113 
State Key Laboratory of Earth Surface Processes and Resource Ecology (Beijing Normal University), College of 
Resources Science & Technology, Beijing Normal University, Beijing 100875, China, 
* Corresponding author: - cyh@ires.cn 
KEY WORDS: Remote Sensing, Urban, Algorithm, Infrared, Estimation 
ABSTRACT: 
Surface emissivity is a measure of inherent efficiency of land surface. It is applied to convert heat energy into radiant energy. In this 
study, an unmixing pixel based algorithm was proposed to compute pixel effective emissivity for Advanced Spacebome Thermal 
Emission and Reflection Radiometer (ASTER) data within Beijing, China. In this paper, vegetation, together with water and 3 kinds 
of manmade materials surface distribution is estimated through a part constrained linear spectral model after PPI (Purity Pixel 
Indices) calculated. Sample emissivities presented in this research were extracted from Jet propulsion laboratory (JPL) spectral 
database. Root Mean Square Error (RSME) results of the whole study area is 0.1012, 0.0952, 0.2178, 0.0941and 0.0951 for ASTER 
5 TIR (8.125-11.65pm) bands, respectively. This study suggests that this model is useful for the estimation of land surface 
emissivity, and it can be used as a rather simple alternative to complex algorithms. 
INTRODUCTION 
Surface emissivity is an important parameter for studies of 
global energy balance. Thus estimation of emissivity is of 
particular interest. 
A number of methods have been explored in order to retrieve 
land surface emissivity, such as Temperature-Independent 
Spectral Indices (TISI) methods (Tian et al., 2006). Some TISI 
methods are described in Becker and Li (Li & Becker, 1993; 
Becker & Li, 1995). This kind of algorithm combines middle 
wave infrared data (MWIR: 3.4-5.2pm) with thermal infrared 
data (TIR: 8-14pm) to measure emissivity. Gilleapie et al. (1998) 
developed this method for ASTER data and carried out a high 
accuracy results. But the accuracy of this algorithm depends on 
some assumptions and ties to atmosphere correction. NDVI 
methods proposed by Caselles et al. (1989) and developed by 
Van et al. (1993) supply a new technique to calculate emissivity, 
and performance successfully in natural surface. But this 
method assumes the land surface is mainly made up of two 
types-vegetations and soil, which is disagreement with urban 
surface. Wan et al. (1998) utilized classification-based 
emissivity method and applied results to split window method, 
which performed well and the accuracy of land surface 
temperature is ±1K (1996). Snyder et al. (1990) also used this 
method to retrieval global emissivity without considering the 
complicated urban surface. 
Urban surface is more complex. And spectral heterogeneity is 
notable due to the spatial resolution, especially for TIR image. 
So it is common for the existence of mixing pixel. And effective 
emissivity should be given more attention. This study utilizes 
Linear Spectral Mixture Analysis (LSMA) and spectral database 
to extract subpixel information and achieve pixel effective 
emissivity. 
2. METHOD 
2.1 Study Area and Data Processing 
Study area is located in Beijing city (Fig.l). Beijing, a 
metropolitan city, is a centre of China. Urbanization is 
continuing to accelerate in this megacity. The pattern of it is 
complex and pixel mixing phenomenon is outstanding. The 
selected region contains most of representation urban land 
features, including vegetation, water, central business distinct, 
and so on. 
Location of Study Area 
Figure 1. Location of study area 
Advanced Spacebome Thermal Emission and Reflection 
Radiometer (ASTER) LIB Metadata and L2B emissivity 
product data acquired on 31 st August 2004 is used in this study. 
ASTER provides images in Visible/Near-Infrared (VNIR) with 
a spatial resolution of 15m, in Shortwave-Infrared (SWIR) with 
a spatial resolution of 30m, and in TIR with a resolution of 90m 
(Abrams, 2000; Yamaguchi et al., 1998). Hence, it is more 
suitable for urban studies at region scales than other satellite 
data, such as the Moderate Resolution Imaging 
Spectroradiometer (MODIS), which has a resolution of 250m, 
500m, 1000m for VNIR, SWIR and TIR, respectively (Lu et al., 
2006).