How climate change can destabilize a stable road cut

 

Road cuts are designed to be boring.
The geometry is fixed, the weakest layer is identified, drainage is installed, and the expectation is that the slope will spend most of its life with a comfortable safety margin. Failures are supposed to be rare, event driven, and localized.

What I kept wondering was simple: what happens to a road cut that is stable today when rainfall becomes more intense and more persistent? Not steeper. Not re excavated. Just exposed to a different climate.

To explore that question, I built a small interactive model implemented in JavaScript as a browser based sandbox, focused on shallow planar failures in road cuts. The goal was not to predict a specific landslide at a specific site, but to isolate the mechanisms that matter and to see how a stable system can change regime under altered rainfall forcing.

What follows is a mechanism driven analysis, not a calibrated forecast.

The setup

The geometry is kept constant across all scenarios. The cut height is 15 m and all parameters can be modified in the interactive sandbox. The cut angle is 60°, and the failure surface is a weak layer dipping at 32° that daylights into the cut. The potential slip surface is located 3 m below the road surface. Unit weight is 19 kN/m³.

The material strength of the weak layer is represented by cohesion and friction angle with uncertainty. Cohesion has a mean of 12 kPa with a standard deviation of 4 kPa. Friction angle has a mean of 28° with a standard deviation of 4°. For each simulated day, 4000 Monte Carlo realizations are run to capture this uncertainty.

The factor of safety FS(t) is computed in the standard way, and the probability of failure Pf(t) is defined as the fraction of realizations where FS < 1. A warning threshold is set at Pf = 0.35, not as a collapse criterion but as a practical indicator of elevated risk.

Rainfall forcing and data origin

Rainfall is generated synthetically using four parameters: mean daily precipitation, storm day frequency, storm persistence, and storm intensity multiplier. This structure allows control over both rainfall intensity and clustering, which is critical for climate related analyses.

The baseline climate parameters are anchored to publicly available Norwegian precipitation statistics for the 1991 to 2020 reference period, using values representative of relatively dry coastal regions such as southern Norway. Future scenarios are not direct climate model outputs, but controlled perturbations of these baseline statistics, consistent with documented Norwegian climate projections showing increased precipitation, more frequent heavy rainfall, and greater event persistence.

Baseline and future runs use the same geometry and the same random structure, so differences in stability arise from changes in rainfall statistics rather than different storm timing.

Hydrological response

Hydrology is intentionally simple. Rainfall drives a wetness index θ(t). Drainage removes a fraction of water each day, and pore pressure u(t) activates only once θ exceeds a threshold. This is not a groundwater model. It is a proxy for how repeated rainfall pushes a weak layer toward saturation and pore pressure generation.

Two optional mechanisms can be switched on to reflect common real world behavior:

• Impaired drainage when very wet, representing perched water, limited outflow capacity, clogged drains, or low permeability shear zones.
• Wet weakening and loss of suction, representing the disappearance of unsaturated strength and a modest reduction of strength parameters under very wet conditions.

In Scenario 1, both optional mechanisms are switched off to represent a well functioning drained slope under typical conditions. In Scenarios 2 and 3, they are switched on to represent how real slopes often behave once the system enters a very wet regime, where drainage becomes less effective and unsaturated strength is gradually lost.

All simulations are run over a period of 365 days, allowing the model to capture seasonal behavior, rainfall clustering, and recovery between wet periods. This time window is long enough to observe whether the slope repeatedly returns to stable conditions or instead transitions into a persistently marginal or unstable regime.

Scenario 1: current climate, 1991 to 2020 like

In the first scenario, rainfall statistics correspond to recent historical conditions. Storms occur, but clustering is limited and recovery between events is common. Baseline rainfall parameters are representative of southern Norwegian coastal climates, such as the Kristiansand region, and are based on publicly available precipitation statistics.

The overall behavior reflects what an engineered road cut is expected to show under present day conditions. The factor of safety remains above 1 and essentially constant because pore pressure rarely activates and the optional wet regime mechanisms are switched off. The probability of failure shows short lived fluctuations, but the maximum Pf remains low, around 7 to 8 percent, and there are zero days above the alert threshold Pf ≥ 0.35.

The risk summary table confirms this interpretation. Baseline and forecast results are nearly identical in this scenario: the maximum Pf differs by less than one percentage point, the day of maximum Pf is the same, and no days exceed the alert threshold in either case. The longest run above the threshold is zero days, and there is no first exceedance day. In practical terms, this means that under current climate conditions the slope never enters a high risk state, even temporarily.

This lack of change is not a bug or a trivial outcome. It validates the baseline configuration. With this geometry and these strength parameters, the slope is conditionally stable and behaves as expected under realistic present day rainfall forcing. The model is not predisposed to failure and does not generate artificial instability in the absence of sustained wet conditions.

One detail is worth noting: the factor of safety appears flat in this scenario. That is a direct consequence of the modelling choices. In this model, FS becomes time dependent primarily through pore pressure activation and wet regime weakening. If neither mechanism activates, FS remains constant by design.

That does not mean risk is zero. FS is a single snapshot computed from mean strength. Pf is a risk profile that accounts for uncertainty in cohesion and friction angle. In Scenario 1 the slope remains stable on average, but there is still a small probability that unfavorable strength combinations produce FS below 1. That is why Pf fluctuates around 7 to 8 percent even while FS looks unchanged.

Scenario 2: mid century worsening, 2050 like

The stable behavior observed under current climate conditions provides an important reference: it shows that the slope is not inherently unstable, and that the changes discussed below are driven by altered rainfall forcing rather than geometry or material properties.

In the second scenario, mean precipitation increases modestly, storms become more frequent, and rainfall persistence increases. In the figures, the black curves represent baseline conditions, while the red curves represent the forecast climate. The change is not dominated by a single extreme event. Instead, rainfall becomes more clustered, reducing the time available for drainage and recovery between storms. In particular, note whether the red curves return to low Pf after storm clusters, or whether they remain elevated, which is the signature of incomplete hydrological recovery.

This difference in rainfall structure drives a clear shift in slope behavior.

In the rainfall time series, both baseline and forecast show similar peak intensities. However, in the forecast case, wet periods occur closer together. As a result, the hydrological system does not fully drain between events. The wetness index remains elevated for longer periods, allowing pore pressure to activate more frequently and persistently, and enabling wet weakening mechanisms to become relevant.

The transition from episodic to persistent instability is clearly visible in the probability of failure Pf(t). In the baseline case, Pf remains low for most of the year, with a short period of elevated risk around mid year. In the forecast case, Pf rises sharply and then remains high for most of the remaining simulation. The visual transition occurs around day 93. Up to that point, the forecast curve behaves similarly to the baseline, with discrete spikes followed by recovery. After approximately day 93, the system enters a sustained wet regime: Pf no longer returns to low values and instead stays close to unity.

The risk summary quantifies the magnitude of this change. Under baseline conditions, the slope exceeds the alert threshold Pf ≥ 0.35 on 19 days, corresponding to about 5.2 percent of the year. Under the forecast climate, the same threshold is exceeded on 256 days, or approximately 70.1 percent of the year. The longest continuous run above the threshold increases from 14 days in the baseline to 255 days in the forecast, indicating that once the forecast scenario crosses into the high risk regime, it effectively remains there for almost the entire remaining year.

The severity of the risk also increases. The maximum probability of failure rises from 82.0 percent in the baseline to 98.3 percent in the forecast, meaning that during the most critical period the vast majority of strength realizations indicate instability. In addition, the onset of elevated risk shifts earlier in the year: the first day with Pf ≥ 0.35 occurs on day 69 in the forecast scenario, compared to day 144 in the baseline. This implies a longer vulnerability season, not just higher peak risk.

The factor of safety plot supports this interpretation. While FS in the baseline fluctuates around a relatively stable level, the forecast FS curve is systematically lower for extended periods. This separation persists even when rainfall peaks appear similar in both scenarios, highlighting that the controlling factor is not peak intensity alone, but the persistence of wet conditions that sustain pore pressure and reduce effective strength.

The most important point is not that collapse becomes inevitable on a specific day. The critical change is that the slope no longer returns to a comfortable safety margin. It transitions from a system characterized by short, event driven vulnerability to one characterized by chronic instability, where safety margins are persistently reduced and the probability of failure remains elevated for months.

From an engineering perspective, this is the dangerous state. When a slope is persistently marginal, small secondary triggers become decisive: partial blockage of drainage, minor toe erosion, freeze thaw damage, traffic induced vibration, or delayed maintenance. Under these conditions, climate change does not need to deliver a single catastrophic storm to increase failures. It increases the time spent near the edge, which is often more important than the height of the edge itself.

Scenario 3: end of century stress test

The third scenario is explicitly a stress test, not a calibrated forecast. It asks a different question than the previous scenarios: if rainfall intensification and persistence increase far enough, does the system undergo a fundamental regime change. The answer from the model is unambiguous.

In this scenario, rainfall is characterized by frequent, intense events with strong persistence. As in the previous scenarios, the black curves represent baseline conditions and the red curves represent the forecast climate. Unlike Scenario 2, however, the distinction between baseline and forecast becomes less important after an early point in the simulation, because both trajectories enter a state of near continuous instability.

The probability of failure Pf(t) reaches 100 percent and remains there for long intervals. In the plot this appears as the red Pf curve saturating near 1 and staying there, while the FS curve collapses early and does not recover. In the forecast case, the first exceedance of the alert threshold Pf ≥ 0.35 occurs on day 26, and by day 26 to 30 the probability of failure has effectively saturated. After this point, Pf rarely drops below unity, indicating that almost all strength realizations predict instability. Recovery essentially disappears.

The risk summary table makes the scale of this regime shift explicit. Under baseline conditions, Pf ≥ 0.35 occurs on 303 days of the year, corresponding to 83 percent of the simulation period. Under the forecast climate, this increases to 333 days, or over 91 percent of the year. The longest continuous run above the alert threshold is 303 days in the baseline and 313 days in the forecast, meaning that once instability is triggered, the slope remains in a high risk state for almost the entire remaining year. The maximum probability of failure reaches 100 percent in both cases.

The timing is equally important. The first day exceeding the alert threshold shifts from day 68 in the baseline to day 26 in the forecast, indicating a much earlier onset of the instability season. This implies not only higher risk, but a dramatically longer period during which the slope operates beyond acceptable safety margins.

The factor of safety plot supports this interpretation. FS drops rapidly early in the simulation and remains well below typical design margins for the rest of the year. Unlike Scenario 2, where FS fluctuates around a reduced but recoverable level, here the system does not meaningfully rebound. The separation between baseline and forecast is small compared to the overall loss of stability, underscoring that the dominant change is a system level transition, not sensitivity to individual storms.

At this point, the slope is no longer conditionally stable. It is chronically unstable. The original design assumptions have been exceeded, even though the geometry and material parameters have not changed. Drainage and maintenance measures no longer act as safety enhancements. They merely delay or redistribute failure.

This scenario provides the clearest illustration of how climate change can affect infrastructure. Rather than causing isolated failures through extreme events alone, it can push engineered systems across a threshold where the dominant behavior changes, from stable with occasional risk to unstable as the default state.

What can be concluded

Several conclusions emerge clearly from these analyses.

First, a road cut that is stable under present day conditions can become unstable without any change in geometry or material properties, purely as a result of altered rainfall statistics. Stability is not lost because the slope is poorly designed, but because the environmental conditions it was designed for no longer apply.

Second, the dominant driver of this transition is not peak rainfall intensity alone, but rainfall persistence and clustering. When wet periods become frequent enough that the slope does not fully recover between events, the system shifts from episodic vulnerability to chronic instability. This finding is critical, because design and maintenance practices often focus on extreme events rather than on recovery time.

Third, risk increases well before deterministic failure becomes obvious. In the simulations, the probability of failure rises sharply while the factor of safety may still appear acceptable. This shows why probabilistic measures are essential for climate adaptation: they capture the growing likelihood of failure during marginal conditions that deterministic checks can miss.

Fourth, once a slope spends extended time in a very wet regime, multiple weakening mechanisms act together. Pore pressure reduces effective stress, while loss of unsaturated strength further degrades resistance along weak layers. The result is a rapid loss of stability and a strong tendency toward persistent high risk states. These are not exotic or speculative processes. They are the same mechanisms documented in many real road cut failures.

The practical implication is clear. Climate change does not need to deliver a single catastrophic storm to increase failures. It increases the time spent near the edge, and that is often more dangerous than crossing the edge once. In such conditions, small secondary triggers blocked drainage, minor erosion, freeze thaw damage, traffic loading, or delayed maintenance can become decisive.

From an infrastructure perspective, this points to the need for anticipation rather than reaction. Prevention becomes more important than remediation. Drainage capacity, inspection frequency, monitoring thresholds, and maintenance strategies must be reassessed in light of longer wet periods and reduced recovery time. Slopes that are safe today may not remain so under future climate regimes, even if they have performed well for decades.

In practice, this argues for climate linked inspection and maintenance triggers rather than fixed calendars. For example, if wetness remains above the pore pressure activation threshold for a sustained period, or if Pf exceeds an operational threshold for several consecutive days, this should trigger an inspection, drainage cleaning, and targeted mitigation, even if the annual schedule would not normally call for it. The key adaptation is not only designing for higher peaks, but managing longer wet seasons with faster response.

About the model and its limitations

The model presented here is intentionally simple and transparent. It uses an infinite slope style formulation focused on a single weak layer, treats strength uncertainty probabilistically, and represents hydrological response through a threshold based bucket model. This simplicity is deliberate: it makes the governing mechanisms visible and testable.

The model does not resolve spatial groundwater flow, progressive deformation, or post failure behavior. It does not replace site investigation, laboratory testing, or detailed numerical analysis. It is not intended to predict the exact timing or location of a specific landslide.

What it does provide is clarity. Every assumption is explicit. Every parameter can be modified. The model is designed to explore regimes, thresholds, and trends, not to produce a single predictive number. That transparency allows users to understand why stability changes, not just that it changes.

In the context of climate adaptation, this distinction matters. Decisions about infrastructure resilience are rarely made on the basis of one precise prediction. They are made by recognizing when systems are approaching conditions for which existing design assumptions no longer hold.

Try it yourself

If you want to explore your own assumptions, you can interact with the model directly. Try changing one parameter at a time: storm persistence, drainage rate, weak layer dip, or the pore pressure threshold. Watch how the duration and timing of elevated risk respond.

The most informative experiments keep the geometry fixed and vary rainfall persistence. That is where the climate signal becomes unmistakable.

You can try the interactive model here.