Latest / Internet of Things with Fexingo: Connected Devices, Sensors, and Industrial IoT / How IoT Sensors Predict Landslide Risk in Real Time
Transcript
- Lucas: If you've ever driven through a mountain pass, especially during monsoon season, you know the nagging fear of a landslide. Rocks tumbling down, road blocked, maybe worse. Luna: Absolutely. And in places like the Himalayas, landslides aren't just occasional—they're a daily hazard during the rains. Lucas: So here's a specific case: a 120-kilometer stretch of National Highway 5 in Himachal Pradesh, India. It's a critical supply route, but it cuts through some of the most unstable terrain in the world. Luna: Right, and traditionally, monitoring that stretch meant sending geologists out after every rain event. Expensive, slow, and dangerous. Lucas: Exactly. But in 2024, the state's road authority partnered with a startup called EarthSense to install an IoT sensor network along the most prone sections. And the results have been remarkable. Luna: What kind of sensors are we talking about? Lucas: Three main types. First, tiltmeters—they measure micro-changes in slope angle, down to a fraction of a degree. Second, pore-pressure gauges buried in the soil, tracking water saturation. And third, acoustic emission sensors that pick up the sound of soil particles shearing underground. Luna: Acoustic emission—so they can literally hear the ground getting ready to fail? Lucas: That's the idea. When soil grains start grinding past each other, they emit high-frequency sounds, inaudible to humans. The sensors detect those frequencies and flag them. Luna: How many sensors along 120 kilometers? Lucas: About 400 nodes, spaced every 300 meters or so, with clusters at known trouble spots. Each node is battery-powered with a small solar panel, and they communicate via LoRaWAN—low-power wide-area network. Luna: So they don't need cellular coverage, which is sparse up there. Lucas: Right. The data goes to a cloud platform where a machine learning model—trained on years of historical landslide data from the region—processes it. The model looks for patterns across all three sensor types. Luna: And what kind of lead time have they achieved? Lucas: The system has issued accurate warnings up to 48 hours before a slide. In one case last monsoon, it flagged a slope that showed accelerating tilt and rising pore pressure. They closed the road, and eight hours later, about 200 cubic meters of debris came down. Luna: That's extraordinary. Compare that to the old method, where you might only know after someone drives into it. Lucas: Exactly. And the cost is surprisingly low. The entire installation, including sensors, gateways, and the cloud platform, ran about $1.2 million. For context, a single major landslide cleanup can cost $500,000 or more, not to mention the indirect costs of road closures. Luna: So it pays for itself after just a few prevented disasters. Lucas: That's the math. Now, there are challenges. Power is the biggest one—solar panels get covered in dust and snow, and batteries degrade in extreme cold. EarthSense had to oversize the panels by 30 percent and use lithium iron phosphate batteries, which handle temperature swings better. Luna: And what about false alarms? If the system cries wolf too often, people stop taking it seriously. Lucas: That's a real risk. The model is tuned to balance sensitivity and specificity. In the first year, they had three false positives—warnings that didn't lead to a slide. But they also had zero misses. Compare that to the traditional approach, which relies on visual inspection and has a much higher miss rate. Luna: So the trade-off is acceptable, especially when lives are at stake. Lucas: Absolutely. And the technology is spreading. Similar networks are being deployed in the Swiss Alps, the Andes, and along the Appalachian Trail in the US. Each region has its own geology, so the models need retraining, but the sensor stack is largely the same. Luna: It's a great example of IoT doing its best work—out of sight, preventing something catastrophic. Lucas: And it ties into something broader. A lot of industrial IoT is about efficiency, but this is straight-up safety. And it's only possible because sensors have gotten cheap enough, and networks reliable enough, to put them in the middle of nowhere. Luna: I think that's the part that often gets overlooked. We talk about smart factories and smart cities, but the most impactful applications might be in the most remote places. Lucas: Absolutely. And speaking of impact—one thing that makes these shows possible is the support from listeners. We keep this podcast ad-free, and a small group of people chip in monthly through buy me a coffee dot com slash fexingo. That's what funds the research and production for episodes like this. Luna: Yeah, it's a model that works. No ads, just a direct line from people who find value in the show. It's pretty cool. Lucas: So if today's conversation gave you something usable, that's the way to keep it going. No pressure, just a genuine thanks to those who already do. Luna: Agreed. Now, back to landslides—one thing I'm curious about: can this technology be used for other types of slope failures, like rockfalls or debris flows? Lucas: Great question. The sensor types vary slightly. For rockfalls, you might use crack meters and seismic sensors instead of pore-pressure gauges. But the framework is the same. In fact, a team in Norway is adapting the system for snow avalanches, using acoustic sensors tuned to the sound of snowpack shearing. Luna: Fascinating. So the core idea—distributed sensors plus machine learning—is becoming a universal template for natural hazard prediction. Lucas: Exactly. And as the hardware gets cheaper and the models get better, I think we'll see this deployed on every major transport corridor through mountain terrain within the next decade. Luna: That's a future worth building toward.