Estimation of Percent Vegetation Cover of Grassland in Monglia using NOAA AVHR Data
3.1 Function for the estimate of percent vegetation cover by VI
The semi-empirical model was used in the study for the estimate of green vegetation cover from NOAA AVHRR data in the semi-arid region. Estimating model was formalized by applying a regression technique to the data where the percent vegetation cover was held as the dependent variable, while the VIs values was held as the explaining variable (Ts Purevdorj et al., 1997). The equations which describe empirical relationships between percentage green vegetation cover and VIs are:
Cover (%) = -4.337 - 3.733*NDVI + 161.968*NVVI2 (1)
Cover (%) = 1.062 + 43.5442* TSAVI + 97.817*TSAVI2 (2)
Our experimental results (Purevdorj et al., 1996) indicated that NDVI and TSAVI perform better result than VIs for the estimate of green vegetation cover. Therefore, in the study these two polynomial equations of second order were chosen as a equation for estimating percent green vegetation cover from the NOAA AVHRR data.
3.2 Field measurement of vegetation cover
Ground measurements of vegetation cover amount for the test sites were carried out during August and September 1996. Sampling sites were chosen in the large homogeneous areas of grasslands so as to be best represent the grasslands and concentrated in the low and middle vegetation densities region which are most difficult and important to detect by AVHRR data. The geographical location was determined by GPS camera "Konica". A total of 6 sites were sampled during the ground truth campaign(Table 1) with one sample plot in each test site.
Sampling scheme was similar for all test sites and carried out as following. In each of the test sites random about 1km x 1km area was chosen. For the site wee sampled 25 points and the color photographs of ground cover of the points we taken. Then vegetation cover of each point was estimated by counting numbers of green vegetation pixels using pixel value of red, green, and blue band of the digitized photographs. The total vegetation cover amount for each test site was determined as mean of the percents vegetation cover estimated for 25 sample points. The estimates of green vegetation cover for the sampling sites are presented in Table 2. Calculated values of standard deviation given in Table 2 show that vegetation density is relatively constant.
Table 2. Ground measurement of green vegetation cover for test sites.
| |
Site 1 |
Site2 |
Site3 |
Site4 |
Site5 |
Site6 |
| Mean |
13.4 |
14.0 |
25.0 |
44.2 |
8.8 |
15.0 |
| Standard deviation |
2.6 |
4.0 |
5.8 |
5.9 |
2.4 |
4.2 |
4. Image processing
In order to compare the ground measurements of percent vegetation cover with the estimate from satellite data, were used NOAA AVHRR High Resolution Picture Transmission (HRPT) data. The images acquired during the field measurement wee processed and visually examined on the basis of false color composites to ensure that only cloud-free data were used for the estimate. From the images, two scenes were selected. These two scenes acquired on August 3, 1996 in receiving station of Ministry of Nature and Environment of Mongolia and on August 12, 1996, from the archives of the EROS DATA CENTER (USA) were as least cloudy image for test sites. Subscenes of 1200 X 1200 pixels for the study images are located between 40 to 50 degree latitude and 100 to 110 degree east longitude.
4.1 Geometric correction
The images were geometrically corrected by "PaNDA" using ground control points. The ground control points wee obtained from Operational Navigation Chart (1: 1 000 000, Defense Mapping Agency Aerospace Center, St. Louis, Missouri, 1988).
4.2 Atmospheric correction
In the study atmospheric correction of the AVHRR Channel 1 and Channel 2 was carried out using "6S" program (Goddard Space Flight Center, USA ) which was adopted by authors in SUN SPARC workstation. After atmospheric correction of the Channel data the NDVI values were calculated and made the histogram. The comparison of the histogram indicates that there is intrinsic differences between the arospherically corrected and uncorrected NDVI values (Figure 1). The results show that maximum value of the uncorrected and corrected NDVI was about 0.62 and 0.79 respectively.
Figure 1: Histogram of atmospherically corrected of NOAA AVHRRR.
