Landslides on highways from rain: anticipate the closure

Landslides on highways from rain: how to anticipate the closure by slope and km

For a highway concession's operations control center (CCO), the worst landslide is not the largest one. It is the one nobody saw coming. The slope fails over the road, the stretch closes, and the field team goes into emergency mode with no window to prepare anything. The decision to close the road happens when the problem is already on the pavement, not before.

In May 2024, on a single 24 km segment of the BR-470 highway in southern Brazil, between Bento Gonçalves and Veranópolis, 102 slope failure points were recorded in one day. At 8 of those points the road was completely destroyed, with craters opening up. Recovery of that BR-470 stretch reached R$ 700 million in contracts, with anchored retaining walls and two new viaducts. It was not an isolated case: in the same season, the BR-116 highway separately reached 26 full or partial blockages in a single day.

This is the kind of event still treated as force majeure. And it almost never is.

The invisible cost: what nobody puts on the spreadsheet

The landslide that makes the news is the infrastructure repair. What does not make the news is the cost chain it triggers before and after.

When a slope fails and the stretch closes, the concession pays on several fronts at once. Emergency mobilization of crews and machinery off-hours. Assistance to stranded users. Risk to the lives of those on the road at the moment of the event. Contract rebalancing and regulatory exposure. And the whole network feels it. Brazil's CNT Climate Resilience Survey (Nov 2025) found that 70.6% of transport companies had financial losses from climate events in the last five years, and nearly one in four reported losses above R$ 1 million.

The number that hurts is not the repair. It is the repair plus everything that stopped around it. This cost is invisible because it arrives fragmented, under different line items, in different months. But it is expensive not to solve, because it tends to repeat every rainy season.

Why weather forecasting does not prevent a landslide

Here is the confusion that costs money: weather forecasting is not climate intelligence.

Weather forecasting describes the atmosphere. It says it will rain in the region. It does not say whether slope T-14 at km 223 is at high risk, or when, or what your team should do about it. For a control center, "heavy rain in the South" is not a decision. It is a bulletin — one of the inputs to climate intelligence, not the final product.

Climate intelligence starts somewhere else. It begins with business knowledge: which slopes are critical, which stretches cannot stop, what can stop with warning and what needs intervention first. Then comes hyperlocalization, because a landslide does not happen "in the region," it happens at the point. Knowing the risk is at km 223, not in the next town, is the difference between mobilizing one crew and mobilizing ten. On top of that comes situational awareness: what triggered the last event, how that impact should have been handled, what could have been avoided. And the output is not a rain alert. It is a recommendation with a clear action for the right owner.

Public forecasting works at a broad scale, typically tens of kilometers. Geotechnical risk lives on the slope. Between the two sits the closure you did not see coming.

The path to anticipate the landslide

Anticipating is not forecasting the rain earlier. It is turning risk into protocol. The path that works for a control center has six steps:

1. Define risk through business knowledge. Register each critical slope as a monitored asset, with coordinates, geometry and event history. What cannot stop comes first.

2. Forecast with lead time and hyperlocalization. Cross accumulated rainfall and soil saturation with the hyperlocal forecast for that point, not the region. Accumulated rain and soil saturation usually open a 24 to 72 hour window.

3. Understand protocols, impacts and resources. For each risk level, know which crew mobilizes, which machinery, which detour to activate and what each scenario can cost.

4. Alert the right owner. The alert reaches the decision-maker with risk level, likely window and recommended action, not a generic warning lost in a group chat.

5. Trigger the predefined action. With the protocol ready, response time stops depending on improvisation. The crew already knows what to do when risk crosses the threshold.

6. Audit. Record what was forecast, what was decided and what happened. This helps calibrate the model and document the value of anticipation to the contract and the regulator.

The practical difference is easy to see. Reacting to a landslide costs emergency mobilization, a closed stretch and stranded users. Anticipating can cost one crew positioned in advance, at the right point, with hours to spare.

Start with your most critical stretch

You do not need to instrument the whole network to start. You need the stretch that keeps you up at night. Map the critical slopes, cross them with event history and see where anticipation changes the math.

Request a free climate exposure diagnosis for your stretch → Request diagnosis

FAQ

Can you monitor a specific slope? Yes. Each at-risk slope is registered as a monitored asset, with coordinates, geometry and history. The model combines accumulated rainfall, soil saturation and operational triggers to generate an alert dedicated to that point.

How much lead time do I get for a landslide? It depends on the trigger. Accumulated rain and soil saturation usually open a 24 to 72 hour window. Severe storms and gusts, 1 to 6 hours. Rain arriving now, 15 minutes to 2 hours.

Does it work on a stretch with no nearby weather station? Yes, and that is where it adds the most. Hyperlocal forecasting does not depend on a ground station. It combines a numerical model, proprietary reanalysis and Machine Learning to generate risk at any point.

Is this weather forecasting? No. It is closure risk per asset. The alert translates rain and saturation into the probability of the stretch stopping, with a recommended action, not into the chance of rain.