i Detection of Smoke and Dust Aerosols Using Multi-sensor Satellite ...

UVAI image; (b) MODIS true color image; (c) Smoke image.
According to the procedure of aerosol selection presented in OMAERUN README
file (http://daac.gsfc.nasa.gov/documents/OMAERUV_README_File_v3.doc), the
aerosol associated with biomass burning usually has the UVAI values larger than 0.7.
Therefore, the OMI UVAI images with the UVAI values larger than 0.7, 1.0, and 1.2 are
displayed respectively in Fig. 6.2, for multiple comparisons with smoke image derived
with multi-spectral algorithm using MODIS measurements. Some obviously
misclassifications (bad or low quality data) in OMI UVAI product, are filtered out to
improve the accuracy of comparisons.
The distributions of identified, unidentified, and misidentified pixels in three
comparisons are given in Fig. 6.2. In the first comparison (UVAI > 0.7; Figs. 6.2a and
6.2b), there are 513,582 pixels are identified correctly (labeled as smoke in both images),
listed in Table 6.2. About 365,853 pixels are unidentified (detected by OMI only) and
1,563 pixels are misidentified (detected by MODIS only) in MODIS smoke image. The
percentage of identified smoke pixels increases to 73.78% with acceptable 4.34%
misidentification in second comparison (UVAI > 1.0; Figs. 6.2c and 6.2d). In the third
comparison (UVAI > 1.2; Figs. 6.2e and 6.2f), the unidentified smoke pixels decreases
further to 11.18% but around 19.05% pixels are labeled as smoke in the smoke image
which is labeled as nonsmoker pixel in OMI UVAI image.
Overall, the core part of smoke is explicitly detected by MODIS smoke image in all
three comparisons. The major difference exists at the edge of plume, which may attribute
to the different spatial resolutions between two sensors. Because of different spatial
81