Skip to main content

Non-Linear Bayesian partition-modeling of the Earth's mantle transition zone

Periodic Reporting for period 1 - NoLiMit (Non-Linear Bayesian partition-modeling of the Earth's mantle transition zone)

Reporting period: 2018-09-01 to 2020-08-31

The Earth’s mantle transition zone is a complex region exhibiting mineralogical phase changes as revealed by sharp increases of seismic wave-speed between 410 and 660 km depths. Because of its potential in filtering chemical elements, the transition zone represents a key region for understanding how efficient is global mantle convection to mix and recycle geochemical heterogeneities. Global sampling of the transition zone is only possible with seismic methods, via the analysis of seismic waves generated by large, distant earthquakes and subsequently recorded by receivers located on the Earth’s surface. These waves propagate and illuminate the Earth’s deep internal interior, and provide critical constraints on the elastic structure. Seismologists and geophysicists have since the 90’s attempted to isolate the effects of temperature and composition on mantle elastic properties. However, a major issue is imperfect seismic data coverage that prevents from reconstructing the multiple length-scales of thermo-chemical heterogeneities. Seismic and laboratory-based data suffer also from large uncertainties, and the relationship between seismic observables and in situ thermo-chemical parameters remains questionable. To overcome these limitations, this project will use a partitioning (multi-scale) approach to isolate the effects of mantle temperature and composition on the most comprehensive databases of seismic waves sensitive to the transition zone. Using a Bayesian probabilistic framework, I will simulate with their uncertainties the multi-scale physical properties of the transition zone, confront the results with high-pressure mineral physics experiments and with predictions from mantle convective mixing models. The interdisciplinary approach of this project relies on using state-of-the-art numerical methods and high performance computing to answer fundamental questions in Earth Sciences, seismology, geodynamics, and mineral physics. The uniqueness of the approach arises from quantifying in a probabilistic sense how conceptual mantle-mixing models explain seismic data. The outcomes of the project will be to provide new databases, procedures, models, and software packages for the analysis and understanding of the Earth's upper mantle structure.
The overall progress is in line with the work plan as defined by the Grant Agreement. The project includes 2 work packages, WP1 Partition modeling, and WP2 Connecting seismological and laboratory-based data. The plan was to achieve WP1 in September 2019, delivering a first software module (S1), and the submission of a manuscript (P1). WP1 has been completed. S1 has been implemented and used in conjunction with a new dataset. P1 has been submitted (Waszek et al, submitted to Nature Geoscience). In addition, anticipated advances have been achieved on WP2. Preliminary versions of the software module dedicated to the inversion for physical parameters have been implemented (S2), and aggregated to the S1 module (S3). These modules have been used for seismic analysis, with results incorporated in the submitted manuscript (P1). Presently, the software modules S1 and S2 that are delivered in WP1 and WP2 are implemented without the Bayesian inference component. This is to account for massive datasets that have been built in collaboration with Lauren Waszek (New Mexico State University and Australian National University). The mining of such data required enhanced performance with a partition-modeling scheme. I developed a hybrid algorithm where partition modeling is achieved through optimization. The modular structure will allow in the remaining time plugging directly the Bayesian inference component to the existing modules. Both software modules S1 and S2 will be updated with the Bayesian component by August 2021. For WP2, the manuscript P2 presenting the technical implementation is in preparation. Four communications have been done according to the plan, at ANU, EGU 2019, AGU 2019, and UCB Lyon 1. Additional communications have been done in South Korea, and in a workshop about mantle structure and composition. The project website is operative ( The data management plan has been delivered. Two outreach communications have been done as planned for Physics Market Day and the 50th birthday of the Moon Landing during Science week. A visit and animation at the French Australian Pre-School has also been done. The career development plan has been done through a Performance and Development Review in May 2018. Weekly meeting were done until the end of the outgoing phase within the group of Prof. H. Tkalčić. Quarterly progress reports are done through presentations within group meetings in Australia until August, in France since September.
The main impact of the project is to provide a better description of convective processes occuring in the mantle. These processes are at the origin of volcanism at the surface of the Earth, which provides a non-negligible component of hazard for populations, as well as an important contribution to the CO2 in the atmosphere. Progresses beyond the state of the art have been:
(1) The development of a complete workflow for predicting seismic waveforms from mineralogy and realistic pressure and temperature conditions, and the possibility of doing a spatial analysis of these synthetic data accounting for real data acquisition geometry and realistic processing
(2) The ability to provide from observed seismic data a more accurate description of the scale of variations of the mantle structure, including uncertainties
(3) The possibility of accounting for the effect of mantle mixing. Preliminary integration of multiple datatypes indeed suggests that unperfect mixing is the rule at ~500 km and longer length-scale
By the end of the project, the addition of a Bayesian component into the main analysis framework will allow:
(i) Inverting seismic data for the first-order mantle thermo-chemistry, provide uncertainties on inverted model parameters, and provide for the first time quantitative estimates on mantle mixing processes (length-scale)
(ii) Self-consistently integrating seismic data of different natures (converted and reflected body-waves, surface waves)
So far, the project has demonstrated that the mantle is more heterogeneous than previously suggested by smooth models of the Earth's interior based on seismic tomography. Heterogeneities exist at multiple scales in the vicinity of the main phase transitions between upper and lower mantle, and this heterogeneity is partly related to phenomena associated to phase transitions (stability of multiple phase assemblages, partial melting and compositional segregation). We have been able to anchor thermal profiles of the mantle down to the base of the transition zone, showing that a ~2100 Kelvin limit exists at ~660 km depth. This temperature can locally be exceeded in very small regions of the mantle (0.6%), in mantle plumes, and this must occur below a ~500 km wavelength. Large regions with a temperature of 1700-1800 Kelvin also exist, in particular beneath the Pacific basin. They appear related with upwelling material, but are not consistent with narrow plumes originating from the core mantle boundary.