Visual digital intermediaries and global climate communication: Is climate change still a distant problem on YouTube?

PLOS ONE Visual digital intermediaries and global climate communication groups dedicated to popular science communication (as opposed to individuals, public intel- lectuals, or school classes). For the visual analysis, we used a method centred on computational image analysis to assess the presence of a distancing visual theme that has been strongly represented in con- ventional climate communication such as mainstream media reporting. We sought a means of analysing local connection in visual repertoires as part of assessing a potential break in convention and move towards relatable climate communication. Because of the difficulty in achieving robust large-scale comparative coding of ‘local’ themes, we focused solely on the opposite to local, i.e. the presence of a classic distancing theme. We selected the distancing theme in its arctic rendition: polar bears, icebergs, and ice cliffs in water. Building on the liter- ature outlined earlier, we assume that many find the arctic themes not relatable. This, because these classic themes are geographically remote (for most), devoid of people, and specific to a particular visual repertoire tradition. Moreover, the use of the classic arctic themes suggests thematic continuity rather than innovation. To prepare the visual data for computational image analysis, we segmented the videos into shots (contiguous frames from an individual camera) and selected the central frame from each shot, resulting in about 150 000 images for a set of top-20 results. Following a procedure

reported in greater detail in the Supplementary material, to identify polar bears, we classified the images using a combination of pre-trained and custom deep neural networks, all with a single corresponding label. We also added polar bear images that we discovered in the Google Vision API search for ice images, and manually eliminated false positives. For icebergs and cliffs, we used Google Vision API to identify frames with ice (and related labels) and manually coded the resulting data according to a visual code book (see supplementary material: coding schemes, visual themes). The inter-coder reliability between two coders on this material was Krippendorf ’s alpha .84. The high reliability score can partly be explained by the decision to only consider unequivocal depictions of the themes. For example, pictures coded as icebergs must show an iceberg in a prominent position, with the ice clearly protruding from the water (the visual codebook and examples of positive and negative images are provided in the Sup- plementary material). We note that this approach yields a conservative assessment of the pres- ence of the themes: allowing broader definitions of these themes would have led to an even higher amount of content being coded as icebergs and cliffs. 4.4 Assessing diversity and proximity in top-ranked results We draw on the channel coding and image analysis just outlined to assess the platform pre- sentation in terms of diversity and proximity (RQ1). To assess diversity, we analyse both the

concentration and composition of channels. We consider concentration of channel types using entropy and classification error as measures to examine whether the platform presents a broad or narrow set of voices in top-ranked results. We analyse the composition of channel types to examine the diversification of voice beyond traditional climate communicators (see supplementary material: measures). To assess proximity in top-ranked results, we analyse the local connection of contributing actors (i.e. channels) and visual themes in the video content. To analyse localness of channels, we refer to channel metadata about source location. For visual themes, as discussed earlier, we focus on classic arctic distancing themes as preeminent non-local motifs. We analyse the pro- portional prevalence of these specific themes both in top-20 videos and across lower-ranked results. PLOS ONE https://doi.org/10.1371/journal.pone.0318338 April 8, 2025 9/ 23