Every Canadian climate article since 1978,
read sentence by sentence.
We collect the climate coverage of 22 Canadian outlets four times a day, then read each article one sentence at a time — some sixty annotation categories, every person, organisation and place named. From that reading we detect the events several outlets converge on, the sustained waves of attention, and the dominant voices of the day, then publish it all in the live observatory. We release the metadata, our annotations, our summaries and our analyses — never the articles’ full text (fair dealing for research, s. 29 of the Copyright Act).
Six chapters, from collection to the live observatory
Each chapter stands on its own and answers one question about the method. The figures quoted here are read from the database on every page load: 287,227 articles from 22 Canadian outlets, 1978 to 2026.
The country’s climate coverage, province by province
This is the observatory’s own map, served here unchanged. Each outlet sits in its home city, CBC.ca and Radio-Canada.ca articles are assigned to a province by our local model, and each detected event is tied to its city.
Map — climate articles by province (count of corpus articles attributed to each province; the darker the tone, the larger the volume). Window: the whole corpus, 1978 to 2026, except city-located articles, which cover the last eight days. Each dot is an outlet placed in its home city; the C and R badges mark CBC.ca and Radio-Canada.ca articles classified by province. Reading it: click a province for its last two days of coverage, its events and cascades; click an outlet for its frame profile and latest articles.
What we publish, and what we do not
We publish each article’s metadata (headline, outlet, date, byline, front-page placement, regional edition, opinion section), the annotations our models produce, the summaries written by our local model, the events and cascades our detection method finds, and the analysis notes built on them. The API and the Data Explorer serve the same material.
Full article text stays with the outlets that published it: it is neither reproduced nor redistributed here. Our reading of that text falls under fair dealing for research, section 29 of the Copyright Act.
Research teams can request access to the API: .
The technical paper and the code
The paper, published in Scientific Data (Nature Portfolio, 2026), sets out how the corpus is built, how the models were trained and how their reading is checked against human annotators. The public repository holds the code that applies them.