period of 1.05 year ( + .87, -.03) preceding August 1989. The sKewed 
uncertainty range is due to the skewed distribution of the dates of the LANDSAT 
images. 
Table I presents the results by individual states, and Figure I 
summarizes the time evolution of the deforested area for the region. 
The average rate of deforestation for the period 1978/89 of 
21,218sq.kmjM0% conflicts sharply with rates that have been used in several 
recent publications on deforestation and its impacts. The World Resources 
Institute (WRI) Report for 1990-1991 (WRI, 1990) uses 80,000sq.km as the 
annual rate for "the 1980’s." Norman Myers (1989) uses 50,OOOsq.km/year as 
the rate as of 1988. Both estimates are based on calculations of the burning 
area derived from the number of fires estimated with the thermal infrared band 3 
(3.5-3.9 micrometer) of the Advanced Very High Resolution Radiometer 
(AVHRR) on the US National Oceanographic and Atmospheric Administration 
(NOAA-9) meteorological satellite. The 80,OOOsq.km/year rate used by WRI was 
that calculated for the year 1987, which apparently had much more 
deforestation and burning than other years due to a combination of dry weather 
and a constitutional debate on confiscating forest areas from large ranchers for 
redistribution in a proposed agrarian reform program. 
The 1987 estimate (Setzer et al.. 1988), as well as the value for 1988 
used by Myers (Setzer and Pereira, nd) suffer from severe (and possibly 
insolvable) methodological problems for estimating areas (reviewed in 
Fearnside, 1990a, nd-a). The AVHRR area estimates were a by-product of 
studies designed to locate and count fires. The correction factors used to 
adjust for partially burning pixels (0.7) and for the proportion of the burning 
attributed to new forest clearing (0.4) could both be high by as much as a factor 
of two. A correction factor for partially burning pixels is difficult to derive 
because of large increases in the proportion of overestimation caused by small 
increases in fire temperature (a highly variable parameter) - theoretical 
calculations show that a fire of only 900sq.m is sufficient to trigger an entire pixel 
of 1,200,000sq.m (Robinson, nd), although practical experience suggests that 
narrow flame fronts up to 2 km in length can escape detection (A.W. Setzer, 
personal communication, 1990). The correction factor for non-forest is high 
because cerrado was apparently included in the numerator but not in the 
denominator when deriving the factor (see Fearnside, nd-a). 
542