Viewshed at the Chaves/Hummingbird Site.

Broad alluvial basins, with deep arroyo cutting, bordered by sandstone ridges and small mesas characterize the landscape of the Chaves Ranch. Within the alluvial basins, areas of extreme sheet wash feed into numerous arroyos, some of which are as deep as 2 meters. The topography of the ranch is highly variable, with elevations ranging between 5400 and 5700 feet. At the end of the southeastern edges of the ranch, sandstone ledges have formed in eroded exposures. Medium-grained aeolian sands, with some clay, characterize the major portion of the soils on the ranch, and alluvial deposits contribute considerably to the soil sediments in and around the Canada de los Apaches, which bisects the ranch, running from northwest to southeast. The Canada del Ojo crosses the northeast corner of the ranch and is also characterized by a high rate of deposition. Deposition in proximity to this area is so extreme that only the upper six inches of a fenceline constructed in the 1930’s is left exposed.

The nearest permanent water source is apparently the Rio Puerco, a perennial tributary of the Rio Grande, and thus far there is no evidence to the contrary for the prehistoric period.

Chaves Ranch biota is best described as a low-shrub desert, with creosote bush, cholla, yucca, agave and various species of grasses typical of the area. The ranch is rather impoverished with trees, though junipers do occur, sparsely distributed across the landscape. Fauna typical of this type of environ include various rodent and deer species, as well as coyotes.

Problem Statement

This project aims to contribute to the research at Chaves Ranch in two ways. First, the problem of illuminating the landscape characteristics that may have contributed to the selection of the site of the Chaves/Hummingbird Ruin as the location of a human settlement which lasted for over 200 years is examined. In particular, the proposition that the Hummingbird Pueblo was located based upon defensive characteristics (Adler 2004) is examined using a combination of site viewshed and aspect analysis to evaluate the defensive position of the site relative to the surrounding landscape.

Second, the limits of human visual perception are combined with the spatial characteristics of surrounding landscape to map the visual domain of the Chaves/Hummingbird site. It is hoped that this visual domain may be used as a means of defining the past perceptual boundaries which defined the settlement limits to its inhabitants. Additionally, it is hoped that these boundaries might be used to define regions of interest for further investigation of cultivation and food production at the site.

The two approaches represent the application of similar methods at different relative scales. The first approach examines the visual and spatial characteristics of the Chaves/Hummingbird Ruin relative to the surrounding landscape. The second examines the spatial characteristics of landscape visibility within boundaries defined by the physiological limits of human vision.

Literature Review

Vision and Visibility

The use of the physiological limits of human vision in the delineation and characterization of landscape visibility has been visited thoroughly, though infrequently, in the past. The most notable examinations, however, have been applied to problems of landscape architecture (Higuchi 1975), or to the inventory of visual aspects of forest landscape (Litton 1968). These studies used an observer based visibility analysis, in which the limiting attributes of the observer (height, range of vision, visual angle of ascent or depression, etc…) interact with the spatial characteristics of the landscape (topography, land cover, etc…) to create mappings of the perceived landscape, delineating between fore and background, and introducing a model of visual distance decay which provides a framework within which to explore the relative cognitive impact of visible surfaces upon the observer (Higuchi 1975).

Higuchi (1975) specifically defines visual space in terms of how we perceive “here” and “there” based upon visual proximity. His indices of viewshed distance divide a landscape into three classes: near-distance; middle-distance; far-distance. The near distance corresponds with the visual extent within which the features of a standard observed object are visible and distinguishable. This viewshed class defines the “intimate” space defined by human vision, within which individuals are identifiable. The middle-distance describes the visible range within which, while not individually distinguishable, the standard observed object is visually distinguishable from the landscape. Within this viewshed class, a human would be visible and distinct from their surroundings, though they would not be individually identifiable. Beyond the middle-distance, the far-distance described the distance beyond which the standard object is no longer discernable from the landscape. Figure 4 shows a photo of the Chaves landscape classified based upon the bushes in the foreground as the standard object, and illustrates the phenomenon of visual distance decay described by Higuchi’s index.

Figure 4 Shows the Chaves Landscape Classed by Visual Quality according to Higuchi (1975).

Archaeological Visibility

The application of viewshed analysis is not widely used in archaeological application to date, though there are some notable examples of the examination of visibility in the field. One examination of viewshed (Fisher and Farrelly 1997) provides an analysis of contemporaneous Bronze Age Cairns on the Isle of Mull and their visual relationship to one another, and to the surrounding sea. By examining the overlap of cairn viewsheds it was revealed that they shared a common view of the sea to the north of the island, suggesting some perceptual relationship beyond the temporal association of the sites.

Lake, Woodman and Mithen (Lake, Woodman et al. 1998) have developed a method called Cumulative Viewshed Analysis which describes the relative prominence of features across a landscape by analyzing the viewshed of the entirety of the landscape. In concept, the idea is to perform a viewshed analysis using each cell of a topographic grid as an observer point, then to obtain the sum of the viewsheds which yields a grid of values describing the total number of points that each cell is visible from. In practice, the authors implement the methods using random sampling and concern themselves more with the process that the application to any archaeological problem.

The methods used in this project were first discussed and suggested for archaeological application by Wheatley and Gillings (Wheatley and Gillings 2000). Though their work concerned itself with creating a methodology of “enriched approaches to the study of viewsheds,” more than application, they describe the used of Higuchi’s indices and provide descriptions of methods for implementing those indices. Specifically, the means of evaluating direction in viewshed was described as it is used in this project, as well as an alternative method of distance classification which was not used in this study.

Data

DEM

The Digital Elevation Model used in this project was created using ASTERDTM, an extension for Research Systems’ ENVI image processing software, which extracts a Digital Terrain Model (DTM) from an ASTER level 1b multispectral satellite image file. ASTER scenes provide a resolution of 15 meters at the relevant spectral bands (3N and 3B).

Figure 5 Digital Elevation Model Based Extracted from Aster Data.

ASTERDTM uses the data of two telescopes; the nadir-looking VNIR (band 3N) and the backward looking-VNIR (band 3B) to extract height information. The band 3B data is acquired approximately 30 seconds after the band 3N data, creating a stereo pair. The extraction process matches the two scenes on a pixel by pixel basis, notes the parallax and then calculates the relative or absolute height (based upon the use of optional Ground Control Points). It should be noted that extraction of elevation values using ASTERDTM is time intensive. At a resolution of 15 meters, the extraction process takes over 2 hours per scene.

The resulting DEM has the following specifications:

Projection: NAD27, UTM zone 13N
15 meter resolution
~10 meter vertical and horizontal accuracy
~60 X ~60 kilometer extent

The ASTER scene used in this project was acquired on June 4, 2001 by The Advanced Spaceborne Thermal Emission and Reflection radiometer (ASTER) onboard the Terra spacecraft, launched in 1999. The scene was obtained from the NASA Jet Propulsion Laboratory through the University of Texas at Dallas, Geosciences Department.

Observer Points

Observer points were collected in June of 2004, using a Garmin ETREX Vista GPS unit with a vertical and horizontal accuracy of 13 feet. For each of the two relevant locations, the site and the mesa, GPS readings were taken at multiple locations. For the site mound points, four readings were collected at the cardinal positions defining the architectural extent of the site mound and a single point collected at the central point (and elevational apex) of the site. The eleven mesa observer points were collected, at arbitrary intervals, walking the perimeter of the mesa-top.

Once collected, these points were imported to ArcGIS as X,Y,Z events, converted to shapefile form and projected to the NAD27 UTM zone 13n coordinate system.

Before use in the viewshed analysis, it was necessary to populate the attribute tables of each observer point shapefile with fields used by Arc visibility analysis as optional limiting parameters. These parameters were used as proxies for the limitations of human vision and are described in detail later in this paper. After adding the viewshed parameter fields to each of the original two observer points shapefiles, these files were used as templates to produce the final observer point files containing the limiting parameters for each of the viewshed analyses used in this project.

Finally, because viewshed analysis is a processor intensive procedure, and to provide batch processing capabilities, these shapefiles were exported to coverage format for use in the ArcToolbox version of the visibility tools, which provide more efficient use of computing resources than ArcMAP.

Methods

Refining Viewshed Analysis Using the Spatial Parameters of Visual Perception

ESRI’s ArcToolbox visibility analysis tools provide a number of optional parameters which may be used to limit and control the output of the process. These optional parameters include:

· OFFSETA – Used to provide the observer height, to be added to the elevation value at each observer point. (Default = 0)

· OFFSETB – Used to provide the observed object height, to be added to the elevation value at each observed mesh point. (Default = 0)

· RADIUS1 – Minimum visible distance from observer point. (Default = 0)

· RADIUS2 – Maximum visible distance from observer point. (Default = infinity)

· VERT1 – Maximum visible angle of elevation from the horizontal. (Default = 90)

· VERT2 – Maximum visible angle of depression from the horizontal. (Default = -90)

Figure 6 Offset as Implemented in ArcToolbox.

Figure 7 Radius as implemented in ArcToolbox.

These parameters are implemented at runtime when the visibility analysis tools encounter the parameters within the attribute table of each observer point coverage (Figure X). When these fields do not exist, visibility analysis tools assume the default values for each parameter.

Figure 8 The Feature Attribute Table of the Observer Point File for a "Middle-Distance" viewshed from the Mesa.

Because, as observed previously, a raw viewshed analysis using the default parameters may yield visibility results outside the physical limits of human vision, visibility must be ‘limited’ by the parameters of human visual perception in order to make meaningful statements about the perception of the targeted landscape. Using the visibility indices identified by Higuchi (Higuchi 1975) as a guide, the attribute tables of the observer point files were populated with the appropriate fields, which were then populated with the parameters for each desired viewshed (Figure X).

Table 1 Applying Higuchi's Visibility Indices in ArcToolbox.

Higuchi Index	ArcToolbox Visibility Parameter	Parameter Value
Standard Object of Observation	OFFSETB	2 meters (~human height)
Standard Observer Elevation	OFFSETA	2 meters (~human height)
Near-Distance Visibility (Minimum)	RADIUS1	0 meters
Near-Distance Visibility (Maximum) (~60 x standard object height)	RADIUS2	120 meters
Middle-Distance Visibility (Minimum) (~60x standard object height)	RADIUS1	120 meters
Middle-Distance Visibility (Maximum) (~1100x standard object height)	RADIUS2	2200 meters
Far-Distance Visibility (Minimum) (~1100x standard object height)	RADIUS1	2200 meters
Far-Distance Visibility (Maximum) (Infinity)	RADIUS2	Infinity (default)

This process resulted in four observer point files each for both the mesa and the site mound, yielding eight observer point files, in all. The final observer point files are described below:

Table 2 Description of the Observer Point Files.

Filename	Description	RADIUS1	RADIUS2	OFFSETA	OFFSETB
mesa_near	Describes the attributes necessary to produce a viewshed based upon Higuchi’s Near-Distance index, from the mesa.	0	120	2	2
mesa_mid	Describes the attributes necessary to produce a viewshed based upon Higuchi’s Middle-Distance index, from the mesa.	120	2200	2	2
mesa_far	Describes the attributes necessary to produce a viewshed based upon Higuchi’s Far-Distance index, from the mesa.	2200	infinity	2	2
mesa_all	Describes the attributes necessary to produce a raw viewshed, from the mesa.	0	infinity	2	2
ruin_near	Describes the attributes necessary to produce a viewshed based upon Higuchi’s Near-Distance index, from the ruin.	0	120	2	2
ruin_mid	Describes the attributes necessary to produce a viewshed based upon Higuchi’s Middle-Distance index, from the ruin.	120	2200	2	2
ruin_far	Describes the attributes necessary to produce a viewshed based upon Higuchi’s Far-Distance index, from the ruin.	2200	infinity	2	2
ruin_all	Describes the attributes necessary to produce a raw viewshed, from the ruin.	0	infinity	2	2

Running the Viewshed Tool

Once all observer point files were prepared, ArcToolbox was used to perform the visibility analysis in batch mode. As noted previously, visibility analysis is a processor intensive process. Running a single viewshed analysis with five observer points on the DEM used in this project took as long as 90 minutes, making attended, one-at-a-time running of the analysis problematic. The batch processing capabilities of ArcToolbox provided the means to “queue” multiple viewshed processes to run consecutively and unattended. Additionally, the ArcToolbox batch processing tools provide the ability to save batch processing schemes to an *.aml file, allowing the analyses to be repeated without the reentry of the batch parameters. This feature is particularly useful when errors occur in unattended processing, and the analyses must be repeated after changes are made to input files.

Figure 9 Example of an ArcToolbaox AML script for Batch Processing Visibility

Visibility analysis produces a grid of cells whose value is the number of observation points in the input coverage that are visible from each individual cell. Because the observer point files used in this case each contain multiple observer points, the visibility grid files yielded by the process are coded with values from 0 (not visible) to (visible), where n is the number of input observer points. These visibility grids were then reclassed to provide the binary grids (0 = non-visible, 1 = visible) desired for obtaining data about directionality of viewshed.

Determining “Directionality” in Viewshed

One characteristic of particular interest in this project was determining whether there was any particular “directionality” of viewshed from either the mesa or the site mound. To quantify this characteristic, a method suggested by Wheatley and Gillings (Wheatley and Gillings 2000) was used. This method involves the creation of a conical pseudo-DEM, with the relevant observer site at its apex. This conical DEM is used to produce an aspect grid, coded with the direction relative to the observer site. This process was repeated for both the mesa observer points, and the ruin observer points.

The first step in the process was to buffer the chosen observation point (a single, central point from each observer point file was used) with 1000 concentric rings, at a distance of 15 meters each. This yields a set of circular buffers with a diameter of 30 kilometers (well beyond the visible range identified previously). This file was then exported to a raster format, using the ToBufDistance as the value field. In order to produce the final conical DEM required, the grid file was then inverted using Raster Calculator by subtracting the maximum cell value from all cells:

conedem = inputgrid – (maximum inputgrid value)

Figure 10 Conical DEM Created from an Observer Point

Using ArcMAP’s Spatial Analyst, the conical DEM was then used as the input grid for an aspect analysis, which yielded a grid coded with cell values of 1-8, corresponding to direction, relative to the observation point used to produce the files.

Figure 11 The Aspect Grid Created from the Conical DEM

This final aspect grid was then used to code the binary viewshed grids with the direction of each cell relative to the observation point. This was done, again using raster calculator, by multiplying the aspect grid by the viewshed grid:

Directiongrid = aspectgrid * viewshedbinary

The final result of this process is a grid file of all cells visible from the given observer point, each cell coded with a value between 1-8 corresponding to its direction relative to the observer point.

Figure 12 Directionality Grid for the Mesa Visual Domain

In order to quantify the directionality of each viewshed, the frequency of each directionality value within the viewshed was obtained from the final directionality grid and plotted graphically using Microsoft Excel.

Figure 13 The Value Table for a Direction Grid of the Site Mound

Results

Raw Viewshed

Raw viewsheds, coded with the number of observer points visible from each grid cell and without limiting parameters, were processed for each of the two sets of observer points. Because these viewsheds are coded with the number of points visible from each cell, they provide a rough measure of the visual prominence of the observed cells. That is, a cell with a higher value is visible from more parts of the feature used as the observation point. One thing that is immediately apparent about these visibility values is that the highest values tend to be further from the feature observed from. This is because even small obstructing features close to the observation point tend to have a greater impact upon the visibility values at the distal side of the obstruction from the observer point.

Figure 14 Obstruction of Visibility by Small Topographic Variations Can Have Large Consequences (after Higuchi, 1975).

This phenomenon also contributes to the fragmentary nature of the viewshed as small topographic variations create visibility shadows, again at the distal side of the obstruction, relative to the observer point.

The raw site mound viewshed illustrates both of phenomenons, with the highest values of visibility occurring to the West-Northwest on the bluffs whose elevations are above the visual angle of elevation of all features more proximal to the observer points. It should be noted, however, that these points lie from 10 to 20 kilometers from the observer points, reducing their visual impact upon the observer. The site mound viewshed also illuminates a pronounced directionality of visibility from the architectural center of the site. Blocked to the northeast by the mesa, and to the east, south and southwest by a rise in topography, visibility from the site mound is concentrated, in particular, upon a stretch of the Canada de los Apaches, to the northwest. The inverse of this observation, of course, is that the architectural center of the settlement is concealed from all but those same areas of visibility. In particular, we observe that the visibility of the site is concealed from nearly the entire approach from the Rio Puerco, along the Canada de los Apaches (what would almost certainly be the main travel route from the site to the Rio Puerco).

Figure 15 Total Classed Viewshed of the Site Mound

Turning to the mesa viewshed, we find very different qualities of visibility. Again we find the viewshed somewhat fragmentary, with the highest values of visibility occurring at great distance from the observation points. There is also a pronounced directionality to the viewshed, in particular along virtually the entire length of the Canada de los Apaches and extending across the Rio Puerco to its far banks. It is interesting to note that while the architectural center of the settlement provides concealment from virtually all approaches to the site, and in particular from the Rio Puerco, the mesa provides a vantage point from which to observe virtually all approaches to the settlement, in particular from the Rio Puerco.

Figure 16 Total Classed Viewshed of Mesa.

Higuchi Viewshed

Of course, lacking the advantage of magnifying instruments, the long distance views observed in the raw viewsheds discussed above would likely have been of little advantage to a human observer given the limits of human visual perception. In order to characterize the viewshed to the extent that it was meaningfully perceived by the observer, it is necessary to employ the Higuchi indices as limiting parameters for examining viewshed.

Taken together, Higuchi’s near-distance and middle-distance viewsheds can be referred to as the visual domain of the observation point. This visual domain can be taken to be the limit of meaningful visibility, in which the features of a standard observed object (in this case a human being), are to some degree variable with distance, resolvable by normal human visual perception.

Figure 17 Visual Domain of the Site Mound

What we observe when we limit the site mound viewshed analysis using the Higuchi indices is that the pronounced directionality observed in the raw viewshed is still apparent. Also still present is concentration of visibility to the northwest, while we find that the proportion of the viewshed concentrated upon the Canada de Los Apaches is considerably reduced.

The visual domain of the mesa provides an interesting contrast to site mound viewshed. While in its raw form the mesa viewshed was characterized by pronounced directionality that followed the Canada de los Apaches, by limiting the viewshed to the extent that visibility would be meaningful to a human observer, we find that the viewshed is characterized by relative homogeneity in all directions same to the south. This almost total visibility in all directions provides a vantage on the landscape surrounding the settlement, while providing concealment for the architectural area from visibility in the direction of the Rio Puerco.

Figure 18 Visual Domain of the Mesa.

Directionality

As noted before, the viewshed of the site mound demonstrates a pronounced directionality, apparently to the northwest, while the mesa viewshed provides a relatively homogenous viewshed in all directions. Additionally, the mesa viewshed provides a greater magnitude of visibility that the site mound. To measure this directionality, an aspect grid with values from 1 to 8, corresponding to the cardinal directions, was created as described previously. By multiplying this aspect grid by the binary visual domain grid, we obtain a grid whose cells are coded with their corresponding direction relative to the observer point.

Figure 19 Directionality of Visual Domain from the Site Mound

Figure 20 Directionality of Viewshed from Mesa.

The frequencies of each of the direction values are then plotted to visualize the dominant directionality of the viewshed. Plotting the two grids on the same scale reveals both the dominant direction of the viewshed relative to the observer, while also revealing the magnitude of the viewsheds relative to each other.

Figure 21 Direction and Magnitude of the Mesa Visual Domain

Figure 22 Direction and Maginitude of the Site Mound Visual Domain

An alternative mapping of the magnitudes on a radial graphs with matching scales is more revealing of the relative magnitudes of the visual domains. Again, we see the greater magnitude and homogeneity of view provided by the mesa, as compared with the site mound.

Conclusions and Future Research

While making definitive claims about the defensive positioning of the prehistoric settlement at Chaves/Hummingbird could not be made solely on the basis of the viewshed characterizations discussed, the results remain interesting and certainly to not preclude the possibility. The simultaneous viewshed and concealment provide by the mesa only a few hundred meters to the northeast of the site are such a strong feature of the landscape that they almost certainly contributed to the selection of the site. It would be an interesting project to examine the viewsheds of contemporaneous settlements in the region for similar suggestions of defensive positioning.

Of particular interest is the limited visual domain of the site mound. This visual domain might be used as a starting point for the design of a systematic investigation of activity areas associated with the occupation of the Chaves/Hummingbird settlement, but not necessarily associated with the architecture of the settlement.

Finally, further quantification of the characteristics of viewsheds will be necessary for the methods to be of utility to archaeologist attempting to maintain rigorous scientific process in their investigations. The use of directionality grids provides the means to quantify, to some degree, one characteristic of viewshed, though this method requires further thought to obtain a single numeric variable that may be used to compare across viewsheds. Numeric measures of fragmentation, prominence, angle of incidence and other properties of the perceived landscape might provide future avenues of interest in the used of viewshed analysis in the characterization of archaeological landscapes.

References

Adler, M. (2004). personal communication. S. Maples. Albuquerque.

Allen, K. M. S. (2003). Considerations of Scale in Modeling Settlement Patterns Using GIS. Practical Applications of GIS for Archaeologists. K. a. R. J. B. Wescott. London, Taylor & Francis: 101-112.

Fisher, P. and C. Farrelly (1997). "Spatial Analysis of Visible Areas from the Bronze Age Cairns of Mull." Journal of Archaeological Science 24: 581-592.

Higuchi, T. (1975). The Visual and Spatial Structure of Landscape. Tokyo, Gihodo Publishing Co. Ltd.

Lake, M. W., P. E. Woodman, et al. (1998). "Tailoring GIS Software for Archaeological Applications: An Example Concerning Viewshed Analysis." Journal of Archaeological Science 25: 27-38.

Litton, R. B. J. (1968). Forest Landscape Description and Inventories. Berkeley, U.S.D.A.

Lock, G., Ed. (2000). Beyond the Map: Archaeology and Spatial Technologies. Oxford, IOS Press.

Wescott, K., R. J. Brandon, et al. (2003). Practical applications of GIS for archaeologists a predictive modeling toolkit, Taylor and Francis.

Wheatley, D. and M. Gillings (2000). Vision, perception and GIS: developing enriched approaches to the study of archaeological visibility. Beyond the Map: Archaeology and Spatial Technologies. G. Lock. Oxford, IOS Press: 1-26.