Narratives, Emotions and the Logic of Disinformation During Natural Disasters: Evidence from X in the 2024 DANA in Spain

Abstract Natural disasters generate intense online communication, where messages providing practical information and mutual support coexist with political conflict and mis/disinformation. However, relatively little is known about how competing narratives evolve over time and how they shape user engagement and information diffusion. This paper examines more than 168,000 original posts published on X to study the interplay between narratives, emotions and engagement throughout the 2024 DANA floods in Spain. Tweets were classified into narrative categories using a supervised multi-label classifier based on multilingual sentence embeddings and logistic regression. We combined this classification with sentiment and emotion measures to estimate a series of regression models aimed at analyzing both user engagement and content diffusion. In the first days after the floods, most conversations focused on help, solidarity and practical information. As the crisis unfolded, however, political blame and mis/disinformation attracted disproportionately high levels of engagement. Anger was the emotion most strongly associated with amplification. Optimism, however, was also linked to higher engagement. Climate-related narratives followed a different trajectory. They peaked immediately after the floods, when discussion centered more heavily on possible explanations for the event, and then declined as attention shifted towards institutional responsibility. Overall, posts combining strong emotional content with political conflict received more engagement, especially when they expressed anger or assigned blame. These patterns matter for how public institutions communicate during crises and for how platforms respond to misleading and highly polarizing content.