Before a fire is hot enough for a thermal camera to see, it already smells. A network of low-cost electronic-nose sensors detects the volatile organic compounds (VOCs) of early combustion and raises the alarm in real time.
Smouldering vegetation releases a plume of volatile organic compounds — the chemical fingerprint of a fire — minutes to hours before there is enough heat or smoke for cameras and satellites to detect. An electronic nose reads that fingerprint at the source, on the ground, where the fire starts.
Need a visible plume or a heat signature — they see a fire once it is already established and line-of-sight allows.
Smells the first volatile compounds of combustion at ground level, in the dark, behind ridges — the earliest possible signal.
A gas sensor array continuously samples the air for the VOC mixture of early combustion, alongside temperature, humidity and air pressure.
An embedded machine-learning model, trained on the specific smell of a Karst wildfire, separates it from ordinary background air to keep false alarms low.
On a positive detection the node sends an alert over long-range LoRaWAN — kilometres of reach through forest, no mobile coverage needed.
The platform pinpoints the node, raises a real-time alarm to responders and drops the location onto the operational map.
Each sensor is mounted on a tree in a strategic, vulnerable zone — and is built to run unattended.
Gas-sensor array for volatile organic compounds plus a micro-climate suite — the same data also enriches the fire models.
Long-range, low-power radio — kilometres of reach, no SIM or mains, mesh-friendly across a roadless plateau.
A small photovoltaic panel and battery keep the node alive through the night and the season — no cabling, no maintenance visits.
No single sensor catches every wildfire. The Karst Firewall approach stitches five complementary layers together — each one good at a different moment — so that whichever raises the alarm first, the operational platform turns it into a dispatch in seconds.
Smells the chemical fingerprint of early combustion at ground level — even behind ridges or in the dark.
Fixed long-range thermal cameras on hill-top masts pick up heat signatures across line-of-sight valleys, day and night.
Operator-launched multispectral drones confirm an alarm, pin the exact perimeter and guide responders to the ignition point.
Sentinel-2, MODIS and VIIRS feeds (Copernicus + EFFIS) detect active fires and map burned area across the whole AOI.
Hikers, farmers and residents report smoke or flame through 112 / 113 and the project app — every call is geocoded and surfaces on the operational map alongside the sensors.
All five layers funnel into the same operational cockpit. The platform de-duplicates concurrent signals, prioritises by KFWI risk in that cell, and routes a single confirmed alert to the responsible civil-protection team — over email, SMS and WhatsApp. How satellites & drones plug in →
A live snapshot of the operational electronic-nose deployment and the LoRaMIP gateways that ferry their data into the platform.
| Sensor | Status | Temp (°C) | RH (%) | Pressure (hPa) | Last seen | Gateway |
|---|---|---|---|---|---|---|
| office | offline | 32.6 | 38 | 980.9 | 40 d | — |
| demo-sgonico-N | offline | 25.7 | 55 | 1,006.8 | 63 d | d0000001 |
| demo-sgonico-S | offline | 25.3 | 58 | 1,004.4 | 63 d | d0000001 |
| demo-lipica-N | offline | 25.4 | 64 | 1,006.9 | 63 d | d0000002 |
| demo-lipica-S | offline | 25.5 | 65 | 1,004.9 | 63 d | d0000002 |
| Test_Ufficio_pcb_v2 | offline | 30.4 | 35 | 974.8 | 25 d | — |
| ufficio ext2 | offline | 27.0 | 41 | 981.2 | 4 d | — |
| ufficio ext | offline | 29.5 | 45 | 982.3 | 20 d | — |
| ufficio ext3 | offline | 31.6 | 34 | 977.5 | 1 d | — |
| test_deploy_5 | offline | 25.0 | 50 | 1,013.3 | 2 d | — |
| test_deploy_3 | never | — | — | — | — | — |
| test_deploy_4 | never | — | — | — | — | — |
| test_deploy_1 | never | — | — | — | — | — |
| test_deploy_2 | never | — | — | — | — | — |
| test_deploy_6 | never | — | — | — | — | — |
| testbench | offline | 29.6 | 33 | 979.5 | 6 d | — |
| 0000012d | live | 28.6 | 36 | 974.0 | 2 min | 000000ae |
| 000000e2 | live | 28.1 | 37 | 973.8 | 24 s | 000000ae |
| 00000131 | live | 28.6 | 34 | 973.8 | 12 s | 000000ae |
| 0000012c | live | 28.4 | 36 | 973.7 | 2 min | 000000ae |
| Gateway | Model · firmware | Network | MQTT | InfluxDB | Paired | MQTT msgs | Last seen |
|---|---|---|---|---|---|---|---|
| KF-gateway-Aurisina 000000ae |
loramip · 02000005 | connected | connected | live | 60 | 3113502 | 9 s |
Source: live inventory in the operational platform (iot_sensor + iot_gateway). Status fields are pushed by kf50-observation-ingest from each gateway every 60 s; freshness compares the latest reading against the sensor's expected reporting interval.
Beyond the instant alarm, the continuous stream of on-the-ground temperature, humidity, pressure and air-chemistry feeds the Karst Firewall digital twin. Real micro-climate from inside the forest makes the fire-spread simulator more precise — and the same network doubles as a dense environmental-monitoring grid.
VOC detection before flame or plume.
Placed where hazard is highest.
Years in the field, unattended.
Trained to know wildfire smell.