Detection of stressed vegetation for mapping heavy metal polluted soil

Detection of stressed vegetation for mapping heavy metal polluted soil I. Reusena , L. Bertelsa, S. Debackerb, W. Debruyna, P. Scheundersb, S. Sterckx...
Author: Evelyn Jordan
1 downloads 2 Views 198KB Size
Detection of stressed vegetation for mapping heavy metal polluted soil I. Reusena , L. Bertelsa, S. Debackerb, W. Debruyna, P. Scheundersb, S. Sterckxa, Wouter Van den Broekb a

Vito, Teledetection and Atmospherical Processes, Boeretang 200, B-2400 Mol, Belgium, email: [email protected] b University of Antwerp (UA), Vision Lab, Groenenborgerlaan 171, B-2020 Antwerp, Belgium

ABSTRACT ‘Maatheide’ in the northern part of Belgium is polluted with heavy metals (Zn, Pb, Cu, Cd, …) due to several decades of historical non-ferrous industrial activities. In the 1970’s the industrial activities ceased and the premises including highly polluted waste were spread over the industrial site. The heavy metal pollution causing vegetation stress introduces subtle changes in the reflectance spectrum of the vegetation. In this project vegetation stress in Pinus Sylvestris L. is used to map heavy metal pollution in the region.In August 2000 an airborne campaign with the CASI-2 (Compact Airborne Spectrographic Imager, Itres) sensor was organized in Belgium. CASI-2 images were acquired with a spatial resolution of 1 m by 1 m in 18 ‘vegetation bands’, mainly concentrated in the red-edge curve (680 nm - 780 nm). Spectral angle mapper classification, using image reference spectra, in combination with multi-band segmentation and ‘highest frequency smoothing’ was used to create a Pinus Sylvestris L. map of the region. Multi-band segmentation delineating individual tree crowns improved the classification accuracy significantly. Based on literature and laboratory experiments 18 potential vegetation stress indices were selected. For each selected index a Pinus Sylvestris L. vegetation stress index map was created. The EGFN index (‘EdgeGreen First derivative Normalized difference’), based on the first derivative of the reflectance in the red-edge curve and the green peak, clearly indicates vegetation stress in the region of ‘Maatheide’ and is correlated with Zn and Cd concentrations in soil and vegetable samples collected near ‘Maatheide’. Keywords: heavy metal pollution, vegetation stress, airborne remote sensing.

1 INTRODUCTION The heavy metal pollution in the northern part of Belgium is caused by several decades of historical non-ferrous industrial activities starting from the early 20th century. In the 1970’s the industrial activities ceased and the premises including heavy metal polluted ashes were spread over the industrial terrain. Due to wind erosion the heavy metals were distributed which resulted in the diffuse character of the heavy metal pollution and high Zn, Pb, Cu and Cd concentrations (Table 1.) [1] spread over an area of 450 km2 in the northern part of Belgium. Moreover, the heavy metal pollution effects the ground water and surface water. For the management of the land and water contamination at regional scale local scale strategies are not suitable. In this work the potential of airborne remote sensing for mapping heavy metal pollution will be tested as a management tool for this heavy metal polluted region. Heavy metal pollution causing vegetation stress introduces subtle in the reflectance spectrum of the vegetation. The method described below detects these subtle changes in the vegetation reflectance spectrum using airborne CASI-2 images with 18 spectral bands and 1 m spatial resolution.

________________________________ Presented at the 3rd EARSeL Workshop on Imaging Spectroscopy, Oberpfaffenhofen, May 13-16 2003

Table 1. Heavy metal concentrations (mg/kg dry soil) after removal of the concrete material (>2 cm diameter).

Heavy metal Zn Pb Cu Cd

Concentration (mg/kg) 10000 1700 1000 10 - 70

2 DATA ACQUISITION AND PROCESSING 2.1 Data acquisition In August 2000 an airborne campaign with the CASI-2 sensor (Itres) was organized in the northern part of Belgium. In total 30 images covering an area of 300 km2 were acquired. The CASI-2 data acquisition was accompanied with airborne GPS and INS measurements of the position and movement of the aircraft and base station GPS, reflectance and irradiance measurements at the ground for dGPS correction of the CASI-2 images and to support the atmospheric correction of the CASI-2 images. The CASI-2 data acquisition mode was spatial mode with 18 ‘vegetation bands’ and 1 m spatial resolution. The flight lines were oriented in SW-NE direction according the main wind direction and thus the expected heavy metal concentration gradient. Table 2. Details of the CASI-2 airborne campaign in northern Belgium in August 2000.

Field of view Altitude Ground speed Spatial resolution Swath width Track length

‘Maatheide’ (northern part of Belgium ) CASI-2 (Compact Airborne Spectrographic Imager), Itres Piper Navajo Chieftain 37.8° 2600 ft 120 knots 1mx1m 511 pixels or 511 m 30 km

Area covered Number of spectral bands Flight line orientation

10 km x 30 km 18 SW – NE

Location Sensor Aircraft

2.2 Processing The geocoded CASI-2 data was atmospherically corrected with ATCOR4 [2] which is based on the radiative transfer model MODTRAN4. Besides atmospheric correction, ATCOR4 allows the determination of radiometric calibration factors based on ground reflectance measurements of natural and/or artificial targets.

2.3 Data processing chain Figure 1. is an overview of the data processing chain from 18 bands CASI-2 images to vegetation stress maps. Due to a spectral shift of a few nm of the CASI-2 sensor, it was not possible to correct properly for atmospheric effects at bands near the O2 and H2O vapor absorption features. For that reason 4 spectral bands were removed. Subsequently a vegetation mask eliminating non-vegetation pixels was built using an NDVI threshold in order to decrease the further processing time. To allow vegetation stress studies, a good vegetation species classification is required. A combination of spectral angle mapper (ENVI), multi-band segmentation (eCognition) and ‘highest frequency smoothing’ is used for the vegetation species classification. Spectral angle mapper (SAM) classification with reference spectra determined from Regions of Interest selected within the image results in a first rough classification (Prod. Acc. 24.07%, User acc. 86.67%). SAM combined with segmentation, delineating individual crown trees, and ‘highest frequency smoothing’, to determine the dominant vegetation species within each segment,

results in a better classification (Prod. Acc. 66.67%, User acc. 87.80%) as shown in figure 2. Since Pinus Sylvestris L. is the dominant vegetation species in northern Belgium a pine mask was deduced from the latter classification result.

18 bands Radiometric, Geometric & Atmospheric corrected CASI data

Remove noise, select the correct bands

14 bands CASI data

Build & apply vegetation mask

14 bands CASI vegetation data

Different Masks per vegetation type

Select ROI’s for vegetation types

Segmentation using 3 bands

SAM classification via ROI spectra

Segmented vegetation image

Highest frequency smoothing

Make unique segment image

Build mask per vegetation type

Unique Segment image

Calculate different vegetation indices

Different vegetation indices

Apply type mask & color table

Vegetation stress maps

Figure 1. Data processing chain from 18 bands CASI-2 images to vegetation stress maps.

Figure 2. Result of SAM classification, multi-band segmentation and ‘Highest Frequency Smoothing’.

Figure 3: Pine mask deduced from the ‘Pine’ class after SAM classification, multi-band segmentation and ‘Highest Frequency Smoothing’.

Subsequently, for each pixel in the pine mask several vegetation stress indices are calculated [3], [4] and averaged per segment. The EGFN, Edge-Green First derivative Normalized difference, values are represented by a 10 percentile color code. Low EGFN values (red) represent high vegetation stress levels, high EGFN values (blue) represent a lack of vegetation stress. The EGFN vegetation stress index map clearly indicates high vegetation stress levels near the former zinc smelting factory indicated by the white arrow in figure 4 and lower vegetation stress levels at more distant locations.

Figure 4: : The EGFN vegetation stress index map for pine in the ‘Maatheide’ track. The location of the former zinc smelting factory is indicated by the white arrow. The vegetation stress detection technique was verified at a reference track where no heavy metal pollution is expected. To visualize, the ‘Maatheide’ track and the reference track are 25° rotated clockwise and divided into 100 m slices. If at least 2% meaningful pixels are present within the slice, the 20% contribution bars are calculated. Figure 5 clearly indicates low vegetation stress levels (high EGFN values, blue) for the reference track as expected and high vegetation stress levels (low EGFN values, red) near the former zinc smelting factory. For comparison, Cd concentrations, determined from soil and vegetables samples collected about 1000 m south of ‘Maatheide’ and analyzed by LISEC, are displayed for the same 100 m slices. Clearly a correlation between the EGFN contribution bars and the Cd concentrations is present: high Cd concentrations correspond to low (red) EGFN values or high vegetation stress levels.

Location of the former zinc factory

100% 90% 80% 70%

Reference track

60% 50% 40% 30% 20% 10% 0% 1

3

5

7

9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 67 69 71 73 75 77 79 81 83 85

100% 90% 80%

Maatheide track

70% 60% 50% 40% 30% 20% 10% 0% 1

4

7

10

13

16

19

22

25

28

31

34

37

40

43

46

49

52

55

58

61

64

67

70

73

76

79

82

85

88

91

94

97 100 103 106 109 112 115 118 121 124 127 130 133 136 139 142 145 148 151 154 157 160 163 166 169 172 175 178 181 184 187 190 193 196 199 202 205

LISEC data

15

Cd concentraction (mg/kg)

10 5 0 1

4

7

10 13 16 19 22 25 28 31 34 37 40 43 46 49 52 55 58 61 64 67 70 73 76 79 82 85 88 91 94 97 100

Figure 5: The EGFN 20 percentile contribution bars per 100 m slice for the reference track where no heavy metal pollution is expected and the polluted ‘Maatheide’ track. For comparison the Cd concentrations in soil and vegetable samples (LISEC) are displayed.

3 CONCLUSIONS Spectral angle mapper classification in combination with multi-band segmentation and ‘highest frequency smoothing’ improves the pine classification result. A pine mask is deduced from the classification result and subsequently vegetation stress indices like EGFN are calculated for each individual pixel and averaged per segment. The EGFN vegetation stress map clearly indicates high vegetation stress levels (low EGFN values, red) near the former zinc smelting factory and a complete lack of vegetation stress (high EGFN values, blue) for the reference track where no heavy metal pollution is expected. The EGFN slice representation was validated against the slice representation of Cd concentration in soil and vegetable samples. This vegetation stress detection techniques was applied successfully for pine, but can probably be used to map vegetation stress in other tree types, crops, …

ACKNOWLEDGMENTS We would like to thank the Flemish Waste Management Agency OVAM for supporting the work reported here.

REFERENCES [1] VANGRONSVELD, J., VAN ASSCHE, F., CLIJSTERS, H., 1995: Reclamation of a bare industrial area contaminated by non-ferrous metals: In situ metal immobilization and revegetation. Environmental Pollution 87, pp. 51-59. [2] RICHTER, R., AND SCHLAEPFER, D., 2002: Geo-atmospheric processing of airborne imaging spectrometry data. Part 2: atmospheric/topographic correction. Int. J. Remote Sensing 23, pp. 2631-2649. [3] PENUELAS, J., GAMON, J.A., FREDEEN, A.L., MERINO, J., FIELD, C.B., 1994: Reflectance indices associated with physiological changes in nitrogen- and water-limited sunflower leaves. Remote Sensing of Environment 48, pp. 135-146. [4] ZARCO, P.J., 1999: Optical indices as bioindicators of forest condition from hyperspectral CASI data. Proceedings 19th EARSeL symposium, Valladolid, Spain.

Suggest Documents