Using AI-enabled wildfire risk maps to communicate risk: the role of labelling, information presentation, perceived trustworthiness and emotion in shaping perceived risk in Veluwe, Netherlands

Rising temperatures and changing climate conditions have increased wildfire risk across the world, including in regions such as The Netherlands that have not historically faced these threats. With this trend expected to continue, understanding risk perceptions among individuals with little to no wildfire experience becomes crucial for mitigating the impacts and designing effective risk communication strategies.

Recent advancements in Artificial Intelligence (AI) wildfire mapping tools have proven highly effective in identifying areas susceptible to wildfires, particularly in detecting low-probability incidents by uncovering subtle patterns often missed by traditional methods. For example, machine learning (ML) wildfire risk maps developed by MEJOR Technologies have accurately predicted wildfire locations in The Netherlands in the past. Despite the potential, the use of these tools as communication instruments to improve wildfire risk perception among the public remains largely unexplored.

Through an online randomised experiment conducted among a sample of residents in the Veluwe area of The Netherlands, we empirically assess how AI-generated labelling (AI label, ML label, or no label) and information presentation formats (map, text, or combined) affect individuals’ perceived wildfire risk. Additionally, we investigate whether perceived trustworthiness in technologies and emotion mediate these effects, providing deeper insights into the cognitive and affective processes that shape how individuals in this area perceive wildfire risk. By leveraging our results, policy makers and AI mapping developers can design effective communication interventions and improve public preparedness in the face of wildfires. While our findings are specific to wildfires in the Veluwe area, they may also hold relevance for understanding the perception of other low-probability hazards among individuals with little to no prior exposure.