Emotional divergence in urban climate discourse: spatial patterns and semantic decoding across global cities

Climate discourse on social media is increasingly influential in shaping public interpretations of climate risks and motivating urban climate action, yet its underlying emotional and semantic architectures remain insufficiently understood. Here, we analyze 104,274 climate-related tweets (Dec 2021-Mar 2023), of which 56,556 are geolocated, using hybrid sentiment classification, latent topic modeling, and spatial clustering to examine spatiotemporal dynamics and urban-scale variation in affective expression across 1129 cities in 78 countries. We find that climate sentiment is highly event-responsive yet structurally constrained, with emotional extremity rarely exceeding 15% even during major climate shocks. Spatially, cities in the Global North show more systematically negative or ambivalent sentiment, whereas cities in the Global South tend toward positive or activist-oriented optimism. Twelve discursive themes are identified. Semantic-affective decoupling further shows that adaptation- and development-related themes correlate with positive sentiment, while denial- or institutional-critique narratives are predominantly negative. Further, spatial clustering reveals that major metropolitan areas converge into shared emotional-discursive regimes regardless of geographic location, whereas smaller and peripheral cities diverge markedly in their group affective tone. Understanding the interplay among emotion, meaning, and urban context can inform climate narratives and support the development of context-sensitive urban climate communication strategies.