WaterSoftHack: Best Practices for Advancing Cyberinfrastructure and Machine Learning Training in Water Science
Abstract
The 2026 cycle added a three-day module on cloud and edge computing, delivered 21 to 23 July to all 17 Fellows. Each day paired a two-hour lecture with a one-hour hands-on workshop, and one scenario ran across all three days: a real-time flood early-warning system built from a network of USGS stream gauges.Day 1, Cloud. Utility computing, service models, containers, storage, cost, and the hydrologic archives that already sit in cloud object storage.Day 2, Edge. Bandwidth, power, and latency in the field, on-device detection, adaptive sampling, and model compression.Day 3, Hybrid. Five patterns that combine the two layers, an MQTT message bus, sensor fusion, and graceful degradation.Every session ran in a browser on a shared Jetstream2 JupyterHub funded by 200,000 NSF ACCESS credits. No Fellow installed software locally, and the three afternoon workshops together cost about one hundred credits.