Two methods are described to determine burned area from Advanced Very
High Resolution Radiometer (AVHRR) data. The first method, or the ''li
near method,'' employs Channel 2 reflectance, R(2), and is based on th
e nearly linear relationship between the fraction of pixel burned, P,
and R(2). The second method, or the ''nonlinear method,'' employs the
Normalized Difference Vegetation Index (NDVI) derived from Channels 1
and 2 reflectances, and is based on the nonlinear relationship P=f(NDV
I), a polynomial of order 2 in NDVI. The coefficients of the polynomia
l are parameterized as a function of the NDVI of the background before
the fire event. Radiative transfer simulations indicate that the line
ar method, unlike the nonlinear method, must be applied to top-of-atmo
sphere reflectances that have been corrected for atmospheric influence
. Sensitivity studies suggest that the methods are subject to some lim
itations. To avoid discontinuity problems, the original background (ju
st before the fire) must be characterized by a Channel 2 reflectance a
bove 0.07 and by a positive NDVI. To separate the useful signal from a
tmospheric effects, the fire scar must occupy at least 20% and 12% of
the pixel area in the case of savanna and green vegetation (e.g., fore
st), respectively When applied to uniform pixels, the mean relative er
ror on the fraction of area burned is about 20% for the linear method
and 10% for the nonlinear method. The linear method gives better resul
ts for nonuniform pixels, but neither method can be used when the pixe
l contains low reflectance backgrounds (e.g., water).