Artificial Intelligence (AI) has become a critical tool in natural disaster management, enabling real-time predictions, risk assessments, and decision-making. However, the effectiveness of AI models can be undermined by spurious correlations—where irrelevant features in the training data influence predictions. These shortcuts, often due to imperfections in training datasets, pose serious risks in high-stakes scenarios.
Wildfire is an integral component of many ecosystems, often necessary for habitat renewal and biodiversity. However, as recent events have shown, wildfires can also inflict severe damage and impacts on communities, infrastructure and the environment. Wildfires pose a growing threat globally, with their frequency and intensity exacerbated by climate change, urbanisation and changes in land use. Predicting fire behaviour is critical for ensuring safety, optimising resource deployment and mitigating damage.
