NLPLabNTUST/Merged-CWA

CWA Benchmark: A Seismic Dataset from Taiwan for Seismic Research

Dataset Description

This dataset includes a larger number of seismic events, especially high-magnitude. A comprehensive set of events collected by the Central Weather Bureau in Taiwan. The CWA benchmark features over 40 attributes and ∼500,000 seismograms, providing valuable data labels for various seismology-related tasks. In the future, we will keep updating the dataset to ensure its relevance and utility.

The entire dataset are released in Seisbench, the dataset can be loaded properly by the following:

1. Install Seisbench & HuggingfaceHub

$ git clone https://github.com/seisbench/seisbench.git

$ cd seisbench
$ pip install .
$ pip install --upgrade huggingface_hub

2. Load the CWA dataset

import seisbench.data as sbd

basepath = <path to metadata.csv and chunks.hdf5>
cwa = sbd.CWA(download_kwargs={'basepath': basepath})
print(cwa.metadata.head())

• Some parameters customizing the CWA dataset
  • (Bool) merge: Whether to load the merged version (CWASN + TSMIP + Noise), default=True.
  • (String) subset: Specify the seismographic network (CWASN, TSMIP, All), default="All".
  • (List) train_year: The range of years used for the training set, default=[2011, 2018].
  • (List) dev_year: The range of years used for the development set default=[2019].
  • (List) test_year: The range of years used for the testing set, default=[2020, 2021].

import seisbench.data as sbd

merge = False
subset = "CWASN"
train_year = [2012, 2017]
dev_year = [2018, 2020]
test_year = [2021, 2021]

basepath = <path to the directory of metadata.csv and chunks.hdf5>
cwa = sbd.CWA(subset=subset, 
              merge=merge, 
              download_kwargs={'basepath': basepath},
              train_year = train_year,
              dev_year = dev_year,
              test_year = test_year
)

print(cwa.metadata.head())

---

Overview of the CWA

Attribute	CWASN	TSMIP	Noise
Events	5,849	12,306	x
Traces	1,237,272	98,665	806,914
p-picks	1,237,272	98,665	x
s-picks	821275	98,665	x
sampling rate	100	200	100
Multi-event	196965	0	x
Measurement	cm/s2, cm/s	cm/s2	cm/s2, cm/s
Time period	2012-2021	2011-2020	2012-2021

---

Number of trace

• Categorized into 5 levels by five different pickers
• Noise samples are also provided
• Number of traces

  Level	Meaning	CWASN	TSMIP
  0	The 3-component data are all zeros	45,244	9,620
  1	None of the pickers predicted correctly	174,720	3,046
  2	The seismic signal is interrupted	714	0
  3	1 or 2 pickers predicted correctly	192,889	3,487
  4	More than 3 pickers predicted correctly	823,705	82,512

File size

2011	2012	2013	2014	2015	2016	2017	2018	2019	2020	2021	Noise	Total
1.7G	40 G	46 G	41 G	53 G	57 G	41 G	83 G	44 G	43 G	50 G	176 G	836 G

---

Station in CWA

---

Event in CWA

---

Acknowledgement

This work was supported by grants from the Central Weather Administration (CWA) of Taiwan and the National Science and Technology Council of Taiwan (NSTC 112-2636-E-011-002, NSTC 112-2628-E-011-008-MY3, and NSTC 113-2640-B-002-005). In addition, the authors acknowledge the support of the “Empower Vocational Education Research Center” at the National Taiwan University of Science and Technology through the Featured Areas Research Center Program within the framework of the Higher Education Sprout Project by the Ministry of Education (MOE) in Taiwan. The authors would also like to express our gratitude to the National Center for High-performance Computing of the National Applied Research Laboratories (NARLabs) in Taiwan for providing computational and storage resources.

---

Citation

@misc{tang2024CWA,
  title={The CWA Benchmark: A Seismic Dataset from Taiwan for Seismic Research},
  author={Kuan‐Wei Tang, Kuan‐Yu Chen, Da‐Yi Chen, Tai‐Lin Chin, and Ting‐Yu Hsu},
  howpublished={https://pubs.geoscienceworld.org/ssa/srl/article-abstract/doi/10.1785/0220230393/650394/The-CWA-Benchmark-A-Seismic-Dataset-from-Taiwan?redirectedFrom=fulltext}, 
  note={Early publication on Seismological Research Letters},
  year={2024},
}