Numerical investigations of the mantle plume initiation model for flood basalt events

Abstract

Recent studies have linked both the eruption of continental flood basalts and the emplacement of large oceanic plateaus to the initial phase of mantle plume activity, i.e., the “plume head” model. Here we develop this concept further through numerical models of a large thermal diapir rising through the mantle and impinging upon the base of the lithosphere. A finite element analysis in axisymmetric geometry is used to model the dynamics of solid state convective flow and heat transport in the mantle, and an anhydrous batch melting model is used to estimate melt volumes and compositions obtained from partial melting in the shallow mantle. We explore the effects of a number of model variables, including mantle viscosity structure, plume temperature and diameter, and the mechanical response of the lithosphere. Our study emphasizes cases in which the lithosphere does not undergo significant extension prior to large‐scale melting. A “standard model” is found to be consistent with some observed characteristics of flood basalt/plume initiation events. This model includes an initial diapir of radius 400 km and initial excess temperature 350°C rising through a mantle of viscosity ∼10 21 Pa s and impinging beneath a model oceanic lithosphere ∼100 Ma in age. If lithospheric extension does not occur, melting is almost entirely sublithospheric, the melt volume produced is ∼3.5×l0 6 km 3 and its composition has high MgO content (15–20 wt %). A time sequence of precursory central (axial) uplift of several kilometers followed by major melting (volcanism) is a robust feature of all the models, regardless of mantle viscosity structure or lithospheric response. Low viscosity in the upper mantle compresses the timescale for these events, and results in a constriction of the plume “head” initially impinging on the lithosphere. The radial extent of the melting region (about 500 km at the time of maximum melt production) is only slightly affected by the upper mantle viscosity. For nonrifting lithosphere, no significant melt generation occurs for initial plume excess temperatures less than about 300°C. The initial plume radius has only a modest effect upon the volume of melt obtained. The most important aspect of lithospheric response is whether or not the uppermost crust and lithosphere (above ∼40 km depth) undergo extension: otherwise, the ductile strength of the lower lithosphere has little effect. If the lithosphere is allowed to extend freely, the melt volume increases by about an order of magnitude, and it is comparable with some of the largest oceanic plateaus such as Ontong Java. If continental or oceanic flood basalt events occur on old, nonrifting lithosphere, the primary melts are likely to be much more MgO rich than the basalts erupted at the surface. This is consistent with abundant evidence for olivine fractionation in many flood basalt provinces, along with the occurrence of picrite lavas, and we hypothesize that fractionation occurs at a density trap at the crust/mantle boundary (Moho). At this stage and following, it is likely that significant lower crustal and mantle‐lithospheric contamination of mantle plume‐derived magmas may occur, both isotopically and in the trace elements. The sequence and timing of both precursory and post‐magmatic events in the vast, well‐exposed oceanic flood basalt terrane known as “Wrangellia” (SE Alaska; British Columbia) are consistent with the predictions of our model. The uplift predicted by our models is of order 2–4 km if we require that mantle plumes are hot enough to melt beneath unrifted lithosphere. Effects not modeled here, such as small‐scale convection and hydrous melting, will tend to reduce the model excess temperature and uplift. In order to better understand the relation between mantle plumes and large igneous provinces, several important advances are required: implementation of a fractional melting model, some realistic assessment of melt transport through the mantle and crust, and a fully three‐dimensional convection model.