Italy 2024 Forecasting Experiment – Open Call for modelers

Preface

“Experiments are the only means of knowledge at our disposal. The rest is poetry, imagination”

Max Planck

In open systems, such as the natural systems, the definition of “experiment” is more challenging than in one laboratory where all factors are under control, but it maintains its unique status in the general scientific enterprise. The Collaboratory for the Study of Earthquake Predictability (CSEP) has been always aiming at the definition of transparent and reproducible experiments to evaluate and compare earthquake forecasting models. The transparency and reproducibility are the two founding pillars of CSEP experiments, which allow any scientist to carry out any kind of statistical tests, at any time. This new short-term forecasting experiment in the Italian region follows the first round of experiments (Marzocchi et al., 2010), whose results have been analyzed by Taroni et al (2018) and Herrmann et al (2023).

This new Italy 2024 Forecasting Experiment has few important new features. The forecasts consist of simulated catalogs provided by the model, allowing a testing phase that is not anymore rooted in the debatable Poisson assumption (e.g., Lombardi and Marzocchi, 2010) and allowing, in principle, a proper treatment of the aleatory variability and epistemic uncertainty. Then, a modified suite of tests has been developed accordingly, taking into account the new format of the forecasts. A major advance is given by the publicly available pyCSEP software that will facilitate the analysis of the results and it will help modelers to familiarize with the CSEP world. One main weakness of the previous CSEP experiment was the lack of information for the whole duration (5 years) of the experiment. For this reason, in this new experiment we plan to release results of the experiment continuously to keep high the interest of modelers and interested actors.

Experiment Description, Documents and Materials

Model Submission

A forecasting model should be submitted as a repository of source code files. Before model submission, each modeler should download and/or review the following items from the experiment’s repository, and follow the instructions of the Modeler’s Instructive :

To initiate the submission and participation, please contact pciturri@gfz-potsdam.de

News

pyCSEP v0.6.3 is now available on PyPI and conda-forge. Visit the GitHub page for more information.

Recent BlogS