SAGE4ARM

ARMing the Edge: Demonstration of Edge Computing

1 August 2020 - 1 August 2024

Lead Scientist: Scott Collis

Observatory: SGP

Earth System Processes span vast spatial and temporal scales. Complex instrumentation (lidar, radar, hyper-spectral imaging, etc.) frequently perform irreversible in-machine processing and have a large parameter space of configurations. Not all processing and configurations are suitable for all phenomena. When it comes to adaptive processing or adaptive sampling (storm tracking with radar, for example), scientists and research institutions tend to "roll their own". This is where Sage differs. By separating out a common requirement for a variety of applications, the cyberinfrastructure ─ Sage ─ aims to provide an open community platform of hardware and software at the edge that can be tailored, reused, and grown by the atmospheric and oceanic sciences community (and far beyond) to perform processing and control right up against the instrument collection point, which is known as the edge. ARMing the Edge: Demonstration of Edge Computing, will deploy two edge computing devices, one to be installed in a 19-inch rack (Sage Blade) and one rugged device with camera systems (Wild Sage Node), and provide a proof of concept (of edge computing) by answering the question: Can we do machine learning on Doppler lidar (DL) spectra in order to classify the sky as clear, cloudy, or rain? Note: Although the final aim of Sage is to control configurable (e.g., scannable) instruments, we are not asking this of ARM instruments. The Wild Sage node will have a pan tilt zoom (PTZ) camera that will act as a “exemplar” scanning instrument. We will address a slew of Edge questions. For example, when we detect rain in the DL spectra we can dispatch the camera system to observe instruments sensitive to wet radome (microwave radiometer, Ka ARM Zenith Radar, etc.). When clouds are detected we can use the PTZ system to perform all-sky scans and use computer vision or machine learning (artificial intelligence/ML) techniques to derive cloud fraction. Finally, this deployment will guide the design and implementation of Sage (a $9M National Science Foundation Mid Scale Research Infrastructure project http://sagecontinum.org) to ensure it will meet the needs of facilities like ARM.