Poseidon Dataset Release
Release of World's Largest Earthquake Dataset
12/25/20253 min read
I’m thrilled to announce the release of “Poseidon”, the world’s largest open-source earthquake dataset, marking the first phase of an ambitious project my team and I have been working on at the Hong Kong University of Science and Technology to develop next-generation earthquake prediction technology.
A Dataset of Unprecedented Scale
Named after the Greek god of earthquakes, Poseidon contains over 2.8 million seismic events spanning 30 years of global earthquake activity from 1990 to 2020. The dataset includes detailed measurements of magnitude, depth, geographic coordinates, energy release calculations, and dozens of quality metrics for each recorded event.
Access the Poseidon Dataset on Hugging Face
This is not just another dataset sample release. Poseidon represents the foundation upon which we will build truly adaptive earthquake prediction systems. The seismic research community has long needed a unified, accessible resource of this scale, and I’m proud to make it freely available to researchers worldwide.
Beyond Static Models: The Dynamic Adaptation Revolution
This release marks the beginning of a two-phase initiative from my AI research group. The second phase, expected in the coming months, will introduce an open-source artificial intelligence model featuring what we describe as a “dynamic physics-informed architecture.” Our approach aims to combine traditional physics-based seismic modeling with adaptive learning techniques, revolutionizing how scientists approach earthquake forecasting.
Throughout my career, I’ve focused on dynamic adaptation — a paradigm that emphasizes creating AI systems capable of modifying their internal structures and decision-making processes in real-time as conditions change. My research portfolio demonstrates a consistent commitment to moving beyond static, rigid computational models toward autonomous systems that can adjust their logic in response to evolving data distributions and environmental contexts.
Building on a Foundation of Innovation
In November 2025, I published work on morphing tree structures in gradient boosting algorithms, demonstrating how traditionally fixed decision tree architectures could be redesigned to dynamically reshape themselves during the learning process. Earlier in January, my collaborators and I introduced the concept of epigenetic memory in evolutionary algorithms, drawing inspiration from biological systems where organisms retain and transmit adaptive responses across generations without altering their underlying genetic code.
The earthquake prediction problem is fundamentally a dynamic adaptation challenge. Seismic patterns are not static. Fault systems evolve, stress accumulates and releases in complex patterns, and the relationships between precursor signals and actual events shift over time. Any AI system that hopes to make meaningful predictions must be capable of adapting its understanding as the Earth itself changes.
From the Arctic to Seismic Zones
My team brings significant experience in applying advanced computational methods to large-scale environmental challenges. Earlier this year, we completed the largest open-source study conducted on the Arctic zone, developing AI-powered algorithmic analysis to process vast quantities of climate and environmental data from the polar region. That project demonstrated our capability to handle massive datasets while extracting meaningful patterns relevant to global scientific challenges.
Built for the Research Community
The Poseidon dataset has been specifically structured to support a wide range of machine learning applications, including:
Earthquake prediction modeling
Aftershock sequence analysis
Magnitude-frequency studies
Tsunami early warning system development
Seismic hazard mapping
The inclusion of pre-computed energy values using the Gutenberg-Richter relation and spatial grid features for heatmap generation reflects our intention to lower barriers for researchers entering the field.
Key Dataset Statistics
Total Events: 2,833,766
Time Span: 1990–01–01 to 2019–12–31
Magnitude Range: 0.0–9.1
Geographic Coverage: Global (-90° to 90° lat, -180° to 180° lon)
Spatial Resolution: 180 x 360 grid bins (1° resolution)
Open Science for Global Impact
The open-source nature of both the dataset and our forthcoming AI model represents a significant departure from the proprietary approaches that have characterized much of recent AI development. By making these resources freely available, my team and I are positioning this work as a public good rather than a commercial product.
A New Era in Earthquake Prediction
As the world grapples with increasing seismic risks in densely populated regions, the potential for AI-assisted earthquake prediction has never been more urgent. With the Poseidon dataset now available and a physics-informed adaptive AI model on the horizon, my team will be laying the groundwork for a new era in humanity’s ability to anticipate and prepare for one of nature’s most destructive forces.
Get Started with Poseidon