Can AI Write a Flood Risk Assessment?

Artificial intelligence can now draft contracts, summarise reports and answer technical questions in seconds. So it is a fair question to ask: can it write a Flood Risk Assessment (FRA) too?

The short answer is that AI can help produce the words, but it cannot produce a valid Flood Risk Assessment. An FRA is not a piece of writing. It is a site-specific, regulated judgement that has to satisfy a planning authority and, often, the Environment Agency, and stand up if it is ever challenged.

This matters for anyone who needs an FRA to support a planning application: property developers, architects, planning consultants and self-builders. Getting it wrong means delay, refusal or, at worst, a scheme that floods.

What a Flood Risk Assessment actually is

An FRA is a technical document that assesses the flood risk to and from a proposed development, and shows how that risk will be safely managed for the lifetime of the building.

To be accepted, it has to do several things properly:

• Assess all sources of flooding, not just rivers and the sea.
• Apply national planning policy, including the Sequential Test and, where needed, the Exception Test.
• Use current, location-specific data and climate change allowances.
• Set out workable mitigation, such as finished floor levels, safe access and drainage design.
• Be prepared by a competent person who stands behind it.

A document that looks like an FRA but fails any of these is not much use to a planner.

What AI can genuinely help with

It is worth being fair about what the technology does well. Used carefully, AI can:

• Draft a first structure.
• Summarise long planning policy documents into plain English.
• Explain technical terms to a non-specialist client.
• Speed up routine, low-risk writing tasks.

These are real efficiencies, and many consultants already use them. The problem starts when the tool is asked to be the assessor rather than the assistant.

Where AI falls short

It cannot visit the site

The single most important input to an FRA often comes from a site walkover, and no model can do this. A consultant on site sees ground levels, a blocked culvert, a sagging ditch or a roof shedding water against a wall.

We see this constantly. In a recent groundwater investigation, the headline data pointed to groundwater, but the site visit revealed that much of the water was rainwater from a gutterless roof, trapped above the damp proof course by raised ground levels. AI would have confidently described a groundwater problem that was only half the story.

It misses sources of flooding

National policy now requires the Sequential Test to consider all forms of flooding, not just rivers and the sea (NPPF, paragraph 175). In practice, AI tends to default to fluvial and tidal risk and underplay surface water, groundwater and sewer flooding.

That is exactly where real schemes come unstuck. Surface water and groundwater rarely appear on the standard Flood Map for Planning, yet they impair applications.

It cannot model what it cannot see

Many sites cannot be assessed from maps at all. They need hydraulic modelling built from survey data.

On a recurring surface water flooding problem at an East Midlands industrial site, we built a hydraulic model using topographic, CCTV drainage and current FEH22 rainfall data, then tested the network with the outfall free-flowing, surcharged and blocked. The flooding turned out to be driven not by rain on the site, but by water unable to leave through a surcharged outfall. No language model can build, run or interpret that model, and the numbers it invents cannot be trusted.

It gets the data and allowances wrong

The Environment Agency’s climate change allowances vary by management catchment, by time period (the 2050s and 2080s epochs) and by percentile. The correct figure depends on the exact location and the development’s lifetime. AI is prone to quoting plausible but wrong values, which is a serious problem in a document a regulator will scrutinise.

It cannot take responsibility

This is the decisive point. An FRA carries professional accountability. It is signed off by a qualified consultant, often a chartered engineer, who carries a duty of care and professional indemnity insurance.

When flooding ends up in dispute, that accountability is everything. In an arbitration over flooding at a farm in northern England, the case turned on independent expert evidence from a chartered engineer who reviewed the documents, walked the site and weighed competing explanations. AI has no liability, no insurance and no standing to give evidence.

The value of qualified judgement and site-specific experience

There is a dimension no AI tool can replicate: the professional knowledge an experienced engineer or surveyor brings. A qualified person doesn’t just slot data into a template. They draw on accumulated experience, recognising when flood zone boundaries don’t reflect actual topography, spotting anomalies a less experienced eye would accept, knowing when a site’s drainage history warrants further investigation. They also ask the right questions and quality-check the data that AI depends on to perform well. And they regularly arrive at bespoke, proportionate solutions that go beyond standard guidance. AI can process information. It cannot exercise professional judgement.

The real-world risk of an AI-only FRA

Relying on AI alone to produce an FRA invites practical consequences:

• Refusal or delay if the Environment Agency or Lead Local Flood Authority rejects it.
• Wasted fees on an application that has to be redone.
• Missed consents, such as a Flood Risk Activity Permit or Marine Licence, that a specialist would have flagged early.
• An unsafe design that under-protects the people who will live there.

A sensible way to use AI

The honest conclusion is not that AI is useless. It is that AI is a drafting aid, not a flood risk consultant. Left to make the judgements, it produces something that reads well and means little.

Conclusion

AI can generate text that resembles a Flood Risk Assessment. It cannot visit the site, model the drainage, apply the right data, weigh the evidence or take responsibility for the outcome. Those are the things that make an FRA valid, and they need a qualified human.

If you need an FRA for a planning application, start with a consultant who can stand behind it. If you would like to discuss a site, get in touch with our flood risk team.