International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B-YF. Istanbul 2004 
  
Field LAI measurements 
Although the accurate measurement of LAT is crucial to relate 
the satellite observed surface reflectance, it is often difficult to 
obtain reliable ground-truth of LAI. Field measurement of LAI 
can be divided into two major approaches of direct and indirect 
methods (Chen et al., 1997; Gower et al., 1999). The direct 
method includes destructive harvest, litterfall collection, and 
application of allometric equations. In recently years, several 
optical devices become available for the indirect measurement 
of LAI. The basic concept of optical LAI measurement is to 
invert a radiation model that describes the amount of light 
penetrating tree canopy as a function of leaf area and 
distribution. In this study, we used the Li-Cor LAI 2000 plant 
canopy analyzer, which is a commercial instrument to 
indirectly measure LAI This device estimates LAI by 
measuring light transmittance under the forest canopy like as 
other optical devices. 
The 30 ground plots were selected to include diverse forest 
types of both coniferous and natural deciduous species. Field 
measurements were conducted from September 15 to 17, 2003. 
Each plot has an area of 20%20m? and includes five subplots for 
LAI measurements within it. LAI was measured three times at 
each subplot and total of 15 measurements were averaged to 
obtain the LAI value for each plot. The exact locations of the 
30 ground plots were obtained using a differential global 
positioning system (GPS). Figure 1 shows the distribution of 
the 30 plots within the boundary of the Kyongan River basin. 
  
  
Figure 1. Distribution of 30 ground plots for the field LAI 
measurement over Kyongan River basin 
Generation of reference LAI and land cover map 
In an attempt to validate MODIS LAI product, we have 
constructed a reference map of LAI surface by relating the 
field-measured LAI to Landsat ETM+ spectral reflectance and 
spectral vegetation indices. To minimize any discrepancies due 
28 
to the phenological variation of leaf development, the ETM+ 
data was captured on September 10, 2002 for this study. 
Although the ETM+ data acquisition was one year earlier than 
the field measurement, we believe that it did not cause any 
problem since the leaf development and the actual canopy 
condition between 2002 and 2003 were not much different. 
Figure 2 showed the overall procedure of ETM+ data to 
construct reference LAI surfaces using the field-measured LAI. 
The ETM+ data were initially geo-referenced to plane 
rectangular coordinate system by using a set of ground control 
points (GCP) obtained from the 1:5,000 scale topographic maps. 
To extract the surface reflectance of pixels, rather than raw 
digital number (DN) value, ETM+ data were further processed 
to reduce topographic and atmospheric effects. DN value of the 
original image was converted to radiance by applying sensor’s 
gain and bias coefficients that were obtained from image 
headers. We used the MODTRAN radiative transfer model for 
the atmospheric correction using a standard atmosphere model 
and atmospheric humidity and visibility that were obtained 
from local weather stations. 
Geo-Referencing i 
Conversion of DN to Radiance Í 
" Atmospheric correction i 
Topographic correction j 
Converting of Radiance to Reflectance 
    
    
    
The statistical relationship 
between reflectance and LAI 
Classification of Forest Type 
Applying of Multiple regression models 
  
Reference LAI map i 
Figure 2. Procedure of constructing a reference LAI map using 
Landsat ETM- and ground measured LAI 
Further radiometric correction was applied to reduce the 
illumination variation caused by topographic slope and aspect. 
To reduce the topographic effect, we used an empirical method 
that normalizes the illumination difference by applying the 
Minnaert's constant calculated from digital elevation model 
(DEM) and digital map of forest stand. Minnert's method have 
shown its effectiveness to correct the topographically induced 
radiometric distortions over the forests in Korea (Lee and Yoon, 
1997). The detailed information on estimating of Minnaert 
constant k can be found in Lee et al. (2003). 
A vector map of 30 sample stands of the field LAI 
measurement was overlaid to the geo-referenced ETM+ 
reflectance data and the pixels corresponding to each plot were 
extracted. Due to the high spatial autocorrelation, the variation 
of adjacent pixels was very low to overcome the problem of the 
sub-pixel error distance of the geometric registration. 
Initial approach to compare the field measured-LAI and the 
image spectral reflectance was a simple correlation analysis.