H108-0001
Seamless Machine Learning Architecture Access to Long-Tail In-Situ Data through EarthCube's BALTO Cyberinfrastructure

Friday, 11 December 2020
Poster
Daniel R Fuka1, Robin R White2, Barbara Roqueto dos Reis3, Roja Kaveh Garna4, Elyce Buell4, Amy Collick5, Agbeli Ameko6, D. Sarah Stamps7, James H R Gallagher8, Scott Dale Peckham9, David W Fulker8 and Zach M Easton1, (1)Virginia Tech, Department of Biological and Systems Engineering, Blacksburg, VA, United States, (2)Virginia Polytechnic Institute and State University, APSC, Blacksburg, United States, (3)Virginia Polytechnic Institute and State University, Blacksburg, VA, United States, (4)Virginia Polytechnic Institute and State University, Biological Systems Engineering, Blacksburg, VA, United States, (5)University of Maryland Eastern Shore, Princess Anne, MD, United States, (6)National Center for Atmospheric Research, GLOBE, Boulder, United States, (7)Virginia Polytechnic Institute and State University, Department of Geosciences, Blacksburg, United States, (8)OPeNDAP, Inc., Butte, MT, United States, (9)University of Colorado, Institute of Arctic and Alpine Research (INSTAAR), Boulder, CO, United States
Abstract:
The EarthCube brokering cyberinfrastructure BALTO (Brokered Alignment of Long-Tail Observations) provides streamlined access to long-tail data using Web Services. BALTO consists of several distinct brokering mechanisms that we have developed since the project’s inception in 2016. We created an extension for the OPeNDAP framework Hyrax, which is software that serves big data from USGS, NASA, and numerous other sources. To develop BALTO we included several new data handlers to allow secure attachment to familiar long-tail data sources. As a use-case for the project, we developed methods to enable career or citizen scientists to make their in-situ IoT based sensor data collection efforts available to the world. In this presentation, we demonstrate extending access of these data to act as a workflow training TensorFlow Lite in-field event and/or anomaly recognition. We share initial capabilities of the BALTO system coupled with machine learning architectures as an AI Edge Processing training system. To do this, we test the coupled environment using high-frequency sensors in the field, passed through the trained networks running on inexpensive microcontrollers, to recognize events or anomalies the recognition of which can be transmitted over low bandwidth LoRa networks from remote locations affordably.
Previous Abstract | Next Abstract >>