Feeling the heat? Preparing for the certainty of an uncertain future

11 November 2026

By Juliane-Paloma Bierich, Sarah Kaber, Katia Mastrodomenico, Guan Schellekens and Manuel Weissenegger

This summer provided us with a strong reminder that the physical effects of climate change and nature degradation are no longer distant concerns. In this context, the ECB brought together experts from supervised banks, public agencies, academia and risk management practitioners in June to discuss the challenges they all face in measuring and managing physical risks.

Over the past few years, banks have made good progress in putting in place the basic architecture needed to manage climate and nature risks. More than 90% of supervised banks now consider physical risks material and have put in place the foundational elements to manage them. The ECB continues to observe good practices throughout its various engagements, and welcomes the diverse initiatives taken by the industry that are helping to address common challenges.

ECB assessments to date have covered physical risks, which have formed an integral part of the strategy it has pursued to build and strengthen climate and nature-related risk. The ECB has observed that severe weaknesses persist in some banks. It has also observed that banks do not always apply available practices across the various business lines and geographies where they operate. That is why the ECB continues to closely monitor banks’ practices, assessing them on a bank-by-bank basis through ongoing supervision. When doing so, the ECB adheres to its usual remediation framework, which may include the use of escalation measures where necessary. Moreover, in performing this work, the ECB has identified a number of ongoing challenges that will require sustained attention going forward.

Physical risks are materialising with increasing frequency, while their impact on banks’ balance sheets has so far appeared relatively muted. Insurance and government support have absorbed some of the losses, but these risk transfer mechanisms are under pressure. The increasing severity and frequency of extreme weather events is prompting insurers to raise premiums, with some even withdrawing from certain markets.^[1] This could widen protection gaps in the EU^[2] and, in turn, affect the bankability and liquidity of certain assets. Governments may also have less capacity to intervene as an “insurer of last resort” due to the limited fiscal space they have available.^[3] Heightened geopolitical risks add to the picture by increasing the likelihood of compound shocks.

At the same time, the medium and long-term effects of physical risks remain difficult to quantify and have probably been underestimated. Forward-looking estimates depend on incomplete data and models with known limitations along every part of the modelling chain. For instance, although their financial impact could be substantial^[4], tipping points are poorly represented in models. Yet some of them have already been reached (warm-water coral reefs), are at immediate risk of being reached (polar ice sheets melting) or may already be approaching (the Amazon rainforest dieback and the collapse of the Atlantic Meridional Overturning Circulation).^[5]

These challenges mean that available estimates of physical risks are likely to significantly understate their full magnitude.

As part of its 2026-28 supervisory priorities, ECB Banking Supervision has announced that it will continue to closely monitor and follow up on outstanding shortcomings and focus more on tackling these challenges. From next year onwards, the ECB will conduct deep dives on banks to assess how they measure and manage physical risks at the client and transaction-level, as well as how they deploy the range of available mitigants to build resilience to these risks. The deep dives will start with a first sample of banks in 2027, which will gradually be expanded in 2028 and 2029. The 2027 EU-wide stress test, where banks will project flood losses before and after insurance and government support, will complement this work.

Moreover, as physical risk events become more frequent and severe, the response of banks to these events has become increasingly important. The ECB will therefore also launch dialogues with banks to discuss their experience in responding to extreme events, including their efforts to support clients, provide liquidity and finance reconstruction.

This closer supervisory focus matters, because assessing banks’ actual exposure to physical risks is far from straightforward and the mechanisms that have until now shielded banks’ exposure are under strain. The remainder of this article explores the main challenges highlighted by the ECB’s dialogue with the banking industry, and the approaches banks are developing to address these risks while acknowledging that a range of issues cannot be solved by individual banks or supervisors alone.

From hazards to banks’ balance sheets: a high-level description of physical risk measurement approaches^[6]

Banks typically measure the impact of physical risks on their clients and portfolios across three dimensions: the probability and intensity of the climate event (hazard), the location of the financed assets (exposure), and the client’s or asset’s capacity to withstand the event (vulnerability).

Banks use asset-level or macro-level modelling approaches, often relying on climate scenarios (e.g. the Intergovernmental Panel on Climate Change (IPCC) scenarios) or macrofinancial scenarios exploring the impacts of different climate futures (e.g. the Network for Greening the Financial System (NGFS) scenarios) to project future hazard patterns and their potential consequences. When integrating hazards into stress-testing models, leading banks use damage functions to consider both the direct and the indirect impacts. The quantified damages are then translated into financial impacts, which is still a challenge for banks.

Banks with even more advanced practices seek to understand the many and complex transmission channels from hazard to financial impact, as illustrated by the recent report from the C-ESG Risk Roundtable, which sets out a wider framework for identifying relevant transmission channels. It highlights how nature-related factors (e.g. water systems, soils, ecosystems, etc.) can either mitigate or amplify physical climate risks, increasing the direct and indirect impacts and thus hitting financial variables even harder.

Discussions from the ECB workshop in June highlighted that banks have been more successful in understanding and measuring the physical risk impact on collateral, where the transmission mechanism is easier to capture. They have plans to make further progress in this area, especially on more complex transmission channels, including business interruption, supply chain disruptions and macroeconomic spillovers. Several workshop participants stressed that the main challenge is not to identify these channels but to measure their effects reliably and consistently to support decisions, while recognising the limits of these measurements. Workshop participants also emphasised that methodologies, scenarios and transmission channels for nature risks remain significantly less mature than those used for physical climate risks.

Understanding past and current levels of impact caused by physical risks^[7]

The direct impact of physical risks on banks’ observed credit and market losses often remains limited compared with the underlying economic damage. Bruegel reports that a large share of losses from extreme weather events is mitigated through insurance coverage and, even when the insurance protection gap is large, through government support mechanisms or ad-hoc government intervention.

As a result, banks’ balance sheets have so far been relatively sheltered from the impact that extreme weather events have on borrowers, collateral values and economic activity, although some signs of deteriorating credit quality and declining asset values are visible. Insurance and government support are therefore key to interpreting the comparatively limited pass-through of damages to banks.

However, the continued effectiveness and availability of insurance and government support cannot be taken for granted. An already large insurance protection gap in the EU and constrained fiscal space mean that, looking ahead, more losses will likely shift on to households, businesses and banks.

Stress from physical risk events is already visible in insurance and reinsurance markets. Research by the Financial Stability Institute (FSI) highlights that after an extreme weather event, reinsurance premiums typically increase as a result of higher demand for reinsurance cover and lower available capital, which affects the ability of insurers to underwrite climate-related risk. Insurability and availability of mitigants also serve a purpose, underpinning lending, investment and asset valuation. Their availability is therefore not only important to mitigate risks, but also to make assets bankable and liquid.

When insurance is not available – or is not sought by banks’ clients – losses can end up being borne by the public sector in the form of an impact on fiscal budgets, as losses from extreme weather events may:

• drive up contingent liabilities, as insurers intervene as the “insurer of last resort”;
• erode the tax base through lower growth for example; and
• slow economic recovery.

This has led to concerns over a re-emergence of the sovereign-bank nexus, where holdings of public debt on banks’ balance sheets mean that fiscal stress and banking stress can reinforce each other. Moreover, evidence also suggests that climate vulnerability increases sovereign borrowing costs, irrespective of an actual event occurring.^[8]

Spotlight: loss tagging and hidden losses

When losses and defaults occur after an event, banks do not always explicitly identify them as physical risk losses. For example, a hotel affected by a severe storm may subsequently experience financial distress due to a reduction in tourism in the region. It may then default on its loan obligations at a later date. However, instead of being attributed to the hazard that actually occurred, the loss may ultimately be recorded as a credit event linked to weak profitability, high leverage or broader economic conditions. This challenge is probably greater for slow-onset events, where the impact materialises gradually and less visibly (e.g. sea-level rise affecting a coastal hotel).

And yet, loss tagging can help banks better understand actual impacts, identify vulnerable sectors and counterparties, strengthen the calibration and validation of their models and progressively integrate physical risks into their broader risk management processes. Several institutions are developing approaches that combine internal credit data with information on extreme weather events to identify the potential links between physical hazards and borrower deterioration, which is considered good practice by the ECB.

Forward-looking physical risks: inherent uncertainties

The assumption that underpins conventional risk models, i.e. that the past is representative of the future, does not adequately capture the physical risk dynamics. Physical risks are forward-looking, and wider climate and nature risks are characterised by what is referred to as “radical uncertainty” or similar concepts: a form of uncertainty in which future events, their probabilities and even the relevant outcomes cannot be measured with any meaningful degree of confidence. For example, neither financial institutions nor supervisors know how to map, let alone measure, all the complex interactions between climate systems, ecosystems, societies and financial markets.

Estimation bias: overestimation of physical risks?^[9]

Physical risk modelling is subject to limitations stemming from data gaps (e.g. location data) and methodological challenges (e.g. supply chain impacts not considered, or hazards assessed in isolation that do not capture compound events). Unlike uncertainty, these limitations are not intrinsic to physical risks and must progressively be addressed.

Measuring how hazards could develop in the future is done using modelling techniques based on climate scenarios. The value of this approach lies in the breadth of the projected plausible futures, rather than in predicting a single outcome. While some scenario frameworks incorporate assumptions about socioeconomic development and adaptive capacity, they generally do not explicitly model broader consequences of climate change and nature degradation, such as migration, conflicts, geopolitical instability or health impacts. Hence, (economic) impacts might be more severe than anticipated in any given scenario.

To translate future hazard developments into financial metrics, banks integrate their geolocated exposures into risk models that require access to highly granular data, such as the precise geolocation of assets as well as vulnerability-specific information, which is often not readily available to banks. Such data limitations can significantly distort risk estimates, for example when headquarter locations are used to work out the approximate location of physical assets, or when insurance and adaptation measures are not fully accounted for. For instance, banks might use corporate headquarters as a proxy for the location of their clients’ production sites, but research estimates that this proxy may leave out up to 70% of expected investor losses compared with asset-level information. On the other hand, imprecise information could also lead to overestimating risks. This can happen, for instance, when mitigants are not fully accounted for (e.g. insurance coverage or adaptation measures to protect a house from floods).

While it is challenging – and perhaps impossible – to precisely determine the magnitude of potential estimation biases for each individual asset and bank portfolio, the likelihood that physical risks are underestimated, combined with inherent uncertainty in their projections, raises prudential concerns.

The main sources of uncertainty and estimation bias arise along the entire modelling chain, from projecting climate hazards to estimating economic damage and translating it into financial risk. The issues below are not exhaustive, and the categories necessarily simplify the interactions between climate, nature, the economy and the financial system. Together, however, they explain why estimates can vary widely and why apparently precise results should be interpreted with caution.

Climate models produce a wide range of outcomes because results depend on climate sensitivity, the choice of parameters and the methods used to translate global projections into local effects. Even within one scenario, outcomes can differ substantially. For example, a scenario centred on 2.7°C of warming can also produce warming of 3.5°C, albeit with a lower probability.^[10] This range is compounded by the climate system’s non-linear and chaotic behaviour, internal variability and sensitivity to initial conditions, making deterministic forecasts impossible.

Important mechanisms may also be simplified or omitted. Climate tipping points can trigger abrupt regional changes that diverge from average global trends. For instance, a collapse of the Atlantic Meridional Overturning Circulation could sharply reduce winter temperatures in northern Europe, even as the planet continues to warm. In addition, models often neglect greenhouse gas emissions released naturally as warming intensifies, including emissions associated with wildfires and ecosystem degradation. These feedbacks could add up to 0.6°C to global average temperatures and shorten the time available before warming exceeds 2°C by about one-quarter.^[11]

Modelling hazards separately can further understate losses. Heat and drought may reinforce each other, repeated floods can cause cumulative damage, and a sequence of smaller events may be as damaging as one single headline disaster. An initial hazard can also cascade across interconnected systems, i.e. like when a sea level rise causes flooding that then damages infrastructure and disrupts business activity. Climate and nature interact in both directions: ecosystem degradation can amplify hazards and lower thresholds for systemic change, while healthy natural systems can mitigate risks. For instance, accounting for the Amazon deforestation dramatically lowers climate risk tipping thresholds: combining 22-28% deforestation with stress from land use lowers the critical global warming threshold for systemic collapse from 3.7-4.0°C to just 1.5-1.9°C.^[12] On the other hand, risks could also be overestimated if the mitigating potential of nature is ignored. Moreover, seven of the nine planetary boundaries have already been transgressed^[13], which means that many of the Earth system processes on which stable development depends are already operating outside of their safe range, leaving less room for climate-nature interactions to unfold without compounding losses. As many modelling approaches ignore these feedback loops between nature and climate, their outcomes may lead to either the underestimation or the overestimation of risks.

Economic modelling adds another layer of uncertainty because future emissions and warming depend on political decisions to which no objective probability can be assigned. Climate policies are often represented through simplified assumptions, such as a carbon tax, without fully capturing political trade-offs or the intended reduction in carbon-intensive production. Models also tend to omit broader social and geopolitical consequences, including conflict, migration and the health effects of heatwaves, even though experts increasingly regard these channels as material.

The damage functions used to convert hazards into economic losses are another potential source of bias. They are often applied across broad areas or similar hazard types, calibrated on limited regions, events and historical losses, and not updated frequently. Some focus mainly on gradual changes in average temperature while omitting persistent multi-year effects or changes in precipitation. These simplifications can materially understate economic damage.

Financial models may then miss how these effects reach banks. Indirect impacts such as supply chain disruption, cascading losses and damage to ecosystems can greatly exceed directly reported disaster losses: direct annual losses of about USD 195.7 billion compare with an estimated total burden of nearly USD 2.3 trillion when wider effects are included.^[14] Banks that have incorporated supply chain disruption into their scenarios have identified potentially material capital implications. Models may also fail to anticipate market repricing before a physical event occurs, including sudden changes in sovereign bond yields and asset valuations when investors recognise the consequences of disasters or tipping points.

Risk transfer mechanisms are also difficult to project. Banks have limited methods for assessing how the availability and affordability of insurance will change and they may overstate the protection they have by overlooking deductibles, coverage caps and exclusions. Future government support is equally uncertain because fiscal capacity, compensation arrangements and political willingness can change. At the same time, external data providers can assign radically different physical risk scores to the same asset or borrower, ranging from high exposure to no exposure at all.

Not every limitation necessarily raises estimated risk. Financial models might fail to recognise investments by clients that strengthen their adaptation and resilience, such as flood protection or drought-tolerant crops. Reflecting these measures could lower the estimated credit risk.

Overall, however, omitted climate mechanisms, incomplete risk drivers and economic transmission channels, uncertain market responses and inconsistent data create a genuine possibility that physical risks are underestimated.

From measurement to decision-making: challenges for traditional risk management options for physical risks

To manage risks, banks can traditionally draw on four risk management options: avoiding the risk by declining the underlying activity, mitigating the risk through measures that reduce its likelihood or severity while retaining the exposure, transferring the risk to a third party (e.g. through insurance or securitisation) or accepting the residual risk and handling it through pricing or provisioning.

For physical risks, each option comes with constraints:

• Banks could avoid risks by reducing lending to highly exposed sectors or regions, but this could erode their business models. It may also lead to unintended consequences, such as restricting access to financing for the very same borrowers who need to transition and adapt.
• Adaptation and resilience measures implemented by clients can help mitigate risks, but methodologies to assess them are still relatively new.
• Risk transfer through insurance and government support in the future is under strain and uncertain.
• Modelling uncertainties and potentially unrecognised tail risks suggest that banks may not be able to accurately quantify residual risks they take on in the near term.

During dialogue with the industry, European supervisors learned that actively mitigating the underlying risks by implementing adaptation and resilience measures, including those implemented by clients, is becoming an increasingly important risk management tool for banks. However, they also learned that banks’ ability to assess the effectiveness and financial benefits of the adaptation measures undertaken by their clients remains limited, partly because of the challenges faced, such as insufficient project pipelines and benchmarks of future cashflows. Methodologies to reflect these measures in processes to grant credit are starting to emerge.

Existing and new tools: early industry practices for managing physical risks under uncertainty^[15]

The dialogue with the industry shows that banks increasingly acknowledge that risk management and decision-making practices need to adapt to a world in which important risks may not be fully captured, quantified or understood with confidence. They appear to be reconsidering how they operate in light of the uncertainties and underestimation of risk described above. What matters will not only be to use scenarios or risk modelling, but to build resilience against surprises. Banks indicate that this will require building out existing layers of risk management, such as risk appetite setting or portfolio steering, as well as regularly testing their reaction capabilities to ensure that governance and decision-making structures can respond swiftly when the risk landscape changes. It will also require new approaches, such as acknowledging that climate mitigation is the only measure that addresses the underlying driver of climate-related physical risks.

Banks are expanding their risk management toolkit to support decisions under deep uncertainty. The preliminary and non-exhaustive approaches below combine enhancements to established practices with newer responses. They are presented for information only and do not constitute ECB guidance, regulatory expectations or requirements.

Climate mitigation addresses the root cause of physical risk by directing finance towards emissions reduction and low-carbon activities before hazards translate into economic and financial losses. The same dynamics applies to nature mitigation, where the root cause of physical risks can be addressed by directing finance towards the reduction of nature loss and the restoration of ecosystems, such as activities halting deforestation or reducing pollution. Banks can also integrate resilience measures into credit assessments, so that borrowers’ adaptation investments are reflected in loan pricing, ratings and capital allocation to capture their actual vulnerability.

Portfolio monitoring using heat maps, geospatial analysis and event-trigger frameworks can identify the impact of emerging physical risk and support rapid assessments after an extreme event. Risk appetite statements can complement this by setting qualitative or quantitative limits on concentration on exposures to vulnerable sectors or locations.

Scenario and resilience analysis can test how portfolios and strategies perform across a range of plausible climate pathways tailored to a bank’s vulnerabilities, rather than relying on a single projection. Banks mention that these tools need to be supported by stronger internal capabilities and governance: decision-makers should understand current uncertainties and measurement limitations, while risk, business, finance and strategy functions need to work together so that emerging evidence leads to timely action.

These practices are still at an early stage: banks and the wider financial community are still grappling with the depth of uncertainty surrounding physical risks. As experience builds up, these approaches are likely to mature. What these practices already show is that banks are starting to expand their toolkit, showing innovative capabilities alongside improvements to established approaches: besides traditional modelling and risk management tools, a broader set of risk management tools are increasingly being used to address decision-making under uncertainty.

Looking ahead

The ECB’s work on climate and nature risks since the publication of its Guide on climate and nature risks in 2020 has established these risks as a core element of banks’ risk management frameworks, which banks are required to manage in line with the requirements set out in EU legislation. It has also laid bare the challenges that persist, particularly in relation to physical risks. The next phase of our supervisory work aims to build a deeper understanding of how these issues impact the risk management of individual banks. It also aims to take stock of the measures taken by banks to build long-term resilience and maintain their financing capacity as physical risk impacts increase, with the aim of ensuring progress is made in a timely manner. Similar to its approach to transition planning, the ECB will pursue this goal in a gradual and targeted way, progressively increasing the intensity of the supervisory engagement and supervisory instruments as needed.

The ECB will also continue its dialogue with the industry and other stakeholders on emerging practices, methodological developments and implementation challenges. The progress already visible across the industry is an encouraging sign that it can take the steps required to build robust and forward-looking frameworks to manage these risks. In a world where physical risks may be greater than current estimates suggest, resilience will increasingly depend not only on measuring risks more accurately, but also on being prepared for what cannot be measured with certainty.