Abstract
Situation awareness (SA) in natural disasters is crucial for timely and effective response efforts. Traditional approaches typically rely on centralized models that lack consideration of regional variability in disaster types, such as floods in coastal areas or earthquakes in seismic zones. A distributed system for situational awareness is crucial because it enhances resilience, redundancy, and reduces reliance on centralized infrastructure by enabling multiple nodes to collaboratively gather and process information, leading to a more robust system. In addition, mainstream machine or deep learning models usually focus on overall accuracy. To address these limitations, we propose a distributed AI framework that can leverage region-specific models to enhance disaster classification and response. Our deep learning model also employs a disaster-specific gating module to selectively focus on and prioritize relevant information. The model with parametric activation functions can help improve feature selection for different disaster types. Our experiments demonstrate that the deep learning model with a disaster-specific gating module can prioritize relevant disasters and improve the accuracy in terms of disaster classification. This paper also highlights the potential of integrating regionally focused disaster models with gated attention deep learning models in a globally perspective distributed framework, enabling more targeted and effective disaster response strategies.