Stochastic Modelling of Krafla´s Magma Bodies

Author: James Catley
Year: 2018
Supervisor: Juliet Newson

Abstract

The Krafla volcanic system in NE Iceland exhibits active bimodal basic and acidic magmatism, with high-temperature geothermal resources that have been exploited since 1977. Deep wells provide a unique insight into Krafla’s geological structure, including two intersections of acidic magma. It is thought that shallow magma bodies play a significant role in heating the geothermal system, as well as being a volcanic hazard. The study evaluates available geoscience and engineering data; generating several conceptual models for the locations and morphology of current magma bodies. These include various combinations of dykes, sills, cone sheets, and magma chambers. Training images derived from these models form the basis for further stochastic simulation of the bodies using the DeeSse multiple point geostatistics algorithm. A supervised machine learning classification model was used to predict magma occurrence based on exhaustive geophysics data, with the aim of guiding local target probability in the simulations. Unfortunately, the data was inadequate to train a robust model, and excessive processing times also resulted when using a local target probability in DeeSse. The resulting realisations are therefore unconstrained by geophysics and quantities of magma in the models remain arbitrary. When analysed for uncertainty using information entropy, phi, and distance clustering – the simulations show that magma probability is defined mostly by body geometry and large-scale patterns constrained by the limited hard data. The study demonstrates the potential for DeeSse to reproduce complex geological patterns, as well as the difficulty in providing appropriate training images.