Romanian Seismic Network Benchmark Dataset (RO-MLDB)

The Romanian Seismic Network Benchmark Dataset (RO-MLDB) consists of seismic data recorded from local events by the Romanian National Seismic Network, including the Vrancea EPOS NFO. The dataset includes only local events with a minimum moment magnitude of 2.5, recorded between 2014 and 2024, with at least 10 manually picked P- and S-wave arrival pairs. All seismic stations represented in the dataset are three-component stations — comprising both seismometers and strong-motion sensors — operating at sampling rates of 50 and 100 Hz.

RO-MLDB uses the HDF5 format for waveform recordings. This choice ensures full compatibility with the SeisBench toolbox, while also facilitating straightforward integration with other software tools and programming environments. For improved storage efficiency, waveform data are stored in integer format (counts).

The dataset is provided as HDF5 files in either counts (_counts.h5) or physical (_physical.h5) units, and consists of:

19,230 earthquakes;

150,931 three-channel waveforms from broadband and short-period seismometers, as well as accelerometers.

The dataset was created in the context of the TRANSFORM² project.