cture in 
aam in 
jectives 
on and 
studies, 
speckle 
contain 
ing the 
as also 
er flow 
tion for 
on with 
, is that 
at all. 
lems is 
ing the 
es until 
satellite 
;ause it 
n. This 
. Radar 
ty. The 
of less 
peckle- 
oped to 
speckle 
duction 
and the 
(2) the 
n water 
cussed. 
section 
Rob Dekker 
  
2 URBAN MONITORING WITH SYNTHETIC APERTURE RADAR AND ERS DATA SET 
The status of urban monitoring with SAR, at the beginning of this study, was clearly reviewed by Henderson and Xia 
(1997). They stated that, beside the view angle and the nature of the illuminated objects, the visibility and detectability 
of human settlements are also dependent on: 
(1) the wavelength and polarisation of the sensor 
(2) the surrounding land cover, which can facilitate or confuse the visibility of settlements 
(3) the covering area, smaller settlements and medium cities are easier to detect and delimit than large ones 
(4) the experience and knowledge of the interpreter (note that this also applies to detection algorithms) 
Their conclusions with respect to the sensor were that it seems preferable to use an incidence angle greater than 35°, a 
shorter wavelength (i.e. X, C and L band) and cross-polarised imagery (i.e. HV or VH). 
Chosen was for ERS SAR data because it is relatively cheap, easy to obtain and operating on a short wavelength, see 
table 1. Unfortunately it is a VV polarised system with a rather steep incidence angle, but in this paper will be shown 
that this is also a very useful combination with respect to urban monitoring in developing countries. It is even more 
preferable to use fully polarimetric imagery (i.e. HH, HV and VV, Hussin 1995, Xia and Henderson 1997, Forster et al. 
1997), but this type of data is not available yet from satellites. It will become available in the future (e.g. ENVISAT). 
ERS SAR imagery is available in all sorts of formats, from raw to fully terrain geocoded imagery. For interferometry 
only one product is suitable, and that is Single Look Complex (SLC) imagery. It is the only product that contains the 
phase information of the radar signal. Besides it has the advantage that it can be processed into imagery with a higher 
number of looks than the other products with a resolution better than 30 m. This was actually done by averaging 6 
adjacent pixels, resulting in Multi Look (ML) images, for the purposes of change detection and classification. A 
disadvantage of SLC imagery is that geocoding still has to be performed. In terrain geocoded imagery (GTC), also 
suitable for change detection and classification, this has already been done. The DEM could have been used to geocode 
the image, but unfortunately it turned out to be not accurate enough, see section 5. Table 1 shows some general 
characteristics of the ERS SAR data. The data set and application can be found in table 2. 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
ERS SAR 
Average altitude 785 km 
Average ground speed 7100 m/s 
Wavelength 5.7 cm (C-band) 
Polarisation VV 
Swath width 100 km 
Range (slant, mid-swath) 840 km 
Incidence angle (mid-swath) 21.5? 
Ground range resolution (SLC) | 30m 
Azimuth resolution (SLC-ML) 5 m-24 m 
Number of looks (SLC-ML) 1-6 
Table 1. Major parameters of the ERS SAR Single Look Complex (SLC) imagery and self generated Multi Look (ML) 
imagery. 
Platform | Orbit | Frame Date Pass DEM | CHD 
  
ERS-1 32516 3738 03/10/1997 | descending ® 
ERS-2 12843 3738 04/10/1997 | descending e 
ERS-2 04061 7049 29/01/1996 ascending e 
ERS-2 19091 7049 14/12/1998 ascending e 
Table 2. The ERS SAR data set and its application. 
  
  
  
  
  
  
  
  
  
  
  
  
3 CHANGE DETECTION 
To detect changes in SAR imagery different techniques are available. For the application of detecting new human 
settlement in developing countries two were studied. The first is a method that makes use of an edge (and point-target) 
preserving adaptive filter (Dekker 1998). This method already proved that it works properly on SAR difference imagery 
in The Netherlands. The second method that was studied, is called Order-Statistics Constant-False-Alarm-Rate (OS- 
CFAR) detection (Novak and Hesse 1991). It performs detection against the local background, which can mean that 
small changes in terms of intensity can possibly not be detected in predominantly-new areas. The method was studied 
because it performs well in military radar target detection. Both methods were applied on the difference of the 
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part Bl. Amsterdam 2000. 63