Computation of Global Vegetation Index (GVI)
Computation of Global Vegetation Index (GVI)

This section provides technical background on the Global Vegetation Index (GVI) and an overview of how the product is produced and changes that were made in the production of the First Generation experimental product. Note: the words "tape" and "cartridge" are used interchangeably throughout this document.

2.1 BACKGROUND

The spectral reflectance of chlorophyll pigment in the visible and near infrared part of the spectrum provides a means of monitoring density and vigor of green vegetation. Green leaves have a reflectance of less than 20 percent in the 0.5 to 0.7 micrometer spectral interval, but about 60 percent in the 0.7 to 1.3 micrometer range. Differential reflectance in these bands has been widely used with LANDSAT Multispectral data to classify land cover, estimate crop acreage, and detect plant stress.

NOAA series operational meteorological satellites provide data that can be used for vegetation monitoring. The Advanced Very High Resolution Radiometer (AVHRR) flown on NOAA-14 and previously on NOAA-7, NOAA-9 and NOAA-11, is a 5-channel instrument which scans continuously at a ground resolution of 1 km. The 1 km data is sampled and averaged to a nominal 4 km resolution and recorded on board the satellite for subsequent transmission. AVHRR Global Area Coverage (GAC) at 4 km resolution is thus available. This coverage is supplemented by 1 km resolution coverage of selected areas using AVHRR LAC (Local Area Coverage) data. The sampling done on board the satellite to produce the GAC data consists of selecting every third 1 km LAC data line, and selecting four of every five samples along the line and averaging these to yield one GAC sample. Thus, the nominal "footprint" of a GAC sample is 1 km by 4 km, with gaps (near nadir) between GAC lines and small gaps between GAC samples. LAC and GAC samples are 10 bit binary "count" values (digitized voltages).

The spectral ranges of the five AVHRR channels are contained in Table 2.1-1.

Table 2.1-1. Range of spectral values for AVHRR channels.

AVHRR Ch. #

Range (micrometers)

1

0.58 - 0.68

2

0.725 - 1.0

3

3.55 - 3.93

4

10.30 - 11.30

5

11.50 - 12.50

The spectral bands used currently for vegetation monitoring are the Channel 1 visible band (0.58 to 0.68 micrometers) and Channel 2 in the near infrared band (0.725 to 1.0 micrometers). The spectral response curves for these channels are similar to those of bands 5 and 7 on the LANDSAT multispectral scanner. Various mathematical combinations of Channel 1 and 2 data have been found to be sensitive indicators of the presence of green vegetation and are referred to as vegetation indices. This effect is due to the differential reflectance of vegetation in these bands. Thus, the difference in the Channel 2 and Channel 1 data values (which are proportional to the reflectance of the scene viewed) computed as Ch2-Ch1, is a measure of the degree to which the scene viewed includes green vegetation.

The basic index is the Normalized Difference Vegetation Index, (NDVI). The NDVI is defined by the equation:

e-ndvi.gif

where Ch1 and Ch2 are count values from Channel 1 and Channel 2, respectively, of the AVHRR instrument. The NDVI is preferred for global vegetation monitoring because it partially compensates for changing illumination conditions, surface slope, and viewing aspect. Clouds, water, and snow have larger reflectances in the visible than in the near infrared, so for these features NDVI is negative. Rock and bare soil have similar reflectances in these two bands and result in vegetation indices near zero. In scenes with vegetation, the NDVI ranges from 0.1 to 0.6; the higher values are associated with greater density and greenness of the plant canopy. Atmospheric effects such as scattering by dust and aerosols, Rayleigh scattering, and subpixel-sized clouds all act to increase Ch1 with respect to Ch2 and reduce the computed vegetation indices.

Research performed with the First Generation GVI product, which included only Ch1, Ch2, Ch2-Ch1, and NDVI values mapped on a Polar Stereographic grid, suggested that additional information would be of help in mitigating the effects of cloud contamination and facilitating year to year comparisons of NDVI values over the same areas. (Appendix A is a bibliography of significant research reports selected to be of most value to the users of the GVI products.) Thus, the Second Generation GVI product also includes Channel 4 and Channel 5 IR data values, which may be of use in cloud screening, and Solar Zenith and Scan angles, which should assist in normalizing NDVI values for year to year comparisons.

2.2 PROCESSING OVERVIEW

The bit and byte (one byte equals 8-bits) numbering convention for GVI data sets is as follows. The bits within each byte or word are numbered from the most significant bit (MSB) on the left to the least significant bit (LSB) on the right, with the MSB identified as bit 31, the next most significant bit as bit 30, and with the LSB as bit 0. Similarly, the byte containing the 8 MSBs of a 32-bit word is identified as byte 4; and the byte containing the 8 LSBs, as byte 1.

Daily GVI maps are produced by processing AVHRR data orbit by orbit and mapping it to a standard base projection. Briefly, this process consists of truncating the original 10-bit GAC data to 8-bit data and then scaling. When the IR channels were added (for the Second Generation product), they were calibrated, converted to temperature and scaled. The standard base projection was Polar Stereographic from May 1982 through March 1985 (First Generation version), and is currently Plate Carrée (or latitude/longitude) for the Second and Third Generation versions of the product (April 1985 through the present). On a weekly basis, Polar Stereographic composites were produced from May 1982 to the present for the First and Second Generation versions, with Mercator composites also produced for data from April 1985 to the present for the Second Generation versions.

The Second Generation GVI contains data only from the daytime orbits of the afternoon satellites NOAA-9 (April 1985 - November 1988), NOAA-11 (November 1988 - September 1994) and NOAA-14 (February 1995 to the present). The NOAA-14 satellite (and previously the NOAA-7, NOAA-9 and NOAA-11 satellites) is in a near sun-synchronous near-polar orbit. Data from the afternoon polar orbiters is preferred for producing the GVI because of the high sun elevation angle (low solar zenith angle). These satellites cross the equator between 1400 and 1700 local time, with the equator crossing time drifting to a later hour as the satellites age (Price, 1991).

Satellite orbit drift results in a systematic change of illumination conditions and local time of observation - one of the main sources of non-uniformity in multi-annual satellite time series. Orbital drift is shown in Figure 2.2-1 (personal communication with Garik Gutman). This figure shows the general degradation over time of the orbits for NOAA-9, NOAA-11 and NOAA-14. Notice that when NOAA-11 was first launched in 1988, it had a 1330 local time equator crossing. By 1994, NOAA-11's orbit had degraded to 1700 local time.

Figure 2.2-1. Orbital degradation of NOAA-9, NOAA-11 and NOAA-14 over time.

f22-1.gif

The NOAA-11 AVHRR instrument unexpectedly failed on September 13, 1994. NOAA-12 (a morning spacecraft) was used briefly to generate the GVI until October 20, 1994, when NOAA-9 was substituted. Although NOAA-9 is considered an afternoon spacecraft, its nighttime observation has drifted (over the six years it was non-operational) to a morning observation (about 10 AM as of October 1996). This situation was temporary, but caused a gap between September 1994 and February 1995, when NOAA-14 became operational. The GVI data generated during this period is considered to be of poor quality and it's use is not recommended.

In addition to the satellite orbit drift, there is also a drift in the AVHRR sensor itself. Figures 2.2-2 and 2.2-3 from Rao and Chen, 1994 demonstrate this drift problem.

Because of the difference of degradation of the two AVHRR channels (1 and 2), NDVI exhibits trends and discontinuities as shown in Figure 2.2-3.

A single day's map consists of a mosaic of the daytime portion of 14 orbital swaths. The Polar Stereographic maps are on a one-sixteenth sub-mesh grid of the standard 65 x 65 Polar Stereographic projection used by the National Centers for Environmental Prediction (formerly National Meteorological Center) and the Air Force Global Weather Center. Resolution in the mapped data ranges from 13 km at the equator to 26 km at the poles. The data are sampled in the mapping process so that the GVI in an array location consists of a value computed from a single AVHRR pixel; there is no averaging.

On any single day, about half the Earth is obscured by clouds. To remove clouds, seven-day maximum vegetation index composites are produced from the daily arrays. For each composite period of seven days, the pixel from the daily data having the greatest Ch2-Ch1 difference (computed from counts, i.e., uncalibrated) is retained at each array location (i.e. the "greenest" of the seven daily values for each array location is retained in the composite). This eliminates clouds from the composite except for areas which were cloudy at the time of satellite overpass for all seven days. When several cloud-free views of an area are obtained over the compositing period, NDVI values computed on days with haze and subpixel clouds are eliminated. The values of Ch1, Ch2, Ch4, Ch5, Solar Zenith and Scan Angles that are also present on the composite products are for the same day's data as the NDVI value selected for the composite.

The weekly cycle for the First Generation GVI product began with Monday and ended on the following Sunday. Beginning with the Second Generation GVI product in April 1985, the weekly cycle was synchronized by day of year (i.e., day of year 1 through 7 were always the first week, day of year 8 through 14 were the second week, etc.). This scheme causes the weekly cycle to vary from year to year depending on what day of the week January 1 falls on (e.g., the weekly cycle for 1985 runs from Tuesday through the following Monday, while 1986 runs from Wednesday through the following Tuesday).

Figure 2.2-2. Isotropic albedo in Channel 1 (top) and Channel 2 (bottom) of the southeastern Libyan desert (from Rao and Chen, 1994).

f22-2.gif
Figure 2.2-3. Normalized Difference Vegetation Index for the southeastern Libyan desert (from Rao and Chen, 1994).

f22-3.gif

When NOAA/NESDIS' Information Processing Division (IPD) took over the GVI processing on April 11, 1988, the weekly cycle was changed again to Monday through the following Sunday. This cycle does not change from year to year.

2.3 CHANGES IN THE FIRST GENERATION (EXPERIMENTAL) GVI PRODUCT

Because the First Generation NDVI was an experimental product, users are cautioned that changes in the computation of the daily arrays were made as more was learned about the cloud contamination problem. This trial and error process went through the following steps (in chronological order):

1. From May 10, 1982 (day of year 100) through August 7, 1982 (day of year 219), the daily NDVI was computed from the last value mapped to each location on the Polar Stereographic output grid. This was regarded as a random sampling.

2. From August 8, 1982 (day of year 220) through September 12, 1982 (day of year 255), the daily NDVI was computed from the Channel 1 and Channel 2 values with the greatest difference mapped to each location on the Polar Stereographic output grid. This resulted in an apparent marked improvement in the removal of cloud contamination, but was not a random sampling, in that it selected the "greenest" pixel for each map location on a daily basis.

3. From September 13, 1982 (day of year 256) to March 27, 1983 (day of year 86), the "greenest" pixel from only every other AVHRR scan line was mapped, to save computer time and restore a degree of "randomness" to the sampling.

4. From March 28, 1983 (day of year 87) to the present, the sampling was changed back to the random method (using the last sample mapped to each grid cell in use from May 10 to August 7, 1982). Data from only every other scan line is being used to save computer time. This method was also used in the Second and Third Generation NDVI product.

Source: http://www.ncdc.noaa.gov