Abstract

Synoptic view of large vegetation area should be very useful tool to many environmental and ecological studies. With increasing number of the earth observation satellites, the use of satellite data has become more practical to regional and global monitoring of vegetation activities. Selection of appropriate satellite data depends on the spatial and temporal scales of vegetation changes of interest. Vegetation index, which is a product of unique spectral reflectance of green vegetation, plays a key role for diverse applications of vegetation-related researches. Theoretical basis of remote sensing derived vegetation index and its applications are discussed.

1. Introduction

The limitation of under-sampling has been a primary controversy in field-oriented environmental studies. In particular, environmental and ecological researches at regional and continental scale require a vast amount of information to characterize the spatial and temporal patterns of landscape dynamics, which are hardly obtained by field survey. Ever since the first launch of civilian remote sensing satellite, the use of remote sensing data has increased for a variety of environmental and ecological studies (Quattrochi and Pelletier, 1991). Certainly, remote sensing can resolve such limitation of under-sampling by providing synoptic view over large geographic area. Satellite remote sensing has made the environmental measurements available at regional and even global scales. Current satellite sensors are being operated at a wide range of spectrum, which is far beyond the visible light. For instance, image data obtained at infrared and microwave spectrum can provide rather valuable information that are usually not available at visible spectrum. Furthermore, satellite remote sensing data can provide continuous measurements over the same area.

Under the circumstances of rapid industrialization and population growth, the status of vegetation cover in northeast Asia is vulnerable to several factors of human-induced developments and environmental perturbations. Beside the direct impacts by land development practice, vegetation is also sensitive to environmental perturbations such as climate change, acid rain, or air pollution. As such, a vegetation community can be a responsive indicator of the consequences of such environmental changes. The monitoring of vegetation changes caused by these disturbance factors is very important for maintaining environmental quality and requires pertinent methodology to observe and survey on time.

Satellite remote sensing can be an effective tool to overcome the limitations on data acquisition and analysis of field-oriented survey. With increasing number of earth observation satellites, remote sensing is becoming rather practical solution to monitor the vegetation condition over wide geographic areas. The objective of this paper is to define the potential of satellite remote sensing technology to monitor vegetation cover over the northeast Asia. Considering the possible characteristics of vegetation changes at different time and spatial scales in this region, I describe the sources of appropriate remote sensing data and appropriate methodology to extract correct information related to various features of vegetation.

2. Vegetation Changes and Remote Sensing

Vegetation change may be a terminology of rather comprehensive definitions, which ranges from the in-growth of a single tree to the entire deforestation by clearcut. Whether we can detect and monitor vegetation changes by remote sensing data depends on the spatial and temporal characteristics of the change and the type of remote sensor data to be used. Therefore, it is important for us to understand the nature of vegetation changes prior to analyzing remote sensing data. For example, it is almost impossible to detect a certain type of vegetation change, such as the annual growth of a small forest stand, by using current remote sensing data. Table 1 lists major types of vegetation changes categorized by temporal and spatial scales.

Probably the deforestation is one of the most common types of vegetation change, which can be readily detected and mapped by satellite remote sensing data. Unlike the tropical regions where timber harvesting is the main reason for clearing forests, the expansion of agricultural use and the land developments for residential and industrial uses are primary factors to force the deforestation in northeast Asia. Since green vegetation has distinct spectral characteristics, detection and mapping of such deforestation can effectively be done without much difficulty using satellite imagery. Desertification is another form of major vegetation changes that can be observed in this region. The exact size and distribution of newly formed desert area can be valuable information for the environmental and natural resource management.

Table 1. Temporal and spatial scales of vegetation changes in northeast Asia.

change categories	temporal scale	spatial scale
deforestation - land developments - conversion to agriculture - timber harvest desertification - degradation of cropland - yellow dust	days ~ years	1 ha ~ 10⁶ ha
Fire	hours ~ days	10 ha ~ 10⁶ ha
disease/insects infestation	months ~ years	10 ha ~ 10⁷ ha
stress - drought - air pollution - acid rain	months ~ years	1 ha ~ 10⁶ ha
species composition - succession	years ~ decades	1 ha ~
Growth	years ~ decades	1 ha ~

Large-scale fire can cause dramatic vegetation change at relatively short time period and, therefore, it is hardly detectable during the fire. Satellite remote sensing data, however, has been frequently used to assess fire damages and to monitor the afterward recovery process. Vegetation changes caused by diseases or insects infestation are often the major concern to natural resource managers. Forest damage caused by diseases or insects varies according to the infestation stages. In early stage, it can be somewhat difficult to find any noticeable symptoms even by human eyes on ground. Physiological changes in leaf organisms might have something to do with the reflectance at certain wavelength bands. Once the infestation is fully developed, the canopy layer is almost defoliated or turns out their color. Such alterations in tree canopy can be clearly distinguished from the healthy canopy on multispectral satellite imagery although the identification of disease or insect type is another subject to be solved.

Climate change, acid rain, and air pollution are now being considered as influential factors to change the structure and dynamics of vegetation. Vegetation change caused by such environmental factors may show various consequences at different temporal and spatial scales. At seasonal event of severe drought, the canopy stress can be obvious from the normal condition of previous years. On the other hands, the effects of environmental factor may have relatively slow responses that reveal very subtle annual changes. Species composition, canopy density, biomass, and primary production are main parameters to describe the structure and dynamics of vegetation community. The detail of information related to the biophysical characteristics of vegetation depends on the type of remote sensors and data analysis methods.

There are several different types of satellite remote sensing data available for ecological studies and, therefore, it is often very important to choose appropriate data set for extracting a certain type of vegetative information. Table 2 lists some of the earth observation satellite data that are widely used for a variety of applications including vegetation, oceanography, earth sciences, natural resource management, water resources, urban studies, etc… Although there have been many other remote sensing satellites, they can be classified into the following four groups by their attributes.

Table 2. Satellite remote sensing data for ecological research.

Satellite	Launch	Sensors	spectrum	spatial resolution (m)	temporal resolution (days)
Landsat	1972	MSS, TM	V, IR	15-80	16
SPOT	1986	HRV	V, IR	10-20	5-26
IRS	1988	LISS, WiFS	V, IR	5-200	5-24
NOAA	1970	AVHRR	V, IR	1100	0.5
OrbView	1998	SeaWiFS	V, IR	1100	1
Terra	1999	MODIS	V, IR	250-1000	2
ERS	1991	AMI	microwave	20	variable
RADARSAT	1995	SAR	microwave	20	“
IKONOS	2000	IKONOS	V, IR	1-4	“
KOMPSAT	2000	EOC, OSMI	V	6-800	“

Satellite remote sensing data are primarily categorized by their spatial resolution, which indicates the minimum level of detail recorded by sensor. Landsat-like satellites, having a spatial resolution of less than 100m, have provided enormous volume of imagery data for vegetation monitoring over relatively large geographic area. Although these satellites have provided relatively fine spatial resolution, their usage has been limited to the area between 60² km² and 180² km². If the study area is larger than the above size, several scenes of imagery should be stitched together. Processing several scenes of Landsat and SPOT data may cause problems related to the data acquisition, data calibration, and data processing. Furthermore, since this type of satellites has an orbit cycle of about 20 days, it is often difficult to obtain continuous measurements over the same area without cloud cover.

The second group of satellite data has relatively poor spatial resolution as compared to the first group. NOAA Advanced Very High Resolution Radiometer (AVHRR) data have been widely used to derive vegetation information at continental and global scales. In recent years, several satellite sensors, such as MODIS and SeaWiFS, similar to AVHRR were added. In fact, these new sensors have much finer spectral characteristics than AVHRR. These sensors have relatively short repetition cycle and large area coverage, they are very effective to analyze the temporal changes of vegetation dynamics over regional and continental scale. Next group of satellite data (RADARSAT, ERS) is active microwave sensors, which are capable to acquire imagery under cloudy or day-or-night conditions. Satellite imaging radar system is capable to penetrate canopy layer, which provides unique information related to the internal structure of standing vegetation. Furthermore, the reflected signal of imaging radar has something to do with the moisture content of the target. Such characteristics of SAR data has a great potential to be used for monitoring vegetation cover. The last group of satellite data has very high spatial resolution. These high resolution data are almost comparable to aerial photographs by their spatial resolution and narrow coverage and primarily designed for mapping of detailed topographic and thematic features. High resolution satellite data can be used for the detailed vegetation mapping of relatively small area.

3. Remote Sensing Vegetation Index and Its Uses

In many cases of vegetation remote sensing studies, vegetation index has been used as a primary source of information related to the biophysical characteristics of vegetation over large geographic area (Eidenshink, 1992; Loveland et al., 1991; Townshend and Justice, 1986). Vegetation index, derived from remote sensing data, is used as a single measure of such canopy characteristics as biomass, productivity, leaf area index, photo-synthetically active radiation, or canopy closure (Larsson, 1993). This technique of vegetation index was developed from the unique spectral characteristics of green vegetation in visible and near-infrared wavelengths. Figure 1 shows the spectral reflectance of normal green vegetation that is quite different from other surface features of soil or waters. Green vegetation has relatively low reflectance in visible wavelength and high reflectance in near-infrared spectrum (0.7 – 1.3 micrometers) while other surface types, such as bare soil and water, have similar reflectance in both spectrum. In addition, the spectral reflectance of green vegetation in these spectrum is very sensitive to the amount of chlorophyll content and canopy thickness (Hoffer, 1978). Healthy and fully developed vegetation canopy tends to have less reflectance at red spectrum and higher reflectance in near-infrared spectrum as compared to under-developed canopy condition.

Figure 1. Spectral reflectance of several surface features: (A) light color soil, (B) dark color soil, (C) green vegetation, and (D) water.

Most satellite multispectral sensors supply image data obtained at those two spectral bands of red and near-infrared spectrums. Vegetation index is a simple form of mathematical transformation to combine the two bands data into a scale to enhance the characteristics of vegetation. Although there are several methods of calculating vegetation index using two spectral bands, the normalized difference vegetation index (NDVI) has been most widely used in many fields of applied remote sensing community. NDVI is calculated by dividing the difference of two spectral reflectance values by the sum of two spectral reflectance values.

NDVI = (NIR – R) / (NIR + R)

Theoretically, NDVI value ranges from -1.0 to 1.0, in which the maximum value 1.0 suggests the most green vegetation. Non-vegetation features, where the reflectance in red and near-infrared spectrum are not much different, have the NDVI value close to zero. Figure 2 shows the seasonal variation of NDVI images over Korean peninsula. As mentioned before, since NDVI is correlated to the relative amount of green foliage, the winter months' images show the coniferous forests with light tone. Once the leaves come out in spring and summer, the deciduous forest has lighter gray level than the coniferous forest while the non-vegetated areas such as urban appear dark throughout the year.

Figure 2. Seasonal variation of NDVI images over Korean Peninsula.

Once we acquire NDVI images throughout the year, we should be able to classify different vegetation types by their growing pattern. It is particularly true for northeast Asia where the phenology of leaf development is quite different throughout the year. Figure 3 shows the temporal profile of NDVI values obtained in Korea (Lee, 1994). As expected, the NDVI of coniferous forest has a maximum value on the months of February, March, and November when no green foliage remains for other vegetation types. The mixed forest having substantial number of conifer trees has second highest NDVI values during the leaf-off season. Hardwood forest is distinguished by the high NDVI values from May to September. Two herbaceous classes of crop and grass have relatively low NDVI values during growing season. As can be seen from the figure 3, grass land has slightly higher NDVI value than the cropland. The NDVI profile of urban, which has minimal vegetation cover, shows the lowest NDVI values throughout all season. Temporal profile of the NDVI for a vegetation class might vary by several factors of geographic location, species composition, canopy density, and tree size.

Since the temporal patterns of NDVI values for each cover types are different, they can be separated by computer classification scheme (Derrien et al., 1992). Currently, it is quite common to produce a continental/global scale map of vegetation cover by using a series of multi-temporal NDVI data derived from coarse resolution satellite imagery, such as NOAA

Figure 3. Temporal variation of NDVI values for the six different cover types.

AVHRR. Figure 5 shows an example of vegetation cover map of northeast Asia, which was produced by a series of multi-temporal NDVI data and (Tateishi et al., 1997). It has about 40 vegetation classes and other land cover types. With increased number of earth observation satellites, this type of vegetation cover map can be constructed more frequently and used for several aspects of environmental management as well as for regional-scale ecological studies.

The use of NDVI is being expanded beyond the monitoring of seasonal and inter-annual variation of vegetation condition and the mapping of vegetation cover. Recent development of data analysis, it is now being used to extract rather quantitative and sophisticated information related to the biophysical characteristics of vegetation. The biophysical variables that are currently achieved from the satellite data derived NDVI data include leaf area index (LAI), net primary production, and a fraction of photo-synthetically active radiation (FPAR). These variables are continually produced with well-refined calibration and analysis procedures from the most recent sensor named as MODIS (Tian et al., 2000). Further, the use of these data is no longer limited and they are distributed freely.

4. Conclusions

Considering the large area coverage and the seasonal and inter-annual variation of vegetation cover in northeast Asia, satellite remote sensing data can be a very attractive alternative to monitor the vegetation conditions in this region. In recent years, several new and improved satellite sensors have been added and the data availability has increased for the use of regional scale environmental and ecological studies. It is, however, very important for ecologists to select the appropriate type of remote sensing data. Satellite remote sensing data vary by the spatial resolution and the repetitive data acquisition cycle. For the monitoring of vegetation cover in northeast Asia, the temporal resolution should be more critical factor than the spatial resolution. AVHRR or MODIS data would be suitable for regional scale studies since they can cover the entire world within two days. Further, the coarse resolution data can be easily accessed and the data analysis methods to extract vegetative information are well established.

Figure 5. Land use/cover map derived from multi-temporal AVHRR NDVI data.

References

Derrien, M., B. Farki, H. Legleau and A. Sairouni, 1992. Vegetation cover mapping over France using NOAA-11/AVHRR. International Journal of Remote Sensing, 13(9):1787-1795.

Eidenshink, J. C., 1992. The 1990 conterminous U.S. AVHRR dataset. Photogrammetric Engineering and Remote Sensing, 58(6):809-813.

Hoffer, R.M. 1978. Biological and physical considerations in applying computer-aided analysis techniques to remote sensor data. In Remote Sensing: the quantitative method, Swain and Davis (edt.), McGraw Hill, USA. pp. 227-289.

Larsson, H., 1993. Regression for canopy cover estimation in Acacia woodlands using Landsat TM, MSS, and SPOT HRV XS data, International Journal of Remote Sensing, 14(11):2129-2136.

Lee, K.-S., 1994. Estimation of regional CO2 fixation by forest vegetation using time integrated NDVI data. Proc. of 3rd Korea-Japan Symposium on "Environmental Monitoring from Space", Korea/Japan Joint Comm. for Basic Scientific Research, pp. 19-23.

Ouattrochi, D.A. and R.E. Pelletier, 1991. Remote sensing for analysis of landscapes: an introdution. In Quantitative Methods in Landscape Ecology, Turner and Gardner (edt.), Springer-Verlag, pp. 51-76.

Loveland, T. R., J. W. Merchant, D. O. Ohlen, and J. F. Brown. 1991, Development of a land-cover characteristics database for the conterminous U.S. Photogrammetric Engineering and Remote Sensing, 57(11):1453-1463.

Tateishi, R., W. C. Gang and L. K. Perera, Land cover classification sytem for continental/global applications,Asian Pacific Remote Sensing Journal, Vol.10, No.1, pp.1-9, July 1997

Tian et al., 2000. Prototyping of MODIS LAI and FPAR algorithm with LASUR and LANDSAT data. IEEE Trans. Geosci. Remote Sens., 38(5): 2387-2401.

Townshend J. R. G. and C. O. Justice, 1986. Analysis of thedynamics of African vegetation using the normalized difference vegetation index. International Journal of Remote Sensing,7(11):1435-1445.