The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B6b. Beijing 2008 
Where e m 6 =manmade material emissivity in band b 
/ v =fraction of vegetation in a pixel 
f m =fraction of vegetation manmade material in a pixel 
Many researches prove spectral curve of different vegetation 
and water is almost same in TIR region (Mao et al., 2007; Nerry 
et al., 1990; Rubio et al., 1997; Sobrino et al., 2001; Sobrino et 
al., 2004; Stathopoulou & Cartalis, 2007). In this research, 
vegetation emissivity e v and water emissivity e w is valued as 
0.985 and 0.990, respectively. Manmade land surface emissivity 
is obtained by using spectral database provided by Jet 
propulsion laboratory (http://speclib.jpl.nasa.gov). After 
analyzing 46 kinds (including 6 kind concrete materials, 17 kind 
General Construction Materials, 5 kind Road Asphalts and Tar 
and 18 kind Roofing Materials) manmade material emissivity 
(Fig. 3), we utilize mean emissivity of concrete, general 
construction materials and road asphalts and tar materials in 
Eq.3. Because central business distinct building and 
surrounding of it in study area are covered by concrete. And 
residential is almost constructed by brick. Road is covered by 
asphalts and tar. 
Concete 
Figure 3. Mean emissivity of manmade sample 
2.4 Results and Discussion 
Band NO. 
Spectral Range (pm) 
RMSE 
Band10 
8.125-8.475 
0.101 
Bandi 1 
8.475-8.825 
0.095 
Band12 
8.925-9.275 
0.218 
Bandi3 
10.25-10.95 
0.094 
Band14 
10.95-11.65 
0.095 
Table 1. RMSE of the whole study area 
Result shown in Tab.l indicates our model has a larger RMSE 
in band 12 and band 10. This maybe because emissivity in 
8.925pm~9.275pm and 8.125pm ~8.475pm is much more 
variable, especial for manmade materials. Sample emissivities 
in these spectral ranges prove it as Fig. 3 seen: 
Bandi0 Bandii 
Band 14 
In order to analyze the emissivity obtained from unmix method, 
results were resampled to 90m resolution (Fig. 4), and then 
compared with ASTER emissivity product. Evaluation of this 
algorithm is done by using RSME (Root Mean Square Error) 
expressed as equation 4: 
RMSEb = 
M N 
m=1 «=1 
(4) 
Where RMSE b = Root Mean Square Error of band h 
e m „=emissivity of pixel location at (m,n) baesd on 
unmixing algorithm 
£ m =ASTER product emissivity of pixel location at 
Figure 4. Emissivity results of the model in different band 
Assumption of this model can also contribute to RMSE. First, 
we describe the most suitable condition in a pixel (without 
scattering between distinct land features and temperature is the 
same of them). The fact is scattering exits between features and 
temperature of them is of difference due to their different 
attributes. Second, just three kinds of manmade materials’ 
(concrete, general construction materials and road asphalts and 
tar) mean emissivity is utilized in the model. But emissivity of 
these materials has a range. Third, soil is not in consideration 
for this algorithm. And it maybe shown in some place, such as 
constructing area, park, and so on. Last, result of this method 
depends on LSMA. LSMA is computed by part constrained 
least-squares approach, leading to fraction of some feature 
larger than 1 and some lower than 0. When this problem 
countered, we will make the fraction equal to 1 or to 0, which 
also can make an error.