Periods of exceptional attention
Cascades describe when a climate narrative accelerates across several outlets and how framing converges during that period.
In depth: media cascades →
THE JOURNAL IN YOUR INBOX
Choose your edition and its cadence.
Receive an edition prepared from the freshest news already processed by the observatory, with a link to every original source.
The CCF Observatory
About Thematic Frames
This visualization shows how 8 distinct climate change frames have been used in Canadian media coverage over time. Click any frame to see its definition and a corpus example.
Climate Frames Over Time (National)
Smoothed monthly trends across all 20 Canadian media outlets (1978–2024)
The CCF Observatory
About Emotional Tones
Beyond thematic framing, we track the emotional tone of climate coverage. Click each tone to learn more.
Emotional Tone Over Time
Positive vs Negative emotional framing in climate coverage
The CCF Observatory
Each outlet’s climate lens
How each outlet distributes its climate coverage across the eight frames — select an outlet, compare two, decade after decade.
Climate Frames by Media Outlet
…
…
The CCF Observatory
Across the country, outlet by outlet
One map, the whole picture: provinces shaded by climate coverage, the corpus outlets placed in their home city, and the CBC.ca / Radio-Canada.ca articles classified into each province by the local language model. Click a province or an outlet to explore its coverage, latest articles, related events and cascades — live from the corpus.
Epistemic Authority Network
2024 SampleWho shapes the climate discourse? This visualization maps the most cited individuals in Canadian climate coverage for 2024. Node size = citation frequency · Colors = role category
-
Individuals
-
Total Citations
-
Articles
Top 10 Most Cited Individuals
The CCF Observatory
…
…
⟵⟶
Data & copyright
CCF publishes metadata, annotations and summaries derived from its analyses. Full article text remains on the publishers' websites and is never reproduced here. This use falls within fair dealing for research under section 29 of the Copyright Act.
The CCF Observatory
The most important events surfaced by CCF
A chronological timeline first, then the twenty biggest events of the last 15 days in detail — real news events, surfaced automatically by semantic clustering of article embeddings whenever coverage converges across outlets, the method of our open cascade-detection framework. Each brief is written by our CCF model, from the source articles themselves.
The events through time
Each bubble is a detected event, coloured by type, larger when coverage is strong. Click to open the profile.
Loading timeline…
The 20 most important events of the last 15 days
Detailed cards, one column per category.
Loading the latest events…
⟵Scroll to see all categories⟶
The CCF Observatory
The newsroom, article by article
Every article the observatory ingests — its CCF summary, full frame profile, named entities, and the events & cascades it belongs to. First a fifteen-day grid, outlet by outlet; then the whole set as richer cards.
Fifteen days, outlet by outlet
…
Outlet by outlet, frame by frame
…
⟵⟶
The CCF Observatory
Our method
We follow the same documented sequence for every article. The corpus is collected continuously, analysed sentence by sentence and compared through semantic representations before events and media cascades are detected.
Our method
The detection stage follows the peer-reviewed CCF media-cascade-detection framework. We publish metadata and derived analyses, never the full text of source articles.
A traceable processing chain
Each operation leaves a stored result for the next one. New articles enter the chain without recomputing observations that have not changed.
-
Step 1
Collect and prepare
New articles from 22 Canadian outlets are collected four times a day, de-duplicated and divided into sentences.
In depth: the outlets we follow → -
Step 2
Annotate each sentence
Models trained and evaluated against human coding identify frames, themes, actors, tone and other analytical categories.
In depth: the 65 annotation categories → -
Step 3
Represent meaning
Each article receives a semantic embedding with bge-m3 and joins the incremental FAISS index used for comparison.
In depth: how the pipeline is built → -
Step 4
Detect events
Articles close in meaning are grouped with average-linkage hierarchical clustering and linked to places when the evidence supports it.
In depth: how events are detected → -
Step 5
Detect media cascades
Temporal concentration, change points, outlet participation and frame convergence identify unusually intense periods of coverage.
In depth: how cascades are detected → -
Step 6
Rank the dominant voices
People and organisations named in the coverage are ranked by influence; crossing names with voice types surfaces the figures of the day — each confirmed, identified and summarized by our model.
In depth: how the figures are detected → -
Step 7
Read the challenges to climate science
Sentences that contest, defend or debate climate science are flagged by our annotations; our model then reads each flagged article, confirms the side, restates the arguments and spots direct attacks on scientists or institutions.
In depth: how the reading works → -
Step 8
Interpret and publish
A local language model on our own servers prepares bilingual titles and summaries for the Observatory, Journal and newsletter.
In depth: the real-time pipeline →
What appears in the Observatory
Distinct events in the news
Events bring together articles about the same occurrence and retain their dates, outlets, places and framing profile.
In depth: detected events →The figures of the coverage
The media pulse ranks the people and organisations dominating climate coverage and names the figure of each voice type — scientist, decision-maker, activist — confirmed by our model.
In depth: dominant voices & figures →Bilingual editorial context
The CCF summaries are written locally in English and French on our own servers. They provide context without redistributing source text.
In depth: the live pipeline →