How AI is Used to Respond to Natural Disasters in Cities
Tthe number of people living in cities has tripled in the last 50 years, which means that when a major natural disaster like an earthquake strikes a city, the lives of many are at risk. Meanwhile, the intensity and frequency of extreme weather events has increased—a trend that will continue as the climate warms. That's fueling efforts around the world to build a new generation of earthquake monitoring and weather forecasting systems to make disaster detection and response faster, cheaper, and more accurate than ever before.
On Nov. 6, at the Barcelona Supercomputing Center in Spain, the Global Initiative on Resilience to Natural Hazards with AI Solutions will meet for the first time. A new United Nations initiative aims to guide governments, organizations, and communities in using AI in disaster management.
The initiative builds on the nearly four-year foundation laid by the International Telecommunication Union, the World Meteorological Organization (WMO) and the UN Environment Programme, which in early 2021 jointly formed a group to focus on developing best practices for the use of AI. in disaster management. This includes improving data collection, improving forecasting, and simplifying communication.
Read more: Leading Cities in the 'Climate and Health Crisis.' New Report Provides Framework for Addressing Its Results
“What I find interesting is that, for one type of accident, there are many different ways that AI can be used and this creates many opportunities,” said Monique Kuglitsch, who chairs the focus group. Take hurricanes for example: By 2023, researchers show AI could help policymakers identify the best places to put traffic sensors to detect road closures after tropical storms in Tallahassee, Fla. And in October, meteorologists used AI weather forecasting models to accurately predict Hurricane Milton. you will arrive near Siesta Key, Florida. AI is also used to warn members of the public more effectively. Last year, the National Weather Service announced a partnership with AI translation company Lilt to help deliver forecasts in Spanish and simplified Chinese, saying it could cut the time it takes to translate a hurricane warning from an hour to 10 minutes.
Besides helping communities prepare for disasters, AI is also being used to coordinate response efforts. Following both Hurricane Milton and Hurricane Ian, the non-profit GiveDirectly used Google's machine learning models to analyze before and after satellite images to identify the hardest hit areas, and prioritize financial grants accordingly. Last year AI analysis of aerial images was deployed in cities such as Quelimane, Mozambique, after Hurricane Freddy and Adıyaman, Turkey, after a 7.8 magnitude earthquake, to aid response efforts.
Read more: How Meteorologists Use AI to Predict Hurricane Milton and Other Hurricanes
The operation of early warning systems is primarily a government responsibility, but AI climate modeling—and, to a lesser extent, earthquake detection—has become a growing private industry. Startup SeismicAI says it is working with civil protection organizations in the Mexican states of Guerrero and Jalisco to deploy a network of AI-enhanced sensors, which will be able to detect earthquakes in real time. Tech giants Google, Nvidia, and Huawei have partnered with European forecasters and say their AI-driven models can produce accurate medium-term forecasts thousands of times faster than traditional models, while being less computationally expensive. Also in September, IBM partnered with NASA to release a general-purpose open-source model that can be used for a variety of climate modeling scenarios, and that runs on the desktop.
AI is improving
Although machine learning techniques have been integrated into weather forecasting models for many years, recent advances have allowed many new models to be built using AI from the ground up, improving the accuracy and speed of predictions. Traditional models, which rely on complex physics-based equations to simulate the interaction between water and air in the atmosphere and require large computers to run, can take hours to produce a single prediction. In contrast, AI weather models learn to recognize patterns by training on decades of weather data, much of which was collected by satellites and ground-based sensors and shared cooperatively between governments.
Both AI and physics-based forecasts work by dividing the world into a three-dimensional grid of boxes and determining variables such as temperature and wind speed. But because AI models are so computationally efficient, they can create grids with very fine grain. For example, the highest-resolution model of the European Center for Medium-Range Weather Forecasts divides the world into 5.5-mile boxes, while Atmo's forecasting initiative offers models that are better than one square mile. This conflict of decision can allow the allocation of resources more efficiently during the worst weather, which is very important for cities, said Johan Mathe, founder and CTO of the company, earlier this year entered into agreements with the Philippines and the island nation. in Tuvalu.
Limitations
AI-driven models are often only as good as the data they're trained on, which can be a limiting factor in some areas. “When you're in a very critical situation, like a disaster, you need to be able to rely on the output of the model,” Kuglitsch said. The poorest regions—often at the forefront of climate-related disasters—typically have fewer and less well-maintained climate sensors, for example, creating gaps in climate data. AI systems trained on this skewed data may be less accurate in disaster-prone areas. And unlike physics-based, rule-following models, as AI models become more complex, they increasingly act as complex 'black boxes', where the path from input to output becomes less clear. The UN's mission is to establish guidelines for the responsible use of AI. Kuglitsch says standards, for example, can encourage developers to expose model limitations or ensure that systems work across regional boundaries.
The program will test its recommendations in this field in collaboration with the Mediterranean and pan-European forecast and Early Warning System against environmental hazards (MedEWSa), a project that emerged from the focus group. “We'll be using best practices from our focus group and getting feedback, to figure out what best practices are the easiest to follow,” Kuglitsch said. One MedEWSa pilot project will test machine learning to predict the occurrence of wildfires in the area around Athens, Greece. One will use AI to improve flood and landslide warnings in the area around the city of Tbilisi, Georgia.
Read more: How the Cement Industry Creates Carbon-Free Building Materials
Meanwhile, private companies like Tomorrow.io want to fill these gaps by collecting their own data. The AI weather forecasting startup has deployed satellites with radar and other weather sensors to collect data from regions without ground-based sensors, including historical data to train its models. Tomorrow.io's technology is being used by cities in New England, including Boston, to help city officials decide when to put salt on the roads before the snow falls. It is also used by Uber and Delta Airlines.
Another UN program, the Systematic Observations Financing Facility (SOFF), also aims to close the climate data gap by providing funding and technical assistance to poor countries. Johan Stander, director of WMO services, one of SOFF's partners, says that WMO is working with private AI developers including Google and Microsoft, but stresses the importance of not giving too much responsibility to AI systems.
“You can't go to the machine and say, 'Okay, you were wrong. Answer me, what's going on?' You still need someone to take that ownership,” he said. He sees the role of the private sector as “supporting shared national resources, rather than trying to take them over.”
Source link