BorisKriuk/Poseidon

---
license: cc-by-4.0
task_categories:
- time-series-forecasting
language:
- en
pretty_name: Poseidon
size_categories:
- 1M<n<10M
tags:
- geophysics
- earthquake-prediction
---

# Poseidon: Global Earthquake Dataset (1990-2020)

This is the official dataset for the paper [POSEIDON: Physics-Optimized Seismic Energy Inference and Detection Operating Network](https://huggingface.co/papers/2601.02264).

## Overview

**Poseidon** is a largest opensource global earthquake dataset containing **2.8+ million seismic events** spanning 30 years (1990-2020). Named after the Greek god of earthquakes, this dataset is designed for machine learning applications including earthquake prediction, seismic hazard analysis, spatiotemporal pattern recognition, and energy-based modeling.

## Dataset Statistics

| Metric | Value |
|--------|-------|
| **Total Events** | 2,833,766 |
| **Time Span** | 1990-01-01 to 2024-12-31 |
| **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 degree resolution) |

## Features

### Core Seismic Properties

| Column | Type | Description |
|--------|------|-------------|
| id | string | Unique USGS event identifier |
| time | string | ISO 8601 timestamp (UTC) |
| latitude | float64 | Event latitude (-90 to 90) |
| longitude | float64 | Event longitude (-180 to 180) |
| depth | float64 | Hypocenter depth in kilometers |
| magnitude | float64 | Event magnitude |
| mag_type | string | Magnitude type (ml, mb, mw, md, etc.) |

### Event Metadata

| Column | Type | Description |
|--------|------|-------------|
| place | string | Human-readable location description |
| type | string | Event type (earthquake, quarry blast, etc.) |
| status | string | Review status (reviewed, automatic) |
| tsunami | int64 | Tsunami flag (1 = tsunami generated, 0 = none) |
| sig | int64 | Significance score (0-1000+) |
| net | string | Contributing seismic network code |

### Quality Metrics

| Column | Type | Description |
|--------|------|-------------|
| nst | float64 | Number of stations used |
| dmin | float64 | Minimum distance to nearest station (degrees) |
| rms | float64 | Root mean square travel time residual |
| gap | float64 | Azimuthal gap (degrees) |
| horizontal_error | float64 | Horizontal location uncertainty (km) |
| depth_error | float64 | Depth uncertainty (km) |
| mag_error | float64 | Magnitude uncertainty |
| mag_nst | float64 | Number of stations for magnitude calculation |

### Temporal Features (Pre-computed)

| Column | Type | Description |
|--------|------|-------------|
| year | int64 | Event year |
| month | int64 | Event month (1-12) |
| day | int64 | Event day (1-31) |
| hour | int64 | Event hour (0-23 UTC) |
| minute | int64 | Event minute (0-59) |
| second | int64 | Event second (0-59) |

### Spatial Grid Features (Pre-computed)

| Column | Type | Description |
|--------|------|-------------|
| lat_bin | int64 | Latitude bin index (0-179) for heatmap generation |
| lon_bin | int64 | Longitude bin index (0-359) for heatmap generation |

### Energy Features (Pre-computed)

| Column | Type | Description |
|--------|------|-------------|
| energy_joules | float64 | Seismic energy release in Joules |
| log_energy | float64 | Log10 of energy (for numerical stability) |

## Energy Calculation

Seismic energy is computed using the Gutenberg-Richter energy-magnitude relation:

log10(E) = 1.5 x M + 4.8

Where E = Energy in Joules and M = Earthquake magnitude.

Example values:

| Magnitude | Energy (Joules) | Equivalent |
|-----------|-----------------|------------|
| 2.0 | 6.3 x 10^7 | Small explosion |
| 4.0 | 6.3 x 10^10 | 15 tons TNT |
| 6.0 | 6.3 x 10^13 | 15 kilotons TNT |
| 8.0 | 6.3 x 10^16 | 15 megatons TNT |
| 9.0 | 2.0 x 10^18 | 475 megatons TNT |

## Usage

```python
import pandas as pd
df = pd.read_csv("poseidon.csv")
df['datetime'] = pd.to_datetime(df['time'])
```

### Example: Filter Significant Events

```python
major_quakes = df[df['magnitude'] >= 6.0]
print(f"Major earthquakes (M6+): {len(major_quakes):,}")

tsunami_events = df[df['tsunami'] == 1]
print(f"Tsunami-generating events: {len(tsunami_events):,}")
```

## Applications

This dataset is designed for:

- Earthquake Prediction Models
- Aftershock Sequence Analysis
- Magnitude-Frequency Analysis
- Tsunami Early Warning
- Energy-Based Models (EBMs)
- CNN/RNN Training
- Seismic Hazard Mapping

## License

This dataset is released under CC BY 4.0.

## Acknowledgments

- USGS Earthquake Hazards Program for providing the source data
- Gutenberg and Richter for the foundational energy-magnitude relation