Water, Mid-ocean Ridges, and Dynamic Geochemical Layering of the Earth's Mantle

A model has been developed that could help reconcile apparently conflicting observations about the extent of geochemical layering of the Earth's mantle. Seismological observations and numerical simulations both suggest that there is significant material exchange across the upper mantle (UM; defined as above the 660 km seismic discontinuity) and the lower mantle (LM). Geochemical observations indicate that the mantle that melts at mid-ocean ridges (MOR) is different from that melted at intraplate hot spots (ocean island basalts, OIB), suggesting that the LM is chemically distinct from the UM. An important additional observation is that typical MOR mantle is "depleted," but only in incompatible trace elements, indicating that it has been affected by removal of only small degrees of partial melt (ca. 1%). We propose that a key component of the mantle differentiation cycle is the effect of water on melting under MORs, and possibly also in island arcs. The presence of ca. 200-500 ppm of H₂O causes melting to begin at greater depth (ca. 100-150 km). The resulting oceanic lithologic column is composed of not only the basaltic crust (4-9km) and depleted harzburgite/lherzolite (DHL; ca. 40-70 km thick) sections, but also a third major section of incipiently melted lherzolite (IML; 30-70 km thick). If during subduction the IML is preferentially retained in the UM since it remains at the ambient mantle temperature, while the BC-DHL package is preferentially carried to the LM (as a consequence of its negative buoyancy due to lower temperature and phase assemblage, as well as high viscosity), the result will be a dynamical geochemical layering in terms of trace elements, isotopes, and heterogeneity. Over 2 billion years a quasi-steady-state UM will develop that is highly "depleted" and relatively homogeneous, and the complementary LM will be slightly depleted to enriched, and more heterogeneous. Application of a simple box model to the Sm-Nd system gives a steady-state UM-LM difference in ɛ _{Nd} of 4 units, successfully reproducing the observed difference between mean MORB and mean OIB. This value occurs for mass fluxes between UM and LM corresponding to about 50% of the current rate of generation of oceanic lithosphere. Therefore, this model accounts for observed \epsilon_Nd values if slab penetration into the LM occurs but is ca. 50% efficient. The LM would also tend to have more variable ɛ _Nd, by about a factor of 2. The results are not sensitive to the relative size of UM and LM, but the overall Nd mass balance for the Earth requires the UM be small.