Modelling Terrain: The Global Positioning System (GPS) Survey at Kerkenes Dag, Turkey
Branting, Scott, Summers, Geoffrey D., Antiquity
This note introduces the methods and results of the GPS survey at Kerkenes Dag, a 271-hectare mountain-top Iron Age city in central Turkey. Kerkenes was a preplanned imperial foundation that was inhabited for only a few generations before being destroyed and abandoned (Summers 2000). Following test excavations over only eight days in 1928 (Schmidt 1929) the site lay dormant until the inauguration of the Kerkenes Project by Geoffrey Summers in 1993 (http://www.metu.edu.tr/home/wwwkerk/). A primary objective is the recovery of the city plan through the integration of mainly balloon photographs, the GPS topographic survey, the mapping of extant surface features, a nearly completed site-wide fluxgate gradiometer survey, more limited resistivity and electromagnetic induction surveys and precision excavations. In 2001 work was completed on a highly detailed and accurate terrain model for the entire area of the site. Using a survey grade Trimble GPS system for a total of only four months spread over four years, from 1997 to 2000, more than 1.4 million discrete elevation data points were collected and processed. These points, each of which has an accuracy of [+ or -] 10-25 cm, form the basis for the terrain model that was produced using ArcView 3D Analyst software.
The GPS survey used continuous kinematic survey techniques with two or four Trimble 4600LS receivers. One receiver always functioned as a base station while one or three receivers, carried by individual team members or local workers, collected data. These 'rovers' walked systematic transects 2 m apart as data readings were taken every two seconds. Participants and supervisors were trained to recognize surface indications of buried architectural, geological and topographical features, where it was advantageous to increase the density of readings. This intentional collection scheme is arguably superior to comparably accurate automated techniques, such as LIDAR (http:// www.airbornelasermapping.com/ALMNews.html), and is much more accurate than models derived from digitized contour maps or even from the forthcoming Shuttle Radar Topography Mission (SRTM) data (http://www.jpl.nasa.gov/srtm/).
Each reading requires subsequent processing in order to achieve optimal accuracy by removing errors resulting from natural phenomena, such as the atmosphere, and intentionally introduced artificial errors, such as Selective Availability (SA) (http:// www.ngs.noaa.gov/FGCS/info/sans_SA/). With the discontinuation of SA in the spring of 2000, much better pre-processed readings were obtained. However, use of the processing software system was still necessary to remove natural errors and the occasional artificial error that seemed to resurface in our data readings, often coincident with reports of military activity in adjacent countries. A more costly real-time GPS solution would have been an alternative possibility that would have decreased the daily time spent on the survey.
Errors that cannot be removed through processing include the vertical movement of the roving devices and their accuracy in kinematic collection mode ([+ or -] 2 cm). These errors are primarily caused by the up-and-down motion of the device during a rover's stride, with taller individuals yielding higher average errors. In our experiments this source of error was a maximum of [+ or -] 8 cm, in line with laboratory estimates of 4.9 [+ or -] 1.1 cm (Murray et al. 1964: 349). This error is less on sloping terrain because of shorter strides and slower speeds (Sun et al. 1996: 684; Kawamura et al. …