A hybrid Zigbee–LoRa sensor network with integrated machine learning for ignition-stage wildfire risk monitoring

Wildfires are increasingly frequent and destructive, making ignition-stage monitoring important for rapid response. This work presents a deployable IoT wildfire-monitoring system based on a two-layer communication design. Zigbee mesh networking connects multiple near-ground temperature and humidity sensor nodes over short distances, while LoRa carries aggregated data over longer distances to a Wi-Fi gateway that uploads readings to a cloud database and web dashboard. The dashboard performs real-time, in-browser Random Forest inference to classify incoming readings into an ignition-risk warning level without relying on a separate machine-learning server. In baseline measurements, Zigbee achieved at least 96% packet delivery up to 50 m hop distance and 91.7% at 80 m in an indoor line-of-sight corridor. LoRa maintained at least 90% packet delivery at 800 m in an open-field line-of-sight configuration. To examine vegetation and non-line-of-sight effects more directly, additional repeated LoRa trials were conducted in open-field and forested settings at 25 m and 50 m. Across 870–872 transmitted packets per outdoor condition, mean packet delivery remained between 99.77% and 100.00%, although the forested trials showed a 10–12 dB reduction in mean RSSI relative to the open field at the same nominal range. With five active sensing nodes, the end-to-end sensor-to-cloud delay averaged approximately 1.8 s, with a maximum of 2.4 s. The classifier achieved a recall of 0.907. These results establish prototype feasibility for near-real-time ignition-risk monitoring, while broader environmental validation, larger-node-count testing, and site-specific model calibration remain necessary before the deployment claims can be extended.