Title: Exploring the Potentials of the Visible and NIR Regions for Vegetation and Soil Spectroscopic Measurements
Eric Ariel L. Salas, PhD
Broadband remote sensing products limit to using only average spectral information over broadband widths resulting in loss of crucial information available in specific narrow bands (Blackburn, 1998a; Thenkabail et al., 2000). However, latest developments in hyperspectral remote sensing or imaging spectroscopy have afforded additional bands within the visible and NIR regions of the spectrum, mostly in less than 10 nm bandwidths from the visible to the SWIR (Asner, 1998). In fact, in the case of field spectroscopy, it has been extensively used in the development and validation of physical models (Barnsley et al., 1997a and 1997b).
Field spectroscopy describes the studies undertaken in the natural environment, particularly those in which the reflectance properties of vegetation, soils, rocks and water bodies are measured under solar illumination (Jung et al., 2005). Information obtained from the reflectance at different wavebands can be used to describe growth and physiological conditions of the plants (Penuelas, 1994; Yanev et al., 1996; Archer et al., 1999) and assess grain and soil qualities (Morra et al., 1991; Reeves et al., 2002; Kamrunnahar et al., 2003; He et al., 2005).
Spectrometers, operating in the 400 – 2400 nm spectral range have the capability to detect sharp absorption features manifesting due to composition and a certain physical, chemical and biophysical condition of materials (Rast, 1991). Potentially, the use of high spectral resolution spectrometers to measure the reflected radiation of vegetation offers new opportunities to estimate important carbohydrates of plants (Elvidge, 1990). Scientists have studied the spectral differences of grass species (Skidmore and Schmidt, 2001; Skidmore and Schmidt, 2003) using the unique spectral characteristics of vegetation and the ability of spectroscopy to detect detailed narrow spectral features. Kooistra et al. (2003) used a field spectrometer to study soil properties, organic matter and clay content in relation to vegetation cover.
Spectroscopy and vegetation indices
The concept of Visible-NIR spectroscopy is exploited in scientific studies for various purposes. Vegetation indices (VIs), for instance, use the concept to take advantage of the red and NIR in order to measure the photosynthetic activity of the plant. Vegetation reflectance is relatively sensitive to changes in chlorophyll content in the green window and red-edge and insensitive in the NIR part (see review by Morris, 2001). Differences in reflectance between healthy and stressed plants often occur at the green peak and the regions between the red and the NIR, while little change may occur in the chlorophyll trench and NIR plateau (Horler et al., 1983; Ahern, 1988). Green vegetation strongly reflects incident irradiation in the NIR; reflectance in this region reaches 40-60% (Gitelson, 2004).
Vegetation indices have been widely used for the phonologic monitoring, vegetation classification and biophysical derivation of radiometric and structural vegetation parameters (Wan, 1999). They are generally formed from combinations of information of the spectral channels, in the form of ratios, or weighted linear combinations (Colwell and Sadowski, 1993).
A number of derivatives and alternatives have been put forward in the professional scientific arena to deal with the disadvantages of the global-based, widely-used Normalized Difference Vegetation Index (NDVI). The main disadvantage of the NDVI is the non-linear relationship with biophysical characteristics such as Vegetation Fraction (VF), Leaf Area Index (LAI) and above ground biomass (Myneni et al., 1995). Studies conducted by Baret and Guyot (1991) and Gitelson et al. (2003) showed how the NDVI asymptotically loses sensitivity under moderate to high biomass conditions and for certain ranges of LAI and VF.
Other vegetation indices were developed to address the limitations of NDVI. To name some: the Perpendicular Vegetation Index (PVI: Richardson et al., 1977), the Enhanced Vegetation Index (EVI: Matsushita et al., 2007), the Soil-Adjusted Vegetation Index (SAVI: Huete 1988), the Atmospherically Resistant Vegetation Index (ARVI: Kaufman and Tanre 1992), the Wide Dynamic Range Vegetation Index (WDRVI: Gitelson, 2004), the Global Environment Monitoring Index (GEMI: Pinty and Verstraete 1992), the Modified Simple Ratio (MSR: Chen and Cihlar, 1996), and the Renormalized Difference Vegetation Index (RDVI: Roujean and F.M. Breon 1995). Each of the indices endeavored to include fundamental correction for one or more disturbing factors of the NDVI.
Although the indices reduces the adverse effects of environmental factors such as atmospheric conditions (ARVI: Kaufman and Tanre, 1992) and canopy background (SAVI: Huete, 1988), it does not look further into the scientific opportunity presented by Horler et al. (1983) and Ahern (1988) that the reflectance between healthy and stressed plants often occur at the green peak and the regions between the red and the NIR, while little change may occur in the chlorophyll trench and NIR plateau. This understanding emphasizes the importance of the green peak in chlorophyll studies. This is further backed by the result of Morris (2001) that the reflectance is relatively sensitive to changes in chlorophyll content in the green window and red edge, and insensitive in the NIR part.
The green and the NIR peaks and the red trench of the vegetation spectral signature play a role in this research. Thus, one of the goals of the study is to develop a new algorithm that can be utilized to improve vegetation studies and monitor vegetation health using the area-between-peaks concept presented by Salas (2004). Vina et al. (2004) modified the NDVI to come up with the WDRVI that has an increased sensitivity to green biomass. Gitelson and Merzlyak (1995) developed algorithms that are sensitive to chlorophyll concentration, while a study by Stark et al. (2000) established a new technique for remote sensing estimation of vegetation fraction. These studies show that vegetation indices have limitations, due to band width and location (Stark et al., 2000) and efforts must be expended to improve them. The new algorithm here aims to improve the indices that previously relate the visible regions green and red to a wide range of Chl (Thomas and Gaussman, 1977; Gitelson et al., 1996a, 1996b). In the absence of Chl data, the new algorithm will be tested against other established VIs.
Spectroscopy and soil color indices
Numerous studies have applied soil spectroscopy to investigate relationships between soil reflectance and moisture content such as those from Baumgardner et al. (1985), Twomey et al. (1986), and Ishida et al. (1991). Vis-NIR spectroscopy provided the best tool for quantitative and qualitative soil color assessment used under laboratory or in situ conditions (Barrett, 2002). Vis-NIR spectroscopy coupled with multidimensional statistical analyses was applied to derive soil groups (Mouazen et al., 2007) and physically characterize soil color (Konen et al., 2003).
Soil color in the human visible range (0.4-0.7 μm) is related to the presence of pigments or chromophores that absorb or scatter radiation in different wavelengths and intensities (Levin et al., 2005). Organic matter, water molecules, iron oxides, carbonates and chemical composition of transition metals in clay minerals are the major chemical components affecting soil color; while grain size, as a physical chromophore, also plays an important role in affecting this color (Ben-Dor et al., 1998).
Norris (1969) showed that there is abundant evidence that many dune sands become reddened with time and that this process is promoted by warm temperatures, oxidizing conditions, and the periodic presence of moisture. Moisture content, through stepwise multiple regression, gave significant equational relationship with aggregate stability index, along with organic matter, bulk density, and pH (Idowu, 2003).
The existence of soil color indices as mentioned by Madeira et al. (1997) and Mathieu et al. (1998) may reveal that soil color indices can become predictors of moisture content in sandy soils. Characterizing the water content through spectroscopic measurements and indices, without the additional effort of conducting laboratory tests, can support any existing information of the stability of sand dunes. Will changes in soil color indices derived from spectroscopic data able to detect the differences in water content of sand?
Water content will be estimated through calculating the water absorption feature of the sand spectra around 970nm. The feature at 970nm is not that pronounced compared to those from 1450nm and 1950nm, but still clearly observable. Thus, this offers possibilities for deriving information on soil water content.
Digital camera photos: a tool to characterize soil and vegetation samples Levin et al. (2005) applied spectroscopy to assess the accuracy of soil color indices derived from a digital camera. The results showed the possibility of using a digital camera as an alternative way for a three-band spectrometer in the Vis range.
As far as this author’s knowledge is concerned, there is no study that is centered on soil and vegetation spectroscopy to examine the various existing soil color indices and vegetation indices, within the visible region, using digital camera photos. Digital cameras have been used in scientific publications: identify damage in maize (Sena et al., 2003), monitor snow cover in Antarctica (Hinkler et al., 2002), measure the color of soil-thin sections (Adderly et al., 2002). Digital cameras have an advantage over 35-mm photography, as they avoid inconsistencies in film development and errors introduced during scanning (Dean et al., 2000). Can digital camera photos aid spectroscopic field measurements in deriving spectral indices that may characterize soil and vegetation samples?
Digital photos were simultaneously taken for each spectroscopic recording from the Sandhills region in Nebraska from several study plots with diverse degree of vegetation cover or percentage of sand. Using Vis-NIR spectroscopy, vegetation indices can be derived as well as soil color indices. Digital photos can be tapped to calculate the same; however, only three bands are available – red, green and blue. These bands can also be transformed into other color coordinates, such as the Commission Internationale de l’Eclairage (CIE) L*a*b*/CIE Luv color coordinates. Further, the three bands can also be converted into reflectance values.
Soil color indices are ratio-based. As such, the author expects that shading and BRDF effects that may be present in the photos are reduced, as also stated by King (1995). According to Levin et al. (2005), ratioed image of a scene effectively compensates for the brightness variation caused by the differences in topography, and emphasizes the color content of the data.
Thus, this study may look into the possibility of using a simple method of measuring soil color indices through digital photos. Assessment of accuracy may be done using the spectroscopic data simultaneously obtained during fieldwork. Vis-NIR vegetation indices may also be calculated using the photos and relate them against those spectroscopically-derived.
Vegetation indices vs. Soil color indices
Vegetation indices were tested several times by different authors on other experimental measurements (Gilabert et al., 1996; Blackburn, 1998b). Le Maire et al. (2004) tested 60 published chlorophyll indices on experimental database and on a larger simulated database at leaf scale level. According to Le Maire et al. (2004), wavelengths in the 550nm or 700nm neighborhood of the absorption maxima are usually used by many studies because reflectance values are more sensitive to high chlorophyll concentration. He found that vegetation indices based on simple reflectance combinations (e.g. ratio-based) have better results than complicated indices based on derivatives. Even the simplest ratio of two visible bands (Datt, 1999; Carter, 1994; Chapelle et al., 1992; Mc Murtey et al., 1994; Vogelman et al., 1993) gave fair performances compared to most indices that utilize the NIR region. The modified normalized difference (mND) indices based on reflectance (Datt, 1999; Maccioni et al., 2001) showed good results as well.
Thus, it is best to test and compare several ratio-based and normalized-difference (or modified) indices as mentioned by Le Maire et al. (2004) against ratio-based color indices that are mostly used in soil classification studies. Other indices not mentioned by Le Maire et al. (2004) such as the Greeness Index (GI), Chlorophyll Index (CI), Enhanced Vegetation Index (EVI: Matsushita et al., 2007), Wide Dynamic Range Vegetation Index (WDRVI: Gitelson, 2004; Henebry et al., 2005) must be studies as well.
The color indices are based on three bands (blue, green and red). In the absence of spectroscopic measurements, these color indices can be applied to the raw DN values of digital photos to describe vegetation variables, such as chlorophyll, if proven to associate well with established vegetation indices. This approach provides a simple method for vegetation characterization.