n of 
| for 
reas 
nage 
Il as 
m & 
IS in 
AS 
r to 
it to 
stect 
cing 
at 
HOLS 
s of 
1sus 
— 
1SUS 
the 
nge 
scur 
ban 
fied 
ator 
| of 
sors 
use 
hen 
ient 
and 
The 
ing 
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B1. Istanbul 2004 
  
I. Detect urban growth 
Identify and classify urban growth 
t2 
3. Quantify urban growth 
3. METHODOLOGY 
The approach that has been used in this methodology for the 
image processing, is straightforward without complexity to 
prepare the images for detecting urban growth and subsequent 
mapping. The images were ortho-rectified using digital ortho- 
photos, and then the merged layers were created, followed by 
the enhancement of imagery. 
Once the images were prepared, the 2001 GIS layer of 
enumerator area of Statistics SA were overlayed on the 
imagery. All urban change were mapped on screen, and 
classified based on visual interpretation and rather than 
complex classification procedures (where pixel reflection 
values are grouped together), which requires visual checking 
and assessment anyway for census applications. Finally the 
quantification for each class was done and compared to actual 
structure counts. 
3.1 Data Sources 
Spot Image supplied the following Spot 5 images in level 1A 
for use during this study: 
e Spot 5: Colour 10m 
e Spot 5: Black & White 5m 
e Spot 5: Supermode 2.5m 
These images were ortho-rectified using the following two data 
sets as ground control points (GCP)'s and elevation layer 
e GCP's: Aerial Photography: Black & White Im 
e Elevation Source: Digital Elevation Model € 20 
metre resolution 
During the study two additional data sets were created by 
merging the following images: 
e Spot 5: Colour 10m with Panchromatic Sm: rgb=321 
e Spot 5: Colour 10m with Supermode 2.5m: rgb=321 
This resulted in a total of five image data sets that were used 
and evaluated in this study. 
3.2 Image Preparation 
All image processing steps listed below was performed in Erdas 
Imagine and Orthobase modules. 
e Import imagery and check imagery for radiometric 
quality 
= Each satellite image was imported and checked 
for radiometric quality which is essential for 
visual interpretation 
*  Ortho-rectify satellite images using the following two 
inputs: 
» collecting ground control points from ortho- 
photos 
= 20m digital elevation model (DEM) 
* Accuracy assessment of ortho-rectified imagery 
" imagery has been checked by overlaying and 
comparing to street maps and enumerator areas 
e Merge of Panchromatic and Multispectral imagery 
= process of merging the 2,5m Supermode and 5m 
Panchromatic images enhance the image contrast 
and easier identification of structures. 
* Enhancement filters 
= edge enhancement filters was passed over all the 
imagery to highlight boundaries between urban 
and rural areas. 
3.3 Processing and Mapping 
Digital mapping technologies was used to create a GIS layer 
that show the change in urban area between the 2001 EA layer 
on the August 2002 Spot 5 imagery. The change is however 
actually from a longer period because the 2001 EA's are 
prepared from imagery that is captured approximately 24 -18 
months before the census. This is done to have enough time to 
prepare for the actual census surveys. 
All mapping is based on visual interpretation and heads up 
digitizing of urban changes. Visual interpretation uses the 
human eye and brain to consider context, shape, proximity and 
texture to identify features on satellite imagery. The images 
were visually scanned using a grid pattern to identify any urban 
features on the imagery that is not covered by urban EA 
polygons. These areas were mapped out, based on the 
parameters below, as polygons to designate change for the 
different periods, using the ArcGIS software modules. 
e Urban areas not covered by urban EA's (smaller than 
100ha) 
e Urban areas in EA's with an attribute of less 20 
household structures for previous census 
* Backdrop imagery showing indications of new 
developments (ie street patterns but no houses yet) 
After the change detection polygons were mapped from the 
Supermode colour enhanced image, the polygons were overlaid 
on all the individual image sets. All the different classes were 
tested on the complete set of images, to determine which of the 
imagery contains enough detail and resolution to allow 
classification into different classes. 
This classification can assist with the update and demarcation of 
the EA's. Each of the mapped polygons showing change was 
classified into the following classes listed below: 
e  townhouses/cluster housing 
e security estate 
e low cost housing 
e informal settlement 
e residential (normal suburban street layout) 
The criteria were that if any of the classes could not be 
classified clearly from any of the five image sets it will be 
considered not classified. 
553