to question.
Standard SSP-based projections only diverge meaningfully after 2050 — unworkable for regulators, insurers, and asset managers operating on 3-to-10-year horizons. The NGFS acknowledged this, releasing its first short-term scenario in May 2025; the field overall has been slower to respond.
● The global insurance protection gap for natural disasters is 60% — most economic losses from climate events are uninsured.
● H1 2025 resulted in $80 billion in global insured catastrophe losses — the second-costliest first half on record, almost double the 2015–2024 average.
● Billion-dollar weather events in the US have risen from about 3 per year in the 1980s to over 20 annually today.
● The proportion of uninsured US households has grown from 5% to 12% since 2019, as insurers pull back from regions their models failed to assess accurately.
These losses aren't simply more assets at risk. They reflect a consistent underestimation of physical risk that better models would have caught years earlier.
The limitations aren't mainly a science problem — climate science is well established. The issues are methodological and commercial: cost-cutting choices made in 2015 that no longer hold up. Climate Risk 2.0 needs to address:
● Hazard scores without uncertainty ranges, hiding tail risk behind a false impression of precision.
● Single-peril models that miss the compound, cascading interactions behind the most severe losses.
● Outputs disconnected from financial metrics — asset value, credit risk, insurability — that institutions can't act on.
● Black-box methodologies that can't be audited, failing supervisory standards now enforced across the US, Canada, and the EU.
● A macroeconomic amplification channel often ignored: a hurricane hitting when banks carry high loan-to-value ratios and thin liquidity buffers can trigger credit cascades well beyond the physical damage itself.
Climate Risk 2.0 isn't a product or a vendor claim — it's a minimum standard spanning five domains.
Up-to-date, consistent meteorological forcing data. Continuously updated, not calibrated a decade back — and the same dataset used to bias-correct both historical and future projections, since mixing datasets across scenarios produces contradictory outputs.
Full physical model runs across every scenario. The climate-change-factor shortcut is removed: every model-scenario pair gets a full simulation, not a scaled estimate — full fire-weather and fuel-moisture modelling for wildfire, dynamical downscaling for cyclones, full urban heat island simulation for heat stress.
Full probabilistic outputs. The fullmultihazard distribution — 10th, 50th, and 90th percentile outcomes across all warming levels — not a median projection, so tail risk stays visible.
Stochastic event modelling. A handful of scenario runs can't represent the full space of plausible outcomes within a warming pathway, especially for rare or compound events with no historical analogue. Climate Risk 2.0 requires large science-based,synthetic catalogues of physically consistent event years, calibrated against observed climate and loss data, to populate the tail of the distribution above.
Short-term scenario capability. The 3-to-5-year blind spot in long-term models is unworkable for financial risk management. Climate Risk 2.0 requires a dedicated short-term layer — annual variability, policy shock transmission, near-term extremes — structurally separate from long-term modelling.
Annual vintage outputs with event attribution. Estimates update at least once a year. For each significant loss event, providers should quantify how its probability and severity shifted relative to a pre-industrial baseline.
Multi-hazard correlation modelling. The most severe losses are usually compound — flood plus power outage, drought plus wildfire, storm surge plus subsidence. Climate Risk 2.0 requires multiple hazards modelled as correlated, not independent.
Infrastructure dependency mapping. Physical risk to one asset doesn't stay contained: a flooded substation can disrupt a data centre, a damaged bridge can cut off a logistics hub. Climate Risk 2.0 requires modelling how risk propagates through these dependencies.
Insurability threshold indicators. Most models price current risk; few show when an asset crosses into uninsurable territory. Climate Risk 2.0 requires a forward curve of that threshold — the transition point banks and asset managers actually need.
Asset value impairment curves. A hazard score alone can't allocate capital, price a loan, or set a premium. Climate Risk 2.0 requires direct conversion of physical risk into expected asset value change over 5-, 10-, 20-, and 30-year horizons.
Credit risk integration. Asset impairment raises default likelihood; collateral deterioration raises loss given default; revenue disruption weakens debt-service capacity. Climate Risk 2.0 requires physical risk to feed credit models directly — loan-level PD uplift, LGD adjustment, portfolio-level concentration by geography and sector.
Regulatory framework alignment. Climate disclosures are mandatory in multiple jurisdictions, with ISSB S2, OSFI B-15, CSRD, and central bank stress tests setting the baseline. Climate Risk 2.0 requires outputs aligned to these frameworks directly.
Explainable model outputs. Every output needs a plain-language explanation of its key drivers — warming level, scenario, physical mechanism, model choice. A risk estimate that can't be explained can't be justified to a board or regulator.
Vintage transparency and comparability. Users need to compare current estimates against previous vintages, to distinguish a climate-signal change from a methodology change.
Open methodology and auditability. Not open-sourcing proprietary algorithms — documenting data vintages, scenarios run, bias correction methods, and core assumptions for independent review.
The table below sets out the core requirements distinguishing Climate Risk 2.0 from the current generation.
| **Capability Area** | **Current Industry Baseline** | **Climate Risk 2.0 Requirement** |
| Data vintage | Meteorological forcing data to \~2015 | Continuously updated through present |
| Physical modelling | Scaled approximation ('climate change factor') | Full model run per scenario & warming level |
| Bias correction | Applied inconsistently across scenarios | Consistent dataset across historical & future |
| Scenario depth | Representative pathways only | Full ensemble across warming levels, supplemented by stochastic event simulation |
| Near-term resolution | Poor resolution before 2050 | Explicit short-term (1–5 year) shock modelling |
| Peril coverage | Primarily flood / hydrology | Multi-peril: flood, wildfire, heat, drought, coastal |
| Compound events | Single-peril, independent treatment | Joint probability of compound & cascading events |
| Financial translation | Hazard scores in isolation | Asset value, credit risk, insurability outputs |
| Uncertainty | Median projections only | Full probabilistic distribution inc. tail risks |
| Transparency | Black-box outputs | Explainable drivers, auditable methodology |
**For insurers and reinsurers: **current-generation tools leave insurers underestimating portfolio risk, short of capital relative to actual exposure, and exposed when modelled forecasts diverge from real losses. Climate Risk 2.0 tools price against current climate signals, catch compound exposures single-peril models miss, and flag the path to insurability before forced retreat.
**For asset managers and institutional investors: **physical climate risk is material across every asset class with tangible exposure — real estate, infrastructure, agriculture, energy, supply chains. Climate Risk 2.0 moves assessment from portfolio screening to asset-level due diligence, surfacing stranded-asset risk ahead of the market and producing auditable disclosures.
**For banks and lenders: **physical risk hits balance sheets through collateral losses, borrower revenue disruption, and geographic concentration. Climate Risk 2.0 lets banks embed physical risk into loan pricing, run the asset-level stress tests regulators require, and produce ISSB S2-aligned disclosures.
**For regulators and policymakers: **the gap between model-reported and loss-revealed risk is itself a financial stability concern. The ECB is furthest along — embedding climate risk into core supervision in January 2026 and issuing its first climate-materiality penalty (ABANCA, November 2025). The Bank of England and OSFI (Canada) have their own frameworks; the US rescinded federal guidance in October 2025, leaving the EU, UK, and Canada as the active jurisdictions on mandatory stress testing.
The gap between modelled risk and actual loss has grown too large to explain as bad luck. It reflects systemic model failure: outdated data, simplified physics, hidden uncertainty, and outputs that don't translate into financial decisions.
Climate Risk 2.0 doesn't require new science — the science exists. It requires the discipline to use current data, run the full model, report the full distribution, model compound risk, and speak the financial language of the institutions that rely on the output. Institutions that adopt it — as buyers or as providers — will be better positioned for the regulatory environment ahead, and better equipped for a climate that has already arrived.
You can download the full paper from here.
A minimum standard for physical climate risk measurement across five domains: current data and full physics, near-term resolution, compound and cascading risk, financial translation, and transparency. Not a product or vendor claim.
Current models calibrate on ~2015 data, use a scaled "climate change factor" instead of full physics, and report a median rather than a full distribution. Climate Risk 2.0 requires continuously updated data, full model runs per scenario, and full probabilistic output including tail risk.
Scenario-based projections show one warming pathway; stochastic modelling generates a much larger set of physically plausible event sequences around it, including combinations with no historical precedent. It's how the full probability distribution above actually gets populated.
2024 was the first year on record to exceed 1.5°C above pre-industrial levels — a shift absent from models calibrated on ~2015 data, making their output an estimate of a climate that no longer exists.
A scaling method that multiplies historical hazard estimates by a scenario-specific factor instead of running a full physical model. Cheaper, but it smooths out extremes and produces inconsistent outputs across scenarios.
Compound risk is multiple hazards amplifying each other — a storm surge combined with subsidence. Cascading risk is damage propagating through infrastructure dependencies, like a flooded substation disrupting a data centre. Most models treat hazards as independent, understating the losses that come from events striking together.
The point where modelled risk exceeds what an insurer can price at any commercially viable premium. Climate Risk 2.0 requires a forward-looking indicator of that threshold, not just current pricing.
Physical risk outputs convert directly into asset value impairment curves over 5-, 10-, 20-, and 30-year horizons, and feed credit models through loan-level PD uplift and LGD adjustment — rather than stopping at a hazard score.
ISSB S2, OSFI B-15 (Canada), CSRD (EU), and central bank stress-testing frameworks including the ECB and Bank of England. The EU, UK, and Canada are currently most active on mandatory stress testing.
No — it's a field-wide minimum standard, not a proprietary framework. RiskThinking.AI built its platform to meet it, but any provider or institution can use the same benchmark.