Deadly Philippines earthquake found to have raised seabed by up to 2 metres - The Guardian

## The Deadly Philippines earthquake That Lifted the Seabed by 2 Metres: A Wake-Up Call for Tech In December 2024, a magnitude 7. 0 earthquake struck the island of Mindanao in the Philippines, killing at least 61 people and causing over ₱1 billion in damage. But the most startling revelation came days later when scientists confirmed that the seabed along the coast had been thrust upward by up to two metres. This isn't just a geological curiosity-it is a stress test for every earthquake prediction model, early Warning system, and structural engineering code in existence. For engineers and software developers, the Philippines seabed uplift is a rare, real-world dataset that exposes the gap between our mathematical models and the chaotic behaviour of the Earth's crust. In this article, I'll analyse what this event teaches us about the technology we rely on to forecast, measure, and survive earthquakes. I'll draw on satellite altimetry data, AI-based early warning systems. And open-source seismic tools to offer a technical perspective that goes beyond the headlines.

If you work with geospatial data, machine learning. Or civil engineering software, this event holds lessons you can't afford to ignore.

---

The Geological Phenomenon That Demands Better Modelling

The uplift observed near Sarangani and Davao Occidental is a textbook example of a thrust fault rupture. When the Philippine Sea Plate subducts beneath the Sunda Plate, stress builds until the crust fractures and the seafloor is suddenly pushed upward. What makes this event so valuable is the precise measurement of that displacement: up to two metres in some locations, as reported by The Guardian and local outlets such as The Guardian's initial report.

Current geodynamic models, like those implemented in software such as PyLith (a finite-element code for crustal deformation) or STATICS from USGS, assume a homogeneous Earth. But real fault zones are heterogeneous, with variable friction, fluid pressure. And rock strength. The Mindanao uplift reveals that standard elastic rebound theory underestimates near-field deformation by as much as 30% in complex subduction zones. Engineers building deep-sea infrastructure-like submarine cables or offshore wind turbines-need to incorporate these higher displacement scenarios into their probabilistic risk assessments.

For software developers in this space, the event is a call to move beyond simple Coulomb stress models. Hybrid approaches coupling finite-difference wave propagation with empirical attenuation relations-similar to the techniques used in the OpenQuake engine maintained by the Global Earthquake Model (GEM)-can better capture the non-linear response of the seafloor.

---

How Machine Learning Is Redefining Earthquake Early Warning Systems

AI-based early warning systems have made headlines in recent years, with platforms like ShakeAlert on the US West Coast achieving detection times of 5-20 seconds before the strongest shaking arrives. But the Philippines earthquake tests a harder scenario: an offshore subduction event where the nearest seismometer might be dozens of kilometres away.

The Google-Philippines AI early warning pilot. Which uses on-device accelerometers in Android phones to detect P‑wave arrival, triggered alerts for more than 500,000 users during the Mindanao quake. Yet the uplift data shows that the peak ground displacement (PGD) was significantly higher than predicted by the real-time inversion algorithms. Why? Because the current models are trained primarily on shallow crustal earthquakes (

To improve accuracy, we need to integrate seafloor geodesy data-from ocean-bottom pressure gauges and InSAR (Interferometric Synthetic Aperture Radar)-directly into the training pipelines. I recommend reading the USGS Real-Time Earthquake Standards documentation to understand the current latency constraints. The Mindanao uplift proves that machine learning models must learn to correlate far-field ground motion with near-field permanent deformation, something that no open-source dataset currently captures adequately.

---

Satellite Altimetry and InSAR: The Eyes That Measured the Seabed Rise

Measuring a two-metre vertical displacement of the seabed isn't trivial. Traditional tide gauges only give point measurements, and ship-based surveys take weeks. The first confirmation of the uplift came from Copernicus Sentinel-1 satellite data processed using InSAR technique.

InSAR compares two radar images of the same area taken at different times, measuring the phase difference to detect centimetre-level ground deformation. For the Mindanao event, researchers at the Philippine Institute of Volcanology and Seismology (PHIVOLCS) used the ESA SNAP toolbox to unwrap interferograms and produce an uplift map that showed maximum displacement near the coastline of Davao Occidental.

But there's a catch: InSAR loses coherence over water. To measure the seabed itself, scientists had to rely on coseismic deformation models that invert onshore GPS and offshore wave data satellite altimetry from Jason-3. The resulting model suggests the rupture propagated all the way to the seafloor, creating a bulge that raised the seabed by up to 2. 0 metres. This finding pushes the limits of what we can model with elastic dislocation theory-software like Coulomb 3. 4 (USGS) assumes a flat, infinite half-space. But the real seabed slope of 5%-15% off Mindanao introduces significant errors.

For developers of geophysical inversion software, the event underscores the need to incorporate bathymetric and tomographic heterogeneity into forward models. The open-source library Pyrocko (which underlies the online tool Grond) is a good starting point for building such custom inversions, as it already integrates with Global CMT catalogues and station databases.

---

Structural Engineering Lessons from the Mindanao Quake

Buildings that collapsed in Sarangani and Davao City (e g., a two-storey school) were mostly non-engineered masonry structures typical of older construction. However, several modern steel-frame buildings in Davao City performed remarkably well, experiencing only non-structural damage. This dichotomy highlights two critical points for structural engineers who write design software.

First, the Philippines National Structural Code (NSCP 2015) prescribes seismic loads based on a spectral response acceleration for a 10% probability of exceedance in 50 years. The Mindanao earthquake produced peak ground accelerations (PGA) that exceeded these design values by 15-20% at frequencies around 1-2 Hz, precisely where multi-storey buildings are most vulnerable. Software like ETABS or SAP2000 often allow engineers to set "code-compliant" response spectra. But these may not capture the near-fault forward-directivity pulses that caused the high-velocity demands observed.

Second, the uplift confirmed that tsunami waves generated by this earthquake (up to 1. 5 metres high locally) were significantly smaller than those that would be expected from a subduction event of the same magnitude. Why? Because the shallow rupture uplifted the seafloor above sea level, partially cancelling the water displacement. This counter-intuitive effect should be built into tsunami early warning codes (e. And g, the MOST model from NOAA) to avoid false alarms in future events.

I recommend engineers working on seismic design review the FEMA P-1051 Guidelines for Seismic Performance of Nonstructural Components for updated acceleration-sensitive component evaluation.

---

Open Data for Seismic Research: Why Transparency Matters

One of the most encouraging outcomes of the Mindanao earthquake has been the rapid release of waveform data by PHIVOLCS and the IRIS-DMC (Incorporated Research Institutions for Seismology Data Management Center). Within 48 hours, raw seed volumes from stations like DAV and MSLP were publicly available, enabling independent researchers to verify the focal mechanism and aftershock sequence.

This openness is a game-changer for Machine Learning in seismology. The STEAD dataset (Stanford Earthquake Dataset) contains 1. 2 million waveforms from global earthquakes. But only 2% come from subduction zones in Southeast Asia. The Mindanao event fills that gap with high-quality, near-field records that can be used to train neural networks for phase picking (e g., EQTransformer or PhaseNet) with greater accuracy in complex tectonic settings.

I urge every developer contributing to open-source seismology software-from Obspy to SeisComp³-to integrate these new waveforms into their test suites. The reproducibility of science depends on it.

---

The Role of Citizen Science in Post-Disaster Mapping

In the hours after the earthquake, the Humanitarian OpenStreetMap Team activated mapping tasks for Mindanao? Over 2,000 volunteers digitised buildings and roads, cross-referencing satellite imagery from Maxar's Open Data Program. This data was used by local government units and the Department of Public Works and Highways (DPWH) to prioritise debris removal.

But there's an underappreciated technology layer here: AI-assisted feature extraction. Models like Facebook's RapidAI4Earth can automatically segment building footprints from post-disaster imagery with 85% accuracy. For the Mindanao event, however, many buildings were obscured by vegetation or had partially collapsed roofs that confused the segmentation algorithm. The manual corrections from citizen mappers created a high-quality labelled dataset that researchers have now released on Zenodo to improve future AI models.

If you work in computer vision for disaster response, this dataset is a goldmine for transfer learning. I suggest looking at the DeepLabv3+ architecture implemented in TensorFlow and fine-tuning it on the Mindanao images-you will likely see a 10-15% improvement in IoU for tropical, low-lying coastal environments.

---

Building Resilient Infrastructure with Real-Time Monitoring

Seismology isn't just about prediction; it's about detection and response. The Philippines has a network of 100 strong-motion accelerometers. But only a fraction are equipped with real-time telemetry. During the earthquake, some stations experienced power outages and lost cellular connectivity, delaying the shaking intensity map by nearly an hour.

Low-cost IoT alternatives exist. The MEMS-based sensor platform "G-Cloud" developed by the National Institute of Advanced Industrial Science and Technology (AIST) in Japan uses 3G/4G to stream data at 200 Hz with a latency under 500 milliseconds. Deploying even 50 such nodes along the coast of Mindanao would have provided ground-truth acceleration data for building a shake map within seconds.

For side-projects, I recommend the Raspberry Shake community-a network of over 1,500 low-cost seismometers that interface with the cloud via Python scripts. While not accurate enough for early warning, these devices can validate shaking intensity in near real-time and help populate the Did You Feel It? system of USGS.

---

AI-Powered Damage Assessment from Aerial Imagery

After the earthquake, the Philippine Air Force deployed UAVs to survey hard-hit barangays in Sarangani. The imagery was processed through a convolutional neural network trained on the xView2 dataset (a collection of building damage labels from past disasters). The model classified damaged buildings into four categories: undamaged, minor, major. And destroyed.

However, the model's performance was significantly lower on the Mindanao images compared to its baseline on xView2 (which uses scenes mainly from the US and Caribbean). The cause: roofing materials in the Philippines (corrugated metal, thatch, clay tiles) differ from the pre-training data. This is a classic domain adaptation problem that can be solved using adversarial training or self-supervised learning with unlabelled local imagery.

For teams building disaster response AI, I recommend incorporating CycleGAN style transfer to normalise the visual features before inference. The improvement in F1-score for "major damage" class could jump from 0. And 55 to 072 based on similar experiments from the 2023 Türkiye-Syria earthquakes.

---

FAQ: What Developers and Engineers Should Know

• Can technology predict earthquakes like the one in Mindanao?
  No-deterministic prediction remains impossible. However, machine learning can improve short-term forecasting by identifying precursory signals (e g., slow slip events) from continuous GPS and InSAR data. For non-technical readers, I suggest USGS Earthquake Hazards Program.
• How accurate are satellite measurements of seabed uplift?

.