📄 earth_science

Sentinel-1 SBAS-InSAR detection of pre-failure ground motion at the 2025 Njintout landslide, central Cameroon Volcanic Line.

This study demonstrates that Sentinel-1 SBAS-InSAR analysis of open satellite data successfully detected pre-failure ground acceleration at the 2025 Njintout landslide in Cameroon approximately 4.5 months before the catastrophic event, despite the absence of ground-based monitoring equipment.

Original authors: Jospen Roméo NEGOU

Published 2026-08-05

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Jospen Roméo NEGOU

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine the ground beneath our feet is like a slow-moving river of rock and soil. Sometimes, this river flows so gently that we can't feel it, but when it speeds up, it can turn into a rushing torrent that destroys everything in its path: a landslide. Scientists have long known that before a landslide crashes down, the ground often starts to creep or "sneak" forward, like a car slowly rolling down a hill before the brakes fail. The big challenge is spotting this sneaky movement in places covered by thick forests or far away from cities, where we don't have sensors or people standing around to watch. This is where a special kind of space detective work comes in. Instead of using eyes, scientists use radar beams bounced off the Earth from satellites. By comparing these radar pictures taken weeks apart, they can measure if the ground has moved by just a few millimeters—about the thickness of a pencil eraser. This is crucial because catching that tiny, slow movement early could give people enough time to get to safety before the big crash.

This paper tells the story of a team of scientists who acted as time-traveling detectives for a landslide that happened in September 2025 in Njintout, Cameroon. They didn't have any ground sensors to watch the hill as it was happening, so they went back to a digital archive of radar images from a satellite called Sentinel-1. Using a clever computer method called SBAS-InSAR (which is like stitching together hundreds of blurry radar photos to find a clear pattern), they looked at the hill for the sixteen months leading up to the disaster. What they found was a "ghost" signal: a slow, creeping movement that started about four and a half months before the landslide actually collapsed. The ground wasn't just moving; it was accelerating, shifting from a near-still state to a faster creep of about +21.8 mm per year just before the failure.

The researchers found that the ground at the disaster site was moving toward the satellite at a speed of +5.85 mm per year on average. But the real story is in the timeline. For the first ten months, the hill was essentially still, moving at a rate indistinguishable from zero. Then, around May 8, 2025, something changed. The ground suddenly started to speed up, and by the time the landslide hit in September, it had moved a total of about +8.57 mm. The team compared this moving spot to a stable patch of ground just 2 kilometers away, and the difference was huge: the landslide site moved six times more than the stable ground, a difference so clear it would happen by pure chance less than once in 5,000 tries.

However, the paper is very careful not to call this a perfect "warning system" yet. The scientists admit that without ground sensors to double-check their radar numbers, they can't be 100% sure of the exact speed. They also note that their method isn't perfect at pinpointing the exact pixel where the slide will happen; instead, it lights up a whole neighborhood of the hill that is unstable. In fact, if you look at the whole region, their method flags about 17.7% of the area as "suspicious," meaning it's good at finding trouble zones but not great at ruling out safe spots. The paper suggests that this technique is a powerful new tool for finding these slow-moving dangers in tropical, forested areas where we usually have no eyes on the ground, but it's a "detection near the sensitivity limit" rather than a fully proven, foolproof alarm clock. It proves that we can see the ground sneaking before it crashes, even in the thick jungles of the Cameroon Volcanic Line, but we still need more tools to make that signal loud and clear enough to save lives.

Technical Summary: Sentinel-1 SBAS-InSAR Detection of Pre-Failure Ground Motion at the 2025 Njintout Landslide

Problem Statement
Landslides in the central Cameroon Volcanic Line (CVL), particularly in the densely populated Foumban district, pose significant risks due to heavy rainfall saturating thick lateritic regoliths. Historically, this region lacks ground-based instrumentation (GNSS, extensometers) capable of monitoring sluggish ground motion prior to collapse. While susceptibility mapping and rainfall analysis exist, they cannot determine if a specific slope is currently deforming. This study addresses the gap in detecting pre-failure deformation in heavily vegetated, poorly instrumented tropical terrain using only open-access satellite data.

Methodology
The research utilizes a retrospective analysis of the 2025 Njintout landslide (19 September 2025), which destroyed approximately 100 homes. The author employed a Small Baseline Subset (SBAS) InSAR processing chain on 36 Sentinel-1A ascending orbit (track 59) scenes acquired between 30 June 2024 and 23 October 2025.

Key methodological adaptations for the tropical environment include:

Coherence Thresholding: Recognizing that the standard literature default for phase unwrapping ( $\gamma > 0.125$ ) would mask most of the vegetated study area (mean coherence $\gamma \approx 0.17$ ), the author empirically lowered the threshold to $\gamma \ge 0.080$ . This preserved 70.4% of pixels and maintained network connectivity, whereas the default threshold would have reduced coverage to 30.9% and disconnected the target slope.
Processing Chain: The workflow involved coregistration, interferogram generation (145 pairs reduced to 50 high-quality pairs), adaptive Goldstein filtering, SNAPHU phase unwrapping, and coherence-weighted inversion.
Validation Strategy: In the absence of ground truth, reliability was assessed internally via:
- Comparison with a stable reference cell 2 km away.
- Comparison with four high-coherence reference points across the district.
- A "noise-only" null hypothesis test using 5,000 resamples of stable cell residuals.
- False-alarm characterization by applying a five-criteria detection rule across the entire processing extent.

Key Results

Pre-Failure Signal Detection: At the failure point (5.7330°N, 10.8897°E), the analysis recovered a positive Line-of-Sight (LOS) velocity of +5.85 mm yr⁻¹ (95% confidence: +2.70 to +8.99 mm yr⁻¹).
Temporal Evolution (Tertiary Creep): A change-point analysis revealed that the deformation was not linear.
- Pre-onset: The first ~10 months showed a rate indistinguishable from zero.
- Onset: Sustained motion began on 8 May 2025 (approx. 4.5 months before the catastrophic failure).
- Acceleration: Post-onset, the rate accelerated to +21.8 mm yr⁻¹ (95% confidence: +12.0 to +31.7 mm yr⁻¹).
- The two-segment model (flat then accelerating) fit significantly better than a single linear trend ( $F=8.4$ , bootstrap $p=0.001$ ).
Statistical Significance: The signal at the failure site separated from the stable reference cell by approximately six times the local temporal dispersion. The probability of a velocity of this magnitude arising from noise alone was calculated to be less than 1 in 5,000.
Spatial Context: The failure point lies within a contiguous zone of positive LOS velocity (+2 to +8 mm yr⁻¹) covering the southern two-thirds of the 4.4 km study window, indicating the signal represents a slope-scale feature rather than an isolated pixel anomaly.
Specificity: While the detection rule matched 17.7% of valid cells across the entire Foumban district (indicating low specificity at the regional scale), within the immediate vicinity of the failure, the matches formed a single continuous feature, suggesting the rule identifies the extent of unstable slopes rather than specific failure cells.

Significance and Claims
The paper positions itself as a feasibility demonstration and a first-order characterization of pre-failure motion in tropical lateritic environments, rather than a proven operational early-warning system.

Methodological Contribution: The study provides the first peer-reviewed InSAR velocity map of the Foumban region and demonstrates that sub-centimeter pre-failure deformation can be recovered in densely vegetated terrain by tuning the coherence threshold ( $\gamma \ge 0.080$ ) below standard defaults.
Scientific Contribution: It documents a pre-failure acceleration signal consistent with tertiary creep theory, detected roughly 4.5 months before the catastrophic event and 2.2 months before the first observed precursor failure.
Operational Implication: The research highlights that satellite interferometry is the only currently available observable capable of linking antecedent failures to slope status in regions lacking ground instrumentation. The derived processing template is transferable to other exposed communities along the CVL.

Limitations Acknowledged by the Authors
The author explicitly states the following constraints:

No External Validation: The signal has not been confirmed by GNSS or leveling; the +5.85 mm yr⁻¹ rate is a single-modality detection near the sensitivity limit.
Geometry: Single-track (ascending) data prevents the separation of vertical and horizontal motion components; the interpretation of eastward creep is provisional based on slope geometry.
Atmospheric Effects: No external atmospheric correction (e.g., GACOS, ERA5) was applied, though a 10-km high-pass filter was used.
Temporal Resolution: The 16-month window covers only 1.3 annual cycles, making the seasonal component underdetermined.
Specificity: The detection rule is sensitive but not specific at the regional scale (17.7% false-positive rate on unrelated terrain), meaning it identifies zones of concern rather than precise failure locations.

In conclusion, the paper asserts that open Sentinel-1 data and an open-source SBAS chain can recover pre-failure slope deformation in tropical terrain previously monitored only post-failure, providing a practical starting point for precursor-based hazard assessment in instrument-poor regions.

More like this