The concept of a Global Triangle Model developed from AVHRR data

The concept of a Global Triangle Model developed from AVHRR data

Jiang Li
Arizona Remote Sensing Center, University of Arizona, USA

Abstract
A global triangle model is proposed based on the characteristics of the data structure observed in spectral space. The research hypothesis is that a triangle shaped spatial arrangement of data can be observed in two dimensional spectral space as long as three conditions are met : (1) a; “Temperature” Channel and a “vegetation channel are used to produce the scatter graph, (2) the image is large enough to include a variety of landform and vegetation types, and (3) the sensors are radio metrically calibrated for vegetation, soils, water bodies, and various earth surface materials. A global triangle model is hereby derived to simulate the amount of green vegetation and the surface condition corresponding to its thermal, or ecological environment. The three corners of conditions corresponding to its thermal, or ecological environment. The three corners of the global triangle are arid terrain, full vegetation, ad deep water. Every site on the earth surface is located somewhere within the triangle according to its thermal, moisture and biomass conditions. The model could be used to help to understand how a basic eco-system is organized and operates. The triangular relationship could also be used for broad vegetation class discrimination. Temporal comparison of the triangle models at the same location may provide a mean of change detection.

Introduction
Technological advancement in remote sensing has given us tool to study the earth as a system. We can now gain comprehensive knowledge, not only of separate components, but also of the interrelation and interaction between them. This study proposes a global triangle model based on the characteristics of the data structure observed in spectral space of remotely sensed data. This model is derived to synthesize surface components and correlate earth system dynamics. The development of the global triangle model is an attempt to examine vegetation and terrain in relation to their ecological and terrain in relation to their ecological environment, and better understand the surface of the earth as an integrated system.

Vegetation and Terrain Studies in Remote Sensing
The spectral band between 0.63 and 0.69mm is known as the chlorophyll absorption band within this spectral range, the soil radiance is at a maximum and vegetation response is at a minimum. This indicates that a band within the red portion o the visible spectrum is a sensitive indicator of green vegetation. Vegetation reflectance increases rapidly and becomes significantly higher than soil reflectance in the near infrared (NIR) region between 0.75 to 1.2mm. Thus, the NIr band can also be used to distinguish vegetation from the soil background Tucker, 1978). The ratio of the NIR band to the red, first defined by Jordan (1966), yields an index that is highly sensitive to band to the red, first defined by Jordan (966), yields an index that is highly sensitive to green vegetation. Deering (1975), proposed the Normalized Difference Vegetation Index (NDVI) : (NIR – Red) / (NIR + Red.). These and other vegetation parameters including leaf are index (LAI), biomass, percentage of vegetation cover and productivity (Tucker, 1979). Another spectral wavelength region used for vegetation and soil studies is the thermal infrared (TIR). Healthy plants are in energy equilibrium with their thermal environment. Their temperature and water status are adjusted constantly in a way best adaptable to environment changes. TIR measurements can indicate differences in vegetation type, cover, turgidity, and morphology s well as background soil moisture and other physical properties (Jackson 1977). The Canopy Temperature Variability Index (CTV) was infrared thermometer sensed canopy temperature within a field. Gardener et al. (1981) found the standard deviation of mid – day canopy temperature is an useful indicator for irrigation scheduling. The aim of studies using thermal indices is to obtain insight into the physiologic condition of plant relation to their surrounding environment, rather than quantification of vegetation biomass.

Methodology
The prime sensor on board the National Oceanographic and Atmospheric Administration (NOAA) earth orbiting satellites is the Advanced Very high Resolution Readiometer (AVHRR). Data from the red and NIR channels are conventionally used to generate vegetation indices as discussed earlier for agricultural and natural resources monitoring. Data from canners sensing in the 105 – 12.5 mm range can be used to determine surface thermal parameters (March 1976). The AVHRR data are provided at full resolution from (1.1km) called Local Area Coverage (LAC) or at degraded resolution form (4 km) called Global area coverage (GAC).

Five LAC AVHRR scenes of West Africa acquired on January 3 and 23, February 11 and 22 and March23, 1989 were used in this study. The West African images, consisting of more than three million pixels 9each pixel 1:1 km x 1.1 km), cover a large portion of the African Continent, including deep ocean, inland lakes rivers, desert and other landforms, and various types of vegetation and crop cover. A location map of the central portion of the image area is shown in fig. 1

Figure 1 Location Map
Digital numbers (DN) stored on computer compatible Tape (CCT) were read into the PC-based image processing package ERDAS and Macintosh-based graphics software. A scatter plot was produced using data from channels 2 and 4 of the AVHRR LAC scene of January 3, 1989. the scatter graph is global view” of a part of west Africa in NIR and TIR space. The data spread appears a triangle. The triangular data distribution was also observed when the whole scene was cut in half, but still contained the variety was also observed when the whole scene was cut in half, but still contained the variety of landscapes. The same phenomena were also observed in January 23, February 11 and 22 and March 23 data sets despite different could conditions which changed the shape of the triangles. AVHRR thermal channels normally have negative calibration slope coefficients that convert the thermal DNs to temperature values. The negative slope coefficients mean that the higher the thermal DN, the lower the real surface temperature.

An subset image of 501 x 519 pixels was extracted from the scene to reduce the extensive portion of the Atlantic Ocean and to avoid heavy clouds. Figure 1 shoes the area covered by the subset on a nearly cloud free day. To reduce data processing, a skip factor of 10 was used to sub sample the subset in horizontal and vertical directions. consequently, the amount of data was reduced to 1% of the original subset. Resulting data are plotted in NIT-TIR space to obtain the familiar triangular distribution (figure 2). The data triangle was also observed using AVHRR bands 2 and 5, e.g replacing had 4 with band5, since bands 4 and 5 are both thermal channels in the adjoining spectral ranges of 10.5 – 11.5 mm and 11.5 – 12.5 mm respectively (see figure 3).

Figure 2 Band 2 Versus Band 4

Figure 3 Band 2 Versus Band 5
NDVI is a widely used vegetation index, and exhibits a positive correlation with vegetation content on a continental scale. Scatter plots of the subset between NDVI and channels 4 and 5 were generated as illustrated in figures 4 and 5. The plots still appear as triangles, but changes in shape and orientation become apparent.

Figure 4 NDVI Versus Band 4

Figure 5 NDVI Versus Band 5

Figure 6 Global Triangle Model
To illustrate the effect of heavy clouds, February 22 data set geometrically registered to the January 3 data set. The same subset of 509 x 519 pixels was extracted from the scene, and normalized to the January 3 data set using the darkest and brightest objects on the subsets. Scatter plots using the processed February 22 data are included as figures 7 and 8. The global triangles on February 22, 1989, strongly influenced by extensive clouds, still keep approximately the same size and locality.

Figure 7 Band 2 Versus Band 4

Figure 8 Band 2 Versus Band 5
Four small sites (deep ocean, bare soil and sparse vegetation, dry grassland on the edge of the Sahara Desert, and dense vegetation close river networks) were selected based on visual interpretation of the image separate scatter graphs. The locations of the four sites relative to the global triangles are indicated in figure 2 and 4.

Results and Discussions
In the NIR-TIR triangle as illustrated in figure 2, soil pixels occur at the bottom left corner of the triangle where minimum vegetation and high temperature are common. The vegetation clusters (site 1 and site 4) lay at the top of the triangle where the pixels are characterized by high vegetation and relative low temperature. Water pixels crowd around the bottom right corner of the triangle where temperature and vegetation are both low. Every pixel of the subset image is associated with the certain location within the triangle according to its thermal and NIR measurements. Discussed earlier in this paper. NIR is usually an indicator of green vegetation, and AVHRR band 4 is designed such that the recorded DN is linearly related to the observed thermal radiance. So, the spectral space of NIR and TIR is, infact, a vegetation temperature space, describing the relation between green biomass admits thermal environment. The hump on the right side of the triangle is due to the effect of clouds, (figure 3). Cool clouds shift the cloudy land and water pixels in eh scatter graph towards lower temperature.

The global triangle is rotated clockwise in the NDVI and TIR spectral plane as shown in figure 4. The high vegetation corner is tilted towards lower temperature regions, which is often true because of the cooling effect of green vegetation. The desert corner is lifted a little above the water corner since some vegetation, more or less, is always found in desert areas. it also can be seem in figure 4 that the clouds effect were minimized in the NDVI and TIR model because NDVI rationing helped to remove a part of the cloud influence (Schowengerdt, 1983). The basic relationship and ditirbution of the triangular structure stays the same, and the thermal and NDVI and TIR space. A possible explanation for this improvement is that the NDVI is a better vegetation indicator than the NIR band alone.

In Figure 4, temperature increases from right to left on the bottom with a range of temperature from deep ocean to hot desert. Vegetation increases from bare ground to full green cover. Understanding the right leg of the triangle requires additional study. It is possibly related to moisture content of the ground surface which increases from top to bottom, corresponding to full vegetation partial canopy with dry and wet background soils, open water etc. these relationships thus suggest a global triangle model to simulate the consistent observations of the triangle relationship of the earth surface components and dynamics. As illustrated in figure 6, the model has three corners of arid desert full vegetation, and deep ocean and three legs of temperature, moisture, and biomass.

Surface temperatures varies most widely in non-vegetated areas (along the bottom line of the triangle), the lowest temperature is normally found in water bodies during and other bare ground lay somewhere in between. In vegetated areas, plants absorb soil moisture through their roots, transport it to leaves, where it evaporates. The evapotranspiration has a cooling effect, which keeps canopy temperature as constant as possible (Ids 1976). This could explain why temperature varies the most at the bottom of the model where vegetation is at a minimum, and why temperature varies the most at the bottom of the model where vegetation is at a minimum, and why temperature varies the most at the bottom of the model where vegetation is at a minimum, and why temperature varies little at the top, where vegetation is high. It can be seen that the length of the bottom leg of the triangle is the range of maximum temperature variability on the surface of West Africa. The temperature variability decreases as vegetation increases, and becomes zero at the point of full vegetation.

The biggest vegetation diversity is usually found at the high and moderate temperature regions where abundant energy and sufficient moisture are easily obtainable to support photosynthesis. Millions of species are actively growing in tropical and temperate lands, forming a vigorous and complex bio-system. In cooler regions, fewer species are usually found in a less diverse environment. This is due to insufficient energy supply and excessive moisture. Similarly, the length of the left leg of the triangle may be related to maximum vegetation diversity of the surface eco-system found in West Africa. The vegetation diversity decreases to about zero as temperature decreases to that of the deep ocean.

Moisture availability is the highest in cool areas, varying fro water body, swamp, and different level of wet to dry soils. In hot areas, strong evaporation leaves limited or no water on the ground surface. This is why less and less moisture variability can be observed towards the desert corner of the triangle. Again, the length of the right leg of the triangle may be related to maximum moisture variability. The moisture variability decrease as temperature increases and vegetation decreases, and reaches zero at arid desert.

The association of the temperature variability, moisture availability, and vegetation diversity is a theoretical base to explain why the global view of earth surface in the spectral has a triangular shape.

Conclusion
This study has investigated the repetitive observations of the global triangle in the two dimensional spectral space. Three conditions are required to identify this triangle: (1) a thermal channel and a ”vegetation” channel are used to produce the scatter graph, (2) the image is large enough to include the global features, i.e. various land cover and vegetation types, and (3) the infrared sensors are radiometrically calibrated to a range corresponding to the radiance level of plants and soils.

The global triangle is nothing more than a project of earth system in spectral space. Every pixel of multispectral image corresponds to a certain location within the triangle according to its temperature, moisture, and biomass conditions. It bears some analogy with the Tasseled cap Model in visible and near-infrared space proposed by Kauth and Thomas (1976).

The model could be used to help to understand how a basic eco-system is organized and operated. The triangular relationship could be used for broad vegetation class discrimination. Temporal comparison of the triangle models at the same location may provide a mean of change detection.

References

Clawson, K.L, and B.L. Blad, 1982, Infrared thermometry or scheduling irrigation of corm, Agron, J., 64:603-608
Deering, D.W. J.W Rouse, Jr., R.H. Haas and J.A. Schell, 1980 Mesuring’ “forage production” of grazing units from Landsaat MSS data, I Proceedings of the sysmposium of Remote Sensing of the Environment, pp. 1169-1178, university of Michigan Press, Ann Arbor.
Gardner, B.F., B.L. Blad, and D.G. Watts, 1981, Plant and air temperatures in differentially irrigated corn, Agric, Meteorol,25:207-217.
Idso, S.B. and Ehrler, W.L. 1976, Estimating soil moisture in the root zone of crops : a technique adaptable to remote sensing, Geophys, Res, Lett., 3:22-25
Kauth, R.J, and Thomas, G.S. 1976, the Tesseled cap-a graphic description of the spectral-temporal development of agriculatural corps as seen by Landsat, Proc. Saymp. On Machine of Remotely Sensed Data Purdue University, West Lafayatte, Indiana, pp. 41-51.
Jackson, R.D., R.J. Raginato, and S.B. Idso, 1977, Wheat Canopy temperature: a Practical tool for evaluating water requirements, water Resources Res., 13:651-656.
Jordan, C.F., 1966,Derivation of leaf area index from quality of light on the forst floor, Ecology, 50:663-666
Marsh, S.E. and R.J. Lyon, 1976 Field Mapping for Heat Capacity Mapping Determinations : Ground Support for Airborne Thermal Surveys, Stanford Remote Sensing Lab Technical Report 76-3.
Schowengerdt, R.A, 1983, Technique for image processing and classification in remote Sensing, Academic Press pp. 154-156.
Tucker, C.J. 1978, A comparison of satellite sensor bands for vegetation monitoring, Photogrammetric Engineering and Remote Sensing, 44:1369-1380.
Tucker, C.J. 1979, Red and photographic infrared linear combinations for monitoring vegetation, Remote Sensing of Environment 8:127-150