Assessment of heavy metal contamination in river ... - CiteSeerX

10 downloads 0 Views 856KB Size Report
... on the spectral sampling interval applied. To overcome. Presented at the 3rd EARSeL Workshop on Imaging Spectroscopy, Herrsching, 13-16 May 2003. 173 ...
Assessment of heavy metal contamination in river floodplains by using the red-edge index J.G.P.W. Clevers & L. Kooistra Centre Geo-information (CGI), Wageningen University and Research Centre, P.O. Box 339, NL-6700 AH Wageningen, The Netherlands E-mail: [email protected] ABSTRACT Heavy metal contamination of soils is one of the major environmental problems of regular flooding of rivers. Various studies have shown that vegetation spectra are influenced by soil contamination. The main objective of this study was to investigate whether the red-edge of vegetation spectra may provide information on soil contamination by heavy metals. The use of the maximum first derivative method, smoothing methods (like polynomial fitting and the inverted Gaussian function), and interpolation methods based on just a few spectral bands were tested. Test site was a floodplain along the river Waal in the Netherlands. Heavy metal concentrations of soil samples and reflectance spectra of the growing vegetation were measured on selected transects using a field spectroradiometer. A significant, negative correlation between the red-edge position (REP) and heavy metal concentration was found using the maximum first derivative method. The first derivative spectra in this study notably showed the presence of a double peak within the red-edge region. This double peak can only be observed if a very fine spectral resolution (1 – 3 nm bandwidth) is used. Other methods for estimating the REP in this study do not make full use of the very fine spectral resolution of the used spectroradiometer. They apply either a function, which severely smoothes the spectral signature in the red-edge region, or they even use only three or four fixed wavelength positions for deriving the REP. As a result, the correlation between these estimated REP values and the heavy metal concentration was not significant. Keywords: red-edge, first derivative, spectroradiometer, heavy metal contamination, river floodplains.

1 INTRODUCTION One of the major environmental problems related to the regular flooding of rivers in Europe is heavy metal contamination of the soils. As an example, sediments of river floodplains in the Netherlands show significant levels of heavy metals such as cadmium (Cd), copper (Cu), nickel (Ni), lead (Pb) and zinc (Zn). As shown in Ref. 1 stress factors like heavy metal uptake will influence reflectance spectra at the leaf and canopy level. A number of field studies have demonstrated that shifts in vegetation spectra were metal-induced due to geochemical stress [2], [3] or due to the occurrence of old waste deposit sites [4] and occurred in both the visible and the near-infrared part of the spectrum. Collins [5] and Horler [6] were among the first researchers pointing out the importance of the wavelength region of the red – near-infrared (NIR) transition for vegetation studies. Both the position and the slope of this red-edge change under stress conditions, resulting in a shift of the slope towards shorter wavelengths (e.g. [6], [7]). The position of the red-edge is defined as the position of the inflection point of the red-NIR slope. This is also denoted as the red-edge position (REP). The inflection point can be studied by plotting dR/dλ, the first derivative of the reflectance (R) with respect to wavelength (λ), as a function of λ. Alternatively, in many studies simple functions have been fitted to the reflectance spectrum in the red-edge region, and subsequently the wavelength belonging to the maximum slope has been extracted from such an analytical expression. The objective of the present study is to investigate whether the REP of vegetation samples may provide information on soil contamination by heavy metals in contaminated river floodplains under natural vegetation.

2 METHODOLOGY FOR DETERMINING THE RED-EDGE POSITION (REP) The position of the inflection point in the region of the red-edge is the maximum first derivative of the reflectance spectrum of vegetation. Accurate determination of the REP requires a large number of spectral measurements in narrow bands. However, the calculated REP will depend on the spectral sampling interval applied. To overcome Presented at the 3rd EARSeL Workshop on Imaging Spectroscopy, Herrsching, 13-16 May 2003

173

this dependency on the sampling interval high-order curve fitting techniques are employed to fit a continuous function to the derivative spectrum [6], [8], [9]. However, existing curve fitting techniques are very complex. In Ref. 10 a technique is presented based upon a three-point Lagrangian interpolation technique for interpolating the REP between fine spectral samples yielding accurate REP estimates within the spectral sampling interval. It fits a second-order polynomial curve to three successive spectral bands, from which the REP is determined. Although an increasing number of airborne imaging spectrometers have become available, their spectral resolution is not that fine as required for an accurate determination of the REP using derivative spectra. Therefore, fitting a mathematical function to a few measurements in the red-edge region is often applied for approximating the REP. In Ref. 11 a simple third-order polynomial function is fitted to the red-edge spectrum. The REP can then simply be calculated by determining the maximum first derivative of this function. Major disadvantage of a third-order polynomial is its symmetry around the REP by definition. In Ref. 12 a sixth-order polynomial is applied to derive the REP. In this way, potential asymmetry of the rededge is captured. The REP is determined from the maximum first derivative in the 690 - 740 nm range. In Ref. 13 a so-called inverted Gaussian fit to the red-infrared slope is described, thus smoothing the reflectance spectrum. Several authors (e.g., [14], [15], [16], [17]) have used this approach. In Ref. 18 a simple linear model to the red-infrared slope is applied. This method assumes that the reflectance curve at the red-edge can be simplified to a straight line centred around a midpoint between the reflectance in the NIR at about 780 nm and the reflectance minimum of the chlorophyll absorption feature at about 670 nm. The reflectance of the REP is then estimated as being halfway this NIR and red reflectance. Subsequently, the REP is estimated by linearly interpolating between measurements at 700 and 740 nm. Finally, in Ref. 19 an a priori polynomial equation is proposed, based on only three spectral bands, as an estimate of the inflection point. Model simulations were used to select optimal spectral bands at 672, 710 and 780 nm, and subsequently to fit a polynomial equation to the maximum first derivative. A problem with some of the previous methods arises when the first derivative shows more than one maximum. In Ref. 6 two maxima in the first derivative spectrum with peaks around 705 and 725 nm have been identified. The red-edge index then represented whichever peak was dominant, and therefore a red-edge shift could involve a jump between the two peaks, creating a discontinuity in the red-edge index. Other authors also found two (or more) peaks in the derivative spectrum [20], [21], [22], [23], [24]. Experimental results suggest the peak at around 705 nm is influenced by chlorophyll concentration, whereas the peak around 725 nm is influenced by a combination of chlorophyll concentration and LAI [6], [20]. The second peak moves to longer wavelengths and becomes more pronounced as the LAI increases.

3 MATERIAL AND METHODS 3.1 Test site and field sampling The study site is located in the Afferdensche and Deestsche Waarden floodplain along the river Waal, the main branch of the river Rhine in the Netherlands (figure 1). This is a 250 ha area, where ecological rehabilitation is being performed by transforming farmland into wetlands. During the past decades, large amounts of contaminated sediment have been deposited in the floodplain. Vegetation composition and coverage are relatively homogeneous and characterized by natural grasslands with patches of herbaceous vegetation. Nine transects of 50 m each were located in areas with natural grassland. Most of the transects were located in areas with heavy metal contamination, as known from previous studies [25]. Three transects (No. 6, 7 and 8) were positioned in an area that has recently been excavated. Six equidistant soil samples were collected along every transect, so they were 10 m apart. Within the field of view (FOV) of the spectroradiometer (section 3.3) three scoops from the soil top 10 cm were taken, homogenized, stored in a plastic bag and taken to the laboratory for chemical analysis. In addition, a description of the vegetation for every sampling point was made. All fieldwork took place within three days in the beginning of August 2001.

3.2 Chemical analysis Soil moisture content was determined by oven-drying the soil samples at 105°C for 24 hours and measuring the weight loss. Soil organic matter content was obtained by the loss-on-ignition method described in Ref. 26. The total concentrations of Cd, Cu, Ni, Pb and Zn were determined by treating dried soil samples (1 g) with a HNO3/H2O2 solution using the microwave digestion method. After mineralisation, total metal concentrations were measured by means of ICP-AES spectrometry [25].

174

Figure 1. Location of the study site and of the transects sampled within the Afferdensche en Deestsche Waarden floodplain along the river Rhine in the Netherlands.

3.3 Field reflectance measurements For all sample points described in section 3.1, detailed spectral measurements were collected with an ASD FieldSpec FR spectroradiometer. The radiometer has a 25° FOV and a 1 nm spectral resolution within the 400 to 2500 nm spectral range. The fiber optics of the spectroradiometer was mounted on a tripod looking from nadir at a height of 1 m above the target. This resulted in a measurement area of 0.15 m2. Every measurement was recorded as the average of 15 readings. Every target measurement was preceded by a reference measurement over a standardized white spectralon panel. To reduce measurement noise a 3 band moving average window was applied to the spectral reflectances, simulating a bandwidth of 3 nm and a sampling interval of 1 nm.

4 RESULTS AND DISCUSSION 4.1 Spectral signatures Figure 2 illustrates the spectral signature of a sample point with a low contamination level (from the excavated area with low vegetation coverage), a vegetation sample with a medium contamination level and a vegetation sample with a high contamination level. The latter two samples were classified as homogeneous grassland. It is striking that the sample with the high contamination level shows the highest NIR reflectance, indicating a higher biomass level than the sample with the lower contamination level. Figure 2 also shows that the sample with the high metal contamination level and the higher biomass level (higher NIR reflectance) has a higher red reflectance than the sample with the medium contamination level. Metals may have a negative effect on the chlorophyll concentration, and thus cause a higher red reflectance. This would be consistent with the commonly reported effect that heavy metal treated plants contain lower chlorophyll concentrations than healthy plants [6], [27]. Samples with low contamination level only occurred in the excavated area. An example of such a sample is also illustrated in figure 2. The spectrum shows a rather low NIR reflectance and a rather high red reflectance, coinciding with the general trend observed in literature for soils with low vegetation cover.

4.2 Derivative spectra Figure 3 illustrates the derivative spectra for the signatures shown in figure 2. First of all, more than one maximum is observed for the individual derivative spectra. The spectrum with low vegetation cover and a low heavy metal contamination level shows a maximum around 700 nm. This can be explained by the fact that spectra with low vegetation cover (low biomass and low chlorophyll content) generally show a rather low REP value. The two other spectra show a peak around 720 nm. The derivative spectrum with the higher derivative values matches the spectral signature with the steeper slope in the red-edge region. This was the grassland sample with the high biomass level (and the high contamination level). Although one would expect a higher REP value for this sample in comparison to the sample with the lower biomass level (medium contamination level), the reverse effect can be observed. This suggests that the REP may be useful as an indicator for the level of metal contamination. As stated in section 2, other authors also found two peaks (around 705 and 725 nm) in the derivative spectra [6], [20], [21], [22], [23]. In Ref. 20 a spectroradiometer with 2 nm bandwidth was used and it was observed that the

175

second peak sometimes consisted of two individual peaks. In Ref. 23 one peak was found between 720 and 730 nm in experimental data. However, in that study a spectroradiometer with a spectral resolution of about 3.5 nm was used. In Ref. 22 only one peak was observed at the position of the second peak, but a spectroradiometer with a bandwidth of 8 nm was used. In Ref. 21 neither the very fine spectral detail was obtained while using a laboratory spectroradiometer with 4 nm bandwidth. 70

60

reflectance (%)

50

40

low medium high

30

20

10

0 600

650

700

750

800

850

900

wavelength (nm)

Figure 2. Example of a spectral signature measured over a grassland area with a high heavy metal contamination level, one with a medium heavy metal contamination level, and a measurement over an excavated area with low vegetation cover and a low contamination level.

first derivative reflectance (% nm-1)

1.4

1st peak

2nd peak





1.2

1

0.8

low medium high

0.6

0.4

0.2

0 670

680

690

700

710

720

730

740

750

wavelength (nm)

Figure 3. Simulated first derivative spectra for a spectroradiometer with 3 nm spectral resolution (based on the reflectance spectra of figure 2).

4.3 Relationship between REP and heavy metal concentrations Eighteen measurement points were classified in the field as being homogeneous grasslands with annual bluegrass (Poa annua L.) and perennial ryegrass (Lolium perenne L.) as dominant species. In order to avoid effects of differences in species composition, REP analysis was limited to the grassland samples. Table 1 gives the correlation coefficients between the various methods described in section 2 for deriving the REP and the heavy metal concentrations using the 18 grassland samples. In all cases highest correlations were found when using the maximum first derivative for calculating the REP. Figure 4 illustrates the relationships between the REP using the maximum first derivative method and the Pb concentration for the grassland samples. The maximum first derivative method shows a significant negative correlation with the Pb concentration (as well as with the other heavy metals, cf. table 1). All other procedures for calculating the REP did not yield significant correlations. They either apply a rigorous smoothing to the spectral

176

measurements or only use a very limited number of spectral measurements (3 or 4). As a result, only the derivative method makes full use of the fine spectral information present in the spectroradiometer data. Table 1. Correlation coefficients between the various REP measures and the heavy metal concentrations for the grassland samples within the study area (n=18). Significance: *p