No. 36 · Seismological Research Letters · 20 February 2025

RPNet: Robust P-wave first-motion polarity determination using deep learning

Jongwon Han, Seongryong Kim, Dong-Hoon Sheen

DOI 10.1785/0220240384 Publisher page Cite PI corresponding author Seismological methods Earthquake properties

Representative figure

RPNet architecture and inception module. A P-wave waveform passes through convolution, inception, LSTM, self-attention, fully connected, and Monte Carlo dropout layers; one hundred iterative predictions are combined to classify polarity as up, down, or unknown. — RPNet combines inception modules, self-attention, and Monte Carlo dropout to determine P-wave first-motion polarity while quantifying prediction uncertainty.

Figure 1 · Han, Kim, and Sheen (2025), Seismological Research Letters, Figure 1. · Author-provided image; reuse not permitted

Citation

Jongwon Han, Seongryong Kim, and Dong-Hoon Sheen (2025), RPNet: Robust P-wave first-motion polarity determination using deep learning, Seismological Research Letters, 96(4), 2405–2417, doi: https://doi.org/10.1785/0220240384.