1Agriculture Victoria, Department of Natural Resources and Environment, Victoria
621 Sneydes Rd, Werribee 3030 (03) 9742 8727-ph, (03) 9742 8700 -fax firstname.lastname@example.org
2Department of Soil and Water, University of Adelaide
Airborne high resolution imagery from the HyMap Hyperspectral Imagery System has been used to identify and map soils at the sub-paddock scale. The key objective of this research is to be able to provide land managers with soil information that can be used for paddock scale management and planning.
The HyMap system is a hyperspectral scanner and measures reflected solar radiation over the visible to short wave infra-red parts of the spectrum (420-2500 nanometers). The sensor makes many spectral measurements over this wavelength region (up to 128) which is considerably more than is offered by a satellite such as Landsat TM which has 6 broad bands over this region. The sensor also offers high spatial (on ground) resolution of 5 metres, again compared to 30 metres for the Landsat TM sensor. For this reason, HyMap, which has been used extensively for mineral mapping by mining companies, offers a lot of information that can be useful for mapping surface soils and their properties.
The imagery used was taken over the Barossa Valley area in April 1998, and much of the current study has focussed on the soil variability around Southcorp Pty Ltd Koonunga Hill vineyard. Detailed fieldwork has involved measuring the spectral properties of 128 soil samples in this area and this information is used to assist with the image processing. Preliminary results from image analysis has shown that HyMap data contains a wealth of information on soils, and this spectral variability can be related to the differences in surface soil texture, soil moisture and soil mineralogy. Information on soils extracted from the imagery has potential to be used in property planning, improved design of soil sampling programs, and to assist in interpreting yield variability maps.
Current landholders require soil information at the sub-paddock level, a scale at which informative land management decisions can be made, but often is not available from traditional soil mapping programs. The Australian developed HyMap Airborne Hyperspectral Imaging System acquires imagery at the resolution and quality that can be used in agriculture for discriminating and mapping soils at the sub-paddock scale.
Hyperspectral imaging refers to the simultaneous collection of images in many narrow, contiguous spectral bands (HyMap has 128 spectra bands) from the visible to the short wave infrared regions of the reflectance spectrum - Figure 1(Goetz et al., 1985). This differs from multispectral images which are collected in broad, often separated spectral bands, such as those acquired from Landsat TM which has 6 broad spectral bands. Hyperspectral imagers are able to remotely acquire a complete reflectance spectrum for each pixel within an image and this can be compared directly to laboratory or field collected spectra (Goetz et al., 1985). As hyperpectral imagers, such as HyMap, are airborne they also offer high spatial resolution (3-10 m on ground), making hyperspectral imaging an excellent tool to more accurately identify and map surface materials such as soils (Boardman & Kruse, 1994).
Figure 1: Hyperspectral imaging concept. A spectral signature can be obtained for every pixel in the scene (modified from Kruse, 1994; Staenz, 1992).
Hyperspectral imaging has been successfully used for mineral exploration and geological mapping in arid environments (see for example Boardman & Kruse, 1994; McCubbin et al., 1998, Keeling et al., 1998; Taylor et al., 1998). However the use of such imagery for mapping soils in temperate climates under a range of agricultural practices has been limited. Research that has utilised hyperspectral imaging for soil discrimination and mapping has largely been confined to the United States and Europe (see Hill et al., 1994; Palacios-Orueta & Ustin, 1996; Accioly et al., 1998).
This research investigates the spectral characteristics of soils over the Barossa Valley study area and evaluates using high resolution imagery captured by the HyMap imaging system for mapping soils and their properties over the landscape.
Study site and background data
The study was conducted in the Barossa Valley region of South Australia, located approximately 75 kilometres north east of Adelaide (Figure 2). The 5 km x 20 km HyMap scene used in the study runs north from the town of Nuriootpa to Bald Hill. The study area incorporates some of Australia's premier wine producing region, as well as supporting dairying, broad acre cropping and sheep farming. Much of the present study focused on Southcorp Pty Ltd's Koonunga Hill vineyard and surrounding areas, in the north central part of the HyMap scene.
Climate in this region is temperate, with an average rainfall of around 500 mm. The majority of the study area is relatively flat with elevations of 270-280 m a.s.l, becoming undulating to hilly in the north with elevations of 400 m in the vicinity of Bald Hill. Geology ranges from Quaternary and Tertiary sediments (south and central) and Palaeozoic and Proterozoic sedimentary rocks of shale, siltstone, sandstone and lesser exposures of limestone outcrop. Soils over much of the study area are Red and Brown Chromosols and Sodosols (Red-Brown Earths) and Black Vertosols (Black Earths) associated with lesser Tenosols and Dermosols.
* Koonunga Hill Vineyard site
Figure 2: Barossa Valley study area.
To gain an understanding of the soil variability over the study area, particularly at the Koonunga Hill vineyard site, a field survey was conducted during April 1999. Information collected from this survey was to be used to aid analysis of the HyMap imagery. 128 surface (0-5 cm) soil samples were collected, including 96 samples over the Koonunga Hill vineyard on a 60 m x 60 m grid and 31 samples representing a range of different soil units in the region (Figure 4).
Figure 3: A Red-Brown fine sandy clay loams at the Koonunga Hill vineyard and B Dark grey medium clay (right side of photo) and Grey-Brown sandy loams (back right).
Figure 4: Location of soil samples, with the HyMap imagery as a grayscale backdrop
Collection and analysis of soil spectra
Reflectance spectra were collected for all 128 soil samples over the study area. This was done to gain an understanding the spectral characteristics of soils and to use this knowledge to assist in interpreting the HyMap imagery. Furthermore an important reason for collecting the spectra was to determine if the spectra of a soil could be used to identify the presence or absence of particular properties that comprise that soil, such as moisture content, texture, mineralogy and colour. If this could be done then there was potential for the HyMap scanner to also identify some of these properties. Spectra were obtained for each soil using two spectrometers, the Lastek VNIR spectroradiometer that measures reflectance in the 600-1000 nm wavelength region and the PIMA II SWIR reflectance spectrometer that measures reflectance in the 1300-2500 nm wavelength region.
Figure 5. The PIMA II spectrometer used to collect reflectance spectra of soils for the 1300-2500 nm region, with an example of the output spectra after processing.
Prior to interpreting the HyMap imagery an analysis of the spectra collected from field samples was conducted. This involved examining the spectra of the soils to determine their main characteristics and to see if spectra varied for different soil types across the study area. Techniques used to do this were a visual inspection of the soil spectra, absorption band analysis, and statistics, including principal component and cluster analysis. Information on the reflectance spectra of the soils could then be used to assist in the image processing. Some of the key spectral features of soils over the study area is presented in the Results section.
HyMap imagery analysis
Hyperspectral imagery was acquired by the HyMap Airborne Hyperspectral Imaging System operated by Integrated Spectronics Pty Ltd over the Barossa Valley area on 3rd April 1998 for the University of Adelaide. Specifications of the HyMap system are presented in Table 1 below. Image analysis was conducted using ENVI image analysis software.
Table 1: Specifications of HyMap imagery for the Barossa Valley scene
20 km x 5 km
Several image analysis techniques have been applied to extract soil information from the HyMap imagery over the study area. These include:
- Principal components transformation was applied to define new uncorrelated bands that maximised the variance in the spectral data, this reduces the dimensionality of highly correlated spectral data and separates noise from information in the data;
- Band ratioing involves dividing one spectral band into another, resulting in an image of relative band intensities. Band ratios aimed to enhance the spectral information contained in key spectral bands and made full use of HyMap's narrow band widths and positions, and;
- Matched filtering is a technique used to separate target spectra (in this case spectra of soils) from the background (all other spectral data).
- Outputs from these image analysis techniques are presented below.
Soil spectral analysis
Results from the spectral analysis shows that overall the soils have a similar curve shape, with the presence of three strong absorption bands at 1400-1450 nm, 1900-1950 nm and 2200 nm (Figure 6). The first two of these absorption bands relates water molecules present in the soil and the 2200nm absorption band is due to the hydroxyl ion as a result of clay minerals in the soil (Baumgardner et al., 1985). Limited XRD results show that the soils contain montmorillonite and lesser kaolinite, which have diagnostic absorption features at this position.
Figure 6: Example of reflectance spectra for different soils over the study area, showing the position of key absorption features.
The most obvious difference in the spectra of the soils is the variation in albedo (overall spectral brightness) which is likely to reflect the moisture content of the soils and is essentially influenced by the texture of the soil. Lighter textured soils (sands-sandy loams) have a lower water holding capacity than the heavier textured soils (clays). Figure 7 shows the reduction in overall brightness due to the oven drying of soils; this is particularly the case in the 1300-2400nm region.
Figure 7: Reflectance spectra of soils with varying moisture contents (name responds to soil unit as mapped by Northcote, 1954 eg. F2, Moppa, Dimchurch; MC = Medium Clay, LS = Loamy sand, SL = Sandy loam).
The other key difference in the spectra of the soils is in the 600-1000 nm region, where the spectra of some soils have a sigmoidal shape (convex between 600-780 nm, concave 780-1000 nm) whilst others are straight (Figure 6). The convave shape centres around a weak absorption feature near 870 nm and is likely to be due to the presence of iron oxides in the soils (Baumgardner et al., 1985). Soils that have these spectra are usually redder (indication to the presence of haematite) in colour than soils that show no sigmoidal shape, and often have ferruginous gravel on their surface.
Comparison of soil and image spectra
A comparison was made before image analysis between field collected soil spectra and spectra extracted from the HyMap image for the same location (for locations refer to Figure 3). Figure 8 shows image and field spectra for selected soils collected over the study area. Comparison of the two spectra at each location found that the image spectra generally had higher reflectance than laboratory spectra. This difference is likely be due to differences in soil moisture content, the field samples were collected a year after the image was captured. Overall however the spectral shapes and positions of the main absorption bands for both field and image spectra are similar.
Figure 8: Comparison of field and image spectra of different soils over the study area.
HyMap image analysis
The research, to date, has applied a number of image analysis techniques aimed to extract soil information at the sub-paddock level. Information from the analysis of soil spectra was used to assist in the image processing. Research into soil feature extraction from the imagery is still in progress and this section aims to present some of the initial results.
A review of the principal component bands after the transformation had been applied indicates that many of these new bands contain information relating to the spatial variability of soils. Figure 9 is a colour composite of PC bands 21, 8 and 1, and have been combined to enhance soil variability over the Koonunga Hill vineyard. This soil information is being extracted from under mature vineyards (Figure 2). The lighter colours highlight areas where the soils, on the whole, have a heavier texture (light-medium clays) than for the rest of the block, areas with blue-purple colours (fine sandy clay loams). Also associated with the heavier textured soils, particularly in the southern part of the block, is abundant calcrete, ferruginous gravel and sandstone pebbles on the surface.
Figure 9: Colour composite image showing PC bands 21 (red), 8 (green) and 1 (blue) over Koonunga Hill vineyard, showing spatial variability of soils under mature vines.
Band ratios were successful in identifying soil variability over the region by enhancing the spectral information contained in key spectral bands. Figure 10 is an RGB colour composite using three band ratios to enhance soil information from vegetation. In Figure 10 the colour red is defined by the ratio 1643nm/603nm and was used to separate lighter textured soils (sands, loamy sands), which are bright due to the presence of abundant fine- medium quartz grains, from other soils and vegetation. Green is defined by the ratio 2152nm/846nm and was used to enhance soils/bare ground from vegetation. Blue is defined by the ratio 2306nm/2204nm and was used to distinguish heavier textured soils (clays), which are dark due to the higher moisture content of these soils, from healthy vegetation. The results from the ratios in this image show more detailed soil variability than previously mapped by regional soil surveys.
Figure 10: Colour composite of band ratios 1643nm/603nm (Red), 2152nm/846nm (Green) and 2306nm/2204nm (Blue). Soil units as mapped by Northcote, 1954 are overlain on the image.
An output from the Matched Filtering process is shown in Figure 11, white colours indicates a closer match between "target" spectra and image spectra. The "target" spectra used in this case is of a medium clay soil (similar to that in Figure 7), and the resultant map relates closely to the known occurrences of this soil as mapped by Northcote.
Figure 11: Distribution of heavy textured soils (light-medium clays) over the study area as defined by the process of Matched Filtering and the known occurrences of these soils as mapped by Northcote, 1954. Mismatched areas still require field verification.
Conclusions and discussion
Collecting and interpreting the spectra of soils in the field has provided valuable information on the spectral characteristics of soils in the Barossa Valley. Information gained from this initial work was important to assist in understanding the wealth of spectral information collected by the HyMap scanner. Overall spectral brightness of the soils, relating essentially to the moisture content, and the presence of key absorption bands reflecting the water, clay and possibly the iron content of the soils, were the main spectral properties that has enabled different soils to be identified.
Several image analysis techniques have been investigated to extract spectral information from the HyMap imagery relating to soils over the study site. Principal components and band ratios have been successful in defining the spatial variability of soils. The high spatial resolution of HyMap allows for these techniques to be able to map surface soil variability at a finer resolution than previous soil mapping. The matched filtering technique using spectra of known soils as "targets", and has been a useful tool for discriminating soils based on their surface texture. Future investigations aim to further establish relationships between HyMap's spectral data and the properties of soils, particularly soil colour, texture, mineralogy and moisture content. Although hyperspectral imaging systems like HyMap only measure reflectance from surface materials, soil information extracted from the surface can be utilised by land holders to assist in classifying their land and managing different soil units at the sub-paddock level accordingly.
HyMap imagery over Barossa Valley has been kindly provided to the University of Adelaide by Integrated Spectronics Pty Ltd. Access to the Koonunga Hill vineyard and to unpublished soil data by Southcorp Pty Ltd is greatly appreciated. Primary Industries South Australia are thanked for the use of their PIMA II spectrometer for measuring soil spectra, and access to unpublished 1:100 000 scale soil mapping.
Accioly, L., Huete, A.R., and Batchily, K. (1998). Using mixture analysis for soil information extraction from an AVIRIS scene at the Walnut Gulch experimental watershed-Arizona. AVIRIS Airborne Geoscience Workshop.
Baumgardner, M. F., Silva, L.F., Biehl, L.L. and Stoner, E.R. (1985). Reflectance properties of soils. Advances in Agronomy 38: 1044.
Boardman, J. W. and Kruse, F.A. (1994). Automated spectral analysis: A geological example using AVIRIS data, North Grapevine Mountains, Nevada. Proceedings of Tenth Thematic Conference on Geological Remote Sensing, San Antonio, Texas.
Goetz, A. F. H., Vane, G., Solomon, J.E. and Rock, B.N. (1995). Imaging spectrometery for earth remote sensing. Science 228: 1147-1153.
Hill, J., Mehl, W. and Altherr, M. (1994). Land degradation and soil erosion mapping in a Mediterranean ecosystem. In J. Hill and J.Megier (eds) Imaging Spectrometery- a tool for Environmental Observations. Dordrecht, Kulwer Academic Publishers 237-260.
Keeling, J. L., Mauger, A.J., Horsfall, C., and Crettenden, P.P. (1998). Defining South Australian magnesite resources using high resolution airborne spectrometry and survey grade GPS. Ninth Australasian Remote Sensing Conference, Sydney.
Kruse, F. A. (1994). Imaging spectrometer data analysis- a tutorial. International Symposium on Spectral Sensing Research '94 (ISSSR)., San Diego, CA.
McCubbin, I., Green, R., Lang, H. and Roberts, D. (1998). Mineral Mapping using partial unmixing at Ray Mine, AZ. AVIRIS Airborne Geoscience Workshop.
Northcote, K.H., Russell, J.S. and Wells C.R. (1954). Soils and land use in the Barossa District, South Australia, Zone 1: The Nuriootpa Area, Soils and Land use Series No.13 CSIRO.
Palacios-Orueta, A. and Ustin, S.L. (1996). Mulivariate statistical classification of soil spectra. Remote Sensing Environment 57: 108-118.
Staenz, K. (1992). A decade of imaging spectrometry in Canada. Canadian Journal of Remote Sensing. 18: 187-197.
Taylor, G. T., Reston, M. and Cocks, T. (1998). Spectral endmembers determined from HyMap imaging spectrometry. Ninth Australasian Remote Sensing Conference, Sydney.