Toward a Universal Sensor: Predictive Inference for Adaptive Sensing in Wildfires

As climate-driven disasters such as wildfires become more frequent and unpredictable, there is an increasing need for sensing systems that can adapt to dynamic, uncertain environments. This paper presents the Universal Sensor, a cognitive-inspired, software-defined architecture grounded in predictive processing. The system operates as a continuous loop of prediction, sensing, comparison, and control, enabling real-time adaptation to changing conditions. It combines predictive modeling, adaptive multi-sensor data collection, and sensor fusion to interpret context and respond efficiently. Feedback-driven control and a modular policy memory support proactive resource management and lay the groundwork for future analogical transfer. We evaluate the system using a wildfire simulation based on data from the 2016 Fort McMurray event, demonstrating improved adaptability and resource use compared to static sensing strategies. In an 8-day simulation with 2,000 sensors, Universal Sensors reduced transmissions, energy use, and data volume by 36%, while preserving over 90% of critical hotspot events. These findings illustrate how predictive adaptation can enable efficient sensing at scale without sacrificing critical event coverage.