Weathering of fluvial sands as a compositional linear process: A new modeling approach

Weathering of fluvial sands as a compositional linear process: A new modeling approach

Boogaerdt, N.M.

Weltje, G.J. (mentor)

Civil Engineering and Geosciences

Geoscience & Engineering

2012-07-05

Geological data is often compositional (i.e. positive and sum to unity). The constrained nature of compositional data can lead to severe problems during data analysis. Transforming from the simplex to centered log-ratio (CLR) space can solve these problems, converting constrained non-linear processes into unconstrained linear processes (transforming perturbation and power transformation into addition and multiplication). Using this transformation, techniques were developed to analyze weathering (a compositional process) in the fluvial system. Detailed knowledge of the weathering process will improve reservoir characterization by providing in- sights into the spatial distribution of reservoir rock, provenance regions, basin evolution and composition of sediment at the time of burial (an important input of diagenesis models). To analyze the CLR-transformed compositional data, several methods were developed. The SVD method uses principal component analysis (PCA), singular value decomposition (SVD) and cluster analysis to extract weathering directions from bedload samples (models usually ignore the bedload due to its complicated relation with time and its high level of noise, even though bedload is the main component of reservoir rock in fluvial reservoirs). The pairwise p-vector method combines directional derivatives with simple geological knowledge to obtain weathering directions and drainage patterns. Combined use of both methods creates the best results. The techniques were tested on a dataset including fluvial bedload samples taken in the Orinoco drainage basin. Two different grain type groupings were used; [Q F R] and [Qms Qmu Qp2 Qp3x F R]. To restrict the influence of non weathering processes, mineral groups with deviating densities or shapes were discarded. The developed techniques are able to extract directional information from CLR transformed data, which can provide a prediction of spatial trends in reservoir quality across a fluvial basin. Cluster analysis techniques can divide samples into distinct lithological regions, of which the upstream ones coincide with different provenance regions. The actual flow directions between the different clusters can be reproduced with a fair amount of reliability. A more detailed drainage pattern can be obtained by creating a flow direction vector plot using a grid over the geographical area. The weathering direction can be obtained with a reasonable degree of reliability with the techniques in this thesis (the ranking of elements in the weathering direction vectors, which reflects their durability, is in accordance with geological knowledge). With the weathering direction vector and the created drainage pattern one can start to predict compositional change across a fluvial basin. The results of this thesis contain the first building blocks towards a compositional linear process (CLP) model for reservoir-quality assessment in fluvial systems.

compositional data
weathering
directional data
centered log-ratio transform
fluvial
Orinoco

http://resolver.tudelft.nl/uuid:dd4aa3e3-2622-421b-83e7-b688ec679a22

Student theses

master thesis

(c) 2012 Boogaerdt, N.M.