SIParCS 2025 - Adebowale Adelekan

radIOTic: An Intelligent, Fault-Tolerant, AI-Enhanced Mesonet for Atmospheric Sensing

Recorded Talk

radIOTic is a scalable, intelligent, and fault-tolerant atmospheric mesonet designed for community deployment. Built on LoRa-enabled microcontrollers with heterogeneous sensor configurations, each station autonomously transmits environmental data through nearby LoRa-enabled Raspberry Pis (gateways) to a cloud-based microservices architecture. The cloud backend, hosted on CIRRUS, includes scalable services like PostgreSQL for structured storage, Redis for fast-access caching, and ThingsBoard for seamless station management and real-time visualization.

At the frontend, a dynamic interactive map powered by Redis reflects the live status of each station—including mobile nodes tracked via GPS. Stations that disconnect are automatically removed, while newly connected ones appear in real time. The map also integrates the open-source Gemma AI model to generate human-readable summaries of sensor readings, enhancing interpretability. For forecasting, radIOTic incorporates the CREDIT model to predict conditions at the grid-cell level.

Each station intelligently selects the optimal gateway using a ping-pong protocol that evaluates signal strength and gateway load, enabling both load balancing and reliable communication. In the absence of a direct Pi connection, stations may relay data through neighboring stations, forming ad hoc chains to ensure uplink availability. Load-aware routing ensures stations can decline relays when overwhelmed, preserving overall system health. A keep-alive mechanism further enables resilience to node failures.

Designed for plug-and-play replication, radIOTic empowers communities to build and extend their own localized mesonets with minimal setup. Its open, modular architecture supports a wide range of deployments—from fixed installations to mobile weather sensors—while providing robust data infrastructure and intelligent coordination at every layer.

Mentors: Agbeli Ameko, Keith Maull, John Schreck

Slides and poster