UCL

6. Project Write up

6.1. Assessment

  • 100% Assessed Practical (3500 words) - submission date: Friday 22th March 2019 (12 noon) via moodle.

NB:

  • Penalties for late submission and over length WILL be applied
  • Different arrangements for JYA/Socrates (make sure you inform the lecturers if this affects you)

6.2. Requirements

A reminder of the project requirements. This is repeated from text given at the appropriate sections in the notes. Note the marking guidelines to give you an idea of how much effort to put into each part.

6.2.1. Part 1: Introduction (20%)

[20% of marks in total for Part 1]

Provide an introduction to the the purpose of the study: [5%]

  • briefly covering why we might want to monitor and model urban change (and urban change in Shenzhen, in particular). Outline previous studies.
  • Outline what is to be done in the rest of the study. Cite literature as appropriate.

For one year (your choice between 1986 and last year , but state the year used):

  • describe a method of manual data download (e.g. using the Google Earth Engine App), with illustrations as appropriate. Note any processing that has been already done to the dataset, and which wavebands are made available to you, with equations as appropriate. [5%]
  • Explore the dataset (histograms, scatterplots) to assess its information content (use figures); [10%]

6.2.2. Part 2: Data processing (60%)

[60% of marks in total for Part 2]

You must include an accuracy assessment for each manual classification you perform (so, for just one year).

The section or sub-section should contain ‘full-sized’ (on the page) pictures of the manual classification results, along with an appropriate table to interpret the colours.

6.2.2.1. Part 2a: Data Exploration and Classification Theory

[30% of marks in Part 2a]

For the data selected for the single year:

  • Choose one supervised and one unsupervised classification approach (we recommend Maximum Likelihood and ISOData
  • Theory: describe how the approaches work (noting similarities and differences) and relate this to the information content of your data. Cite literature as appropriate. [10%]
  • Perform a supervised classification and an unsupervised classification using envi, relating the training information (e.g. class seperability) to the material presented above;
  • Present the results of the classification and assess the accuracy of these classifications; [15%]
  • Discuss the issues raised and how this might translate to unsupervised classification of the whole time series. [5%]

6.2.2.2. Part 2a: Time Series Classification

[30% of marks in Part 2b]

You should report the number or proportion of pixels of each class, plotted as a function of year and present any other results you feel appropriate.

  • Download Landsat annual datasets for 1986 to present (or some suitable subset of at least 18 years): you can copy the data from the GEOG0027 archive
  • Perform an unsupervised classifications (clustering) of the time series of Landsat data, using an envi program that you will be provided with (classy.pro);
  • apply suitable class labels, and modify the number of classes as appropriate;
  • Calculate the area of urban land use for Shenzhen for each year
  • Estimate the area of agricultural land use for Shenzhen for each year (if possible, not critical)
  • Try to assign a value of uncertainty to the derived data (from earlier accuracy assessment)
  • Write up this section of work, describing:
    • the tasks undertaken (materials and method) [5%]
    • the experiments conducted (e.g. with varying class number/waveband) [10%]
    • the results and uncertainty [10%]
    • discussion of the results (in context of text above) [5%]

6.2.3. Part 3: Modelling (15%)

[15% of marks in total for Part 3]

Following the general approach of Seto & Kaufmann (2003), we will build a multi-linear model to attempt to describe the urban land use change per year (the ‘y’ variable) as a function of a number of key socioeconomic factors (e.g. capital investment, land productivity, population, wage rates, etc) (the ‘\(x\)’ variables).

Equation 1:

\[y = p_0 + p_1 x_1 + p_2 x_2 + + p_3 x_3 + p_4 x_4 + p_5 x_5\]

The model relates socio-economic variables (constant, plus x1, x2, x3, x4, x5), weighted by model parameters (p0, p1, p2, p3, p4, p5) to predict the rate if change of urban area per year (du_dy).

We have taken a set of observations of du_dy, derived from Landsat land cover classifications for the years 1986 to present (or a subset). Along with estimates of the x variables from the Guangdong yearbook, we have then seen how to produce an esrimate of the model parameters (the p terms).

This forms the basis of the modelling section of this coursework: As noted above, you need to perform a model calibration, plot results, and describe and interpret summary statistics. Your interpretation of the statistics is vital here as it will show your understanding of the terms printed. Your plots should be neatly done, with full axis labelling, titles etc, noting any units or scaling factors used.

  • Introduce the data and modelling task, referring to the contextual information in the introduction section (part 1), and the urban/agricultural area information from part 2. Introduce the ideas of calibration and validation to outline the approach taken here. [5%]

Using the data derived above, calibrate a model that describes urban land use change as a function of a set of socioeconomic factors, following the approach of Seto et al. (2002, 2003). You are provided with R code and appropriate datasets to achieve this.

Analyse the statistics of the model and experiment to try to find an improved model with fewer parameters.

  • Write up the results of the modelling and your interpretation of the statistics [10%]

We have given you a set of questions to help guide your statistical interpretation.

You are then required to see if you can come up with a model with fewer parameters. The original model has 6 parameters, but it could well be the case that we can develop a more robust model with fewer parameters. One way we can judge ‘better’ here is to take a measure of goodness of fit that accounts for the model degrees of freedom: ‘better’ then is a balance of these things.

You are free to perform additional experiment, with the expectation of higher marks, provided (i) you have done the basic requirements well enough, and (ii) you show clarity of thoiught and understanding of what you are doing in your experiments.

6.2.4. Part 4: Discussion and Conclusions (5%)

[5% of marks in total for Part 4]

  • Discuss your work and your findings, and draw conclusions. Try to relate these to the motivation for the project that you outline in the introduction. [5%]

6.3. Write up

Your write-up should include figures and diagrams relevant to describing the approach you have taken and sufficient to demonstrate your results.

The write-up should be 3500 words or less. It should include a declaration of word length (as usual), although we consider that the ‘word limit’ does not include:

  • computer codes
  • data tables
  • any figure or table text

The write-up should be in the style of a scientific experimentation report.

It should be possible to obtain some good results within this experiment, but that is not actually the critical factor in our assessment. We are more interested in you demonstrating that you have carried out the work sensibly and fully, and conducted a good set of experimentation within the tools and time available, and that you have written it up clearly and concisely, with relevant and sufficient reference to the literature. We will be looking for a clear discussion of the information content of the data and its relationship to the classifications, of the model and the results, as well as a set of relevant conclusions. As always, the highest marks marks can be gained for clarity, originality, application and demonstration of depth of understanding.

You may wish to deviate from this structure in your write-up, but you are strongly advised to seek advice from the course tutors before doing so.

Course tutors will normally be available during practical classes, as well as during office hours.

All graphs, figures, etc. must be correctly labelled, and your citations must be done correctly.