Barbosa, Paulo 
  
P Ton is the top-of-atmosphere reflectance, 
L ; (mW.em^.sr '. um) is the spectral radiance, 
d (AU) is the Sun to Earth distance in astronomic units, 
Eo. (mW.cm^.sr '. um) is the exo-atmospheric solar irradiance, 
cos 0 is the cosine of the solar zenith angle. 
The Sun to Earth distance was retrieved from tabular data (Igbal, 1983) using the image acquisition date and time. The 
solar exo-atmospheric irradiance was retrieved from Slater (1980). The use of ToA reflectance units instead of the 
original DN units allows a better comparison of images through time, since at least a part of the scene specific 
characteristics like differences in sun illumination geometry and sensor characteristics can be accounted for in the 
conversion process. 
* Three sets of ancillary data were used: burned area maps, pre-fire land cover maps (1990), and ortho-rectified aerial 
photography from 1995. The Portuguese Forest Service (Direcçäo-Geral das Florestas) funds the production of burned 
area maps based on Landsat 5-TM data. This is available for the entire country and contains all the burned areas larger 
than 5 ha since 1995, and larger than 15 ha before 1995. Maps for the period from 1991 to 1995 were used in this study. 
Approximately every 10-years, land cover maps are produced for the entire country by visual interpretation of aerial 
photographs. This information is available at 1:25 000 scale with a minimum mapping unit of 1 ha. For this study, land 
cover maps from 1990 were used to characterize the land cover before the fires. For the post-fire land was cover, it 
performed a visual interpretation of the ortho-rectified aerial photographs from 1995 in order to obtain test areas. 
4 METHODS 
The methodology used for the post-fire change detection was based on vegetation index differencing. The vegetation 
index used was the Atmosphere Resistant Vegetation Index (ARVI). In our study area, this index showed high values 
for vegetation cover, low values for forest burned area, and even lower values for terrain mobilisation (Santos et al, 
1999). 
ARVI was computed using Landsat TM near-infrared (NIR), red (R), and blue (B) bands in ToA reflectance units 
(Kaufman and Tanré, 1992): 
_ NIR-RB 
NIR + RB 
RB=R-y(B-R) 
ARVI 
In this study, y - an atmospheric self-correcting factor which depends on aerosol types - was set to 1 since it permits a 
better adjustment for most remote sensing applications, when the atmospheric data are unknown (Kaufman and Tanré, 
1992). 
The ARVI difference images were computed using the year of the fire as compared with the two years following the 
fire. For 1991 the difference images were computed for 1991-1992 and 1991-1993, for 1992 they were computed for 
1992-1993 and 1992-1994, and for 1993 they were computed for 1993-1994 and 1993-1995. This allowed the 
identification of no change areas, as well as 2 types of change areas: 1) Re-growth of vegetation and, 2) Terrain 
mobilisation/reforestation. 
Regrowth of vegetation after fire should appear negative in the difference image, because ARVI for vegetation is higher 
than for burned areas. In the case of mobilisation, the difference image should be positive, because the ARVI value is 
higher for burned area than for bare soil. To identify the terrain mobilisation areas thresholds were selected based on the 
mean and standard deviation of the image difference histogram. This is a common procedure in land change use 
detection (Mas, 1999). The critical element is to decide the thresholds between change and no-change. A higher number 
of standard deviations will originate higher omission errors but lower commission errors. 
  
128 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000.