Dear all, Posting forwarded from Kerry Gallagher, Imperial College, London. Sheila Peacock, list co-owner > From [log in to unmask] Wed Mar 13 15:03:06 2002 > Date: Wed, 13 Mar 2002 15:14:25 +0000 > From: Kerry Gallagher <[log in to unmask]> > X-Accept-Language: en > MIME-Version: 1.0 > To: [log in to unmask] > Subject: Industrial CASE PhD project with Total-Fina-Elf > Content-Transfer-Encoding: 7bit > > Dear Russ/Shelia > Could you please post this, with the subject above > Thanks > Kerry > > ======================================================================== > > Imperial College of Science, Technology and Medicine > Dept. of Earth Sciences and Engineering > > Fully funded PhD studentship available from October 2002 > > Nonlinear statistical methods for 4D spatial-temporal problems in the Geosciences > > Supervisors : Dr Kerry Gallagher and Dr. David Denison (Mathematics > Dept., Imperial College), Dr. Paul Williamson (Total-Fina-Elf) > > > BACKGROUND > Geoscience addresses problems on a wide variety of scales and types. > Often, these involve both noisy data and spatial-temporal > interpolation/extrapolation issues and many of these impact directly on > exploration and production strategies. For example, geologists working > on oil reservoirs try to characterise the fluid flow properties > (permeability, porosity) in heterogeneous rock types from a limited > number of cm-scale samples obtained from drill holes, and this need to > be interpolated over length scales of km to predict trends in oil > production from the reservoir. Similarly, exploration geophysicists are > interested in characterising the noise associated with seismic data in > 3-D and more recently in time-lapse seismic monitoring of producing > reservoirs to obtain optimal images of physical changes in the reservoir > over time. Interpolation and extrapolation of irregularly distributed > spatial and temporal data is also a major issue for more academic > applications such as long term denudation on the sub-continent scale > inferred from a limited number of apatite fission track samples for > example. This project aims to explore the application of recently > developed non-linear statistical techniques which provide a robust means > of characterising unknown spatially and temporally varying fields > (including noise). > > RATIONALE AND METHODOLOGY > Kriging represents the canonical method of interpolating spatial fields > in the Earth Sciences (and the more general area of geostatistics). This > approach assumes a stationary covariance structure and that nearby data > are more likely to be similar than widely separated data. However, > determining a suitable correlation function that can adequately > represent the true relationship between correlation and distance, even > if one exists, is difficult. Further, it is common practice to use the > data to guide this decision and then estimating its parameters using the > same set of data. This introduces model biases that are difficult to > quantify and may be severe. > In contrast, we propose the use of partition models, which are a newly > proposed method that has shown a great deal of promise, especially in > the modelling of disease risk (a classic spatial problem in medical > science). Partition models work by splitting up the area of interest > into a series of disjoint regions within which local models are fitted. > These local models are usually simple so are assumed to have a > stationary covariance structure. However, by using the data explicitly > to find the regions, partition models allow for non-isotropic > correlation functions, as well as discontinuities such as faults. Such > models provide a means to estimate directly the spatial process > representing, for example, the deviations of the mean level of the > process from the empirical mean of the observed data. Partition models > are readily cast, and particularly tractable, in a Bayesian framework > where the implementation can be carried out with Markov Chain Monte > Carlo (MCMC) methods. Bayesian modelling allows for the explicit > incorporation of a priori information when making inferences about the > spatial (and temporal) correlation between observed data. Integration of > the MCMC models also provides explicit smoothing where discontinuities > are not required, whilst maintaining them if they are. > The objective of this project is to investigate the application of > partition modelling for the first time to some of the geoscience > problems discussed above. In particular, the project will use these new > methods to address the exploration-related issues of reservoir > characterisation and model building and 4D (time-lapse) seismic data, > where we deal with relatively scarce and irregularly distributed hard > data (from boreholes) but large amounts of soft data (from seismic > reflection surveys). The aim is to test the partition model approach for > both a robust characterisation of noise and real geological structure, > and for producing reliable interpolation/extrapolation procedures as a > key step in model building. The application of the methods will focus on > large, high quality data sets, obtained from the CASE partner, > TOTAL-FINA-ELF, such as time-lapse seismic data from the Oseberg Field, > in the Brent province of the North Sea. > > TRAINING > The student will join the Geophysics Group in the Dept. of Earth > Sciences and Engineering at Imperial College, and also be affiliated to > the Statistics section in the Dept. of Mathematics. The student will > receive appropriate training in statistical analysis and computer > programming, as well the key concepts of the exploration-related > applications (time-lapse seismic, reservoir simulation and model > building) as part of the Geophysics group's ongoing industry-funded > research and PhD projects, and the close links with the Petroleum > Geoscience and Engineering groups at Imperial College. The student will > also attend the research student training courses run within our > department, as specified later. Finally, the student will work closely > with TFE staff in London to develop new strategies for dealing with > exploration/production data, learning the industrial side of the > problems on site. > > > > Please contact Kerry Gallagher ([log in to unmask], 020-7594-6424), and see > the departmental website (http://www.ese.ic.ac.uk/) > > > -- > Kerry Gallagher > Reader in Geophysics > Dept. of Earth Science and Engineering > Imperial College of Science, Technology and Medicine > South Kensington > London > SW7 2AS > England > > Ph : 44-(0)20-7594-6424 > Fax : 44-(0)20-7594-6464 > Email : [log in to unmask] > > http://www.huxley.ic.ac.uk/research/Comp&Geophys/geophystmp/people/kerry3.html >