MAPPING AND MONITORING OF ARID LAND VEGETATION IN JORDAN USING ATSR-2

ABSTRACT

The narrow wavebands (0.545-0.565, 0.649-0.669, 0.855-0.875, 1.58-1.64 m) and dual look angle (nadir and 55 forward) of ATSR-2 should enhance the ability to detect vegetation of very low densities from remotely sensed data. This paper looks at the application of ATSR-2 to the mapping and monitoring of sparse vegetation in the eastern desert region of Jordan, an area used extensively for grazing. Imagery was compared with contemporaneous field data. Results indicate that indices utilising combinations of near infrared/red reflectance are of limited use for differentiating vegetation due to bright soil backgrounds; this limits the ability to extract a characteristic spectral response of dry chenopod vegetation, particularly at vegetation densities of less than 10 %. An alternative approach is to simulate the bi-directional reflectance distribution function (BRDF) of ATSR-2 pixels over dryland surfaces in order to develop a model which can be inverted to estimate vegetation characteristics. Ongoing research considers the use of linear mixture modelling incorporating different look angles as data layers and the potential information held in the directional dependence of red/near infrared feature space.

INTRODUCTION

The Along Track Scanning Radiometer (ATSR-2) was launched on-board ERS-2 in April 1995. Designed as a research satellite, ATSR-2 provides the opportunity to study the effects of off-nadir viewing particularly in relation to vegetation mapping and monitoring. Using a conical scan mechanism, locations on the Earth's surface are scanned twice - at nadir and at a forward look angle of 55. Narrow wavebands in the visible and near infrared (0.545-0.565, 0.649-0.669, 0.855-0.875 m) provide detailed information on those areas of the electromagnetic spectrum where the influence of vegetation is most commonly detected.

This paper describes current research examining the usefulness of ATSR-2 data for the study of vegetation in arid environments. In these regions, vegetation is characteristically of low densities exhibiting wide spatial and temporal variations. Off nadir viewing of such areas using ATSR-2 should enhance the detection of sparse vegetation. Furthermore the coarse resolution (1km), relatively large swath (500 km) and 6 day coverage cycle should help capture temporal and spatial variability over a large area. Overgrazing and land degradation are problems found in many arid areas. These are areas that equally face population growth and increased pressure on finite resources. It is therefore important to study these areas effectively mapping their past histories, understanding their development and making informed predictions for the future.

VEGETATION MAPPING AND MONITORING

The use of vegetation indices to detect and map vegetation is widely accepted. Such indices focus on the use of red and near infrared wavebands where differential reflectance by photosynthetically active vegetation gives a low red and a high near infrared response. Indices range from a simple red/near infrared ratio to more complicated indices that include terms relating to background soil conditions and atmospheric effects. The Normalised Difference Vegetation Index (NDVI) is one of the most commonly used vegetation index and numerous examples can be found of it's application in drylands. Kennedy (1989) used the NDVI calculated from Advanced Very High Resolution Radiometer data to study vegetation in Tunisia, whilst Ringrose and Matheson (1987) considered its application to rangeland in Botswana. The Soil-Adjusted Vegetation Index (SAVI), a modified version of the NDVI was proposed by Huete (1988). It 'calibrates' a vegetation index, effectively normalising differences in soil substrate, thus allowing a more accurate estimate of vegetation cover. Equations for the NDVI and SAVI are given below:-

NDVI = SAVI =

where L = 1 for low vegetation densities.

Results by Gutman (1991) show an increase in NDVI in a forward look direction using AVHRR data. Coincidence of the forward look of ATSR-2 with the direction of forward scatter should increase vegetation indices. An oblique view will also mean that proportionally the sensor views more vegetation - vegetation indices should be higher.

An alternative to the use of vegetation indices is to adopt a more direct modelling approach. Modelling the spectral response of a surface can take a number of different forms from theoretically based models and the use of radiative transfer and geometric optics to more empirically based models.

Modelling the spectral response of vegetation has been attempted by a number of authors. Many such as Rosema et al. (1992), Hall et al. (1995) and Li and Strahler (1985) consider forest canopies. The methods and assumptions they use relate well to trees but are more difficult to apply to the shrubby vegetation found in arid regions. Modelling vegetation in an arid environment should have advantages over a similar task in a tropical or temperate environment. The vegetation is characteristically sparse and fairly uniform in shape. A common geometric shape (e.g. spheres or cones) can be assumed and the proportion of shadow is easily calculated. Due to single plants on a soil background, multiple scattering between different canopy layers can be ignored. Having said that, in this environment, background soil conditions are considerable and pure homogenous pixels from which to extract pure endmembers are difficult to find.

In this study, an initial attempt at modelling the spectral response of vegetation in arid regions has focused on the method described by Jasinski (1996). Jasinski's model is essentially a hybrid model using a physical modelling approach to interpret the information held in the red/near infrared feature space derived from a single multispectral satellite image. The canopy is considered as geometric elements randomly positioned on a horizontal soil surface. Using a reflectance model, the reflection of a given pixel can be described in terms of canopy transmittance and proportions of illuminated and shaded canopy, and illuminated and shaded background (see equation 1 below).

() = [m_i + m_S ()]{()[1-²()] + _gi ()²()} + _gi()[g_i + g_S ()] (1)

where = the wavelength of the band centre

m_i = fraction of illuminated canopy

m_S = fraction of shaded canopy

g_i = fraction of illuminated ground

g_S = fraction of shaded ground

= bulk canopy transmittance

_gi = illuminated ground reflectance

= canopy reflectance at zero transmittance

Within this equation, each reflectance term is considered to have a random or Poisson distribution. This moves away from the commonly made assumption of constant endmembers and takes account of spatial variability relating to differences in soil texture, organic content, soil moisture, canopy height and leaf area. Treating plants as randomly distributed objects on a planar surface allows the estimation of ground shadow, g_S, in terms of , a nondimensional solar geometric similarity parameter equating to the ratio of the mean shadow area cast by a single tree to the mean projected area.

g_S = 1 - m - (1 - m ) ^{+ 1} (2)

m_S is similarly defined in terms of , a constant that is derived from plant geometry and the solar zenith angle. The values for bulk transmittance and zero transmittance are estimated using values held in the red/near infrared scattergram. For more details on parameter derivation, see Jasinski (1989, 1996). Jasinski's model estimates the variables in equation 1 and uses them to predict lines of constant canopy cover. This method is invertible and little or no ground information is needed.

THE STUDY AREA

Research is being carried out in association with the Jordan Badia Research and Development Programme, a collaborative project between the Royal Geographical Society in the U.K. and the Higher Council for Science and Technology in Jordan. This programme focuses on an area of 11,210 km² in northeast Jordan between Syria to the north and Saudi Arabia to the south. A basalt regolith overlies the northern half of the area whilst the south is covered by cherts, limestone and other sedimentary rocks. The Badia experiences a generally arid desert climate with a mean annual rainfall of between 200 mm in the north and 50 mm in the south, and a potential evaporation of 1500 mm to 2000 mm per annum (Al-Homoud et al. 1995). Vegetation in the area is spatially variable occurring naturally on very low angle gravel fans that extend out from wadis (known as marabs) into seasonally inundated mudflats (known as qaas).

METHODOLOGY

FIELD DATA COLLECTION

Fieldwork was carried out in November-December 1995 and March-May 1996. A Landsat TM image was used to locate large homogeneous vegetated sites - seven vegetated sites were visited on each occasion and measurements taken of percentage vegetation cover, biomass, plant dimensions, density and spacing, soil moisture, soil type and ground radiometric properties. Three random 30m quadrats were sampled within a central 1 km² sampling frame. Within each quadrat, vegetation measurements were taken along ten 20m transects. The length of transect covered in vegetation in relation to the total transect length was used to give an estimate of percentage vegetation cover.

IMAGE ACQUISITION AND PRE-PROCESSING

An ATSR-2 image dating from 1st December 1995 was used in this work. The image was obtained as gridded brightness temperature (GBT) products. It is 512 * 512 km and contains information on seven spectral wavebands in the nadir and forward look directions. The geolocation information supplied with the product was used to apply a geometric correction. Having converted digital counts to top of atmosphere reflectances using supplied calibration tables, an atmospheric correction was carried out. A detailed description of the method of atmospheric correction can be found in Mackay and Millington (1997).

Following geometric and atmospheric correction of the imagery, vegetation indices were calculated for the areas of field study. Each index was calculated using values in a 3 * 3 grid of pixels. This includes the central 1 km sampling frame and ensures accurate location of field sites (Justice and Townshend 1981).

In order to apply Jasinski's model of pixel reflectance, a computer programme was written in C to solve equation 1. Equation 1 was used to calculate red and near infrared reflectances at different vegetation coverages and soil reflectances. Table 1 gives details of the input parameters used. As an initial form of analysis, many of the figures are taken from Jasinski's own work in Walnut Gulch, Arizona. Vegetation in Walnut Gulch is shrubby with very little herbaceous understory. It is structurally similar to that of the Badia. Running the computer model gives a series of data pairs as output. These can be plotted and superimposed on the red/near infrared scattergrams giving lines of increasing vegetation cover.

RESULTS AND DISCUSSION

Figure 1 shows the results obtained from the application of vegetation indices to the ATSR-2 data. Correcting for the soil effect using SAVI decreases the index but both the NDVI and SAVI show that no obvious relationship exists between the indices and percentage vegetation cover. Ringrose and Matheson (1987) describe how an increase in vegetation cover in arid environments is often associated with a darkening effect due to leaf absorbency and the effects of shadowing. In this case, vegetation amounts are too small to manifest a relationship in any direction. Figure 1 does however indicate a difference between the forward and nadir bands of ATSR-2. This difference is not always positive. As discussed in Edwards et al. (1996), many factors (e.g. resolution, soil reflectance and the nature of the vegetation) influence the sensor recorded signal. Using vegetation indices, the effect of vegetation on the off nadir signal cannot be ascertained from the imagery at the very low coverages examined here.

Figure 2 shows the red/near infrared scattergram derived from an ATSR-2 image of Jordan. Percentage vegetation cover lines imposed are those derived using Jasinski's model. The scattergram shows a characteristic triangle, and a soil line such as that described by Baret et al. (1993) can be identified. With reference to the image, values lying to the right of the soil line represent areas of snow accumulation on the Jebel Drouz and water such as the Dead Sea, whilst those to the extreme left correspond to sites of high vegetation cover in the Jordan Valley. Field sites in the Badia where vegetation cover is known all correctly lie between the soil and 10 % cover lines.

The question arises as to what extent the location of points between 0 and 10 % cover can be attributable to vegetation. As Jasinski (1989) describes, the soil line pivots around a mean according to such factors as soil moisture, soil mineralogy and shadow. These are factors, which vary across the Badia region. Future work will aim to follow Jasinski's model further, calculating probability functions relating to the many combinations of soil reflectance, vegetation cover, transmittance and reflectance that yield the same red/near infrared data pair. Scene simulation should make it possible to establish the point at which vegetation reflectance dominates over soil reflectance.

Paramount to this work is the hypothesis that viewing in a forward look direction will enhance vegetation detection. The acquisition of more ATSR-2 data will allow Jasinski's model to be extended to compare scattergrams derived from forward and nadir looks. Considering viewing geometry in relation to AVHRR, Gutman (1991) stresses the need for bidirectional surface models that incorporate the fact that vegetation is anisotropic. Further fieldwork will study the nature of plant response in arid environments and gain field measurements of factors such as canopy transmittance to input into Jasinski's model. The accuracy of the parameters derived from feature space must be tested.

The influence of resolution will also be studied. Jasinski's work was carried out with reference to a Landsat TM scene. Transferral of this method between field measurements at the metre scale, Landsat TM at the 30m scale and ATSR-2 and AVHRR at the kilometre scale needs to be investigated. Advantages to vegetation detection gained by using the forward look at ATSR-2 may be negated by the coarse resolution of the sensor.

Research using ATSR-2 data and field sites in the Badia region of Jordan has shown that the sparse vegetation cannot be detected using conventional vegetation indices. Through sampling on a pixel by pixel basis, vegetation is not identified. Consideration of subpixel processes is needed.

There are many ways of modelling arid land vegetation. Work presented here gives the results from the initial consideration of Jasinski's model (1996). Future work needs to take this further considering feature space boundary probabilities and the application of the model to the forward look bands of ATSR-2. It must also be recognised that modelling simplifies complex processes. There will be a degree of error involved. Future field work must aim to test the accuracy of assumptions made comparing simulated data with actual ground measurements and imagery.

ACKNOWLEDGEMENTS

Marianne Edwards is supported by a Leicester University Studentship. Imagery was obtained under an ESA PI to White and Millington (A02.UK.125). The authors wish to express their thanks to Michael Jasinski at the NASA Goddard Space Flight Center, Washington for his advice and help with model implementation, to George Mackay, Kevin Tansey and Kevin White for their help with fieldwork and to all at the Higher Council for Science and Technology, Amman, the University of Jordan and Safawi Field Centre for their assistance with field campaigns.

BIBLIOGRAPHY

Al-Homoud, A.S., Allison, R.J., Sunna,B.F., and White,K., 1995, Geology, geomorphology, hydrology, groundwater and physical resources of the desertified Badia environment in Jordan. Geojournal 37 pp 51-67.

Baret, F., Jacquemoud, S., and Hanocq, J.F., 1993, The soil line concept in remote sensing. Remote Sensing Reviews Vol. 7. pp 65-82.

Edwards, M.C., Al-Eisawi,D., and Millington,A.C., 1996, The use of ERS ATSR-2 data for monitoring rangeland vegetation in the eastern Badia, Jordan. Proceedings of the 22nd Annual Conference of the Remote Sensing Society 11-14th September 1996. University of Durham.

Gutman, G.C., 1991, Vegetation Indices from AVHRR: An update and future prospects. Remote Sensing of Environment Vol. 35. pp 121-136.

Hall, F.G., Shimbakuro, Y.E., and Huemmrich,K.F., 1995, Remote sensing of forest biophysical structure using mixture decomposition and geometric reflectance models. Ecological Applications Vol. 5 (4). pp. 993-1013.

Huete, A.R., 1988, A soil-adjusted vegetation index. Remote Sensing of Environment Vol. 25 pp. 295-309.

Jasinski, M.F. and Eagleson, P.S., 1989, The structure of red-infrared scattergrams of semivegetated landscapes. IEEE Transactions on Geoscience and Remote Sensing Vol. 27. No. 4. pp. 441-451.

Jasinski, M.F., 1996, Estimation of subpixel vegetation density of natural regions using satellite imagery. IEEE Transactions on Geoscience and Remote Sensing Vol. 34. No. 3. pp. 804-813.

Justice, C.O. and Townshend, J.R.G., 1981, Integrating ground data with remote sensing, in Townshend , J.R.G, (ed), Terrain analysis and remote sensing, London, George Allen and Unwin.

Kennedy, P.J., 1989, Monitoring the phenology of Tunisian grazing lands. International Journal of Remote Sensing Vol. 10. Nos 4 and 5. pp. 835-845.

Li, X, and Strahler, A.H., 1985, Geometric-optical modelling of a conifer forest canopy. IEEE Transactions on Geoscience and Remote Sensing Vol. 32. No. 5. pp. 705-720.

Mackay, G, and Millington, A.C., 1997, Application of a dual angle atmospheric correction for ATSR-2 solar reflecting channels, submitted to International Journal of Remote Sensing February 1987.

Ringrose, S and Matheson, W., 1987, Spectral assessment of indicators of range degradation in the Botswana hardweld environment. Remote Sensing of Environment Vol. 23. pp 379-396.

Rosema, A., Verhoef, W., Noorbergen,H., and Borgesius, J.J., 1992, A new forest light interaction model in support of forest monitoring. Remote Sensing of Environment Vol. 42. pp. 23-41.

Table 1: Input parameters used to apply Jasinski's model

Figure 1: Percentage vegetation cover plotted against NDVI and SAVI

Figure 2: Red/Near infrared scattergram derived from ATSR-2

Keywords: ESA European Space Agency - Agence spatiale europeenne, observation de la terre, earth observation, satellite remote sensing, teledetection, geophysique, altimetrie, radar, chimique atmospherique, geophysics, altimetry, radar, atmospheric chemistry