Mapping Landslides with InSAR Satellite Data and AI

Geotechnical engineers are increasingly turning to a combination of Interferometric Synthetic Aperture Radar (InSAR) satellite data and artificial intelligence (AI) to predict and map rainfall-induced landslides. By merging millimeter-precise ground movement measurements from space with advanced machine learning algorithms, engineers can analyze how terrain responds to heavy precipitation. This article explores the step-by-step process of how these two technologies are integrated to create accurate landslide susceptibility maps, from initial data collection and environmental analysis to AI model training and hazard forecasting.

Capturing Ground Deformation with InSAR

InSAR is a remote sensing technique that uses radar images from Earth-orbiting satellites to measure ground deformation. Satellites emit radar waves toward the Earth and record the signals that bounce back. By comparing the phase difference of the waves between two or more satellite passes over the same area, engineers generate interferograms.

Advanced techniques, such as Persistent Scatterer InSAR (PS-InSAR), allow engineers to track specific stable points on the ground—like boulders, outcrops, or infrastructure—over several years. This provides a time-series dataset showing slow, millimeter-level slope creep. Because radar waves penetrate clouds and work in darkness, InSAR provides continuous monitoring even during severe rainstorms when optical satellites fail.

Gathering Environmental and Geotechnical Data

InSAR data alone only reveals where the ground is currently moving, not why it is moving or where it might fail next. To predict rainfall-induced landslides, engineers combine InSAR displacement rates with a variety of conditioning and triggering factors:

  • Rainfall Data: Historical precipitation records and real-time weather radar data to understand rainfall intensity, duration, and frequency.
  • Topography: Digital Elevation Models (DEMs) to calculate slope angle, aspect, curvature, and elevation.
  • Geotechnical Properties: Maps detailing soil types, bedrock lithology, fault lines, and soil moisture indices.
  • Hydrology and Land Cover: Proximity to rivers, drainage density, and vegetation indices (like NDVI), which indicate how much water the soil can absorb and how well plant roots anchor the slope.

Processing and Integrating the Datasets

Before AI can analyze the information, geotechnical engineers must standardize the data within a Geographic Information System (GIS). The study area is divided into a grid of pixels, and every pixel is assigned specific values from the collected datasets.

During this stage, engineers align the spatial resolution of the satellite radar data with the topographic and rainfall maps. They also filter out atmospheric noise from the InSAR data and use historical landslide inventories—records of where landslides have previously occurred—to label the data for machine learning.

Training the Artificial Intelligence Models

Once the spatial database is built, engineers feed the data into AI algorithms, commonly using Machine Learning (ML) models like Random Forest, Support Vector Machines (SVM), or Deep Learning frameworks like Convolutional Neural Networks (CNNs).

The training process involves teaching the AI to recognize the complex, non-linear relationships between ground movement, terrain features, and rainfall. For example, the AI learns that a 30-degree slope composed of clay soil, showing a 5-millimeter InSAR creep rate, has a high probability of failing when subjected to 100 millimeters of rain in 24 hours.

Engineers typically split their data: 70% is used to train the model, and 30% is held back to test and validate the model's accuracy. The AI adjusts its internal parameters until it can reliably distinguish between stable ground and landslide-prone areas.

Generating Dynamic Susceptibility Maps

After validation, the trained AI model is applied to the entire region to generate a Landslide Susceptibility Map (LSM). The AI calculates a failure probability score for every pixel on the map, categorizing zones from very low to very high risk.

Crucially, combining InSAR with AI makes these maps dynamic rather than static. Traditional maps only assess fixed terrain features. By continuously feeding the AI new InSAR satellite passes and incoming weather forecasts, the susceptibility map updates in near real-time. If a slope shows accelerated movement via InSAR right before a forecasted typhoon, the AI immediately elevates the risk level for that specific zone. This allows geotechnical engineers and local authorities to reinforce infrastructure and issue targeted evacuation warnings before the landslide occurs.