A flash flood hits a rural town, a typhoon strikes a coastal village, and rescuers are stretched thin as authorities scramble to identify the areas worst affected.
As climate change and extreme weather events become more frequent in the modern day, new initiatives are popping up in an effort to minimise their damage and save lives.
With the development and rising adoption of artificial intelligence (AI)-driven prevention and preparedness systems, a safer future amidst rising natural disasters might just be within sight.
One such example is DeepMind’s WeatherNext 2 forecasting model, which draws on decades of atmospheric data for AI analysis, allowing it to achieve greater accuracy than previous forecasting models.
In a study published in the journal Nature, DeepMind researchers said the model can produce accurate forecasts up to three days ahead, surpassing previous models, which could forecast accurately up to two days ahead.
Data points used for the AI model’s training include global weather dynamics and expert-curated historical cyclone observations, encompassing almost 20TB worth of data and nearly 5,000 historical storms.
This results in WeatherNext 2 being a single AI model capable of accurately predicting the track of a tropical cyclone, its intensity, and wind structure.
The model had already proven its worth last October, when the United States National Hurricane Center (NHC) used it to predict that a storm would intensify from a Category 1 to a Category 5 hurricane an entire five days in advance.
The storm was Hurricane Melissa, which went on to become the strongest hurricane to hit Jamaica in recorded history. This early prediction gave local authorities more time to evacuate communities and allocate resources ahead of the storm’s arrival.
Forecasts made by the NHC cover the Atlantic and Eastern Pacific Oceans, encompassing support for almost 30 countries, including Jamaica.
WeatherNext 2, a model designed to provide rapid weather forecasts, and WeatherNext Cyclone, a version tailored for hurricane forecasting, have since been made open source.
This means the models are now available for organisations, researchers, developers, and other users to access, modify and run locally.
Meanwhile, the US National Aeronautics and Space Administration (Nasa) has similarly been leveraging AI to improve flash flood warnings. Like DeepMind’s WeatherNext, Nasa’s Transient Artifact and Continuous Learning System (TACLS) draws on more than 30 years of historical data.
The data comes from the Global Navigation Satellite System (GNSS), capturing information on phenomena such as atmospheric rivers, monsoonal convection and tropical cyclone remnants. TACLS analyses this data to identify signs of potential flooding and flags them for meteorologists to assess.
In this way, the system acts as a sieve for the vast amounts of data collected, helping meteorologists identify potential threats that might otherwise be overlooked.
The technology is also being incorporated into existing warning systems in the US and will be made open source.
Besides the obvious role such forecasting systems may have in enabling early evacuations and other preventative measures to minimise damage, they also have the potential to do even more good, based on the efforts being made by the United Nations Children’s Fund (Unicef).
In January this year, the agency launched the Ahead of the Storm initiative, which centres on using AI to predict how extreme weather events such as hurricanes, landslides and floods could affect children.
The initiative builds on insights from forecasting systems, including DeepMind’s and Nasa’s AI prediction models. Unicef has prototyped a system that combines this data and forecasting with maps of locations such as schools, health, nutrition, water, sanitation, and hygiene facilities.
It also takes into account factors that indicate how vulnerable communities may be, including child population density and poverty levels.
Rather than simply predicting where a disaster may occur, the system aims to answer questions that can often slow down emergency response and humanitarian aid, such as which areas are likely to suffer the most damage and how many students are in an affected school.
As the AI model receives natural disaster forecasts, it assesses which areas, services and populations are most likely to be severely affected.
This could allow first responders and humanitarian organisations to anticipate needs more effectively and proactively make decisions on where to direct resources before disaster strikes.
Unicef is also supporting a variety of other natural disaster preparedness and prediction initiatives around the world, which seek to integrate AI in one form or another to improve how they are identified and responded to.
Countries around the world have also kicked off their own initiatives aiming to tackle natural disasters prediction and early warning with AI.
Back in July this year, China and Thailand launched a joint laboratory aimed at applying AI in meteorology for greater preparedness, damage mitigation, and to respond to climate change.
The Philippines on the other hand allocated 1bil pesos (RM64.5mil) at the end of August to enhance the country’s AI-powered disaster management programme, Project Noah (Nationwide Operational Assessment of Hazards).
The funds are aimed at further research and deploying equipment and support systems under the programme, with Project Noah utilising technologies such as AI-enhanced modelling, Light Detection and Ranging (Lidar) mapping, and data analytics for flood forecasting.
Back home in Malaysia, the country is also getting in on the use of AI in disaster response. A report from April this year highlighted plans by the National Disaster Management Agency (Nadma) to use AI to locate disaster victims more quickly.
One application being explored is the use of AI-equipped drones to identify people who may be trapped, helping first responders locate victims more efficiently. But the technology could also play a role beyond simply finding those in distress.
When combined with predictive AI data, it could help triangulate areas where rescue teams and other assets should be deployed, enabling agencies and rescuers to respond more quickly.
This predictive capability could also be applied before a disaster even occurs. AI is already being used to anticipate weather conditions during Malaysia’s wet and dry seasons for use in early warnings.
The data is then combined with information from other government agencies to support analysis, decision-making, and response planning.
