Bareback riding on two horses

1
Bareback riding on two horses

• A case study of a cross-disciplinary PhD

2
Bareback riding on two horsesa case study of a
cross-disciplinary PhD

• Overview
• Describe an academic literacy consultation
• Explain the problem complicating factors
• Outline remedial steps taken
• Highlight findings about texts, audiences and
  disciplinary writing practices

3
Bareback riding on two horsesa case study of a
cross-disciplinary PhD

• Problem
• PhD student who cant write
• thinks in a different way
• doesnt put things in context
• cant organise thoughts and ideas
• doesnt get the message across
• Native speaker of English
• Writing an abstract for a conference presentation

4
Bareback riding on two horsesa case study of a
cross-disciplinary PhD

• Complicating factors
• world class department
• policy of encouraging bright students from other
  fields, e.g. Mathematics, to do PhDs
• expectations for professional communication very
  high
• students under pressure to perform
• supervisor recently moved from industry to
  university position

5
How we imagine oil is extracted
6
How oil is actually extracted
7
Errors in predicting performance of reservoirs

• Two kinds of error
• Sampling error
• due to difficulty of collecting samples from the
  reservoir
• Modelling error
• due to trade off between computer processing
  time and degree of detail for the calculations

8
Problem as presented by supervisor

• In your view which of these abstracts is better
  (whatever you take
• better to mean)?
• Error Models for Reservoir Performance Prediction
• In the petroleum industry, the standard method
  for parameter estimation uses the least squares
  method which gives biased predictions. An
  improvement to this method based on statistical
  models for solution error is made by matching
  numerical models to observed data, allowing
  construction of an error model. This error model
  should have the properties of relatively low
  computational cost, simple to construct and be
  robust.
• Error Models for Reservoir Performance Prediction
  
• Accurate reservoir performance prediction relies
  on the accuracy of uncertainty quantification
  methods. In the petroleum industry production
  forecasts are rarely checked and the cases which
  have been were found to be inaccurate. The
  standard approach to uncertainty quantification
  produces biased and overconfident predictions. An
  error model greatly improves the quality of
  performance prediction by significantly reducing
  the effects of bias and accurately quantifying
  uncertainty. The error model should have the
  properties of relatively low computational cost,
  simple to construct and be robust.
• What criteria did you use to evaluate better in
  this writing?

9
Problem as presented by supervisor

• Error Models for Reservoir Performance Prediction
• In the petroleum industry, the standard method
  for parameter estimation uses the least squares
  method which gives biased predictions. An
  improvement to this method based on statistical
  models for solution error is made by matching
  numerical models to observed data, allowing
  construction of an error model. This error model
  should have the properties of relatively low
  computational cost, simple to construct and be
  robust.

History
What we are doing
Properties
10
Problem as seen by the literacy consultant

• Supervisor commented on the rhetorical moves
• in the text
• Non-linguistic issues
• Ownership
• Competition?
• Gender?
• Two important questions to answer
• Who are you writing for?
• Why are you writing?

11
Presenting research to a non-specialist

• My work is concerned with quantifying
  uncertainty to aid in accurate forecasting. In
  the petroleum industry predictions are made far
  into the future. This needs to be accurate as
  there is a lot of money involved in the oil
  recovery process. Usually a number (a p90) is
  quoted meaning there is a 90 chance a particular
  level of oil will be produced by a given time. In
  the oil industry, the p90 is rarely verified at
  the end of production, and the few times it has
  been done it has been found to be very
  inaccurate.
• There are valid reasons why the oil industry is
  bad at making predictions. The first is
  collection of data. As the properties of a
  reservoir (porosity, permeability etc) are in
  general difficult to measure, a few samples will
  be collected, and data in between is inferred.
  Typically around 10(-17) of the reservoir is
  sampled (very very small!!) and the structure of
  a reservoir changes dramatically on a small
  scale. Thus, the input data to a prediction model
  is inaccurate. Currently my work disregards this
  error and focuses on the other main type or
  error, that is model error.
• Does this student really have a writing problem?

12
Presenting research to different audiences

• Rewrite for a petroleum engineer
• Model error refers to the errors incurred by
  solving the flow equations across the reservoir
  in a particular way. The reservoir is divided
  into grid blocks and the system solved via a
  simulator. The geological model can in practice
  take many hours, even days to solve on a standard
  workstation. Using small grid blocks, such as in
  the model produced by the geologist, lengthens
  the run time. The reservoir model is upscaled to
  produce a less detailed model. The upscaling
  process creates more inaccuracies in the
  prediction, but is favoured due to the reduced
  run time.
• Rewrite for a mathematician
• Model error refers to the errors incurred by
  solving the flow equations across the reservoir
  in a particular way. The reservoir will be
  divided into grid blocks and a numerical method
  of a given order is chosen to solve the system of
  partial differential equations. The geological
  model can in practice take many hours, even days
  to solve on a standard workstation. The level of
  detail (size of the grid blocks) consumes the
  bulk of the run time, and a smaller fraction is
  attributed to the order and type of numerical
  method. However, methods of first order are the
  usual choice, leaving the number of grid blocks
  to have the biggest influence. The reservoir
  engineer 'upscales' the geologists model to
  produce a less detailed model. The upscaling
  process creates more inaccuracies in the
  prediction, but is favoured due to the reduced
  run time.

13
Presenting research to different audiences

• Rewrite for a petroleum engineer
• Model error refers to the errors incurred by
  solving the flow equations across the reservoir
  in a particular way. The reservoir is divided
  into grid blocks and the system solved via a
  simulator. The geological model can in practice
  take many hours, even days to solve on a standard
  workstation. Using small grid blocks, such as in
  the model produced by the geologist, lengthens
  the run time. The reservoir model is upscaled to
  produce a less detailed model. The upscaling
  process creates more inaccuracies in the
  prediction, but is favoured due to the reduced
  run time.
• Rewrite for a mathematician
• Model error refers to the errors incurred by
  solving the flow equations across the reservoir
  in a particular way. The reservoir will be
  divided into grid blocks and a numerical method
  of a given order is chosen to solve the system of
  partial differential equations. The geological
  model can in practice take many hours, even days
  to solve on a standard workstation. The level of
  detail (size of the grid blocks) consumes the
  bulk of the run time, and a smaller fraction is
  attributed to the order and type of numerical
  method. However, methods of first order are the
  usual choice, leaving the number of grid blocks
  to have the biggest influence. The reservoir
  engineer 'upscales' the geologists model to
  produce a less detailed model. The upscaling
  process creates more inaccuracies in the
  prediction, but is favoured due to the reduced
  run time.

14
Students comments about audiences in the two
disciplines

• You have to tell people things they know dont
  confuse them.
• Have to explain more of the petroleum side to
  the mathematicians.
• Petroleum Engineers want more results and how
  its going to be useful for them want things to
  be easy.
• Mathematicians like to see the working through.

15
Students comments about purpose for writing
Purpose of conference abstracts? Persuade people
to attend your presentation rather than some
other presentation

• Mathematicians persuade by logical proof and show
  working out
• Petroleum Engineers persuade rhetorically using
  problem and solution

16
Bareback riding on two horses
In terms of discourse practices, being
cross-disciplinary is like riding two
horses.Sooner or later you have to choose to go
with one horse or the other
17
Investigation of discourse practices

• Examine a text for its consideration of audience
• Rewrite students text in the same style
• Mathematical techniques are essential for the
  efficient exploitation of hydrocarbon reserves
  even if a complete solution is beyond current
  capabilities.
• Where are the people in this statement?

18
Investigation of discourse practices

• To recover oil as economically as possible
• reservoir engineers
• must choose between different operating
  strategies.
• This
• requires a method of predicting oil flow in the
  reservoir and resulting oil production.
• Computer simulation
• is the principle tool for managing oil
  reservoirs and optimising the recovery.
• After suitable development
• a reservoir simulator
• can be run, many times.
• at relatively low cost,
• What are the links between these four sentences?

19
Investigation of discourse practices

• To recover oil as economically as possible
• reservoir engineers
• must choose between different operating
  strategies.
• This oil recovery? This choice?
• requires a method of predicting oil flow in the
  reservoir and resulting oil production.
• Computer simulation an operating strategy? a
  method? a tool?
• is the principle tool for managing oil
  reservoirs and optimising the recovery.
• After suitable development
• a reservoir simulator simulation for managing
  reservoirs?
• can be run, many times.
• at relatively low cost,
• The links between these four sentences are not
  made explicit.
• Presumably because the audience can make the
  connections.

20
Investigation of discourse practices

• Modelling
• Typical reservoirs are about 108 square feet in
  area and 100ft thick.
• A field simulation of 100,000 grid blocks (each
  100 x 100 x 10ft) would use several hours of Cray
  processing time to simulate a few years of field
  life and would only resolve horizontal features
  greater than 100ft in length. Figure 1 shows the
  variations that can occur over distances of tens
  of feet
• A common approach is to run a selection of
  models with finer grids and use the results to
  determine the rock and fluid properties
  appropriate to a coarse grid
• (possibly reducing the number of spatial
  dimensions)
• What is the rhetorical pattern of this part of
  the text?
• What effect do the words in bold have on the
  reader?

21
Rewrite in the style of the article

• Petroleum reservoir data is inherently
  uncertain, causing reservoir models to be
  uncertain. Part of the data is obtained from
  cores (10e-17 of reservoir volume) collected at a
  finite set of wells. Other data consist of
  time-averaged responses over larger scales, or
  data derived form incomplete knowledge of the
  subsurface geology. Thus the input data to a
  prediction model is inaccurate. Uncertainty
  quantification is essential for gaining accurate
  predictions within the petroleum industry.

Problem
Reasons for the problem
Result
Evaluation
22
Rewrite in the style of the article

• Standard techniques such as upscaling and fast
  low order numerical methods for solving flow
  equations across the reservoir, incur unavoidable
  model errors. Yet these techniques are favourable
  to running the full geological model with the
  most advanced techniques in order that fast
  results are produced. Simple error model
  construction gives accurate determination of
  parameter values, leading to a more accurate
  forecast.

Current solution
problem
Reasons to use this solution
Better solution
23
Comments from student about writing this way

• I was trying to keep sentences brief getting
  to the point in a shorter amount of time.
• It wasnt more difficult to write I felt it
  should have been more difficult.
• I put less explanation in the text because less
  is needed for the audience.

24
Final interview with the supervisor(student not
present)

• His article was actually targeted at physicists
  and mathematicians not petroleum engineers
• He acknowledged that it was not easy for its
  intended audience to understand (i.e. no-one is
  perfect)
• He liked the style of the students original
  email and didnt necessarily agree that the
  nominalised version was better.
• However,
• Nominalization is for abstracts
• Narrative is for oral presentations
• Situation-problem-solution-evaluation is key for
  writing in Petroleum Engineering.