Issue
Table of Contents

Remote Sensing in Paleontology:
MALAKHOV, DYKE, & KING

Plain-Language &
Multilingual  Abstracts

Abstract

Introduction

Rationale, Background, and Analysis

Spectral Analysis

Conclusion

Acknowledgments

References

 

Print article

 

 
 

RATIONALE, BACKGROUND, AND ANALYSIS

Terminology. Landsat ETM +- Landsat Enhanced Thematic Mapper: sensor that provides imagery at 28.5 meter spatial resolution (14.5 meters after image sharpening in three visual [blue, green and red] and four infrared bands). RSI ENVI: software developed for processing satellite images. ISODATA: algorithm that classifies pixels evenly distributed in the dataset and combines them into separate classes based on statistically significant minimum distances. Spectral Angle Mapper - SAM: physical spectral classification that uses the n-dimensional angle (one pixel characteristic) to match pixels to their corresponding reference spectra. This algorithm determines the spectral similarity between two spectra by calculating the angle between them, treating them as vectors in a space with dimensionality equal to the number of bands. Supervised classification: a set of methods for image classification that require a “region of interest” (ROI), or spectral library, created by the researcher. All pixels within the image are considered along with those included in the ROI. Unsupervised classification: does not refer to any predefined criteria to make a statistical consideration of all pixels and to distribute them into classes.

Study rationale. In contrast to the recent study of Oheim (2007), who analyzed a series of environmental factors considered useful for the prediction of likely fossil-bearing sites using GIS software (ArcInfo 9.1), we have focused our attention on the characterization of the spectral parameters of already known fossil sites identified by our earlier field work in the Syrdarya area (Dyke and Malakhov 2004). Once characterized, spectral parameters can then be re-applied to other regions to predict locations of fossiliferous sites. There are a number of logistical reasons for this: (1) Our study area is remote (the closest village is 90 km away; Figure 1) and not easily accessible; and (2) Previously collected fossil remains have been found at high abundances but concentrated in small areas scattered across the field area (approximately 17,000 square kilometers). These factors highlight the need for precise determination of likely sites rather than indeterminate prospecting effort (at high cost). For large land areas in particular, the development of spectral libraries as an aid to identifying likely fossiliferous strata seems a reasonable approach, albeit not as precise as the analysis of local environmental variables using GIS (Oheim 2007). Spectral analysis, however, can produce sets of ROIs for prospecting paleontologists especially in distant and poorly known land areas.

Thus for us to repeat the analysis of Oheim (2007) in the Syrdarya field area would require a complete database of environmental conditions, and these kinds of data simply do not exist for this region of Kazakhstan (which has very poorly developed local infrastructure). In addition, we do not believe that environmental factors significantly affect the presence and/or preservation of fossils across the Syrdarya area: for more than 17,000 square kilometers across this region of Kazakhstan, the terrain is very smooth and uniform (without rivers or significant water-bodies) and is covered with poor xerophilic vegetation that does not change significantly throughout the summer and thus does not affect the spectral reflectance of the area.

Paleontological and geological background. Since 2002 we have spent a series of field seasons in the Syrdarya field area and so have ‘ground-truthed’ fossil collection and distribution data (e.g., Dyke and Malakhov 2004; Averianov 2007). These data have enabled us to test the predictions of remote spectral analyses via exact GPS coordinates from earlier fossil finds. We have also used geologic maps (1:500 000) (e.g., Figure 2) to validate the results of image classification. Note that original geological maps for this area of Central Asia are of little value for planning field work because they indicate that Upper Cretaceous sediments are developed across a wide area: this is not the case.

The geological and paleontological context of the Syrdarya area has been reported in earlier papers (Kordikova et al. 2001; Dyke and Malakhov 2004; Averianov 2007) (Figure 3). In general, the fossiliferous strata across the area are color-mottled alluvial clays (Figure 4) of the lower Bostobe Formation [Bostobinskaya Svita] interbedded with thin overbank siltstones and channel-filling fluvial sands and sandstones (Figure 3). The fossil fauna of the Bostobe Formation comprises many common vertebrate groups (Dyke and Malakhov 2004) and includes abundant dinosaur bones, crocodile teeth and turtle carapaces as well as other taxa represented in smaller quantities (i.e., sharks, lizards, pterosaurs and urodelans as well as bivalves and fragments of petrified wood). Faunal composition also varies through the Bostobe Formation as the sediments change from semi-aquatic in the lower portion (i.e., bony fish, sharks, amphibians, turtles, crocodiles, small theropods and hadrosaurs) to more terrestrial (i.e., wood fragments, large theropods) in the upper portion of the sequence. Cretaceous sediments in this area are mostly almost flat-lying and are overlain by Paleogene marine clays and localized Neogene sediments of the Paratethys in some upland areas. Quaternary alluvial and fluvial sediments cover extensive low-lying areas across the Syrdarya Uplift.

Analytical protocol. All procedures, except compilation odataaf basic maps, were performed with RSI ENVI 4.2. The following manipulations were used to preprocess the original Landsat fileset: (1) Bands 1 - 7 (including thermal bands 6.1 and 6.2), originally represented separately in GeoTIFF format, were stacked into a single multiband file with pixel size set to 28.5 m; (2) The file was further ‘pansharpened’ to 14.5 m resolution; (3) The total image was cropped to delimit the study area; and (4) Several band combinations were visually tested (compare Figure 5 and Figure 6). Note that recovered images were subject to atmospheric and geometric correction.

 

Next Section

Remote Sensing in Paleontology
Plain-Language & Multilingual  Abstracts | Abstract | IntroductionRationale, Background, and Analysis
Spectral Analysis | Conclusion | Acknowledgments | References | Appendix
Print article