An Open Source Repository for Earthquake Early Warning Performance Results
Abstract
In the deep learning domain, publicly and easily accessible benchmark data sets have been a crucial ingredient for model development and improvement (e.g. ImageNet, Deng et al. 2009). In Earthquake Early Warning (EEW), algorithm performance evaluation is complicated, among other things because the model predictions of earlier studies are not easily available and cumbersome to reproduce. Here we propose to establish a publicly available repository for real-time estimates and predictions made by EEW algorithms.
We suggest a simple, unified data format for storing time series of real-time ground motion predictions and source parameter estimates. Each submitted data set to the repository can be read directly by provided example code in either Python or Matlab. If an algorithm developer uses a data set for which predictions from another study are available in the repository, they can easily load those predictions and use them for comparison with their own predictions. The goal is to drive EEW algorithm improvement and innovation, by making it easy for algorithm developers to evaluate the added value of new algorithms, or algorithm variants. The proposed performance repository folds in well with ongoing efforts to store ShakeAlert algorithm predictions in a unified database, and with existing toolboxes for EEW algorithm evaluation, such as GMCOMP (github.com/crowellbw/GMCOMP). References Deng, Jia, et al. "Imagenet: A large-scale hierarchical image database." 2009 IEEE conference on computer vision and pattern recognition . Ieee, 2009.- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2020
- Bibcode:
- 2020AGUFMS046.0017M
- Keywords:
-
- 7299 General or miscellaneous;
- SEISMOLOGY