Tomographic filtering of shear and compressional wave models reveals uncorrelated variations in the lowermost mantle

• Published in: Geophysical journal international Vol. 234; no. 3; pp. 2114 - 2127
• Main Authors: Su, Jun, Houser, Christine, Hernlund, John W, Deschamps, Frédéric
• Format: Journal Article
• Language: English
• Published: Oxford University Press 27.04.2023
• Subjects: Statistical methods; Inverse theory; Seismic tomography; Composition and structure of the mantle
• ISSN: 0956-540X; 1365-246X
• DOI: 10.1093/gji/ggad190

SUMMARY Seismic tomography models reveal differences in the geographic distribution and magnitude of P- and S-wave velocity variations (VP and VS, respectively) below ∼2200 km depth in the Earth’s mantle. In particular, large low shear velocity provinces (LLSVPs) beneath the Pacific and Africa exhibit a distinct low velocity population in the distribution of VS that does not stand out in VP models, carrying important implications for the origin of these features. However, it is possible that the absence of a distinct low velocity feature in VP models is an artefact of VP models having lower resolution compared to VS models owing to differences in coverage. Here, we use ‘tomographic filters’ computed from the singular value decomposition of the sensitivity matrices for a pair of VP and VS models in order to test whether such low velocity features are suppressed in VP models. Our ‘cross-filtered’ results show that resolution alone cannot explain the absence of a corresponding low VP population. We additionally apply the joint VP and VS tomographic filter technique to thermochemical mantle convection models to show that cases with distinct phase and/or composition may be differentiated from cases where only temperature varies. We then develop a new proxy for exploring uncorrelated VP and VS more broadly using the difference between the observed VP model and the filtered VS model input. Our results show that ‘large uncorrelated modulus provinces’ (LUMPs) extend beyond the boundaries of LLSVPs, and exhibit anomalies in both fast and slow regions.