AI Takes Disaster Planning From Macro Level to Micro Leave a comment

//php echo do_shortcode(‘[responsivevoice_button voice=”US English Male” buttontext=”Listen to Post”]’) ?>

With help from Evergreen, a Toronto-based nonprofit, and Gramener, a Princeton, N.J.-based data science firm, city planners in Calgary, Alberta, can now use an app that relies on multiple datasets and machine learning to identify where heat islands exist in the city. They can then take steps to mitigate the effects of high temperatures.

Short term, city leaders might open cooling centers and first responders can be on alert to health threats in certain neighborhoods posed by high temperatures. Longer term, planners can use the app and its future-looking simulation feature for neighborhood redevelopment efforts, such as determining where to locate parks or paint expanses of concrete white.

Gramener and New Delhi-based partner SEEDS (Sustainable Environment and Ecological Development Society) use a similar AI-powered strategy to help keep people in India safe from natural disasters, such as typhoons (the same weather phenomenon as cyclones and hurricanes), floods and earthquakes. The Indian government used the app in the lead-up to Cyclone Yaas in 2021, ensuring timely evacuation of thousands of people.

“The best part of the technology is the ability to go to a granular level, an individual house level, and inform the inhabitants whether they are at risk,” Sundeep Reddy Mallu, head of analytics and environmental, social, and corporate governance at Gramener, told EE Times.

As recently as five years ago, government warnings of impending threats covered large geographic areas, Mallu said. Therefore, they were less precise and less helpful in protecting people.

Several factors led to the change from macro to micro disaster planning.

Lower cost drives widespread use

Chief among the differences that make the use of Gramener’s technology more widespread is reduced cost. The typical cost for procuring satellite imagery for a 1-square-kilometer (0.6-square-mile) region is less than $10, while close to a decade ago it was over $100, according to Mallu. Cloud computing has also come down in price.

“An NC6 machine on Azure today can be provisioned at $4 per hour,” he said. “Compare these costs to what they were—at least 10 times more—even a couple of years ago.”

Those cost reductions, along with the availability of open-source, moderate-resolution land data from Landsat, a program of NASA and the U.S. Geological Survey, and Sentinel-2, high-resolution land data from the European Space Agency, have also made applications like Evergreen’s possible.

Gramener also uses satellite data from a commercial company.

Ahead of its time on climate action

For the project with Calgary—which lies roughly 300 miles north of Great Falls in Montana—Gramener and Evergreen worked with funding from Microsoft to use data to identify urban heat islands where temperatures can be significantly higher than outlying areas. These islands are a result of buildings, roads and other infrastructure absorbing and emitting heat. The datasets Gramener and Evergreen used to create a low-code app included Landsat imagery, as well as ones for weather, infrastructure, vegetation, pervious/impervious surfaces, census and socio-demographics.

“Calgary has always been very, in my personal opinion, ahead of their time when it comes to taking climate action and making moves to invest in those opportunities to protect its communities,” said Josh Welch, an Evergreen program officer.

The city is using the app to develop neighborhood action plans and for redevelopment purposes. Other cities might also benefit from using this data-driven approach.

“The purpose is to help support and drive investments and policy decisions based on the tool for municipalities who are facing the threat of the climate crisis, and specifically looking at addressing extreme heat and urban heat islands,” Welch said.

A feature added after initial development of the app is a prediction function that Evergreen calls scenario modeling. For example, scenario modeling could simulate the effects of changing the color of roofing to white and identify the expected change in urban heat island trends for the community.

After the success with Calgary, Evergreen worked with the Region of Peel Municipality in Ontario, which encompasses the cities of Brampton, Caledon and Mississauga, to create its own application with additional money from the Royal Bank of Canada Foundation’s RBC Tech for Nature fund.

“They’ve been actively using it to support all kinds of higher-level strategic documentation,” Welch said. “We’re now in conversations with some of those local local cities, like the city of Mississauga, to understand if the tool provides value to them and in ways that they can use it.”

Next is to try to scale the program throughout Canada, he said.

Unpredictability a challenge

Mallu sees potential for an app similar to the one used in Calgary for other disasters, including cyclones, earthquakes and heat waves. The app could be deployed in Japan in the event of an earthquake like the 7.5 magnitude temblor that struck on Jan. 1. Earthquake-prone Japan is also subject to resulting tsunamis.

“The same solution can be applied in Japanese geography because the solution currently is built to accommodate two hazards,” he said.

Another potential use is in risk assessment for commercial buildings and for potential relocation of citizens whose homes are in harm’s way.

The app also has potential for upgrades.

“One of the limitations of the solution is that, even though we are in a position to predict the risk of the house based on historical data and other information, we still do not have the ability to tell whether the house is at risk as of this cyclone,” Mallu explained. “Cyclones have different direction patterns until landfall happens, so we won’t know whether the intensity of a cyclone is category 1, 2, 3, 4 or 5, or whether the direction of the cyclone will affect a particular house or not. So we are not yet there to incorporate real-time direction of the cyclone or hurricane into the model prediction.”