Harnessing AI for more effective disaster risk reduction

MANILA, Philippines — Artificial intelligence has the potential to enhance disaster risk reduction efforts, especially for countries heavily impacted by natural hazards, global experts said.

At the Asia Pacific Ministerial Conference on Disaster Risk Reduction and Resilience (APMCDRR), several professionals in the climate space shared how AI benefits countries by improving weather forecasts and early warning systems on Wednesday, October 16.

Climate and Energy Policy Director Lindy Fursman of the Tony Blair Institute for Global Change said that AI helps people better identify correlations and patterns from huge chunks of data for predicting disaster impacts. 

What kind of AI initiatives exist?

An example Fursman gave was how high-resolution satellite imagery could maximize AI through machine learning to identify early signs of impending disasters such as wildfires. 

Policymakers could also use AI to observe potential impact of disasters through simulated events, Fursman added.

Another speaker shared how their AI startup Spectee Pro has been using the technology to provide people with verified visualizations of disasters worldwide. 

Spectee Pro COO Satoshi Negoro said the firm used AI to source posts from social media containing images of different disasters such as fires, floods and road crashes. 

They use AI to examine the sentences in a post’s caption to determine where the disaster occurred, as well as analyze the different symbols in an image to identify what kind of disaster it is. 

Negoro explained that AI-generated outputs are not delivered to the public without verification, which is carried out by a team of humans.

“AI is very great but it is not perfect, but human intelligence is very flexible and great. We need to multiply those two … [to] give credible information,” he said. 

Philippine initiatives. PAGASA Weather Specialist Michael Samora provided similar efforts in the Philippines where AI is used to detect, monitor, analyze and forecast hazards during typhoons and earthquakes. 

This includes GeoRisk Philippines, a platform featuring interactive, open and accessible information for risk and hazard assessment using satellite data. 

The initiative highlights HazardHunter Philippines which allows Filipinos to determine whether an area is at risk of earthquake, volcano and hydrometeorological hazards.   

Samora also said that the Department of Science and Technology (DOST) currently has a project focused on training PAGASA staff on “developing AI-based solutions” in forecasting weather for up to two weeks.

The agency has also partnered with the Korean Ministry of Environment and K-Water to create an AI-based flood model as part of the country’s river basin monitoring system. 

Language-wise, Samora mentioned that AI would be beneficial to countries with several languages like the Philippines as translating warning signs could be more efficiently done with AI. 

Open science. For Freshteh Rafieian, a United Nations Educational, Scientific and Cultural Organization (UNESCO) science policy program specialist, these efforts to make science available to the public should also “go beyond the output.”

She stressed that open science is also about “opening the process of science” where society can contribute to data gathering and designing scientific research. 

This is why it is essential to have open educational resources and open data, Rafieian added.

How countries should deal with AI in DRR

While there are so many ways to use AI to enhance local DRR efforts, Fursman emphasized that a country first needs public officials who genuinely champion DRR efforts and are aware of what AI is and how it works for a country to benefit from the technology.   

She said that governments must establish foundational frameworks on how AI should be used before investing more in developing the necessary infrastructure to use AI in DRR programs. 

When the speakers were asked about the risks of AI, Samora explained that AI-based weather models pose fewer environmental risks compared to numerical weather prediction models that require large high-performance computing systems.

“In AI systems, you only need big computing resources when you train it, but it is reusable [after],” he said. 

Samora added that scientists and researchers should collaborate with local government units in the collection of data to ensure the accuracy and credibility of the information disseminated to the public.

Lauriane Chardot, Assistant Director of Community Engagement Earth Observatory of Singapore, also highlighted this point, saying that training models need more local data, especially in Southeast Asia. 

“We need more local data to really train the models, otherwise the outputs will be totally biased towards countries where we have more data such as maybe the US or maybe Europe,” she explained. 

In leveraging AI for better risk and hazard assessments, Rafieian said that a country will need a combination of policies, investment and infrastructure, including incentives to encourage participation. 

“So having this climate and generally disaster risk data open and knowledge open and understandable to the public is the first step for [the] prevention of disasters that everyone should consider,” Chardot added.