EA River Modelling Technical Standards (2026): What Flood Risk Assessments Must Demonstrate

Hydraulic modelling sits behind almost every detailed flood risk assessment, and the Environment Agency sets the standard for how those models are built and reviewed. The EA river modelling technical standards define what a credible model looks like and what evidence has to come with it. They are not a software manual. They do not tell you which package to buy. What they tell you is what a reviewer will look for, and getting that right is increasingly what separates an FRA the EA accepts from one it sends back. This guide explains what the standards cover, what they ask of hydraulic modelling work, and what they mean in day-to-day FRA practice.

First published in August 2021 and last updated on 8 May 2026, the standards now run to five separate documents covering source, pathway, receptor and model approach.

What the EA’s river modelling standards cover

The river modelling technical standards and assessment are guidance for the whole flood risk management industry on how to build and review hydraulic models, and how to provide evidence for flood risk management decisions. They are written for hydrologists, modellers, reviewers and project managers who already know the subject, and they assume that knowledge throughout. The idea underneath them is simple. Model the flooding using the source, pathway and receptor concept, then document it well enough that someone who has never seen the site can pick the model up, understand what was assumed, and trust the numbers that come out of it.

In practice the guidance splits into four categories. Source covers how you estimate the water arriving: fluvial flows, rainfall runoff and downstream influences. Pathway covers how you represent the route it takes, through channels, floodplains, structures and defences. Receptor covers the properties and infrastructure at risk. Model approach covers the design decisions that sit over all three, including dimensionality, discretisation and calibration.

• A clear model approach. Built on the source, pathway and receptor concept, matched to the question the model needs to answer.
• A Model User Report (MUR). The technical record of how the model was built, its assumptions and its limitations, so a reviewer can understand and adopt the outputs.
• Calibration and verification. Tested against observed flood data where it exists, with the basis stated where it does not.
• Sensitivity testing. Roughness, inflows and climate-change allowances varied so the reviewer can see how robust the result is.
• Reporting that follows the logic. Inputs, method and results presented so the reviewer can trace the conclusion rather than take it on faith.

The EA also publishes an Excel-based fluvial model assessment tool that lets a reviewer gauge a model’s quality without running a full technical review. It is worth knowing what that tool is for, and what it is not: the EA is explicit that using it “does not exempt you from needing a technical review” before a model is issued. Weak models are easier to spot than they were. If a scheme depends on getting this right, Unda’s 1D and 2D flood modelling service builds and reviews models to these standards.

1D, 2D and linked models: choosing the right approach

Different sites need different approaches. A 1D model represents the channel as a line. It calculates in-channel flow accurately and runs quickly, but it needs a good deal of post-processing before it produces anything resembling a flood map, and it cannot show you water spreading sideways across a floodplain because it has no sideways to spread into. A 2D model represents the ground as a grid and lets water spread in any direction, which suits complex floodplains and direct rainfall. A linked 1D-2D model pairs the two and is now the usual starting point for a detailed site assessment.

The standards are deliberately package-agnostic. The EA removed software-specific guidance in 2023 and now asks only that whatever you use can be shown to fit the question, properly verified and clearly documented. Common industry packages include Flood Modeller, TUFLOW and HEC-RAS, and our guide to flood modelling tools and techniques covers how they differ. The EA’s own benchmarking of hydraulic modelling software tests packages against best practice, but it does not crown a winner.

Choosing the wrong dimensionality for the site is a far more common failing than choosing the wrong software.

What a detailed FRA has to demonstrate

Not every application needs a model. The EA’s guidance on flood risk assessments for planning permission is proportionate about that, and a minor extension will not be judged against these standards at all. Where a model is needed, though, for a Level 2 or Level 3 flood risk assessment for planning, the practical implications are consistent. Justify the modelling approach against the site and the question. Establish the hydrology by the right method for the catchment. Calibrate and verify against observed data where any exists. Run and report sensitivity tests. And document all of it, so the reviewer can follow the logic rather than take it on trust. The EA’s own guidance on using modelling for flood risk assessments makes the same point from the reviewer’s side.

The hydrology comes first. The standards are specific about method here in a way they deliberately are not about software. The FEH statistical method suits catchments with a reasonable record of peak flows; the ReFH rainfall-runoff method suits catchments where something other than peak flow drives the hazard, and ReFH2 is particularly recommended on permeable catchments with a BFIHOST above 0.65. Where a site is ungauged, QMED can be adjusted using donor catchments, and estimating it from catchment descriptors alone is treated as a last resort. Reporting follows the EA’s flood estimation report template.

How accurate does a calibrated model have to be?

A well-calibrated model should match reliable recorded peak data to within about 0.15 metres, tested against at least three flood events where the record allows. Verification then means running a further observed event with no changes to the coefficients, ideally from the same season. Where no observed data exists at all, the standards ask for validation instead: checking model performance against whatever reliable evidence there is, and saying plainly what that evidence was.

A model calibrated to a summer flood on a catchment whose real risk comes from winter storms has not been calibrated at all.

Sensitivity testing is where a lot of otherwise good models come unstuck. The standards ask for a 20 per cent adjustment applied to inflows, hydraulic roughness, the downstream boundary and structural and spill coefficients, with the effect reported rather than filed. Roughness deserves particular care. Selecting Manning’s n is an openly subjective judgement, and the ranges are wide enough to matter: channel values run from around 0.025 for a clean straight channel to 0.150 where vegetation is heavy, while floodplain values span 0.025 on short grass to 0.200 under dense trees, which is a spread easily capable of moving a modelled level by more than the tolerance the calibration is judged against.

Receptors matter as much as the channel. In a 2D model the standards set out four ways of representing buildings, from a blanket increase in urban roughness through to solid blocks, and for detailed work they favour the “stubby building” approach: raise the building threshold, typically by 0.25 to 0.30 metres where no threshold survey exists, and increase the roughness over the footprint. On a flood mapping study the advice reverses, because raising thresholds removes floodplain storage and can push the modelled extent out artificially.

Input data is half the battle. A model is only as good as the ground it sits on, which is why LiDAR and terrain data, GIS and, on sensitive sites, a proper topographical survey matter so much. Thin or mismatched input data is one of the quickest ways to lose a reviewer’s confidence, and it is usually cheaper to fix at the start than after a model has been built. This is one of the clearest places where expertise in an FRA earns its keep.

What goes in a Model User Report

The Model User Report is the document that lets someone else adopt your model, and it is the single most common thing missing from a submission. It is not a summary of results. It is the record of every decision taken during the build, set down in enough detail that a reviewer can reconstruct what you did without having to ring you up and ask.

1. The method statement, agreed before the work started, setting out the approach and the software version.
2. The data used and where it came from, including survey, terrain, gauged records and any existing model.
3. The model build itself, with schematics and a decision log covering how the channel, structures and floodplain were represented.
4. Calibration and verification, with the events used, the coefficients arrived at and how closely the model matched.
5. Sensitivity testing and its results, showing how far the answer moves when the inputs are varied.
6. Stability measures, model run records and a version-controlled development log showing how the model changed.
7. Assumptions and limitations, stated plainly, alongside the post-processed outputs the FRA relies on.

Alongside the report the EA expects a model passport, summarising methodology, extent, scenarios, receptors and limitations at a glance. Both exist for the same reason. A model only its author understands cannot be adopted as evidence.

Can you use an existing Environment Agency model?

Often, yes. It is usually the sensible starting point, and on a main river there may already be a perfectly good model of the reach. But that model was built to answer the EA’s question rather than yours, and the guidance is unusually direct about what follows from that.

Before relying on one, check that it represents current risk, uses current datasets, meets the current standards, works at the right scale, captures the site detail you need, and carries up-to-date climate change allowances. The EA’s own position is that even a recent model needs review and may need updating. Existing models and the data behind them can be requested as Product 4 flood data, and a model that genuinely contradicts the published mapping is the basis on which you would challenge the Flood Map for Planning.

Where modelling meets planning policy

Modelling does not sit apart from policy. The outputs feed straight into how a site sits against the flood zones, whether it can pass the Sequential and Exception Tests, and how safe-for-its-lifetime is demonstrated. That link has tightened since the September 2025 update to the planning practice guidance on flood risk and coastal change and the May 2026 Flood Map for Planning changes, both of which lean harder on modelled surface-water and climate-change data. Where a model result is the evidence a test rests on, it is worth commissioning the Sequential and Exception Test reports and the modelling together rather than in sequence.

A model that is technically sound but presented without the policy thread tends to generate more questions, not fewer.

Climate change allowances are where the two most often collide. Apply one at the wrong epoch, or against the wrong development lifetime, and an otherwise sound model is undone. It is also the sort of error a planning officer can spot without opening a single result file, which is why it so often surfaces late and publicly rather than quietly during the technical review. Our guide to which climate change allowance applies sets out how the choice is made.

The parallel coastal modelling standards

Alongside the river standards, the EA now maintains separate coastal modelling standards for open-coast and tidal modelling. They were published on 8 May 2026, the same day the river page was updated to link to them. Most of Unda’s work is inland, but for estuarine or coastal-fringe sites the same principles carry over: a fit-for-purpose method, verification, and documented evidence.

Common reasons a model gets sent back

Across the FRAs we see, the modelling problems that draw EA questions are rarely exotic. The EA’s own guidance names three grounds for finding a model unsuitable: the quality of the model, the quality of the reporting, and missing information. In practice those three cover almost everything.

• The wrong dimensionality. A 1D model on a site where the floodplain spreading is the whole question, or a 2D model where a structure needed representing properly.
• No reported sensitivity testing. The result may be right, but the reviewer cannot tell how much weight it will bear.
• A thin or missing Model User Report. The reviewer cannot reconstruct what was done, so cannot adopt the outputs.
• Out-of-date climate change allowances. An otherwise good model measured against superseded guidance.
• Structures represented loosely. Afflux at bridges and culverts is where small modelling shortcuts turn into real level differences.
• Survey that does not support the model. Cross-sections too widely spaced, or a survey old enough that the channel has since changed.

None of these is hard to avoid. All of them cost time once an application has stalled. An Environment Agency objection on modelling grounds takes far longer to clear than it would have taken to prevent, because the clock is now running on a determination date rather than on your own programme.

How Unda can help

Unda’s flood modelling team builds and reviews hydraulic models to the EA’s standards, and presents the evidence in a form that holds up under EA scrutiny. We work alongside the Environment Agency as a statutory consultee on main river sites and with lead local flood authorities elsewhere. If your scheme depends on a model result, whether that is a fresh model, a review of someone else’s, or defending a result the EA has questioned, our flood modelling service is the place to start. Get in touch with the site details and what stage you are at, and we will tell you what the modelling needs to show.

Frequently asked questions