AI model flags 1.2 million undefended UK buildings at flood risk

TL;DR

• A Snowflake-Ordnance Survey AI model has identified up to 1.2 million English buildings at flood risk but outside current defence and planning frameworks
• 68% of these buildings are also socially vulnerable; 84% were built before planning rules consistently accounted for flood risk; 85% are exposed primarily to surface water rather than rivers or coast
• The findings challenge the river-and-coast flood narrative and argue for building-level rather than area-based risk assessment in UK policy

A new AI-driven flood model developed by Snowflake using Ordnance Survey geospatial data has identified up to 1.2 million buildings in England that face flood risk despite sitting outside current flood defence and planning frameworks. The Intelligent Flood Readiness Model combines detailed building-level data with government datasets and statutory Flood Risk Management Plans to produce what the partners call a new “structural intelligence” layer for policymakers.

What the model reveals

Three findings cut against conventional flood planning assumptions. First, vulnerability compounds: 68% of at-risk buildings flagged by the model are not only exposed to flooding but also lack the social and economic resilience for recovery, with concentrations in deprived areas. Second, age is the dominant risk factor: 84% of flagged buildings predate 2001, when planning rules began consistently accounting for flood risk — 15% date from before 1919 and 23% from 1919-1959. Third, the nature of risk has shifted: 85% of at-risk undefended buildings face surface-water flooding rather than river or coastal inundation.

The surface-water dominance is particularly significant for urban policy. High-density housing rather than riverside or coastal properties may account for the larger share of at-risk households. Geographic concentration appears along Yorkshire and the Humber east coast, but 37% of English neighbourhoods contain at least one such building — the risk is genuinely dispersed.

Why AI changes the planning picture

Flood planning cycles typically update every six years and rely on relatively coarse data; the built and natural environment change faster than that. The model integrates six data streams — OS building datasets, deprivation indices, Environment Agency flood data, AI analysis of more than 3,000 pages of FRMP documents — to produce near real-time insights at neighbourhood and potentially street level.

The findings point toward a policy shift: from area-based planning to building-level risk assessment, with targeting clusters of vulnerability across administrative boundaries, heavier investment in surface-water infrastructure, and factoring social deprivation into resilience planning. The partners argue that protecting every at-risk property using current methods is unrealistic — implying more explicit triage in investment decisions.

Looking forward

The model fits a broader UK public-sector pattern of integrating fragmented datasets using AI: the Environment Agency deployment sits alongside HMCTS’s Justice Transcribe trial, MoJ’s transcription study, and the AI for Science Strategy backed by Russell Group universities this week. The harder question is political: if 1.2 million buildings are identified as at-risk outside the current framework, expectations will rise for protection or relocation support. That is a fiscal conversation government may not yet be ready for. Expect the model’s recommendations to land in Defra policy consultations over the next 12 months.