List of Tables
4. REMOTE SENSING AND ITS APPLICATION
4.4 REMOTE SENSING USING LANDSAT
4.4.1 Reflectance characteristics of surface features
It is important to understand the reflectance characteristics of different surface features as this allows the most suitable bands to be chosen for image classification. For this study, the most important surface features are those associated with vegetation, wetlands and water.
Understanding how these surface features are measured by the Landsat Thematic Mapper will ensure that the most appropriate bands are used for classification.
The design and choice of Landsat TM bands has been largely influenced by two factors, namely, (a) the presence of atmospheric windows, and (b) the reflectance characteristics of the earth's surface features. Atmospheric windows are spectral regions in which the atmosphere is relatively transparent and where solar radiation can pass through the atmosphere without being absorbed (Schowengerdt, 1997). The location of these windows is determined by the absorption characteristics of the different atmospheric constituents
(e.g. oxygen, water vapour, ozone). The spectral bands of Landsat TM were designed to take advantage of these windows.
The reflectance characteristics of surface features have also been an important consideration in sensor design. For example, locating band 3 in the 0.63 - 0.69 urn region allows the chlorophyll absorption of plants to be determined, while band 4 (0.76 - 0.90 um) covers an area where water is a strong absorber and is therefore useful for delineating water bodies (Wang et al, 1998). Numerous studies have reported on the reflectance characteristics of different surface features and how these relate to the different TM bands.
(a) Vegetation
Many applications of vegetation mapping and monitoring have been described in the literature. Shoshany (2000) found that TM bands 3, 4, 5 and 7 are most suitable for differentiating between vegetation types because of the chlorophyll absorption and mesophyll reflections that occurs at these wavelengths. These reflections are most noticeable in the near-infrared band (TM4), which shows high reflectance values for riparian woodlands (Ringrose et al, 2003), forests (Eastman, 2001) and green vegetation (Lillesand & Kiefer, 2000). Most green vegetation shows a pattern of low reflectance in the blue (TMl), slightly higher in the green (TM2), lower again in the red (TM3) and very high in the near-infrared (TM4) (Eastman, 2001). Dead vegetation, on the other hand, shows high reflectance in both the red (TM3) and near-infrared (TM4) and can therefore be distinguished from living vegetation (Jensen et al, 1984). The near-infrared region, in addition to having high reflectance values for vegetation, is also a region where there is a large variation in reflectance by different vegetation types (Eastman, 2001). These variations are very useful for discriminating between different types of vegetation.
(b) Water
Water has a spectral reflectance pattern that is not often confused with other land cover types (Wickware et al, 1991). Reflectance by water is low in all TM bands, with the lowest values occurring at the longer wavelengths (TM4, TM5 and TM7) (Eastman, 2001; Braga et al, 2003). Indeed, in the infrared bands water has such a low reflectance that it appears to be almost black (De Haan et al, 1991). This property of water allows for the relatively easy delineation of water bodies using bands TM4 or TM5 (e.g. Wickware et al, 1991;
Frazier et al, 2003). Water does show some reflectance, albeit low, in the visible part of
the spectrum with highest values in the blue (TMl) and successively lower values occurring in the green (TM2) and red (TM3). This property has been used to study features in shallow waters (e.g. Braga et al, 2003; Shaghude et al, 2003).
(c) Wetlands and mires
The spectral reflectance patterns of wetlands and mires are generally rather complex due to the varying proportions of water and vegetation that occur. Wickware et al. (1991) used TM bands 3, 4 and 5 for classifying wetlands while Edgley & Werstak (1998) found that a combination of these same bands was useful for contrasting emergent vegetation with open water. Of these three TM bands, band 5 has been found to be the most useful in wetland detection (e.g. Lunetta & Balogh, 1999; Munyati, 2000; Boresjo Bronge & Naslund- Landenmark, 2002).
(d) Urban areas
Urban areas show varied reflectance patterns due to the different types of surfaces that occur in these areas. Non-vegetative features like buildings and concrete structures tend to have high reflectance across all the TM bands while gardens, trees and parks exhibit the reflectance patterns characteristic of vegetation (Eastman, 2001). In addition, the spatial heterogeneity of urban areas in relation to the 30-metre spatial resolution of Landsat TM means that many pixel reflectance values actually comprise a mixture of reflectances from different surface features (Small, 2001).
(e) Bare soils
Exposed sandy soil has high reflectance values across all the Landsat TM bands (Ringrose et al, 2003). This allows for relatively easy separation of bare sand from vegetation and water during image classification, but can cause some confusion with artificial structures that have similar reflectance characteristics. Organic soils and soils saturated with water show different reflectance patterns to dry sandy soil. Lunetta & Balogh (1999) reported that Landsat TM band 5 is useful for studying soil moisture content while bands 2 and 3 can be used to distinguish saturated surfaces from organic soils.
(f) Burnt areas
Newly burnt areas can have very low reflectance values in the visible and near-infrared bands. This can lead to classification confusion with open water, which has a similar
reflectance pattern (Munyati, 2000). This problem can be averted by visual inspection of the image or by comparison with images from another date.