A Shelby Eagleburger – Contextual Survey of Archaeological Walls in Ka’a’awa Valley


[ hide ]

    Presentation: A Contextual Survey of the Southern Walls

    Final Map

    Abstract: The purpose of this research was to determine if there are any commonalities between the walls that run perpendicular to the ridge on the southern slope of the Ka’a’awa Valley in terms of position, context, and construction that may be useful for the eventual production of a predictive model intended to locate similar features in other analogous areas based on a common set of attributes. The secondary goal was to create a more complete understanding and map of what features were present in the area and finding a way to effectively represent them visually since traditional methods proved non beneficial due to many environmental constraints. A number of geographic mapping techniques were used for both the collection and analysis of the data.




    The research was conducted in Ka’a’awa Valley located in Kualoa Ranch. Kualoa Ranch is positioned on  the North Eastern portion of Oahu, Hawai’i. The valley itself is part of a larger mountain system and the wide mouth of the valley faces the coast. It is somewhat uncharacteristic in comparison to its neighbors and counterparts on different parts of the island due to the wide flat bottom that it possesses. Our research area was focus on the inner part of this valley, however my own research ended up taking me outside the initial grid. The valley has a long history of historical human use including sugar cane growing and cattle ranching. These practices are obviously destructive simply by their very nature as most any human activity is. It is presumable that any prehistoric evidence of occupation or use is at least somewhat disturbed. That being said it is also true that there are most likely areas that have not seen as much use true to accessibility or other factors and may contain more intact prehistoric anthropogenic features.

    After being a part of the archaeological group focusing on surveying the valley for the first week while producing the common product and surveying the majority of the vast landscape the very stone walls seemed to be the most distinctive. They are located on the south facing slope immediately on the other side of the prescribed study area. They seemed relatively undamaged in comparison to most of the features in the main valley. Those within the main grid looked to have been moved by people, cattle, or both.





    A systematic survey was the first step in order to collect relevant data. After defining a starting point where it was accessible transects were taken and point collected with the Trimble at both the start and end points. Points indicating other non-wall features were also made for things such as possible platforms, terraces, or rock scatter, even though these were not a part of my study focus. In addition to the X, Y, Z points, slope reading at three points on each wall using the clinometer were taken. A tape measure was also utilized in order to obtain the height and width of each wall segment in its seemly most complete portion. Each clinometer reading was accompanied by a picture of the instrument display which in itself was geotagged. In order to later stitch them together in a three dimensional rendering a large set of photos, often numbering in the hundreds, were taken at opportune locations with the least vegetation and most maneuverability. Extensive sketches and notes were another very important piece of data documenting the context of each individual feature. In order to systematic approach walls were classified as features that prescribed to three basic traits: linear in shape, composed of unworked rock, and exist in a stacked agglomerated structure. Only if it followed these rules was it classified as a wall.




    There were many obstacles both during the data collection and analysis stages. Vegetation was probably the most persistent problem during the collection stage of research. I both hindered travel, making moving forward a struggle, but also blocked any visible features almost completely. It was impossible to spot a wall until it was immediately under you. The trees also added to the problem of gaining enough satellite signal to take a point at all, much less an extremely accurate one. Having more than four or five satellites at any given time was extremely rare and it often took some creative measure to find a break in the canopy including climbing some of the trees. The aerial visibility was even worse on the World View II imagery. They are completely invisible and would never be identifiable with the typical true color representation.

    Identifying what was a wall was in itself an issue. Having no previous experience in the area, it was difficult to assign names to what was on the ground. The classification system was intended to combat this problem, however the system itself is inherently biased and somewhat arbitrary. The time constraints were also a significant impeding force. It was difficult to be thorough over such a large area in the time period allotted.





    After taking the points they were imported and differentially corrected using the base station data. A line was then created connecting each start and end point and added the relevant attribute fields. Although the line features were visible on the map they were not a very good visual representation of what was on the ground so a terrain model was created in ArcScene using a DEM and the World View II Imagery. The first attempt was less than satisfactory due to the 30 meter resolution of the DEM. Once a 5 meter digital terrain model (DTM) was acquired the results were much more accurate. The LiDAR data that was provided also encompassed the area in question and therefor it was added to the terrain model in order to achieve even better resolution. When wall layer was added and extruded, floating it to the same base heights as the DTM it became gave a much clearer picture of the area. Using the DTM, slope, elevation, and aspect were able to be found. Comparing these layers to the position of the walls yielded interesting results. Two other main processes were ran on the lines: Average Nearest Neighbor and Linear Directional Mean. Average nearest neighbor was used to calculate the average distance of each line to the nearest line and from that determine if the group was seemingly random or associated. The directional mean function which extracted the mean length, compass angle, and circular variance (how much orientations deviate from the mean) between each line. This is useful to determine directionality patterns. The average height and width were calculated manually. The many photographs that were taken were processed in both Photoscan and Photosynth with varying results.



    Results and Conclusions


    Six walls were surveyed in total. The slope and elevation analysis yielded very interesting results with all of the walls falling consistently within a range of 100 to 200 meters above sea level according to the contour lines and DEM, although it is subject to debate whether this was an accurate data set. It seems that there may be some error within the data as far as Z values go. The walls were also built on the faces with a slope between 15 and 24 degrees. These numbers corresponded with the clinometer measurements taken while in the field.  The mean length of the walls was approximately 56 meters, mean height: 1.24 meters, and mean width: .76 meters. These numbers were relatively consistent with no extreme outliers. The mean compass angle (clockwise from due north) was 90.4108 degrees. The Circular Variance (how much orientations deviate from the mean) was 0.042463. This small number indicates that the directionality of each of the walls were very similar to one and other. Three of the walls had a northern aspect, two northeastern, and one southeastern. The average nearest neighbor function also produced some potentially useful information. The mean distance between each wall was about 36 meters. It yielded a Z-score of 5.15449. The Z-score represents the relative distribution such as how clustered, random, or evenly distributed items are. A score above 2.58 indicate an even non-random distribution. The Photoscan and Photosynth products were useful in different ways. Photosynth was good for getting a larger picture, however it was not particularly good in the 3D realm. Photoscan was troublesome at first but after some trial and error it was able to render a good interactive model of a wall section.

    All in all, this research project has brought to light some interesting correlations that may be used to create a preliminary predictive model for locating similar features in other areas. To create a model and test it would be the next logical step to take in the continuation of this project. Looking at historical maps could also be useful to see if there are any that depict the walls in a historical context.


    Thanks to Dr. Wechsler, Dr, Lee, Dr, Becker, Dr. Lipo, Paul Nesbit, Emily Allen, Michelle Baroldi, Briton Voorhees, Mike Ferris, Ted Ralston, Dr. Hunt, Scott Winslow.


    Sponsored by the National Science Foundation

    Grant Number: 1005258