---
language:
- en
license: apache-2.0
tags:
- seismology
- geophysics
- earthquake-detection
- alaska-peninsula
pretty_name: PS_Alaska
size_categories:
- 1TB<t<10TB
task_categories:
- other
configs:
- config_name: ManualPick
  data_files:
  - "ManualPick/**/*"
- config_name: ManualPick_ai4eps
  data_files:
  - "ManualPick_ai4eps/**/*"
- config_name: PNTFIter1
  data_files:
  - "PNTFIter1/**/*"
- config_name: PNTFIter1_combined
  data_files:
  - "PNTFIter1_combined/**/*"
dataset_info:
  features:
    - name: network
      dtype: string
    - name: station
      dtype: string
    - name: phase_time
      dtype: string
    - name: phase_type
      dtype: string
  splits:
    - name: train
---

# Dataset Card for PS_Alaska

`PS_Alaska` is a high-precision regional seismological dataset spanning 2018–2022 along the Alaska Peninsula. Built using an iterative, semi-supervised deep learning pipeline (PhaseNet-TF, GaMMA-1D, and tomoDD), it provides continuous waveform windows and phase picks resolving 117,151 events to map background seismicity and the complex aftershock sequences of the 2020–2021 earthquake triplet.

The repository is structured into four primary sub-datasets that drive this pipeline:

### 1. ManualPick
* **Description:** The ground-truth reference baseline.
* **Content:** Zipped 3-component seismic waveforms sliced around 7,242 local earthquakes manually picked by the Alaska Earthquake Center (AEC) between May 2018 and August 2019. Used for initial model retraining and regional optimization. Whenever utilizing this baseline, please reference both the core deep-learning catalog study and the primary AEC data architecture paper.

### 2. ManualPick_ai4eps
* **Description:** Standardized machine learning benchmark subset.
* **Content:** The same expert-verified AEC manual picks structured and formatted specifically for seamless out-of-the-box compatibility with AI4EPS (AI for Earthquake Parametric Systems) deep-learning frameworks. Users of this dataset should provide attribution to the core catalog methodology paper alongside the original AEC reference paper.

### 3. PNTFIter1
* **Description:** The expansion dataset from Iteration 1 of the pipeline.
* **Content:** Waveforms centered around the 110,147 newly discovered, machine-predicted events identified after running the initial retrained model over continuous regional data.

### 4. PNTFIter1_combined
* **Description:** The heavy-duty training matrix for the final pipeline pass.
* **Content:** A ~692 GB consolidated dataset merging accepted Iteration 1 model picks with remaining reference human picks. This combined array optimizes deep-learning performance on challenging, high-noise records from Ocean-Bottom Seismometers (OBS).

## References

```
@article{jie2026deep,
  title={Deep-Learning-Based Catalog of Background Seismicity and Aftershocks of the 2020--2021 Large Earthquakes Along the Alaska Peninsula},
  author={Jie, Yaqi and Wei, Songqiao Shawn and Zhu, Weiqiang and Freymueller, Jeffrey and Elliott, Julie},
  journal={Seismological Research Letters},
  volume={97},
  number={1},
  pages={187--203},
  year={2026},
  publisher={Seismological Society of America},
  doi={10.1785/0220250072}
}

@article{ruppert2022enhanced,
  title={Enhanced Regional Earthquake Catalog with Alaska Amphibious Community Seismic Experiment Data},
  author={Ruppert, Natalia A and Barcheck, G and Abers, Geoffrey A},
  journal={Seismological Research Letters},
  volume={94},
  number={1},
  pages={483--493},
  year={2023},
  doi={10.1785/0220220226}
}