The 2025 Statistical Method: What Changed and Why It Matters

Why the 2025 change matters

A statistical flood estimate is a dependency chain. Catchment descriptors feed rural QMED; donor evidence modifies that index flood; pooling supplies growth-curve L-moments; urbanisation changes the index flood and pooled variability; and a fitted distribution turns those quantities into return-period estimates. Changing one stage while leaving another on an older basis creates a calculation that is difficult to name, reproduce or review.

The official FEH 2025 update therefore matters as a coordinated method change. This report uses three evidence labels deliberately. Official method means a statement supported by FEG v10 or UKCEH. Demonstrated locally means repository code or a gated lock-down test exists at the evidence cut-off.Analyst inference means a professional interpretation that has not been calibrated as a new performance result.

Descriptor lineage is part of the method

The 2025 chain uses SAAR9120, FARL2015, URBEXT2015 and BFIHOST19scaled. AREA remains the catchment-scale input to the rural regression. These fields must be treated as a coherent descriptor vintage: substituting a legacy rainfall, attenuation, urban or soil field changes the method even when the surrounding equation is labelled “2025”.

Descriptor custody begins with the catchment boundary. Automated retrieval cannot decide whether the boundary follows the effective drainage area, whether reservoir behaviour is represented adequately, or whether the mapped urban extent describes the analysis date. Boundary and descriptor review therefore precede numerical adoption.

Rural QMED and the urban chain

The official 2025 rural descriptor equation implemented locally is:

6.8247 × AREA^0.8499 × 0.1780^(1000/SAAR9120) × FARL2015^3.0450 × 0.0321^(BFIHOST19scaled²)

The equation returns a rural descriptor estimate. URBEXT2015 enters the separate 2025 urban adjustment, whose UAF has a factor-of-ten cap. The cap is a method control, not a promise that estimates near it are well constrained. A large adjustment should intensify review of the boundary, descriptor date, drainage changes and the published uncertainty scope.

Donor custody: geography starts the review

The eight geographically close candidates are a starting point for analyst review. Geographic order is useful because nearby gauges can carry relevant regional information, but proximity does not establish hydrological comparability, data quality or freedom from regulation and other local effects.

Transfer is performed on a rural basis: donor information undergoes de-urbanisation before adjustment, and the resulting subject estimate undergoes re-urbanisation using the subject catchment. The adopted donors should record inclusions, exclusions, sensitivity and the local evidence behind each decision. This geographic and local judgement is distinct from the descriptor-based similarity used to construct a pooling group.

Pooling: selection order is not influence

FEH 2025 uses five-component pooling similarity and accumulates a target of 800 station-years. Similarity supplies a defensible order for assembling candidate stations, but the pooled L-moments use weights that respond to record length and descriptor distance. Rank is not weight. A station’s place in the list must not be reported as its numerical contribution to the pool.

The subject hydrology, station suitability, record quality, unusual flood mechanisms, forced inclusions, exclusions and heterogeneity remain review questions. The automated group is a reproducible proposal; the adopted group is an analyst decision with a recorded reason.

Urban and non-flood adjustments are different operations

Pooling first places member evidence on the required rural basis and then applies the subject urban treatment. Under the 2025 method, pooled L-CV is adjusted; pooled L-skew is unchanged. That rule should remain visible because carrying forward an earlier L-skew adjustment would alter tail shape while appearing to be a minor implementation choice.

Non-flood processing addresses unusually low annual maxima within a station record. The local implementation detects the method-defined low-flow condition and can adjust that member’s L-moments before pooling. It is not an urban adjustment, and it is not permission to delete inconvenient years. The affected station, rule and change in pooled evidence remain reviewable.

Candidate distributions are evidence, not a queue

Hydrometric calculates GLO, GEV, PE3 and KAP3 candidates from the relevant L-moments. The first computed candidate is not automatically adopted. Fitting success shows only that a numerical candidate was obtained; it does not establish plausible upper-tail behaviour for the subject catchment.

Adoption should consider L-moment fit, high-return-period behaviour, sensitivity to pool membership, influential records, physical plausibility and consistency with the stated study purpose. Where candidates diverge materially, the divergence is evidence to report rather than a nuisance to hide behind a default.

Uncertainty factors have a published scope

FEG v10 factors should be used only for the estimation route, catchment category and return period covered by the relevant table. They describe the published method evidence for that case. They do not automatically include errors in the chosen boundary, supplied descriptors, local rating, donor decisions, pooling membership, distribution choice or future land-use assumptions.

There is no published uncertainty table for heavily urban catchments. The absence matters: substituting a factor from another category does not create evidence for the missing case. The report should state the gap, show relevant sensitivities and explain how it affects confidence in the adopted estimate.

Scoped local verification

Demonstrated locally, the gated lock-down suite uses RB_02, RL_03 and RD_06. When the licensed fixtures and database are available, it enforces <3% for final donor-adjusted QMED, <0.1% for UAF and <1% for rural descriptor QMED. The matrix records the same thresholds without exposing the licensed reference values.

These checks protect specific calculations against local regression. Three sites do not compare the whole WINFAP or UKFE workflow, do not provide an independent validation cohort, and do not establish every published equation. Database and licence gating also means a normal public test run can skip the comparisons; a skipped gate is not a passing comparison.

Where analyst judgement remains

Analyst judgement remains at the points where evidence is adopted, not merely calculated. The practitioner reviews the effective catchment boundary and descriptor lineage; accepts, rejects or qualifies donors; reviews the proposed pool and its influential members; adopts a distribution; tests sensitivity; and records the limitations attached to the final estimate.

The relevant distinction is between reproducibility and automation. A reproducible calculation makes each input and consequence visible. It does not remove responsibility for local hydrological reasoning or convert a default into a decision.

Implementation status

At the evidence cut-off, Hydrometric implements the 2025 descriptor QMED equation, UAF cap, donor ruralisation and transfer, five-component pooling similarity, the 800 station-year target, record-aware pooling weights, per-member non-flood processing, the 2025 pooled urban rule, and GLO, GEV, PE3 and KAP3 candidate calculations. It exposes intermediate evidence for review in a structured statistical calculation output.

The verification evidence is narrower than that implementation list. The three-site lock-down thresholds cover the quantities in the verification matrix; known differences in pool membership, L-moments or design flows must remain visible rather than being absorbed into a broad equivalence statement. Research prototypes on other branches are not described here as current functionality.

What we do not conclude

• We do not claim approval, certification or accreditation by the Environment Agency, UKCEH, WINFAP or WHS.
• We do not claim end-to-end equivalence with WINFAP or UKFE, an independent national validation, or proof across every equation.
• We do not present Hydrometric as a replacement for WINFAP, ReFH2 or the FEH Web Service.
• We do not claim that the current structured calculation output is a complete guideline-compliant Flood Estimation Report.
• We do not extend published uncertainty factors beyond their named categories or invent a tabulated factor for heavily urban sites.

Disclosure and licensing boundary

This publication discloses the official equation, descriptor lineage, method stages, implemented calculation scope, named local test sites, acceptance thresholds, gating conditions and evidential limits. Its external citations are the Environment Agency and UKCEH primary sources recorded with the report.

It withholds licensed fixture values, paid exports, restricted database records, customer catchments, private analyst notes and code-level workflow logic that would reconstruct the proprietary engine. The disclosed thresholds make the local claim auditable without redistributing commercial material.