34 
GLOBAL LAND COVER MONITORING BY NOAA GVI DATA 
Ryutaro Tateishi and Koji Kajiwara 
Remote Sensing and Image Research Center 
Chiba University 
1-33 Yayoi-cho, Chiba 260 Japan 
Abstract 
NOAA/AVHRR Global Vegetation Index(GVI) data of Asia in 1983 and 1987 were used to evaluate their usefulness 
for global land cover monitoring. Color composite images of monthly GVI data or color composite images 
of principal components from 12 monthly GVI data were found to be useful for visual interpretation of 
seasonal vegetation dynamics. The results of cluster analysis applied to monthly GVI data for a one-year 
period, indicate that unsupervised classification method is useful for global or continental land cover 
classification due to lack of ground truth data. In order to detect land cover changes, the difference 
of 12 monthly GVI data between 1983 and 1987 was calculated. The results show that it is difficult to 
detect land cover changes due to cloud contamination in monthly GVI data and poor registration of GVI 
products. 
Key Words: Global land cover monitoring, NOAA GVI data 
1. INTRODUCTION 
Considerable amount of papers have been reported 
using NOAA/AVHRR GVI data. Some of them were 
applications to agriculture, phenology, 
deforestation or meteorology in rather small 
limited areas (Tucker et al 1986; Johnson et al 
1987; Kennedy 1989; Kerr et al 1989; Malingreau et 
al 1989). Some others were applications to 
phenology or land cover in global or continental 
areas (Justice et al 1985; Goward et al 1987; 
Justice et al 1989; Townshend et al 1989). Their 
studies were based on comparisons of GVI data with 
land cover types at point locations or in limited 
areas. A few papers were on characteristics of GVI 
data themselves (Holben 1986; Singh 1988). 
This study is based on image processing of GVI 
data in approximately whole Asia. This paper tries 
to, produce enhanced images for land cover 
monitoring, classify land cover types and detect 
land cover changes in almost whole Asia. However 
the method of this study can be extended to global 
applications without any change. The results of 
the processing were visually compared with 
geographic images such as land cover data from 
existing maps (Wilson et al 1985), elevation data 
or snow cover data. 
2. THE DATA 
2.1 NOAA/AVHRR weekly GVI Data 
Weekly GVI data of Polar Stereographic Projection 
in full years of 1983 and 1987 were used in this 
study. The explanation of NOAA satellite and 
NOAA/AVHRR weekly GVI data have been well described 
in many papers(SDSD 1986; Johnson et al 1987; 
Goward et al 1987; Justice et al 1985). This paper 
does not repeat the explanation except some 
comments on sampling method and cloud contamination 
of GVI data. 
IFOV of the AVHRR sensor is 1.1km x 1.1km at 
nadir and 6.9km x 2.4km at the maximum off-nadir 
angle. Local Area Coverage(LAC) data have the same 
spatial resolution as the IFOV of the sensor. 
Global Area Coverage(GAC) data are produced by 
sampling 4 LAC pixels in 5 x 3 LAC pixels. This 
sampling is carried out by averaging 4 of every 5 
pixels along a scan line and processing only every 
third line. Therefore GAC data corresponds to 4 
LAC pixels (nominally 1.1km x 4.4km at nadir) and 
a real area of a GAC pixel corresponds to 4/15 of 
an area represented by a GAC pixel. 
Weekly GVI data are produced from GAC data as 
follows. 
(1) GAC data are mapped into Plate Carree 
Projection (latitude/ longitude system) with a 
calculation of Normalized Vegetation Index(NVI) 
from channel 1 and channel 2 of GAC data. 
Actually, Scaled NVI(SNVI) is used to represent NVI 
in 8 bits digital count as shown in Table 1. These 
data are called as daily GVI data in this paper. 
The Plate Carree array covering the latitude range 
of 75°N to 55°S consists of 2500 by 904 pixels. This 
means that an area of Plate Carree pixel at nadir 
is 16km(longitude) by 16km(latitude) and the one at 
75°N is 16km(longitude) by 6.8km (latitude). Since 
the resolution of GAC data is much finer than the 
resolution of Plate Carree array, a number of GAC 
pixels are mapped to each location in the Plate 
Carree array. The selection of GAC data for one 
location of Plate Carree array can be regarded as 
random because each GAC value replaces the one 
previously written. 
(2) Weekly GVI data are produced from daily GVI 
data by selecting the maximum value of NVI for a 
seven-day period to reduce cloud and to reject 
pixels from off-nadir positions which have lower 
NVI value by atmospheric effect. 
(3) Weekly GVI data in Plate Carree Projection are 
transformed to Polar Stereographic projection or 
Mercator Projection. These are the weekly GVI 
products. 
The above explanation for the production of 
GVI data is the second generation version which 
started from April 1985. The first generation 
version before March 1985 has a different 
procedure(SDSD 1986). 
Figure 1 shows comparisons of a real area of 
a GVI pixel to an area represented by a GVI pixel 
at the equator or the north latitude of 75° in 
Plate Carree Projection or Polar Stereographic 
Projection. As shown in Figure 1, a real area of 
a GVI pixel corresponds to only a part of an area 
represented by a GVI pixel. For example, the ratio 
of a real area of a GVI pixel to an area 
represented by the pixel is 1:33 in the case of a 
GVI pixel in Polar Stereographic Projection 
produced from GAC data at nadir in the equator 
region. 
The selection of daily GVI data for the 
production of weekly GVI data is based on the 
assumption that daily GVI images for a seven-day 
period are well registered. However even if two