3. Data Processing Flow
The AVHRR raw data of NOAA-14 afternoon passage collected daily at Tokyo Universty, Kuroshima (Japan). Asian Institute of Technology (AIT) (Thailand) and Ulaanbaatar (Mongolia) receiving stations were used in the composite image. On considering computing resource and data processing method, the AVHRR raw data for the each receiving station was processed separately according to the fixed mapping area (Table 1) and subsequently the image were mosaicked into final composite image.

Table 1. Mapping area of the each receiving station

Receiving Station	Left top corner	Right bottom corner
Tokyo University	130°E, 70°N	170°E, 15°S
Kuroshima	110°E, 45°N	170°E, 0°S
AIT	60°E,35°N	140°E, -20°S
Ulaanbaatar	50°E, 80°N	170°E, 30°S

The laboratory of Remote Sensing Data Analysis of Iwate University receives raw data of AVHRR on 8-mm cartridges from the ground receiving stations. The ground receiving stations have different data format from the input format for "PaNDA" software that is to be used to preprocessing. Therefore, before the processing routine, the AVHRR raw data was converted to the format compatible with input format for "PaNDA" software.

At present the data processing flow include following steps and first prototype composite image was tested:

1. Systematic correction
2. Geometric registration
3. Composting
4. Atmospheric correction

a. Systematic correction
The systematic correction performed by "PaNDA' software includes calibration of five channels, ground control point (GCP) matching with coastline and resampling.

At first, the percent albedo measured by the sensor channel 1 and 2 is computed as linear function of digital number (DN) as follows:

A_i = S_i* DN + I_i
Where A_i is the percent albedo of channel I, S_i and I_i are the scaled slope and intercept values for channel I respectively. The pre-launch slope and intercept values for AVHRR channel 1 and 2 were used from source calibration information of NOAA/NESDIS. The calibration coefficients for channel 3,4 and 5 are derived onboard blackbody and deep space reference. The calibration process converts raw DN to linear radiance using following equation:

RLIN = M_i* DN+L_i
where RLIN is the linear radiance, M_i and L_i are gain and intercept coefficients of channel I respectively. The quantity M_i, L_i (in units of radiance/count) is calculated for each channel from onboard blackbody and deep space reference. The linear radiance calculated was corrected the non-linearity as follows:

R = A_i*RLIN +B_i*RLIN * RLIN + C_i
where R is corrected radiance, A_i,B_i and C_i are correction coefficients for channel I. Then the radiance value was converted to brightness temperature using sensor spectral sensitivity weighted method.

Second step of the systematic correction is the GCP matching and correction for terrain elevation. The digital elevation data used for the correction is GTOPO30 (Global 30 Arc Second Elevation Data, USGS). The control points and coastline image were warped into the satellite projection for the correlation process. The ground control matching uses automatic correlation of image segments of the AVHRR data with ground control points identified from hydrological feature of Digital Chart of the World (DCW0 and then integrates the adjustments derived from the correlation with the orbital model.

Third, the image was resampled using the nearest neighbor interpolation method and radiometric and geometrically corrected (systematic) image was produced. The resulting image includes all five channels and sun target geometry angles data.

b. Gemetric registration
A systematic correction using satellite model alone can not achieve required multi-temporal accuracy in the composite image. The correction result of the "PaNDA' software showed that there is still an insufficient accuracy in the output images and it causes "blurring" in the composite images. Therefore, in order to prevent "blurring" in the compositing process, an image registration procedure was done. An image from coastline image with image segment derived from the AVHRR data. Here we used line matching for coastline and pattern matching for rivers. The image segment for the line matching is created in the following way. At first, image of 64 by 64 data is extracted from the AVHRR channel 1 and 3 centered at the ground control points (GCP) position and then a binary image is created from the image by thresholding Normalized Difference Vegetation Index (NDVI) value. The pixels of the binary image have a value 1 for land area and 0 for water area. Then the edge image generated from the binary image by edge extraction was correlated with template image. The image segment for pattern matching is produced as previous but edge line was not extracted.

The coast and shoreline data was collected from DCW and hydrological feature of rivers was generated from water mask data (EROS DATA Center). Then coastal data was transformed to coastline image with same size of the output image. The template image of 32 by 32 data for the matching is generated from the coastline image centered GCP position. Also GCP data set was developed from DCW. The matching result of the template and edge image was selected according to threshold matching value. Then the resampling is performed using the nearest neighbour algorithm.

c. Compositing
At first, the length of the compositing period of 10 days was used, which has the advantage of a common reporting period for agronomic and biophysical characteristics. The different compositing techniques for land and sea area are respectively proposed.

The compositing image for land area are commonly produced using the maximum value compositing (MVC) tedhnique based on NDVI (Holben, 1986). When the MVC approach was originally proposed, it was argued that the method would preferentially select near-nadir views over larger scan angles. The work that led to that recommendation was based on a model using an assumption of Lambertian reflectance from the earth surface. Numerous studies have found that off-nadir viewing can produce greater NDVI values than at nadir viewing angles (Cihlar et. al., 1994). We have observed too, that the MVC algorithm preferentially selected forescattering views over nadir for a variety of cover types in north-east region of Asia, specially when using atmospherically corrective AVHRR data (Kawada et. al., 1998). Therefore we propose to use the multiple criteria composting technique by using maximum NDVI and minimum scan angles (MaNMis). The composite technique was performed in the following steps as follows:

Step 1: NDVImas (1,p) = MAX{NDVI(1,pd)} d = 1,2,……. 10
Step 2: If NDVI(1,p,d)/NDVImax(1,p) >0.85; (1,p,d_i) pixels retained
Step 3: Pixel (1,p) is selected when smin(1,p)= MIN{S(1p,d_i)}

Where NDVI (1,p,d) si the NDVI calculated from apparent reflectance of Channels 1 and 2 for pixel (1,p) on data d; NDVImax(1,p) is the maximum NDVI value of the pixels (1,p) for the compositing period; di refers to the value NDVI(1,p,d) found within 15% of NDVImax(1,p), smin (1,p) is the minimum scan angle among the all retained pixels di from step 2. The result from this multiple criteria compositing techniques is compared with the other two compositing techniques of maximum brightness temperature of Channel 4 and MVC. In comparision with those two methods, MaNMiS technique is directed to preferentially to select the nearest nadir pixel over larger scan angle. Moreover, a compositing technique for sea surface area is developed using AVHRR Channel 2 and Channel 4. At first, the sunlit pixels were removed using threshold value (10%) of apparent reflectance of Channel 2. Then the pixel with maximum brightness temperature of Channel 4, which is less than the threshold value of Channel 2, was selected for the composite.