Near-real-time detection of co-seismic ionospheric disturbances using
machine learning
Abstract
Tsunamis generated by large earthquake-induced displacements of the
ocean floor can lead to tragic consequences for coastal communities.
Ionospheric measurements of Co-Seismic Disturbances (CIDs) offer a
unique solution to characterize an earthquake’s tsunami potential in
Near-Real-Time (NRT) since CIDs can be detected within 15min of a
seismic event. However, the detection of CIDs rely on human experts
which currently prevents the deployment of ionospheric methods in NRT.
To address this critical lack of automatic procedure, we train
machine-learning models (random forests) over an extensive ionospheric
waveform dataset to (1) classify ionospheric waveforms between CIDs and
noise, (2) pick arrival times, and (3) associate arrivals across a
satellite network in NRT. Our model shows excellent classification and
arrival-time picking performances (~95% recall, average
error < 10 s). This model is the first automatic CID detector
which paves the way for the NRT imaging of surface displacements from
the ionosphere.