Experimental Validation for Early Earthquake Detection Using Transfer Learning

We apply transfer learning with an LSTM-attention model, trained on simulated earthquake SOP data and tested on experimentally emulated data over a 38 km deployed fiber link, achieving $98 \%$ accuracy in early earthquake detection.