IM 
the 
ing. 
age 
fant 
[WO 
| be 
nge 
r to 
ver 
th 
are 
ned 
ire. 
ent 
nd, 
of 
of 
AW. 
the 
hin 
ver 
est 
ain 
of 
ire 
ire. 
Barbosa, Paulo 
  
2 STUDY AREA 
The study area is located on central Portugal and comprehends an area of 2240km° 
(Figure 1). This area has been severely affected by vegetation fires in the last 
  
f nit decade. Maritime pine (Pinus pinaster Ait.) and Tasmanian blue-gum (Eucalyptus 
t globulus Lab.) are the most representative forest species, along with large patches 
yr of Mediterranean shrubs. This site was chosen based on the following conditions: 
(1) to be a forest area largely affected by fires and (2) to be an area where specific 
land cover changes have occurred in the past - such as pine stands (before fire) to 
eucalyptus stands (after fire). 
  
analyzed was 1990-1998, with one image per year obtained in spring or early 
summer. The pre-processing of this data set included geometric correction and 
radiometric correction. However, since we were dealing with a multi-temporal 
data set, some particular aspects had to be considered. 
/ 
id 3 DATA SET 
2 ; 
Fr. A multi-temporal Landsat 5-TM data set was assembled for this study. The period 
A 
  
  
All images were geometrically corrected not only to eliminate geometric 
Figurel- Study area in distortions present in the images but also to register the satellite images to ground 
Portugal. data. Ground Control Points (GCP) were extracted from topographic maps (1:25 
000), using geographic features such as road intersections, land/water interfaces or 
field patterns. Polynomials of 2™ order were used in each registration. The nearest 
neighbor re-sampling method was used, because it is the only one that keeps the original brightness values. All images 
were re-sampled to a 25 m pixel grid. Considering that the study was carried out in more than one image acquired at 
different periods we had to guarantee that the images were not only registered to the reference map but also co- 
registered. For this reason, an image was referenced to a map (target image) and all the others were then registered to 
this one. An assessment of geometric correction results is presented in Table 1. 
  
  
Acquisition date # GCP used RMSE 
4-5-90 50 0.66 pixel 
26-7-91 43 0.63 pixel 
12-7-92 42 0.59 pixel 
13-6 -93 64 0.68 pixel 
29 -4-94 61 0.70 pixel 
5-7-95 69 0.70 pixel 
5-6-96 58 0.75 pixel 
24 -6 - 97 120 0.92 pixel 
27-7-98 40 0.58 pixel 
  
Table 1 — Imagery acquisition dates, number of ground control points (GCP) used for each image and Residual mean 
squared error (RMSE) obtained in the geometric correction procedure of the Landsat 5-TM data set (path: 204; row: 
32). 
The radiometric correction involved the conversion of digital numbers (DN) of each image to Top-of-Atmosphere 
(ToA) reflectance units. In this procedure DN were converted to radiance units using the sensor calibration coefficients 
retrieved from the image headers and corrected to account for the sensor drift over time using the approach proposed by 
the Canada Centre for Remote Sensing (1999). To perform the conversion of radiance units to ToA reflectance, the 
following model was used for each wave band (A): 
„ln 
Pros à = 
s E,, : cos0 
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000. 127