Collaborative Research: Global mapping of the Earth's D" layer using machine learning and implications for deep mantle structure, properties and dynamics

Nearly 1,800 miles below Earth’s surface, the rocky mantle meets the liquid iron core. Heat escaping the core stirs the slow churning of the solid mantle above driving plate tectonics, earthquakes, and volcanoes at the surface. Just above the Earth’s core sits a thin, mysterious layer known as D″ (“D double prime”). Mapping the characteristics of this layer is key to learning how much heat leaves the core and understanding how the planet’s engine works. This project will use machine learning, a form of artificial intelligence, to find faint echoes from this layer in millions of earthquake recordings. The outcome will be the most complete global map of this hidden layer. The project will also train graduate students, and all new software will be shared freely. The D″ discontinuity in the lowermost mantle is commonly attributed to a mineral phase transition from bridgmanite to post-perovskite. A local back-transition to bridgmanite near the core-mantle boundary is used to constrain lowermost-mantle temperatures and to estimate the heat flux across the core-mantle boundary. Yet direct observations remain incomplete and inconsistent among regional studies. For example, D″ has been detected in some regions, while it appears absent in others, furthermore its velocity contrast changes sign from place to place. This project will deliver a global characterization of the D″ discontinuity by combining three-component seismic data in the analysis. First, the investigators will use a seismic dataset of more than 17 million seismograms collected from 25 seismic data centers. Second, the team will develop a new, flexible machine-learning model optimized for picking subtle phases in the global seismic wavefield, primarily D″-reflected P and S waves, and designed for ready application to other deep-Earth phases such as SPdKS or S3KS. Third, detected reflections will be characterized using full-wave simulations to constrain the depth of D″ reflectors and the sign and magnitude of the impedance contrast across them. The results will address the mineralogy of the deepest mantle, the heat flux from the core that helps drive mantle convection, lateral variations in D″ properties, and how D″ relates to large-scale structures seen in seismic tomography, such as subducted slabs and large low-velocity provinces. The project will train two graduate students at the collaborating institutions, and its results and open tools will be incorporated into education and outreach activities. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.