3.1 Data Acquisition

A number of NOAA AVHRR HRPT and LAC data were acquired from different sources including NOAA NESDIS (USA), EROS Data Center (USA), NRCT (Thailand) and SMA/SMC (China) (Table 5). In general, at least four scenes for the harvest season and four for the summer season were acquired for each country covering two time frames 1985-1986 and 1992-1993. Afternoon pass of NOAA-9 for 1985-1986 and NOAA-11 for 1992-1993 were selected for the study. Sample images of pre-processed 1985/86 and 1992/93 NOAA AVHRR data has been presented in the Figure 2 and 3 respectively.

Table 5. Acquired NOAA AVHRR Data for Thailand
Harvest Season Source Summer Season Source

	Harvest Season	Source	Summer Season	Source
1985-86	10December ’85  12December ’85  27December ’85  15 January ’85  18 January ’85  27 January ’85	NOAA/NESDIS   NOAA/NESDIS  NOAA/NESDIS  NOAA/NESDIS   NOAA/NESDIS  NOAA/NESDIS	9 April ’85  8 May ’85  22February'86  22 March '86	NOAA/NESDIS   NOAA/NESDIS  NOAA/NESDIS  NOAA/NESDIS
1992-93	18December ‘92  31 January ’93	EROS  EROS	5 February ‘93  8 February ’93  5 March ’93  3 April ’93  1 May ‘93	EROS  EROS  EROS   EROS  EROS

Phenological characteristics of the vegetation and hence the seasonality were given due consideration in procuring the AVHRR data. Basically data representing two seasons viz. harvest season and summer season were selected for each country. Acquiring summer season data has its clear advantage that it facilitates discriminating deciduous and evergreen forests. The selection of these data sets that exhibit complementary information was found to be informative in distinguishing different forest types and also in distinguishing forest from agricultural lands.

NOAA AVHRR HRPT data were analysed using PC ERDAS and IDRISI image processing software. Further analysis were performed in GIS (ARC/INFO software) environs. In-house software has been written for the down loading, band selection, calibration, geometric correction and cloud masking of AVHRR data as pre-processing steps. Fig. 2 shows the flowchart of the methodology adopted.

3.2 Pre-processing

AVHRR data pre-processing mainly consisted of: data extraction and noise removal, radiometric calibration, geometric correction, and cloud masking procedures.

The HRPT Level-1B AVHRR data received in packed format were converted from BIL to BSQ format using appropriate programs. The original radiometric resolution of 10 bit pixel values was maintained by using two bytes for each pixel for all 5 channels. The bad/noisy lines were then identified by visual inspection of each channel of an image. All such bad/noisy lines were marked as being areas of "no data" by assigning zero values.

Radiometric calibration were performed based on the procedures outlined by European Space Agency (ESA) Handbook on "SHARP LEVEL-2 : Development Procedures and Format Specifications" and by NOAA Technical Memorandum NESS 107 on "Data Extraction and Calibration of TIROS-N/NOAA Radiometers".

Due to the lack of readily available atmospheric data in South and South East Asia, atmospheric correction was not performed. Besides, although several possible approaches for the correction of water vapour absorption and aerosol scattering exist, there is presently no agreement on an acceptable method for atmospheric corrections. Some methods need further validation and are far from straight forward (IGBP, 1992).

The bi-directional reflectance effect caused by the viewing geometry and surface angular anisotropy also affects the AVHRR channels 1 and 2 (Gutman 1990). The bi-directional effect depends upon the vegetation type and could differ from one type to another. In order to correct the effect of viewing direction, angular corrections should be developed for different vegetation types and different seasons. However, images taken at large view angles (off-nadir views) which fall at the extreme of the scan line was excluded and thus, such effects, due to atmospheric scattering and absorption, and viewing geometry, are partly reduced.

A two step procedure has been used for the geometric correction of AVHRR images. The images were first resampled to a reference map projection based on location data generated by orbital navigation model and then further corrected by a linear first order rectification based on ground control points.

Interactive visual cloud masking procedure was used to identify the threshold value for clouds. Use of such an interactive cloud screening procedure proved to be highly effective in removing the clouds without losing useful data.

Finally, country mask was generated by rasterizing the vector boundaries of the study area obtained form the World Data Bank II.

3.3 Classification

Unsupervised classification was performed followed by interacting labelling. Secondary information were fully utilised during the analysis. Field trips were organised to collect secondary data and for results validation.