The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part Bl. Beijing 2008 
97 
fitting results were y=0.804612x+0.050988(R 2 =0.77), 
y=0.862697x+0.034432(R 2 =0.75) and 
y=0.851294x+0.012003(R"=0.71), and the mean relative errors 
were decreased to 16.4%, 18.2% and 19.3% respectively for 
550, 658 and 860nm channels. 
Indexes 
550nm 
658nm 
860nm 
Gosan SNU 
0.159 
0.134 
0.108 
RMSE 
Shirahama 
0.028 
0.037 
0.045 
All Sites 
0.151 
0.128 
0.103 
Gosan SNU 
89.4% 
90.2% 
78.0% 
Mean ( 
T M T A> 
) 
Shirahama 
8.0% 
13.2% 
20.9% 
ta 
All Sites 
81.8% 
83.0% 
72.6% 
Percentage of 
Gosan SNU 
58.6% 
41.4% 
48.3% 
retrievals within 
Shirahama 
100% 
100% 
33.3% 
expected 
All Sites 
62.5% 
46.9% 
46.9% 
uncertainties 
Table 1. AOT match-up error statistics for 25km radius and 
±0.5h 
Using 45km radius and ±1.0h as the sampling window, there 
were 616 match-up. Excluding those with ground-based 
measurements less than 4 and satellite pixels less than 9, 120 
matched data were left, in which ground-based data were from 
sites of “Gosan SNU” (74 match-ups), “Hong Kong PolyU” 
(12 match-ups), “NCU Taiwan” (13 match-ups), “Shirahama” 
(9 match-ups) and “Taichung” (8 match-ups) with the averaged | 
AOT at 550nm of 0.35. Figure 3 displays the scatter diagrams 
of match-ups. Table 2 lists the corresponding error statistics. 
From the AOT linear analysis results, the validation was not 
satisfying with large data dispersion and R 2 all less than 0.3. 
When data from “Taichuang” were discarded, results were 
better with R 2 all exceeding 0.6. 
The RMSE was 0.23, 0.20 and 0.17 respectively for 550, 658 
and 860nm channels. The result of “Hong Kong PolyU” was 
best with RMSE of 0.11, 0.09 and 0.07 respectively, and the 
result of “NCU Taiwan” was the worst. From the mean value 
distribution of match-ups shown in figure 4, except for “Gosan 
SNU”, the MODIS retrievals were lower than AERONET 
measurements, especially at ”NCU Taiwan”. It indicates to 
some extent that the retrieval is not largely influenced by cloud, 
and calibration may have the main impact. The large errors and 
AERONET AOT at “NCU Taiwan” indicate that this site may 
reflect the aerosol condition over land more than over ocean. 
The mean relative errors of 3 channels all exceeded 55%. The 
result of “Hong Kong PolyU” was best with errors of 23.8%, 
24.0% and 25.7% respectively for 550, 658 and 860nm 
channels. The result of “Gosan SNU” was the worst with lowest 
ground-based measurements. 
The percentages of retrievals within the expected uncertainties 
were all less than 50%. If the uncertainties were 
At=±0.05±0.05t for 658 and 860nm channels, the percentages 
were only increased to 47.5% and 50.8% separately. The result 
of “NCU Taiwan” was the best with 61.5% for 3 channels, and 
the result of “Taichung” was the worst. When data from 
“Taichuang” were discarded, the percentages of 550, 658 and 
860nm were 49.1%, 37.0% and 40.7%. 
Chen & Yang (2005) validated the NASA MODIS (Terra/Aqua) 
aerosol product from 2000 to 2004 using ground-based data 
from AERONET (2 sites) and SKYNET (4 sites) around 
Taiwan straits and adjacent area. With 34 match-ups mainly 
from “NCU Taiwan” and “Chen Kung Univ” of AERONET, 
the percentages of retrievals within the expected uncertainties 
were 58.8% and 64.7% for 550 and 660nm separately*, with 
the ground-based averaged AOT at 550nm of 0.6. It was close 
to our result of 61.5% at “NCU Taiwan” (13 match-ups) with 
AERONET averaged 550nm AOT of 0.50. Because the RMSE 
and averaged relative error at “NCU Taiwan” were relatively 
high (0.44 and 35.6% for 550nm) and the AOT linear 
relationship between AERONET and MODIS was not good, it 
was considered that the validation at this site was still not 
satisfying. 
AERONET t(870) 
Figure 3. AOT match-up scatter diagrams at 550nm, 658nm and 
860nm for 45km radius and ±1,0h 
Indexes 
550nm 
658nm 
860nm 
Gosan SNU 
0.180 
0.152 
0.113 
Hong Kong PolyU 
0.109 
0.086 
0.066 
RMSE 
NCU Taiwan 
0.437 
0.414 
0.401 
Shirahama 
0.201 
0.163 
0.135 
Taichung 
0.327 
0.266 
0.181 
All Sites 
0.231 
0.202 
0.172 
Gosan SNU 
81.3% 
82.0% 
68.8% 
Hong Kong 
23.8% 
24.0% 
25.7% 
Mean ( 
T M -T A ) 
: Poiyu 
NCU Taiwan 
T A 
35.6% 
36.3% 
36.8% 
Shirahama 
53.0% 
59.6% 
66.7% 
Taichung 
56.4% 
60.4% 
61.3% 
All Sites 
64.8% 
66.1% 
58.9% 
Gosan SNU 
44.6% 
31.1% 
37.8% 
Percentage 
Hong Kong 
58.3% 
33.3% 
33.3% 
of retrievals 
PolyU 
within 
NCU Taiwan 
61.5% 
61.5% 
61.5% 
expected 
Shirahama 
55.6% 
55.6% 
44.4% 
uncertainties Taichung 
0.0% 
0.0% 
0.0% 
All Sites 
45.0% 
34.2% 
37.5% 
Table 2. AOT match-up error statistics for 45km radius and 
±1.0h 
* The sampling window was 9x9 pixels and ±1.0h, same data 
elimination method as in this paper.