AI could help us see danger earlier, understand consequences more clearly and act before a hazard becomes a catastrophe, writes Mark Coates, VP of Infrastructure Policy Advancement, Bentley Systems
After London Climate Action Week, one word kept surfacing in almost every conversation about infrastructure, climate and resilience: AI. It is often overused and sometimes overpromised. Yet beneath the noise is a serious question for the built environment. If AI is a powerful tool for recognising patterns, could it help us detect risk earlier, understand consequences more clearly and act before a natural hazard becomes a human and economic disaster?
The relevance for the UK could hardly be clearer. During the same week, the country provisionally broke its June maximum temperature record on three consecutive days, reaching 36.1C at Gosport, Hampshire, on 24 June, 36.7C at Merryfield, Somerset, on 25 June and 37.3C at Santon Downham, Suffolk, on 26 June. The previous UK June record, 35.6C, had stood since 1976 and 1957.
Across western Europe, Copernicus Sentinel-3 imagery also showed land surface temperatures above 50C in parts of France and northern Spain. That is ground surface temperature, not air temperature, but it is a stark indicator of heat stress on people, transport networks, power systems, water infrastructure, agriculture and ecosystems.
We often describe disasters as sudden, as if the flood, wildfire or storm arrived without warning. In reality, many disasters are the result of exposure, vulnerability, weak assets, fragmented planning and missed signals. The hazard may be natural, but the disaster is usually systemic.
AI can change the conversation
The promise is not perfect prediction. Earthquakes remain resistant to precise forecasting, and flash floods, wildfires and storms can still move faster than institutions. The real opportunity is earlier insight from combining weather, climate, satellite, sensor, asset and operational data that can turn a warning into practical decisions.
In weather forecasting, progress has been striking. Google DeepMind’s GraphCast model can generate a 10-day global forecast in under a minute and, in published testing, outperformed the European Centre for Medium-Range Weather Forecasts’ (ECMWF) deterministic system on 89.3% of tested variables and lead times. ECMWF is also exploring machine-learning models alongside physics-based forecasting.
That speed matters. A forecast produced faster, tested across more scenarios and updated more frequently gives emergency planners time to close a bridge, move pumps, protect a substation, reroute transport or warn a community before the water arrives.
Catastrophe modelling is another important frontier. Governments, insurers, banks and utilities have long used these models to estimate exposure to floods, storms, wildfires and other extreme events. AI can help generate thousands of plausible scenarios, downscale climate outputs into local risk maps and test tail risks that traditional models struggle to capture. For infrastructure leaders, the question is not only what happened last time, but what could plausibly happen next.
Flooding is one of the clearest examples. Google’s Flood Hub provides AI-driven flood forecasts, including water trends, inundation maps and alerts up to seven days ahead, covering more than 240,000 locations across river basins in roughly 150 countries. This is more than a better map. Traditional flood prediction depends heavily on gauges, hydrological models and local data. Many vulnerable regions lack dense monitoring networks. AI can learn from global rainfall and river behaviour then apply that learning where local data is sparse. It does not replace local knowledge it makes local action possible where the data foundation has been weak.
The same applies to wildfires. AI can combine satellite imagery, weather, vegetation dryness, wind, terrain and historical fire behaviour to identify elevated risk. The point is not just where a fire might start but where it could spread, which communities are exposed, which access routes may fail and which assets need protection first.
Shift to systems thinking
For the construction and infrastructure sectors, the crucial shift is systems thinking. A flood is not only a water problem it can close roads, cut power, isolate hospitals, disrupt housing and halt supply chains. A heatwave is not only a weather event. It tests schools, railways, roads, power networks, water systems, housing quality and workforce resilience. AI is most useful when it helps leaders see those dependencies before the crisis arrives.
The best disaster prediction systems will not be judged by the sophistication of the model but by whether the right people act in time. That means connecting hazard forecasts to asset data, digital twins, transport networks, utilities, hospitals, schools, vulnerable populations and emergency response capacity. A warning that sits in a dashboard is not resilience it is simply a well-designed anxiety machine.
That is why early warning must be treated as an infrastructure and governance challenge, not just a technology challenge. The World Meteorological Organisation and UNDRR’s Early Warnings for All initiative aims to protect everyone with early warning systems by the end of 2027. UNDRR’s 2024 report notes progress on multi-hazard warning systems but also warns that the human and economic impact of disasters continues to grow.
That tension matters
Better prediction does not automatically create better outcomes. Warning systems fail when agencies do not share data, local authorities lack capacity, the public does not trust the message or infrastructure cannot absorb the shock. AI can improve the forecast but governance determines whether the forecast saves lives.
There are risks. AI models can be opaque, struggle with events outside their training data and create false confidence if users forget that a forecast is still a probability. In disaster management, a confident wrong answer can be worse than uncertainty honestly stated.
The answer is not to choose between traditional science and AI. It is to combine physics-based models, meteorological expertise, local observation, satellite data, ground sensors, emergency planning and machine learning. The future is not machines replacing forecasters, engineers, planners or responders. It is those professionals working with better tools.
For infrastructure leaders, the lesson is clear the value of AI prediction connected to action. Which railway embankments are exposed to flooding? Which bridges are vulnerable to scour? Which communities lose access if one road fails? Which pumping stations need backup power? Which port operations are disrupted by storm surge? Which substations, schools, hospitals and care homes are least able to cope during extreme heat? That is the move from weather forecasting to infrastructure resilience.
Natural hazards will continue to test the UK and the wider world. Climate change is increasing the frequency and intensity of some extreme weather events, while urban growth places more people and assets in harm’s way. But infrastructure owners and public authorities are no longer limited to reacting after the event. Used well, AI offers a better chance to anticipate, prepare and intervene.
The technology is impressive. The harder task is institutional. We must connect prediction to planning, investment, operations and public trust. Otherwise, we will have world-class warning systems and very familiar disasters.
AI will not stop the storm forming, the river rising, the hillside moving or the heat building over a city. But it can help us see danger earlier, understand consequences more clearly and act before a hazard becomes a catastrophe. That is the real transformation. It does not replace human judgement. It gives human judgement more time to matter.
*Please note, this is a commercial profile.
The post From forecasts to resilience. How AI can help infrastructure act before disaster strikes appeared first on Planning, Building & Construction Today.
