MODIS (moderate resolution imaging spectroradiometer) airborne simulator (M
AS) data acquired during the SCAR-B (Smoke, Clouds, and Radiation-Brazil) e
xperiment provide a test bed to evaluate aerosol retrievals for MODIS measu
rements over land. The SCAR-B MAS data covered forest and cerrado regions w
ith varying smoke concentration as a result of fires in August-September, 1
995. Excellent agreement is obtained by comparing the retrieved aerosol opt
ical thickness from the MAS data with AERONET (Aerosol Robotic Network) gro
und-based Sun photometer observations. The evaluation of MODIS aerosol algo
rithm was performed for a range of smoke optical thickness from 0.1 to 2 at
0.66 mu m wavelength over cerrado and forest sites. All the retrieved valu
es of optical thickness are found within the anticipated retrieval errors o
f Delta tau = +/-0.05+/-0.2 tau. This confirms for smoke aerosol the validi
ty of the MODIS algorithms and the use of the new dynamic aerosol models in
this algorithm (the algorithm was already validated previously for urban/i
ndustrial aerosol). The retrieved optical thickness is found to be very sen
sitive to the assumed value of the single-scattering albedo but not sensiti
ve to the aerosol refractive index (real part). The single-scattering albed
o of 0.90 suggested in the aerosol model yields the best results. Test of t
he MODIS use of a grid box of 10 x 10 km in the aerosol algorithm shows tha
t the standard deviation of the retrieved aerosol should generally not be m
ore than 10% of the average value for a large range of optical thickness va
lues from 0.2 to 2.0, close and far from fires.