Latest / Internet of Things with Fexingo: Connected Devices, Sensors, and Industrial IoT / How IoT Sensors Are Preventing Landslide Disasters
Transcript
- Lucas: You know how some disasters feel like they come out of nowhere — a sudden wall of mud and rock tearing through a village? Landslides are like that. But the thing is, the ground almost always gives warnings. We just haven't been listening closely enough. Luna: Until now, you mean. Because IoT sensors are basically putting a stethoscope on the hillside. Lucas: Exactly. And there's a pilot project in the Swiss Alps that's become the gold standard. It's above the village of Brienz — a community of about a hundred people that sits directly below a slope that's been moving for decades. In 2023, the movement accelerated dramatically. The local authorities had to evacuate the entire village twice. Luna: Twice? That's a huge disruption. Did the sensors trigger those evacuations? Lucas: They were a major part of the decision. The Swiss Federal Institute for Forest, Snow and Landscape Research — it's called WSL — they'd installed a dense network of sensors on that slope. We're talking about soil moisture probes, tiltmeters that measure changes in angle down to a thousandth of a degree, and acoustic emission sensors that literally listen to the sound of rocks cracking underground. Luna: So they had real-time data on water pressure and ground movement. That's a lot more precise than sending a geologist out once a month with a clipboard. Lucas: Right. The key metric they track is something called pore water pressure — basically, how much water is filling the gaps between soil particles. When that pressure spikes, it lubricates the layers and reduces friction. The slope becomes unstable. The sensors in Brienz detected a sustained rise in pore pressure days before the first evacuation. Luna: And that gave authorities enough lead time to move people out safely. Lucas: Exactly. Now, the slope didn't actually fail in 2023 — it slowed down after heavy rain stopped. But the system proved its concept. And it's running continuously, feeding data into a machine learning model that's trained to distinguish between normal seasonal variation and a precursor to failure. Luna: I want to dig into that machine learning piece. How do you train a model on landslides when landslides are rare events? Lucas: That's the classic class imbalance problem. The WSL team used data from controlled experiments — they actually built test slopes in the lab and triggered failures while monitoring the sensor signatures. They also fed in decades of historical data from other Alpine sites. The model learned patterns like: when tilt rate exceeds X degrees per hour AND acoustic emissions spike above Y decibels simultaneously, that's a high-risk signal. Luna: So it's a multi-sensor fusion approach. One sensor alone could give a false alarm — a cow bumps into a tiltmeter, for instance — but when three different types agree, you pay attention. Lucas: Precisely. And false alarms are a real concern. If you evacuate a village three times and nothing happens, people stop listening. The researchers in Brienz claim their model has a false-positive rate below five percent. They haven't missed a real acceleration event yet. Luna: That's impressive. But what about power? These sensors are on a remote mountainside. You can't run extension cords. Lucas: Great question. Most of the sensors are low-power and run on small solar panels with batteries. The tiltmeters draw about 50 milliwatts when transmitting, and they only send data every 15 minutes unless they detect an anomaly — then they switch to one-minute intervals. The acoustic sensors are a bit hungrier, but they're paired with larger battery packs that can last through a Swiss winter with minimal sun. Luna: That's a classic IoT engineering problem: power budget versus data resolution. Lucas: And they solved it with edge computing. The sensors do some basic processing locally — they calculate running averages and standard deviations — and only send the raw waveform data when the numbers exceed a threshold. That cuts data transmission by roughly ninety percent, which saves battery life and reduces cellular data costs. Luna: Cost is a big factor for scaling this beyond wealthy Alpine nations. I've read that a single sensor node can run a few thousand dollars. That's not cheap for a developing country. Lucas: You're right, and it's a real barrier. But there are projects in Colombia and Nepal that are adapting this approach with cheaper components. In Colombia, near the city of Medellín, researchers from the National University have deployed a network using off-the-shelf soil moisture sensors that cost about fifty dollars each, paired with Arduino-based data loggers. They're measuring rainfall and ground movement in informal settlements built on unstable hillsides. Luna: And the trade-off is less precision, but more coverage. Lucas: Exactly. You don't need sub-millimeter accuracy if you're looking for big shifts. The Nepali project, run by ICIMOD, uses a similar low-cost approach in the Himalayas. They've installed about forty sensor nodes along the Kali Gandaki valley, an area prone to landslides during the monsoon. The data is transmitted via LoRaWAN — long-range, low-power radio — to a central hub in the nearest town. Luna: LoRaWAN can go several kilometers in open terrain. That's perfect for a valley. Lucas: Right. And the hub has satellite internet, so the data reaches researchers in Kathmandu in near-real time. The system has already detected a slope deformation event in 2025 that led to a road closure before a major slide. No casualties. Luna: That's the kind of outcome you want. But I want to go back to the machine learning side for a second. In these low-cost systems, are they using the same kind of multi-sensor fusion, or do they rely more on a single data stream? Lucas: Mostly single-stream for now — primarily soil moisture and rainfall intensity — because the cheap sensors don't include tiltmeters or acoustic emissions. But the Colombian team is experimenting with smartphone accelerometers mounted on poles as a crude tilt sensor. It's not as sensitive, but for a budget of under a hundred dollars per node, it gives you a useful signal. Luna: So the trade-off is acceptable when you're trying to protect communities that otherwise have zero warning. Lucas: Exactly. And that brings up an interesting philosophical shift in how we think about natural hazards. Traditionally, landslide risk is assessed through geological maps and historical records — static data that might be years out of date. IoT sensors turn it into a dynamic, real-time system. You're not just saying 'this area is high risk,' you're saying 'this area is moving right now.' Luna: That's a huge difference for emergency managers. They can allocate resources based on current conditions, not just probabilistic models. Lucas: And it's not just landslides. The same sensor types are being used for sinkhole detection, dam stability monitoring, and even avalanche prediction. The underlying principle is the same: measure the slow, quiet changes that precede a catastrophic failure. Luna: It makes you wonder how many disasters we could prevent if we deployed these networks widely. The cost of a single landslide in terms of lives and infrastructure is enormous. Lucas: That's the promise. The Swiss system in Brienz costs about five hundred thousand dollars to install and maintain for a decade. The alternative — a full-scale engineering stabilization project — would run into the tens of millions. And even then, you can't stabilize every slope. Sometimes the smartest move is just to listen and get out of the way. Luna: Listen and get out of the way — that's a good summary. So if this episode gave you a new way to think about technology and safety, maybe it was worth the price of a coffee. If so, you can find us at buy me a coffee dot com slash fexingo. We're ad-free and listener-supported, and that support keeps conversations like this coming. Lucas: Absolutely. Every bit helps us cover topics that don't make the headlines but matter deeply. Now, back to Brienz — one detail I find fascinating is that the sensor network there also monitors groundwater chemistry. When rock fractures open, they expose fresh mineral surfaces, which changes the pH and dissolved ion content of the water. Luna: So you can detect fracturing before the tiltmeters even register movement? That's almost like a chemical early warning. Lucas: Exactly. The WSL team added a few conductivity and pH sensors to the monitoring wells, and they've seen spikes in calcium and magnesium ions up to 48 hours before acceleration events. It's still experimental, but it could add another layer of lead time. Luna: That's incredible. So we're moving from 'the ground is moving' to 'the ground is about to move'? Lucas: That's the direction. And with lower-cost chemical sensors coming onto the market, that kind of multi-parameter monitoring could become accessible to places like Nepal and Colombia. The technology is evolving fast. Luna: What about integration with other data sources? I know satellite-based InSAR can detect ground deformation over large areas. Could you combine satellite data with ground sensors for a kind of early warning hierarchy? Lucas: Absolutely. In fact, that's exactly what the European Ground Motion Service is doing. They use radar satellite data to identify hotspots of deformation, then deploy ground sensors in the most active areas. It's a triage approach: satellite for broad surveillance, IoT for targeted monitoring. The Brienz project actually started because InSAR data showed the slope was moving several centimeters per year. Luna: So the satellites tell you where to look, and the sensors tell you when to act. Lucas: That's the vision. And it's spreading. The US Geological Survey is piloting a similar system in Washington state, along the Stillaguamish River valley where the 2014 Oso landslide killed 43 people. They've installed a network of 15 sensor nodes that measure rainfall, soil moisture, and ground vibration. Luna: Oso was a tragic reminder that traditional hazard maps aren't enough. That slide was in an area that wasn't considered high risk. Lucas: Right. And the sensors there have already detected a few minor slides that didn't reach populated areas, but they validated the system. The data is publicly available, so researchers can refine their models. Luna: Public data is huge for building trust and advancing the science. Lucas: It is. And it's one reason why the push for open IoT standards in geohazard monitoring is gaining momentum. The Open Geospatial Consortium has a standard called SensorThings API that allows different sensor networks to share data in a common format. That way, a city in Colombia can use the same visualization tools as a research institute in Switzerland. Luna: Interoperability. That's where IoT really delivers its value — not just the sensors themselves, but the network effect of connecting them. Lucas: Exactly. And I think that's the takeaway. Landslides are a natural hazard, but with the right sensor network, they don't have to be a natural disaster. We have the tools to give people the time they need to get to safety.