Sharing and reproducing experiments of earthquake forecasting: designing open Floating Testing Experiments
Testing is essential to extend our knowledge of earthquake processes and to validate seismicity forecasting models. Although the legitimacy of testing is attained through the experiment’s reproducibility, this is seldom exempt from limitations. To improve current scientific practices, the Collaboratory for the Study of Earthquake Predictability (CSEP) released an open-source software, known as pyCSEP, to facilitate testing procedures and make their implementation transparent. Based on it, we designed two prospective Floating Testing Experiments, which are completely open, accessible and reproducible by the scientific community, and also waives the need of a complex computational architecture: (i) A global experiment to understand the limits of the forecasts’ spatial resolution based on the available information, and its detriment on performance, by using a QuadTree approach (ii) An experiment in Italy to investigate key assumptions and the parameter space of short-term forecasting methods, and to validate their use in Operational Earthquake Forecasting. The experiment results are periodically updated on-demand, and they are set public during and after the entire experiment’s duration. Based on these setups, we expect that multiple experiments can be easily implemented and made open by the seismology community to test the hypothesis they consider relevant for seismicity forecasting.