Low Power, Rugged Edge Computing provides a low cost, powerful solution for on the ground remote sensing in extreme environments

Ground based remote sensing devices increasingly incorporate low cost single board computers such as a Raspberry Pi to capture and analyze images and data from the environment. Useful and cheap as these devices are, they are not designed for use in extreme conditions and as a consequence often suffer from early failure. Here we describe a system that incorporates a commercially available rugged Edge Computer running embedded Linux that is designed to operate in remote and extreme environments. The AVERT system  (Anticipating Volcanic Eruptions in  Real Time) developed at Columbia University in New York and funded by the Moore Foundation uses solar and wind powered Sensor nodes configured in a spoke and hub architecture currently operating on two volcanoes overseen by the Alaska Volcano Observatory in the Aleutian Islands, Alaska. Multiple Nodes distributed around the volcanoes are each controlled by an Edge Computer which manages and monitors local sensors, processes and parses their data via radio link to a central Hub and schedules system components to wake and sleep to conserve power. The Hub Edge Computer collects and assembles data from multiple Nodes and passes it via satellite, cellular modem or radio links to servers located elsewhere in the world or cloud for near real-time analysis. The local computer enables us to minimize local power demand to just a few watts in part due to the extremely low power sleep modes that are incorporated into these devices. For instance, a Node incorporating a webcam, IRCam, weather station, Edge Computer, network switch, communications radio and power management relays draws only 4.5W on average. In addition, this level of local computing power and a mature Linux operating environment enables us to run AI algorithms at source that process image and other data to flag precursory indicators of an impending eruption. This also helps to reduce data volume passed across the network at times of low network connectivity. We can also remotely interrogate any part of the system and implement new data schemes to best monitor and react to ongoing events. Future work on the AI algorithm development will incorporate local multisensor data analytics to enhance our anticipatory capability.