How climate risk disclosures explain methods and limitations

Climate disclosures can use similar warming labels while answering different questions. The important comparison is between the analytical questions: what was assessed, over which period, by what method and with which exclusions? This study examines two UK public-department disclosures and then uses an invented example to show why those details change what a result can support.

The sources form a purposive pair, chosen for visible differences in scope and explanatory detail. This is a reading of disclosed methods, not a representative survey, a validation of climate models or a performance ranking. The fictional exercise below is an independent teaching example and supplies no missing facts about either department.

A proportionate operational analysis

The Government Actuary’s Department 2024–25 report describes physical-risk analysis using 2°C and 4°C scenarios, supported by UK Climate Projections information. It considers the 2030s, 2050s and 2080s; the displayed comparison concerns two operational risks in the 2080s: heat affecting homeworking and storms affecting IT or travel. Transition risks were excluded from this analysis. The department reports strong physical-risk resilience and says no action was taken directly because of the analysis. That is its stated conclusion within the disclosed scope, not an independently validated assessment.

An asset metric whose footnote changes the reading

The Ministry of Justice’s 2024–25 performance report describes a five-yearly physical climate-risk assessment using UKCP18. Its table 5 reports 65% and 73% of assets at high or very high overheating risk under the respective 2°C and 4°C labels. Crucially, footnotes 69 and 70 specify RCP2.6 and RCP8.5 calculations for 2020–2039. Its stated next steps include exploring transition scenarios and quantifying financial exposure. These prospective tasks should retain that status when the report is summarised.

The reported percentages differ by 8 percentage points: 73 minus 65. This compares two scenario-labelled results for the stated period. It is not an observed eight-point deterioration between reporting years, an event probability or a loss estimate.

The comparison stops before a resilience league table

Our reading of the pair is that a long-horizon operational judgement and an asset-level risk percentage answer different questions. Their shared warming labels cannot align the horizons, assessed populations or output types. The physical-risk analysis also cannot be expanded silently to cover transition risk.

A useful comparison therefore preserves the description attached to each result. A scenario name is only one part of that description. Keep the modelled period, geographical coverage, asset or activity boundary, variable examined and form of output with the conclusion. Without those links, a neat side-by-side chart can make unlike evidence look commensurable.

Distinguish three dates in the reading record: the reporting year, the period represented by the climate data and the horizon of the business decision. They can differ without being erroneous. The reader’s task is to understand the relationship, including whether the analysis concerns a time slice, an end-century condition or an asset’s expected operating life.

A fictional comparison that fails the matching test

All organisations, records and figures in this exercise are invented. Report A screens twenty locations for a selected hazard in the 2030s and flags eight for further investigation. Its threshold is a screening rule, and the exercise has not assessed building vulnerability, adaptation measures or financial consequences. Eight divided by twenty gives 40% of the screened locations flagged.

Report B examines two operational dependencies under a selected 2080s scenario. A workshop assigns one a high disruption rating and the other a moderate rating, using its own stated criteria. The records contain no location-level count, calibrated event probability or financial model. Neither record claims to reproduce a method used by either public department in the source pair.

It would be misleading to set 40% beside “one high” as if they measured the same property. The first output is a share of screened locations. The second is a qualitative judgement about a dependency. Assigning the words high and moderate arbitrary numbers would not create a shared measurement scale or a justified average.

The correct next step depends on the decision. To compare location exposure, ask for aligned location populations, hazards, periods and thresholds. To compare continuity implications, investigate dependency vulnerability, alternative arrangements and service consequences. To compare financial exposure, define the valuation method and assumptions. None of those conversions can be supplied merely by renaming the output.

If the eight flagged locations are the only ones examined in detail, the remaining twelve should retain their screening status. They are not automatically confirmed safe. Conversely, a flag requests further work; it does not establish that disruption will occur. The distinction protects the interpretation of both sides of the screening boundary.

Original fictional exercise, unrelated to the two departments’ results. Eight of twenty screened locations equals 40%. A separate qualitative dependency assessment has a different period, population and output, so the two results cannot form a performance ranking.

Read limitations as part of the result

A limitation is useful when it tells the reader which inference is unavailable. An excluded risk class, an unassessed location, a qualitative output and a planned future analysis create different boundaries. Recording them together as “uncertainty” loses the practical question: which additional evidence would be needed before this particular conclusion could be extended?

An explicit exclusion and silence in a summary also deserve different descriptions. Write “excluded from the stated analysis” when the source says so. Where information was not found in the passages reviewed, describe that review limit rather than claiming the organisation never performed the work. A brief public narrative may not contain its underlying model or working papers.

A reader can then separate methodological transparency from methodological adequacy. Naming an input or horizon makes an assessment easier to interpret; it does not establish that the input is appropriate or the model captures every relevant mechanism. Testing that requires a defined technical scope and access to the underlying evidence. This article does not perform such validation.

Keep the research conclusion bounded

The practical finding is that scenario labels should travel with their analytical question. For each selected disclosure, retain the exact source passage, period, population, method, output and exclusions. Identify whether the conclusion is the reporting institution’s own statement or the reader’s inference. Preserve future work as future work when preparing a shorter management summary.

The separate climate risks and assumptions guide addresses the construction of an individual risk record. This study instead examines how published analytical outputs can be misread during comparison. It does not assess IFRS S2 compliance or determine which reporting requirements apply to an Indian business.

Sources checked 14 October 2026. GAD: TCFD scenario-analysis section.

MoJ: scenario analysis, table 5 and footnotes 69–70.

Contains public-sector information licensed under the Open Government Licence v3.0. Source publications are linked above. Original analysis, fictional arithmetic and artwork are Nitium’s. No government logos, source figures or third-party images are reproduced; no institutional endorsement is implied.