/RED-PAN

This is the official implementation of Real-time Earthquake Detection and Phase-picking with multi-task Attention Network

Primary LanguagePythonMIT LicenseMIT

RED-PAN

This is the official implementation of Real-time Earthquake Detection and Phase-picking with multi-task Attention Network

trg_all.mp4

Summary

Installation

To run this repository, we suggest to install packages with Anaconda.

Clone this repository:

git clone https://github.com/tso1257771/RED-PAN.git
cd RED-PAN

Create a new environment via conda & pip

conda update conda
conda create --name redpan python=3.9
conda activate redpan
pip install --upgrade pip
pip install -r requirements.txt

Project Architecture

.
├── temp_01_gen_sample_from_INSTANCE      ### *Temp01. Generate template samples with different data augmentation stratigies*
│   ├── auto_implement.sh                 # automatically implement program P00~P04*.py to generate samples from INSTANCE
│   ├── P00_slice_INSTANCE.py             # make SAC files from INSTANCE<br />
│   ├── P01_TFRecord_SingleEQ.py          # generate tfrecord files of single-earthquake waveform
│   ├── P02_TFRecord_SingleEQ_zeropad.py  # generate tfrecord files of zero-padded single-earthquake waveform
│   ├── P03_TFRecord_MMWA.py              # generate Marching Mosaic Waveform Augmentation samples
│   └── P04_TFRecors_EEWA.py              # generate Earthquake Early Warning Augmentation samples
├── temp_02_model_pretrain                ### *Temp02. Fine-tune the REDPAN(60s) with samples obtained with temp_01 outputs.* 
│   ├── P01_train_REDPAN_dwa.py           # a template code of multi-task model training using dynamic weight average (dwa) strategies
│   └── tflite_converter.py               # convert the trained model to tflite format (optional)
├── temp_03_continuous_predict            ### *Temp03. Make prediction on continuous data using Seismogram-Tracking Median Filter (STMF) strategy*
│   ├── P01_continuous_pred.py            # load a 1-hour-long sac file and make predictions
│   ├── P02_pred_result_csv.py            # write the earthquake waveform detection and picking results from the REDPAN output funtions
└── temp_04_realtime_predict              ### *Temp04. Do real-time detection simulations*
    ├── P01_rt_pred_tflite.py             # do real-time detection using tflite model and plot the result at every time step of sliding prediction
    ├── P02_make_avi.py                   # make the animation using ffmpeg package (optional)
.

Run template codes

In this repository, we provide four template scripts:

(1) temp_01_gen_sample_from_INSTANCE
generate some samples using Earthquake Early Warning Augmentation (EEWA) and Marching Mosaic Waveform Augmentation (MMWA) using INSTANCE dataset.

cd temp_01_gen_sample_from_INSTANCE
bash auto_implement.sh

(2) temp_02_model_pretrain
fine-tune the trained model using few samples generated in temp_01_gen_sample_from_INSTANCE

cd temp_02_model_pretrain
python P01_train_REDPAN_dwa.py

(3) temp_03_continuous_predict
process continuous data using Seismogram-Tracking Medium Filter (STMF) 

cd temp_03_continuous_predict
python P01_continuous_pred.py

(4) temp_04_realtime_predict
triggering P arrivals using information-clipped earthquake waveform, simulating the conditions of real-time data processing.

cd temp_04_realtime_predict
python P01_rt_pred_tflite.py