GISdevelopment.net ---> AARS ---> ACRS 1992 ---> Mapping from Space

Generation of Cloud Free Vegetation Index Map

Toshiaki Hashimoto, Shunji Murai
Institute of Industrial science, Univ. of Tokyo
Roppongi 7-22-1, Minato-ku, Tokyo, Japan


Abstract
NDVI ( Normalized Difference Vegetation Index ) calculated from NOAA AVHRR is very suitable and used widely for monitoring the changes of vegetation conditions of a large scale. In an original AVHRR imagery, however, clouds will exist so often which have to be eliminated. A cloud free mosaic is generated from some images acquired successively in a proper period. In this paper, the method of generating a could free vegetation index map is introduced. The method is based on the simulation by LOWTRAN 7 and the examination of real AVHRR images. A cloud free mosaick of channel data 1 & 2 as well as NDVI is generated by the method.

Simulation by LOWTRAN 7
The values of NDVI as well as channel data 1 and 2 reflected land covers. But they are affected by many factors such as the anisotropy of reflectance, atmospheric conditions. Solar position, satellite position, etc. It was examined how big effects these factors have on channel data. The sensor receiving radiances and CCT counts of channel 1 and 2 were calculated by LOWTRAN 7. The simulation was made under the conditions of four types of vegetation coverages (i.e. reflectances), four types of atmospheric conditions (humid & clear, humid & turbid, dry & clear, dry & turbid) and different solar zenith angles (00-850)1). Fig. 1 (A) and (B) show the solar zenith angle versus total radiance and (total radiance – path radiance), respectively. In the figure, ‘D’, ‘H’, ‘C’, ‘T’, ‘P’ and ‘Veg.’ express dry, humid, clear, turbid, path radiance and vegetation coverage ratio, respectively. To avoid complicatedness, ‘Veg.’ of only 98% & 0% are drawn. Fig. 2 shows the NDVI calculated from the results of Fig. 1. The followings are concluded from these figures.
  1. Aerosol has big effect on path radiances of both channel 1 and 2, but moisture has little one.
  2. Total radiance of channel 1 is much effected by acrosol and not so much by moisture, while that of channel 2 is effected by both moisture and aerosol.
  3. The values of path radiance subtracted from total radiance is not so related with atmospheric conditions in channel 1, but related with moisture in channel 2.
  4. Channel data are much dependent of solar zenith angle.
  5. NDVI from total radiance (or CCT counts) is much dependent of solar zenith angle and atmospheric conditions.
  6. NDVI from ( total radiance – path radiance ) is not so related with solar zenith angle or aerosol concentration, but still dependent of moisture.
Examination of real AVHRR images

1 Considerations of AVHRR scanning geometry
The orbit of NOAA satellite is shifted gradually day by day. In case of NOAA-11, the daily shift of orbit is about 3 in longitude which corresponds to about 15 in scan angle, and the orbit is revolved after about nine days. This fact leads to that the ground surface is observed sometimes by forward-looking and other times by backward-looking and the solar zenith angle at observation time is different. The significant channel data 1 and 2 are observed from NOAA satellite of odd number in ascending mode ( on afternoon pass ) and of even number in both ascending and descending modes in a day. The data from the satellite of odd number is generally observed at the smallest solar zenith angle among them. The AVHRR scan angle of 55.4 is so big that the ground resolution in marginal regions of the image is about 12 times larger than that of sub satellite region and the optical length at off-nadir viewing is much longer than that at nadir viewing. The margins of the image are much distorted both geometrically and radiometrically. So such data cannot reflect well the true ground conditions. This fact has been pointed out by some scientists ( e. g. Tucker et al2) ).

2 Color balancing
The results of simulation proved that channel data should be corrected for solar zenith angle and NDVI calculated from ( total radiance – path radiance ) is only related with atmospheric moisture. One of the problems to get such an NDVI is the evaluation of path radiance from real images. The minimum values of channel data are assumed to be path radiances as the same as dehazing procedure. We supposed that channel data could be corrected by the following equation.


where L, L’ , Z and n is uncorrected channel data, corrected channel data, solar zenith angle and coefficient, respectively. After some examination of real AVHRR data, it was found out that the value ‘n’ of 0.8 - 1.0 leaded to good result. But there still remained the dependence of channel data as well as NDVI on atmospheric condition. To correct the dependence, the linear stretching method was adopted. The channel data of target image is converted by the followings.


The statistics ( mean and standard deviation ) are obtained from the land areas with no clouds and within scan angle of less tan 300 in both reference and target images. The NDVI calculated from these data also corrected.

3 Cloud elimination
In case of mosicking, cloud elimination can be made by selecting maximum NDVI. The detection and elimination of clouds are necessary for calculating the sates tics of images. But this process is not so strict for cloud climination, so thresholding is enough.

4 Sample of mosiacking
Fig. 3 shows a sample of mosaicking. The mosaic image is generated from four images which are acquired on 20, 21, 22 and 24 October 1990. They were all observed from NOAA-11 on ascending mode.

Procedure for mosiacking
Trough the investigations as mentioned above, the following procedure is proposed for generating a cloud free vegetation index map from serial AVHRR images ( see Fig. 4 ).
  1. The images acquired within nine days are used for mosiacking.
  2. The data observed under the following conditions are used in prior. In other words, such data form primary images and the other data form supplemental images. The minimum values and statistics are obtained from primari images.
    • observed from NOAA satellite of odd numbers in ascending mode, which corresponds to the smallest solar zenith angle in a day.
    • observed within scan angle of less than 300.
    If a certain area is covered with clouds in all of these images, such an area will be compensated with the images from other satellite, other passes or wider scan angle (i.e. supplemental images).
  3. The minimum values of channel data 1 and 2 are subtracted from channel data, respectively. Then, they are corrected for solar zenith angle.
  4. The corrected channel data are stretched linearly between images.
  5. NDVI is calculated from the stretched channel data.
  6. The maximum NDVI value of the images is adopted as the output of vegetation index map.
Conclusion
The method for generating a cloud free vegetation index map is introduced. In the method, geometric and radiometric characteristics of AVHRR images are taken into account.

Reference
  • M. Matsumoto et al, ‘investigation of solar zenith angle dependence of NOAA GVI data and its correction (in Japanese)’, Journal of thejapn Society of Photogrammetry and Remote Sensing, 1990, Vol. 30, No. 3, pp. 34-41.
  • C.J. Tucker et al. ‘Satellite Remote Sensing of Total Dry Matter Production in the Senegalese Sahel’, Remote Sensing of Environment, 1983, No. 13, pp. 461-474







Fig. 4 Procedure for cloud free mosaicking