Researcher Samira Kahou is using AI for disaster response
As temperatures rise due to the climate crisis, more extreme weather events, such as hurricanes, cyclones and wildfires, will threaten communities around the world. Predicting them before they happen will be critical to our safety and survival.
Canada CIFAR AI Chair Samira Kahou is using artificial intelligence (AI) to detect and predict extreme weather events to aid in disaster management and bring awareness to the devastating reality of the climate crisis.
Predicting and preventing wildfires and hurricanes
Even a single degree rise in the temperature each year can have a dramatic impact on the number of wildfires that erupt, and their intensity, according to the United States Department of Agriculture. Kahou is developing a multimodal and semi-supervised deep reinforcement learning algorithm, a type of machine learning that learns through unlabeled data, to detect and forecast wildfires in North America.
“The goal is to predict which regions are most at risk so that families can be notified and warned when events such as wildfires and hurricanes are imminent,” says Kahou, a faculty member of Mila, an associate professor at École de technologie supérieure (ÉTS) and an adjunct professor at McGill University.
She is working with different sources of data including satellite images and meteorological data. Once the machine learning agent is able to recognize and identify patterns that represent wildfires over a period of time, Kahou will use temporal data to forecast and predict regional areas at risk of wildfire.
Preventing locust swarm attacks in Africa
Rising temperatures also mean more cyclones — extreme, large-scale weather systems that cycle wind inwards. Cyclones contain a high percentage of humidity which stimulates hormones in locusts, causing as many as millions to swarm together, travelling an upwards of 50 kilometers per day in search of food. They typically attack farms and devastate crop yields, devouring the food of up to 300,000 people, according to the Food Agricultural Organization of the United Nations. Cyclones are common in Asia, Africa and the Middle East, and are to blame for locust swarm attacks that are ravaging farms and food supplies in these areas.
Being able to predict future events could support agricultural communities in mitigating against loss and preventing food insecurity.
Kahou recently published research predicting locust swarms in North Africa, India and the Middle East. The research has implications not just for understanding and studying the effects of climate change on locusts, but also on disaster management and food security, particularly in vulnerable communities.
Using locust swarms and vegetation data from FAO locust hub, Kahou and her collaborators developed a long short-term memory, a type of recurrent neural network that processes sequences of data to predict the regional distribution of locust swarms through the creation of a time series. The team’s deep learning model was able to predict locusts swarm with 65 per cent accuracy, a promising result despite a sparsity of data, but one that Kahou is eager to improve upon.
“Coming from the Middle East, I’ve witnessed the devastating impact that locust swarm attacks can have on a community. This field of research is important for understanding the effects of the climate crisis and protecting the food security of vulnerable communities, but it is also personally important to me,” says Samira Kahou. “As a Canada CIFAR AI Chair, I get to benefit from a network of talented AI researchers who are passionate about using AI to understand and mitigate the impacts of climate change.”