This text is part of the special Research section
Researcher Renée Sieber and her colleague Frédéric Fabry examine the social repercussions of disruptive weather events using artificial intelligence. The data obtained could make it possible to better understand these climatic upheavals which are constantly multiplying.
After focusing on extreme phenomena, such as heat waves and hurricanes, Renée Sieber and her team are now looking at the social impacts of disruptive weather events. The latter are perhaps less impressive, but they have significant consequences on the daily lives of the humans who experience them.
“An elderly woman who is afraid to go down the stairs and break something during a freezing episode in Montreal is a typical example of a disruptive event, illustrates the associate professor in the Department of Geography at McGill University. It can cause a lot more injuries, road incidents and anxiety than you think. »
The researcher also mentions the gradual increase in temperatures in summer, at night. “Normally, the temperatures drop at night and we are able to recover from the heat of the day,” she illustrates. When this is not the case, the quality of sleep is then greatly altered and this can create fatigue, stress, and even discomfort.
“It’s interesting to see how governments, historically, have never really listened to the public when it comes to weather,” says Renée Sieber. By focusing on the social dimension and the testimonies of people who have experienced these events, we may be able to better prepare for them. »
Create a social narrative
“Breaking the equation ‘disruptive weather events = extreme events’ with artificial intelligence and social narratives of the past and present”. This is the name of the project led by the McGill team, which seeks human testimonies of these events in newspaper articles and on social media, thanks to artificial intelligence, and more specifically to natural language processing.
A bit like on Twitter, the algorithms are trained to extract concepts from the association of words, in order to determine trends. In the world of AI, a trend is an example of unsupervised classification, that is, methods used to find a type of emotion that relates to an observation, according to pre-established characteristics.
Some two million words will be analyzed in this way from publications that have appeared since the beginning of the industrial revolution until today. It will be a question of distinguishing what makes people vulnerable or resilient in the face of disruptive phenomena. This, thanks to the testimony of people complaining of the impossibility of going to a medical appointment because of a snowstorm, for example, or others, blocked on the road after the overflow of a river .
It is also an opportunity to understand how citizens used to adapt, by installing ropes on the side of hills to facilitate climbing during periods of frost. The professor also refers to an article from the 1960s explaining the steps to follow after a residential basement floods. “We look at how people reacted, how they dressed, changed certain habits, she lists. This is not always negative, on the contrary. »
Multiple challenges
Even though artificial intelligence is very useful for analyzing such large amounts of data, it has its limits. “It’s not magic, tempers Renée Sieber. AI is a good tool, but there’s a lot of human interpretation involved in trying to figure out how people actually felt. »
For some newspaper articles, it is necessary to go to library archives and work in formats that cannot be transcribed and coded directly. This poses some technical challenges. “It’s really a joy to work with archivists and librarians to advance our respective fields,” explains the researcher.
In a related project, Data Rescue Archival and Weather (DRAW), the team is also attempting to recover more than 100 years of historical Montreal weather data, recorded at the McGill Observatory since 1863. Volunteers are encouraged to participate in the transcription of information into a digital format – a long and tedious job – so that it can be used for scientific research. Once exploitable, these data are very useful to the McGill team, to compare them with the collected testimonies.
“A third of Pakistan is under water, as we speak, it’s terrible, underlines Renée Sieber. I think there is an urgent need to protect the most vulnerable. For the professor, it is therefore fundamental to be interested in the best way to respond to climatic upheavals.