Flatfiles for the detection and classification of pulse-like ground motions (PULSEFLAT)

PulseFlat is a curated dataset of near-fault ground motion records containing velocity pulses and related parameters, which are key features in applied seismology and structural response analysis. The flatfiles include detailed waveform metadata, pulse characteristics, and classification labels, making them ready for use in machine learning applications. The core pulse identification and classification algorithms are implemented in Python 3, based on a multi-method framework that builds upon traditional wavelet-based approaches, including those by Baker (2007), Shahi and Baker (2014) and Ertuncay and Costa (2019). The PulseFlat flatfiles are built upon and extend the near-source flatfiles from NESS2 (Sgobba et al., 2021) and BB-SPEEDset (Paolucci et al., 2021), which include empirical and 3D physics-based simulated ground motion records, respectively. These datasets have been enriched with newly derived pulse parameters and classification labels. PulseFlat enables studies in ground motion modeling and hazard assessment, structural performance under pulse-like excitation and data-driven and machine learning analysis.