Page12

To obtain a quantitative evaluation of urban growth, we are going to look at the difference between the land cover maps of 1992 and 2009.
 
a) Change detection by Supervised Classification

In order to create a land cover map for 1992, you can follow the steps given in Exercise 2, using the image of 1992 (1992_cordoba). You must however create new training fields, since changes may have occurred between 1992 and 2009, making the training fields of 2009 not valid for the 1992 image.

Open 1992_cordoba_B1 to 1992_cordoba_B7 and create a false colour composite with band 4 (red), band 5 (green) and band 3 (blue), just like you did in the previous exercise.

Go to Supervised classification and define the three classes (Urban, Vegetation, Bare soil) remembering the typical colours of these land cover classes from the previous exercises. Select new training fields in each land cover class.

Define several fields for every land cover class. Before you decide on the training samples, study the results of 2009. Look at how vegetated area are presented (colours), how bare soil looks like in general, and how you can recognise urban areas. After that you can proceed with a maximum likelihood classification.  
 

The method described above mostly depends on the quality of the training fields and the automatic classifier used. You may be aware of many errors where the class 'Urban' extends too much in the countryside and is wrongly classified as Bare soil.
 
 

Page12