Borstad Associates Ltd. - Comparison of Mineral Maps Produced from SFSI Data with Those from AVIRIS Data

Comparison of Mineral Maps Produced from SFSI Data with Those from AVIRIS Data

Gary Borstad
G. A. Borstad Associates Ltd., Sidney, British Columbia, Canada

Robert Neville
Canada Centre for Remote Sensing Ottawa, Ontario, Canada

Phoebe Hauff
Spectral International Inc. Arvada, Colorado, United States

INTRODUCTION

Electronic and vibrational processes at the atomic scale create narrow diagnostic features in the visible and particularly in the Short Wave Infra-Red (SWIR) reflectance spectra of rocks (e.g. Hunt and Ashley, 1979) and SWIR spectroscopy can be used to identify sulfates, carbonates, chlorites, silicates and phosphates. Minerals, such as kaolinite, dickite, alunite, muscovite, jarosite and illite, which are present in gold-bearing alteration systems can be identified through their absorption spectra. These minerals are important constituents of altered rocks and there are now several commercially manufactured hand-held SWIR spectrometers in common use on the ground (e.g. PIMA©). A number of airborne SWIR sensors have also appeared, with the aim of producing detailed image maps of much larger areas than can be produced by a geologist walking the terrain. Much of the science of airborne mineral mapping has come from a NASA program involving the AVIRIS (Airborne Visible/Infrared Imaging Spectrometer). AVIRIS acquires 224 channels in the Visible and Near Infra-Red (VNIR) and SWIR, and has been shown in a long series of scientific papers to be a very effective mineral mapping tool (e.g. Goetz, et al., 1985). It has become the standard by which all other SWIR instruments are measured.

The canada centre for remote sensing (ccrs) of natural resources Canada has designed and constructed another SWIR airborne sensor, called the SWIR Full Spectrum Imager (SFSI). SFSI was originally intended to provide an alternative source of SWIR data for research and to provide higher spatial resolution imagery than the 20 m pixels produced by AVIRIS (Neville et al, 1995). SFSI also operates in the Short Wave range, from 1220 nm to 2420 nm and has the ability to simultaneously acquire the full spectrum at high spatial (to 20 cm) and spectral (10.4 nm) resolution. The instrument utilizes a two-dimensional detector array, refractive optics and a transmission grating with an angular field of view of 9.4 degrees. Any number of bands between a minimum of 22 and the full set of 115 can be recorded. . Because it was designed as a research instrument, not as a survey tool, the system had a narrow field of view and did not write to tape. The system is currently being modified for a 35-degree field of view and continuous recording more suitable for survey operations.

The purpose of this note is to compare mineral maps produced by the SFSI at 4 m ground resolution, using a process of spectral unmixing, with maps produced from 20 m resolution AVIRIS data, using the TRICORDER 3.3 algorithm.

METHODS

The SFSI data discussed here was acquired 1.7 km north of the village of Cuprite, Nevada on June 21, 1995 at approximately 2348 GMT [1548 solar time] with a relatively low sun elevation [38 degrees]. The co-ordinates are 37° 32' 38" N and 117° 11' 2" W. Motion effects due to aircraft attitude were minor and therefore not corrected. The target area is of moderate relief, with the highest points near 5200'. The lowest altitude is where the watercourse, which runs from the north east corner through the centre, exits to the south west at 4950' (Figure 1).

The SFSI image data was radiometrically calibrated, then corrected for atmospheric effects and converted to pseudo-reflectance at the Canada Centre for Remote Sensing, using a modified MODTRAN atmospheric code. Classification of the SFSI data was accomplished by a Linear Unmixing. Six Spectral End Members were extracted, corresponding to shadow, alunite, kaolinite, buddingtonite, and calicite as recognized by the shape of their spectra (Neville et al., 1997). The sixth end member has not yet been identified. The original end members were continuous tone images. For this comparison, these continuous tone images have been reduced to a single value polygon above a threshold. The threshold was selected in such a way as to isolate what appeared to be a separate population in the histogram. Although the original SFSI data was acquired and processed at approximately 1 m pixel size, the data presented here has been resampled by block averaging to 4 m pixel spacing. A true simulation of the planned pixel resolution of the sensor after modification for surveying, using the block averaged data for classification has not yet been carried out.

The AVIRIS data used here comes from the United States Geological Service Website at speclab.cr.usgs.gov/map.intro.html and represents data acquired on June 23, 1995 just before solar noon, at 18:45 GMT with a solar elevation of 70 degrees. The USGS procedure includes radiometric calibration, atmospheric correction using ATREM, and classification using the Tricorder 3.3 software (Clark and Swayze).

Neither data set was geo-located, so the first step in comparing them was to bring them to a common geolocation. A 1:24,000 scale Topographic map was scanned and geo-referenced. A true colour simulation and the classified mineral map(s) from each scanner were then brought to this reference using Ground Control Points picked from the topographic map and the imagery. The result is a common database with both sets of data in it.

RESULTS

Figure 2 illustrates the SFSI end members for Alunite, Buddingtonite, Kaolinite and Calicite overlaid on the navigated SFSI image. Figure 3 illustrates the same classes derived from the AVIRIS data. At this time, the SFSI classification is relatively simple compared to the very fine discrimination of the AVIRIS data. For this comparison, the AVIRIS sub-classes have been grouped into classes similar to those derived from the SFSI data. For example, our 'silicate' class corresponds to the AVIRIS 'Silicate' and 'Calcite + Kaolinite' classes. Note that the SFSI classes shown do not extend into the shaded areas. At this stage in the development of our classification, one of the end members is a Shadow class.

DISCUSSION

Neville et al. 1997, have shown very good agreement between SFSI spectra and PIMA ground spectra after atmospheric correction and conversion to reflectance (Figure 4). Figures 2 and 3 illustrate that there is also good overall agreement spatially. The location of the four minerals mapped here is strikingly similar given that the data was acquired by two different instruments, on different days, and at different times of the day, and that the data processing and classification methods are rather different. Some of the differences are clearly a result of the finer spatial resolution of SFSI compared to AVIRIS. In order to do a more accurate comparison, we need to block average the SFSI data before classification - some of the smaller features in the SFSI classification may not occur when the pixels are larger and the signals are mixed.

It is obvious that the SFSI "End Member" classes do not map the shaded side of the hills, whereas for the AVIRIS data (which was acquired at a much higher solar elevation), the classes do. This is primarily a reflection of the classification algorithms used. There is a good separation of the SFSI mineral classes, and very little confusion of the classes. At this point, the SFSI classes are derived from the Spectral End Members by simple, rather subjective thresholding. With more sophisticated algorithms we might hope to do considerably better.

REFERENCES

Agar, B. 1996. "Multispectral remote sensing". Mining Opportunity Bulletin, Supplement. Randol International., v. 3, 2: b1-b2.

Clark, R, and G. Swayze, 1997. "Imaging Spectroscopy Material Maps: Cuprite". At speclab.cr.usgs.gov/cuprite.html

Goetz, A. F. H., G. Vane, J. E. Solomon and B. N. Rock., 1985. "Imaging Spectrometry for Earth Remote Sensing". Science, v. 228, p. 1147-1153.

Hauff, P., P. Kowalcyzk, M. Ehling, G. Borstad, G. Edmundo, R. Kern, R. Neville, R. Marois, S. Perry, R. Bedell, C. Sabine, A. Croasta, T. Miura, G. Lipton, V. Sopuck, R. Chapman, M. Tilkov, K. O'Sullivan, M. Hornibrook, D. Coulter and S. Bennett. 1996. "The CCRS SWIR Full Spectrum Imager: Mission to Nevada", June, 1995. Eleventh Thematic Conference on Applied Geologic Remote Sensing., Las Vegas, Nevada, February 27-29, 1996. p. 417-425

Hunt, G. R. and R. P. Ashley, 1979. "Spectra of altered rocks in the visible and near infrared". Economic Geology v 74, p. 1613-1629.

Neville, R. A., K. Staenz, T. Szeredi and P. Hauff. 1997. "Spectral Unmixing of SFSI imagery in Nevada". Proceedings of the 12th International Conference on Applied Geological Remote Sensing. 17-19 November 1997, Denver Colorado, v.II, p. 449-456.

KEYWORDS: SFSI, AVIRIS, remote sensing, mineral mapping, Cuprite

Figure 1 (left). The continuous tone SFSI image.

Figure 2. (centre) The continuous tone SFSI image, with its Spectral End Member maps for Alunite (red), Kaolinite (yellow), Buddingtonite (pink) and Silicate (white) overlaid.

Figure 3 (right). The continuous tone SFSI image with the same classes derived from AVIRIS mineral maps of the same area. The blue class is "calcite and kaolinite".

Figure 4. Comparison of Spectral End Members derived from airborne SFSI data with PIMA reflectance spectra obtained from ground samples
(Neville et al., 1997).

Click HERE to return to Papers on Geology and Mining