README.md

---
license: mit
---

## CEED: *C*alifornia *E*arthquake *E*vent *D*ataset for Machine Learning and Cloud Computing

The California Earthquake Event Dataset (CEED) is a dataset of earthquake waveforms and metadata for machine learning and cloud computing. 

Detailed statistics about the dataset are available in this [arXiv paper](https://arxiv.org/abs/2502.11500).

### Acknowledgments

The seismic data used in this study were collected by (1) the Berkeley Digital Seismic Network (BDSN, doi:10.7932/BDSN) and the USGS Northern California Seismic Network (NCSN, doi:10.7914/SN/NC); and (2) the Southern California Seismic Network (SCSN, doi:10.7914/SN/CI).
The original waveform data, metadata, and data products for this study were accessed through the Northern California Earthquake Data Center (doi:10.7932/NCEDC) and the Southern California Earthquake Center (doi:10.7909/C3WD3xH1).
Please include acknowledgments and citations of the original data providers when using this dataset.

The dataset structure is shown below, and you can find more information about the format at [AI4EPS](https://ai4eps.github.io/homepage/ml4earth/seismic_event_format1/)

```
 Group: / len:60424
  |- Group: /ci38457511 len:35
  |  |-* begin_time = 2019-07-06T03:19:23.668000
  |  |-* depth_km = 8.0
  |  |-* end_time = 2019-07-06T03:21:23.668000
  |  |-* event_id = ci38457511
  |  |-* event_time = 2019-07-06T03:19:53.040000
  |  |-* event_time_index = 2937
  |  |-* latitude = 35.7695
  |  |-* longitude = -117.5993
  |  |-* magnitude = 7.1
  |  |-* magnitude_type = w
  |  |-* nt = 12000
  |  |-* nx = 35
  |  |-* sampling_rate = 100
  |  |-* source = SC
  |  |- Dataset: /ci38457511/CI.CCC..HH (shape:(3, 12000))
  |  |  |- (dtype=float32)
  |  |  |  |-* azimuth = 141.849479
  |  |  |  |-* back_azimuth = 321.986302
  |  |  |  |-* component = ENZ
  |  |  |  |-* depth_km = -0.67
  |  |  |  |-* distance_km = 34.471389
  |  |  |  |-* dt_s = 0.01
  |  |  |  |-* elevation_m = 670.0
  |  |  |  |-* event_id = ['ci38457511' 'ci38457511' 'ci37260300']
  |  |  |  |-* instrument = HH
  |  |  |  |-* latitude = 35.52495
  |  |  |  |-* local_depth_m = 0.0
  |  |  |  |-* location = 
  |  |  |  |-* longitude = -117.36453
  |  |  |  |-* network = CI
  |  |  |  |-* p_phase_index = 3575
  |  |  |  |-* p_phase_polarity = U
  |  |  |  |-* p_phase_score = 0.8
  |  |  |  |-* p_phase_status = manual
  |  |  |  |-* p_phase_time = 2019-07-06T03:19:59.422000
  |  |  |  |-* phase_index = [ 3575  4184 11826]
  |  |  |  |-* phase_picking_channel = ['HHZ' 'HNN' 'HHZ']
  |  |  |  |-* phase_polarity = ['U' 'N' 'N']
  |  |  |  |-* phase_remark = ['i' 'e' 'e']
  |  |  |  |-* phase_score = [0.8 0.5 0.5]
  |  |  |  |-* phase_status = manual
  |  |  |  |-* phase_time = ['2019-07-06T03:19:59.422000' '2019-07-06T03:20:05.509000' '2019-07-06T03:21:21.928000']
  |  |  |  |-* phase_type = ['P' 'S' 'P']
  |  |  |  |-* s_phase_index = 4184
  |  |  |  |-* s_phase_polarity = N
  |  |  |  |-* s_phase_score = 0.5
  |  |  |  |-* s_phase_status = manual
  |  |  |  |-* s_phase_time = 2019-07-06T03:20:05.509000
  |  |  |  |-* snr = [ 637.9865898   286.9100766  1433.04052911]
  |  |  |  |-* station = CCC
  |  |  |  |-* unit = 1e-6m/s
  |  |- Dataset: /ci38457511/CI.CCC..HN (shape:(3, 12000))
  |  |  |- (dtype=float32)
  |  |  |  |-* azimuth = 141.849479
  |  |  |  |-* back_azimuth = 321.986302
  |  |  |  |-* component = ENZ
  |  |  |  |-* depth_km = -0.67
  |  |  |  |-* distance_km = 34.471389
  |  |  |  |-* dt_s = 0.01
  |  |  |  |-* elevation_m = 670.0
  |  |  |  |-* event_id = ['ci38457511' 'ci38457511' 'ci37260300']
  ......
  ```

## Getting Started

### Requirements
- datasets
- h5py
- fsspec
- pytorch

### Usage
Import the necessary packages:
```python
import h5py
import numpy as np
import torch
from datasets import load_dataset
```
We have 6 configurations for the dataset: 
- "station"
- "event"
- "station_train"
- "event_train"
- "station_test"
- "event_test"

"station" yields station-based samples one by one, while "event" yields event-based samples one by one. The configurations with no suffix are the full dataset, while the configurations with suffix "_train" and "_test" only have corresponding split of the full dataset. Train split contains data from 1970 to 2019, while test split contains data in 2020.

The sample of `station` is a dictionary with the following keys:
- `data`: the waveform with shape `(3, nt)`, the default time length is 8192
- `begin_time`: the begin time of the waveform data
- `end_time`: the end time of the waveform data
- `phase_time`: the phase arrival time
- `phase_index`: the time point index of the phase arrival time
- `phase_type`: the phase type
- `phase_polarity`: the phase polarity in ('U', 'D', 'N')
- `event_time`: the event time
- `event_time_index`: the time point index of the event time
- `event_location`: the event location with shape `(3,)`, including latitude, longitude, depth
- `station_location`: the station location with shape `(3,)`, including latitude, longitude and depth

The sample of `event` is a dictionary with the following keys:
- `data`: the waveform with shape `(n_station, 3, nt)`, the default time length is 8192
- `begin_time`: the begin time of the waveform data
- `end_time`: the end time of the waveform data
- `phase_time`: the phase arrival time with shape `(n_station,)`
- `phase_index`: the time point index of the phase arrival time with shape `(n_station,)`
- `phase_type`: the phase type with shape `(n_station,)`
- `phase_polarity`: the phase polarity in ('U', 'D', 'N') with shape `(n_station,)`
- `event_time`: the event time
- `event_time_index`: the time point index of the event time
- `event_location`: the space-time coordinates of the event with shape `(n_staion, 3)`
- `station_location`: the space coordinates of the station with shape `(n_station, 3)`, including latitude, longitude and depth

The default configuration is `station_test`. You can specify the configuration by argument `name`. For example:
```python
# load dataset
# ATTENTION: Streaming(Iterable Dataset) is complex to support because of the feature of HDF5
# So we recommend to directly load the dataset and convert it into iterable later
# The dataset is very large, so you need to wait for some time at the first time

# to load "station_test" with test split
ceed = load_dataset("AI4EPS/CEED", split="test")
# or
ceed = load_dataset("AI4EPS/CEED", name="station_test", split="test")

# to load "event" with train split
ceed = load_dataset("AI4EPS/CEED", name="event", split="train")
```

#### Example loading the dataset
```python
ceed = load_dataset("AI4EPS/CEED", name="station_test", split="test")

# print the first sample of the iterable dataset
for example in ceed:
    print("\nIterable test\n")
    print(example.keys())
    for key in example.keys():
        if key == "data":
            print(key, np.array(example[key]).shape)
        else:
            print(key, example[key])
    break

# %%
ceed = ceed.with_format("torch")
dataloader = DataLoader(ceed, batch_size=8, num_workers=0, collate_fn=lambda x: x)

for batch in dataloader:
    print("\nDataloader test\n")
    print(f"Batch size: {len(batch)}")
    print(batch[0].keys())
    for key in batch[0].keys():
        if key == "data":
            print(key, np.array(batch[0][key]).shape)
        else:
            print(key, batch[0][key])
    break
```

<!-- #### Extension

If you want to introduce new features in to labels, we recommend to make a copy of `CEED.py` and modify the `_generate_examples` method. Check [AI4EPS/EQNet](https://github.com/AI4EPS/EQNet/blob/master/eqnet/data/quakeflow_nc.py) for an example. To load the dataset with your modified script, specify the path to the script in `load_dataset` function:
```python
ceed = load_dataset("path/to/your/CEED.py", name="station_test", split="test", trust_remote_code=True)
```
 -->