Near Real-Time Tsunami Early Warning System Using GNSS Ionospheric Measurements

Tsunamis produced by powerful earthquakes have devastating consequences for coastal communities. Tsunami early warning systems are critical to characterize threatening waves to allow for timely evacuations and significantly reduce casualties and economic damage. As Earths oceans mechanically couple into the atmosphere, tsunamis produce atmospheric acoustic/gravity waves, which propagate up to the ionosphere. Such travelling ionospheric disturbances (TIDs) may be detected using GNSS signals through the integrated total electron content (TEC) along lines-of-sight between the satellites and GNSS receivers. We present a prototype operational tsunami early warning system: the GNSS-based Upper Atmospheric Real-time Disaster Information and Alert Network (GUARDIAN). Our real-time architecture relies on three complementary building blocks. The Data Collection (DC) block aims to collect 1 Hz dual-frequency GNSS data and process them to produce TEC time series. The Deep Learning (DL) block assesses in real-time whether the TEC time series is the result of TIDs. Finally, the goal of the Inverse Modelling (IM) block is to extract tsunami characteristics from the TEC data using ensemble modeling. Based on real-time GNSS data from JPLs Global Differential GPS (GDGPS) System, we compute TEC time series along transmitter-receiver paths using GPS, Galileo, GLONASS and Beidou constellations with GNSS receivers located along the Pacific ring of fire. The architecture of this block is optimized to output time series of differential TEC in near real-time, while also correcting for cycle slips. In addition, we present the first results from a simulated real-time scenario for a proof-of-concept implementation of the DL block. We provide a design and update on our progress towards further extending GUARDIAN's capabilities with a real-time deep-learning system that is adapted to tsunami detection. Lastly, we introduce the IM block. Coupling UCARs data assimilation test bed (DART) and the physics-based model WP-GITM, ensemble runs are conducted, driven by a variety of tsunami parameters. For previously identified TID observations, the IM block is expected to output the optimal set of corresponding tsunami parameters.