From Lidar Change Detection (LCD) to Automated Landslide Intelligence

Teams can now routinely quantify terrain movement across entire corridors — pipelines, railways, highways — with centimeter-scale precision. Subtle deformation that would have gone unnoticed in traditional inspections is now measurable, repeatable, and defensible.

Geohazards rarely fail without warning. The signals are there. Subtle displacements, gradual slope movement, early-stage instability. LCD makes those signals visible.

The Next Bottleneck: Interpretation at Scale

As LCD adoption has grown, a new challenge has emerged.

Large-area LCD programs generate enormous datasets. Thousands of change features. Complex terrain signals. Noise from vegetation, construction, and seasonal variation. 

It’s become a signal-to-noise challenge. Somewhere in that data are the slopes that matter.

Identifying them still requires expert review — scanning LCD, interpreting patterns, delineating polygons, debating whether movement is operationally relevant.

It is important, yet tedious (and monotonous) work. And it takes time (and coffee).

After extreme rainfall, flooding, or seismic activity, every day matters and awareness of what just happened accelerates the road to recovery.

Why Stop at Change Detection?

LCD answers a critical question:
Where has the ground moved?

But data alone are not useful. Operators need to answer a second, equally important question:
Which of that movement is actually landslide activity?

This is where many programs can stall. The challenge is separating signal from noise, and doing so consistently, defensibly, and at corridor scale.

Grading. Deposition. Vegetation effects. Instrument noise. All of these can appear in change datasets.

Introducing Machine Learning into the Workflow

Cambio Landslide Movement Mapping builds directly on LCD to address this gap.

Instead of manually scanning change datasets, machine learning models automatically identify patterns of terrain deformation consistent with landslide activity.

The workflow applies deep learning and semantic segmentation to classify movement at the pixel level, generating high-resolution landslide polygons with associated confidence metrics.

This is not generic AI applied to terrain data. It’s not a threshold method that tries to find change that is “significant” (what does that word mean again?). It is a science-grounded screening tool trained on over 8,000 painstakingly mapped landslides and designed to reflect how landslides actually look in lidar and lidar change. 

From Weeks of Review to Hours of Insight

Traditionally, generating a landslide inventory from LCD required:

• Manual scanning of change surfaces 
• Expert interpretation of deformation patterns
• Delineation of landslide boundaries
• Iteration and validation across teams

For large corridors, this process can take weeks, months, or in extreme cases, years.

With machine learning-assisted mapping, that same first-pass inventory can be generated in hours to days. We’re not replacing our engineering judgement, we are just applying it in more targeted ways. A consistent, repeatable baseline that allows teams to focus their time where it matters most: validation, prioritization, and decision-making.

A More Complete Picture of Risk

LCD remains the foundation. It provides the measurement — precise, repeatable quantification of terrain change across space and time.

Landslide Movement Mapping adds interpretation.

Together, they move teams from:

• Raw change surfaces → to mapped landslide features
• Data-overload→ focused decision making
• Broad screening → to risk-based prioritization

This combination is particularly powerful in three scenarios:

1. Corridor-Scale Programs - Screen hundreds to thousands of kilometers for active landslides. Standardize how hazards are identified across regions and teams. Reduce variability in interpretation.

2. Post-Event Response - Rapidly identify newly activated or accelerating landslides following extreme weather or seismic events. Prioritize field inspections when access and time are limited.

3. Inventory Development & Updates - Build or refresh landslide inventories with a defensible, repeatable methodology grounded in measured terrain change.

Innovation That Respects the Discipline

Geohazard management has always required a balance between data, experience, and uncertainty.

LCD was a step-change because it introduced reliable measurement into that equation. Machine learning is the next step — not by replacing engineering judgement, but by scaling it. 

A common concern with machine learning in geoscience is loss of transparency and that concern is valid. Cambio Earth’s approach is different. The models are trained on real, manually interpreted landslides. The results are designed to be reviewed, validated, and challenged.

From Detection to Action

Infrastructure operators are dealing with aging assets, more frequent extreme weather, and greater scrutiny on risk and compliance.

Reducing the gap between data review and action is critical. Combining LCD with machine learning-enabled landslide mapping enables operators to act faster and with a defensible basis.

The real value of data is not in what it shows. It is in what it allows you to do next — before that surface change becomes a problem.

Because we fully stand behind the rigor of this model, we’re in the process of publishing this work in a fully peer-reviewed journal. Stay tuned and we’ll share the paper when it’s ready.

‍