Hyperspectral alteration mineral mapping using ... - Semantic Scholar
Hyperspectral alteration mineral mapping using the POSAM method Taro Yajima
Kazumichi Ohkawa and Sinzi Huzikawa
Japan Oil, Gas and Metals National Corporation (JOGMEC) 1310 Omiya-Cho, Saiwai-Ku, Kawasaki-shi, Kanagawa, Japan 212-8554 [email protected]
Geotechnos Co., Ltd. 5-10-5 Shimbashi, Minato-Ku, Tokyo, Japan 105-0004
Abstract— The POSAM method identifies each pixel in the hyperspectral data to create an alteration mineral map. An IDL program for “ENVI” was developed. Data processing using the POSAM method was conducted on AVIRIS and Hyperion data. The mineral map created from the AVIRIS data showed better results compared to the general processing method proven by ground truth survey, but only few pixels were identified from the Hyperion data. Keywords- POSAM; mineral mapping; AVIRIS; Hyperion
I.
INTRODUCTION
In general hyperspectral data processing methods, the end member spectra are obtained from the hyperspectral data (i.e. Minimum Noise Fraction Rotation, Pixel Purity Index and nDimensional Visualizer). They are identified and then mapped by spectral matching within the whole data. The difficulty in this method is to identify the end member spectra accurately. Skills are needed for interpreting the spectra. Identification of the end member spectra may be subjective and mistakes may occur. Since each pixel in the hyperspectral data possesses continuous spectral data, it is possible to apply mineral identification on each pixel. An IDL program for the remote sensing data processing software ENVI was developed in order to identify the alteration mineral for each pixel. An alteration mineral identification equipment named POSAM (Portable Spectroradiometer for Mineral identification) has been developed by the Metal Mining Agency of Japan (former organization of JOGMEC) in 1993 [1], [2]. POSAM consists of a spectrometer that measures the reflectance spectra in the short wave infrared (1300-2500 nm) and software to identify alteration minerals (Fig. 1). The POSAM identification was applied on each pixel in the hyperspectral data to create an alteration mineral map. II.
DESCRIPTION OF THE POSAM METHOD
A. Processing The POSAM method identifies each pixel in the hyperspectral data to create an alteration mineral map. The processing steps for the POSAM method are as follows.
Figure 1. View of POSAM and mineral identification result screen. POSAM can identify over 40 minerals at over 80 % of accuracy.
1) Extracting spectral data from pixel (2000-2500 nm). 2) Spectral correction. ･Normalize (spectral enhancement) ･Hull (base line correction) 3) Peak position finding 4) Scoring POSAM has a data file of approximately 60 alteration minerals (Table I). The pixel spectrum is compared with the alteration minerals in the data file and scoring is conducted. Score is added if absorption matches for an alteration mineral and subtracted if there is no match for an important absorption feature. 5) Display Pixels with high scores will be shown in brighter tone and low scores will be shown in darker tone. Low scored pixels under the threshold will not be shown. The steps will be repeated until all the pixels are identified. B. Interface The IDL program that conducts the POSAM identification to all pixels runs on ENVI. The interface of the program is shown on Fig. 2. The user is able to customize the processing. 1) Expected minerals can be chosen for mapping. Mineral maps will be created for each chosen mineral. 2) Spectral corrections can be chosen (“normalize” and “hull”). Normalization will emphasize the absorption. Hull quotient will conduct base line correction (Fig. 3).
Figure 2. Data processing interface of the POSAM method.
100%
Reflectance
80% Test spectral data Shift to 100% Kubelka-Munk Background Normalize Hull quotient
60% 40% 20% 0%
1200
1600
2000
2400
Wave length (nm)
Figure 3. Example of the spectral corrections conducted in the POSAM method.
3) Correlation of the absorption dips can be considered for scoring. The total shape of the spectra will be considered. 4) Threshold depth of the absorbance for the identification can be defined (default is 93 % from 100 % reflectance). 5) Threshold of the calculated scores for mineral mapping can be defined. 6) Output absorption list can be obtained. III.
RESULTS
Processing using the POSAM method was conducted on AVIRIS and Hyperion data. Both data were atmospheric corrected before processing (AVIRIS; ATREM, Hyperion; FLAASH). A. AVIRIS data High altitude AVIRIS data of the Mountain Pass area (California, USA) were used for processing. The data were acquired at 20 km altitude and the size of pixels are 20x20 m. Alteration mineral mapping by the general processing method and the POSAM method were compared (Fig. 4).
Figure 4. Processing results of the high altitude AVIRIS data from the Mountain Pass area (2000), California, USA. The upper mineral map was obtained from the general processing method and the lower was from the POSAM method (purple: calcite, cyan green: dolomite, light green: chlorite, orange: monmorillonite).
The difference in the mapping results can be noticed between chlorite and montmorillonite. A ground truth survey was conducted and the alteration mineral was checked by Xray diffraction method. The identification by the POSAM method was correct. Low altitude AVIRIS data of the Cuprite area (Nevada, USA) were also processed. The data were acquired at 3 km altitude and the size of the pixels are 3x3 m. Alteration mineral maps were created successfully by the POSAM method. B. Hyperion data Hyperion data of the East Tianshan area (Xinjiang, China) were used for processing. Hyperion data were acquired at 705 km altitude and the size of pixels are 30x30 m. Processing using the POSAM method was conducted, but there were only few pixels that were identified as alteration minerals. Satisfactory mineral maps could not be created by the POSAM method from the Hyperion data.
C. Discussions Accurate mineral maps were obtained from the AVIRIS data using the POSAM method. Spectral enhancement and base line correction enabled the accurate mineral identification.
H2O H2O
There were only few identified pixels from the Hyperion data. There are two possible causes for this result. 1) Hyperion data at 2002 and 2012 nm are disturbed due to CO2 in the atmosphere. Though atmospheric correction is conducted, the effect of CO2 is not totally excluded (Fig. 5). A sharp absorption of CO2 still remains in the spectral data. This was interpreted as the strongest absorption feature and identified as null. This CO2 absorption was not present in the AVIRIS data due to the difference of the data acquisition altitude. 2) The signal to noise ratio of Hyperion spectral data is low (noisy). The noises were emphasized as false absorption. In order to obtain a proper mineral map from the Hyperion data by the POSAM method, 1) masking the atmospherically affected wavelengths and 2) spectral smoothing may improve the mineral identification. Also using not only the 2000-2500 nm but the 1300-2500 nm region may lead to more accurate mineral identification since the POSAM regards the 1300-2500 nm region. If the hyperspectral data are carefully corrected, the POSAM method can work effectively as shown in the AVIRIS data. It can provide an objective and accurate alteration mineral maps. The POSAM method can be an effective method additional to the general methods in order to process hyperspectral data for alteration mineral mapping. ACKNOWLEDGMENT This work was performed under a contract with the Ministry of Economy, Trade and Industry (METI), Japan. REFERENCES [1] [2]
http://www.geotechnos.co.jp/subglob/data/posam/index_e.html S. Huzikawa, K. Ohkawa, and S. Tanaka, “Automatic Identification of Alteration Mineral Using a Portable Infrared Spectralmeter,” Jour. Rem. Sen. Soc. Japan, vol. 21, pp. 206–209.
H2O CO2
Figure 5. Atmospherically corrected spectral data of Hyperion. Spectral data from a kaolinite alteration of the East Tianshan area, Xinjiang, CHINA (scene ID EO1H1380292002311110PZ).
TABLE I.
MINERALS FOR IDENTIFICATION IN THE POSAM METHOD IDL PROGRAM.
No. Minerals 1 Alunite 2 Analcime 3 Anhydrite 4 Calcite 5 Chabazite 6 Chlorite 7 Chrysotile 8 Clinoptillolite 9 Dickite 10 Dolomite 11 Epidote 12 Gibbsite 13 Gypsum 14 Halloysite 15 Heulandite 16 Illite 17 Jarosite 18 Kaolinite 19 Laumontite 20 Magnesite
No. Minerals 21 Montmorillonite 22 Mordenite 23 Muscovite 24 Natrolite 25 Nontronite 26 Phlogopite 27 Pyrophyllite 28 Sericite 29 Stilbite 30 Talc 31 Thomsonite 32 Vermiculite 33 Wairakite 34 Yugawaralite 35 H2O 36 Actinolite 37 Forsterite 38 Hornblende 39 Nepheline 40 Wollastonite
No. Minerals 41 Natroalunite 42 Topaz 43 Buddingtonite 44 chlorite_mont 45 chlorite_sericite 46 Anhydrite_chlorite 47 anhydrite_dolomite 48 anhydrite_epidote 49 calcite_gypsum 50 calcite_mont 51 chlorite_dolomite 52 dolomite_epidote 53 chlorite_laumontite 54 chlorite_wairakite 55 dolomite_gypsum 56 dolomite_mont 57 dolomite_sericite 58 epidote_sericite 59 calcite_chrysotile
Kazumichi Ohkawa and Sinzi Huzikawa
Japan Oil, Gas and Metals National Corporation (JOGMEC) 1310 Omiya-Cho, Saiwai-Ku, Kawasaki-shi, Kanagawa, Japan 212-8554 [email protected]
Geotechnos Co., Ltd. 5-10-5 Shimbashi, Minato-Ku, Tokyo, Japan 105-0004
Abstract— The POSAM method identifies each pixel in the hyperspectral data to create an alteration mineral map. An IDL program for “ENVI” was developed. Data processing using the POSAM method was conducted on AVIRIS and Hyperion data. The mineral map created from the AVIRIS data showed better results compared to the general processing method proven by ground truth survey, but only few pixels were identified from the Hyperion data. Keywords- POSAM; mineral mapping; AVIRIS; Hyperion
I.
INTRODUCTION
In general hyperspectral data processing methods, the end member spectra are obtained from the hyperspectral data (i.e. Minimum Noise Fraction Rotation, Pixel Purity Index and nDimensional Visualizer). They are identified and then mapped by spectral matching within the whole data. The difficulty in this method is to identify the end member spectra accurately. Skills are needed for interpreting the spectra. Identification of the end member spectra may be subjective and mistakes may occur. Since each pixel in the hyperspectral data possesses continuous spectral data, it is possible to apply mineral identification on each pixel. An IDL program for the remote sensing data processing software ENVI was developed in order to identify the alteration mineral for each pixel. An alteration mineral identification equipment named POSAM (Portable Spectroradiometer for Mineral identification) has been developed by the Metal Mining Agency of Japan (former organization of JOGMEC) in 1993 [1], [2]. POSAM consists of a spectrometer that measures the reflectance spectra in the short wave infrared (1300-2500 nm) and software to identify alteration minerals (Fig. 1). The POSAM identification was applied on each pixel in the hyperspectral data to create an alteration mineral map. II.
DESCRIPTION OF THE POSAM METHOD
A. Processing The POSAM method identifies each pixel in the hyperspectral data to create an alteration mineral map. The processing steps for the POSAM method are as follows.
Figure 1. View of POSAM and mineral identification result screen. POSAM can identify over 40 minerals at over 80 % of accuracy.
1) Extracting spectral data from pixel (2000-2500 nm). 2) Spectral correction. ･Normalize (spectral enhancement) ･Hull (base line correction) 3) Peak position finding 4) Scoring POSAM has a data file of approximately 60 alteration minerals (Table I). The pixel spectrum is compared with the alteration minerals in the data file and scoring is conducted. Score is added if absorption matches for an alteration mineral and subtracted if there is no match for an important absorption feature. 5) Display Pixels with high scores will be shown in brighter tone and low scores will be shown in darker tone. Low scored pixels under the threshold will not be shown. The steps will be repeated until all the pixels are identified. B. Interface The IDL program that conducts the POSAM identification to all pixels runs on ENVI. The interface of the program is shown on Fig. 2. The user is able to customize the processing. 1) Expected minerals can be chosen for mapping. Mineral maps will be created for each chosen mineral. 2) Spectral corrections can be chosen (“normalize” and “hull”). Normalization will emphasize the absorption. Hull quotient will conduct base line correction (Fig. 3).
Figure 2. Data processing interface of the POSAM method.
100%
Reflectance
80% Test spectral data Shift to 100% Kubelka-Munk Background Normalize Hull quotient
60% 40% 20% 0%
1200
1600
2000
2400
Wave length (nm)
Figure 3. Example of the spectral corrections conducted in the POSAM method.
3) Correlation of the absorption dips can be considered for scoring. The total shape of the spectra will be considered. 4) Threshold depth of the absorbance for the identification can be defined (default is 93 % from 100 % reflectance). 5) Threshold of the calculated scores for mineral mapping can be defined. 6) Output absorption list can be obtained. III.
RESULTS
Processing using the POSAM method was conducted on AVIRIS and Hyperion data. Both data were atmospheric corrected before processing (AVIRIS; ATREM, Hyperion; FLAASH). A. AVIRIS data High altitude AVIRIS data of the Mountain Pass area (California, USA) were used for processing. The data were acquired at 20 km altitude and the size of pixels are 20x20 m. Alteration mineral mapping by the general processing method and the POSAM method were compared (Fig. 4).
Figure 4. Processing results of the high altitude AVIRIS data from the Mountain Pass area (2000), California, USA. The upper mineral map was obtained from the general processing method and the lower was from the POSAM method (purple: calcite, cyan green: dolomite, light green: chlorite, orange: monmorillonite).
The difference in the mapping results can be noticed between chlorite and montmorillonite. A ground truth survey was conducted and the alteration mineral was checked by Xray diffraction method. The identification by the POSAM method was correct. Low altitude AVIRIS data of the Cuprite area (Nevada, USA) were also processed. The data were acquired at 3 km altitude and the size of the pixels are 3x3 m. Alteration mineral maps were created successfully by the POSAM method. B. Hyperion data Hyperion data of the East Tianshan area (Xinjiang, China) were used for processing. Hyperion data were acquired at 705 km altitude and the size of pixels are 30x30 m. Processing using the POSAM method was conducted, but there were only few pixels that were identified as alteration minerals. Satisfactory mineral maps could not be created by the POSAM method from the Hyperion data.
C. Discussions Accurate mineral maps were obtained from the AVIRIS data using the POSAM method. Spectral enhancement and base line correction enabled the accurate mineral identification.
H2O H2O
There were only few identified pixels from the Hyperion data. There are two possible causes for this result. 1) Hyperion data at 2002 and 2012 nm are disturbed due to CO2 in the atmosphere. Though atmospheric correction is conducted, the effect of CO2 is not totally excluded (Fig. 5). A sharp absorption of CO2 still remains in the spectral data. This was interpreted as the strongest absorption feature and identified as null. This CO2 absorption was not present in the AVIRIS data due to the difference of the data acquisition altitude. 2) The signal to noise ratio of Hyperion spectral data is low (noisy). The noises were emphasized as false absorption. In order to obtain a proper mineral map from the Hyperion data by the POSAM method, 1) masking the atmospherically affected wavelengths and 2) spectral smoothing may improve the mineral identification. Also using not only the 2000-2500 nm but the 1300-2500 nm region may lead to more accurate mineral identification since the POSAM regards the 1300-2500 nm region. If the hyperspectral data are carefully corrected, the POSAM method can work effectively as shown in the AVIRIS data. It can provide an objective and accurate alteration mineral maps. The POSAM method can be an effective method additional to the general methods in order to process hyperspectral data for alteration mineral mapping. ACKNOWLEDGMENT This work was performed under a contract with the Ministry of Economy, Trade and Industry (METI), Japan. REFERENCES [1] [2]
http://www.geotechnos.co.jp/subglob/data/posam/index_e.html S. Huzikawa, K. Ohkawa, and S. Tanaka, “Automatic Identification of Alteration Mineral Using a Portable Infrared Spectralmeter,” Jour. Rem. Sen. Soc. Japan, vol. 21, pp. 206–209.
H2O CO2
Figure 5. Atmospherically corrected spectral data of Hyperion. Spectral data from a kaolinite alteration of the East Tianshan area, Xinjiang, CHINA (scene ID EO1H1380292002311110PZ).
TABLE I.
MINERALS FOR IDENTIFICATION IN THE POSAM METHOD IDL PROGRAM.
No. Minerals 1 Alunite 2 Analcime 3 Anhydrite 4 Calcite 5 Chabazite 6 Chlorite 7 Chrysotile 8 Clinoptillolite 9 Dickite 10 Dolomite 11 Epidote 12 Gibbsite 13 Gypsum 14 Halloysite 15 Heulandite 16 Illite 17 Jarosite 18 Kaolinite 19 Laumontite 20 Magnesite
No. Minerals 21 Montmorillonite 22 Mordenite 23 Muscovite 24 Natrolite 25 Nontronite 26 Phlogopite 27 Pyrophyllite 28 Sericite 29 Stilbite 30 Talc 31 Thomsonite 32 Vermiculite 33 Wairakite 34 Yugawaralite 35 H2O 36 Actinolite 37 Forsterite 38 Hornblende 39 Nepheline 40 Wollastonite
No. Minerals 41 Natroalunite 42 Topaz 43 Buddingtonite 44 chlorite_mont 45 chlorite_sericite 46 Anhydrite_chlorite 47 anhydrite_dolomite 48 anhydrite_epidote 49 calcite_gypsum 50 calcite_mont 51 chlorite_dolomite 52 dolomite_epidote 53 chlorite_laumontite 54 chlorite_wairakite 55 dolomite_gypsum 56 dolomite_mont 57 dolomite_sericite 58 epidote_sericite 59 calcite_chrysotile
