Predictive and invertible models of sediment geochemistry from catchment to continental scales

Alexander George Lipp, Gareth Roberts, Alexander Whittaker, Natural Environment Research Council (Great Britain)
2022
The elemental geochemistry of sediments is studied to understand a wide range of topics in the Earth sciences, e.g., the evolution of the continents, the history of the carbon cycle, and the location of mineral deposits. A challenge underlying all these applications is that sedimentary compositions are affected by a wide range of geochemical processes such as weathering, sorting, and mixing. These competing processes make it hard to extract useful signals from a sediment's elemental
more » ... . In this dissertation I will develop two predictive and invertible models for the elemental composition of a sediment. The first model describes a fine-grained sediment in terms of the composition of its protolith and the intensity of weathering it has experienced. A novel benefit of this model is that it can 'unweather' the composition of a sediment to reveal the theoretical protolith from which it derived. Using this unique approach I deduce the composition of the Archean continents from a large compilation of sedimentary rock compositions. I find the Archean exposed continents to be more silica rich than previously considered, comparable to the present day continents. The second model I develop describes fluvial sediments as a conservative mixture of the geochemistry of the sediments in their upstream drainage basin. This model is tested using a uniquely high-resolution geochemical survey covering the Cairngorms, UK, along with 67 new downstream sediment samples gathered for this dissertation. I demonstrate that the composition of these downstream sediments can be reasonably predicted assuming just conservative mixing of the upstream survey dataset. Given the success of this mixing model, I proceeed to invert it. That is, I develop a model to unmix fluvial sediments to generate a geochemical map of their source regions. I find that this unmixing model, when applied to the downstream dataset, generates upstream predictions that compare favourably to the independent survey dataset. The success of both these models sugges [...]
doi:10.25560/100202 fatcat:ekfwqc6jqnbs3dicuaus7uuesu