New Geospatial Approaches to the Anthropological Sciences
edited by Robert L. Anemone and Glen C. Conroy
New Geospatial Approaches
                to the Anthropological Sciences

Chapter 4. Assessing Unsupervised Image Classification as an Aid in Paleoanthropological Explorations


GC Conroy, A Chew, KD Rose, TM Bown, RL Anemone and GF Gunnell

Additional Chapter details here

Figures

Fig. 1. A map of the Bighorn and Great Divide Basins, Wyoming.

Fig. 2. An unsupervised classification (Iso Cluster).

Fig. 3. (a) A satellite view of Bighorn Basin with known Wasatchian fossil localities (black dots). (b) A satellite view of Bighorn Basin with known Wasatchian fossil localities (black dots), draped over a topo map of the Bighorn Basin. (c) An unsupervised classification of Bighorn Basin divided into five groups or clusters. (d) An unsupervised classification of Bighorn Basin divided into fifty groups or clusters.

Fig. 4. A dendrogram showing attribute distances between sequentially merged classes in teh signature file. Potential fossiliferous sediments were mainly associated with spectral signature clusters 44 and 46-50.

Fig. 5. Higher-resolution (a) topo, (b) satellite, and (c) unsupervised classification images of the Paradise Alley region (scale = 1:24,000). Eroded sediments that seem to be good targets for fossil prospecting tend to be associated with the "browner" signature clusters 44 and 46-50 (red dots = known fossil sites).

Fig. 6. Spectral signature clusters 44 and 46-50 (red) draped over satellite image of Bighorn Basin.

Fig. 7. Spectral signature clusters 44 and 46-50 (red) and Wasatchian fossil sites (black dots). Note the westernmost arc of spectral signature clusters 44 and 46-50 between the black curved lines.

Fig. 8. Fossil localities discovered in 2014 (red dots). Note that most newly discovered sites are in the westernmost arc of spectral signature clusters 44 and 46-50, as predicted by the model.

Fig. 9. A pan-sharpened Landsat 8 image of the Oregon Buttes and Continental Peaks region of the GDB "remotely" explored in this study.

Fig. 10. (a) The total extent of Landsat 8 satellite coverage of the great Divide Basin "remotely" explored in this study; (b) The results of unsupervised image classification over the satellite coverage area (each individual spectral signature cluster is based solely on the spectral signatures of the landscape - without a priori knowledge of what these land covers actually represent; (c) Note the light-colored sediments associated with the Tim's Confession site; (d) An unsupervised classification of same image as (c) (note that these light-colored sediments associated with Tim's Confession fall within spectral signature cluster #9 (dark blue).

Fig. 11. (a) Blue pixels represent areas in the "remotely" explored area most likely to yield Eocene fossils based on an unsupervised classification model using the spectral signature of Tim's Confession (these highlighted pixels fall within spectral signature cluster #9, see fig. 4.10); (b) The actual location of Eocene fossil localities in the "remotely" explored area (red dots are sites discovered by Gunnell's teams; black dots are sites discovered by Anemone's teams: note the close approximation of all known fossil localities to the predicted pixels within spectral signature cluster #9); (c) Even those localities not actually within cluster #9 pixels are still within 30 m of the predicted pixels, as seen by the 30 m circular buffers drawn around each fossil locality (note that 30 m is the size of 1 pixel in these Landsat images); (d) The area of spectral signature cluster #9 (approximately 12,500 km2; blue pixels) within the total area of the GDB covered by the Landsat 8 image in fig. 4.2a (approximately 38,000 km2)