Depending on the detail of the legal description, site locations might be off by between 285 m (quarter-quarter-quarter section reports) and 1,140 m (section center reports). According to the U.S. Geological Survey (USGS), 25 percent of its 7.5-minute quadrangles are evaluated for accuracy and are not passed unless 90 percent of the arbitrarily selected map points fall within 40 feet (12 m) of their actual position. Vertical accuracy standards require that 90 percent of the points fall within half of the contour line distance. Thus, it is possible for site UTM coordinates to err by at least 24 m due to USGS error. Site locations may have errors greater than the standard since they often are not near points, such as well-defined road intersections, that are used to check errors. And 10 percent of the tested points can have errors greater than 24m.
For most Nebraska survey projects, site location is reckoned with reference to map features; the site UTM coordinates are then estimated from the map. Coordinate errors may be introduced during the initial reckoning or during estimation. Finally, the journey of site information from the field to its final residence in a database may involve several transcription phases, with the possibility of error occurring each time. Such errors may involve transposing numbers or reporting the wrong UTM zone, resulting in database site locations that may be off by meters or continents.
In a 1993 assessment of site locational information at Mesa Verde National Monument, a report from the National Park Service (NPS) Cultural Resources Geographic Information System (GIS) Facility gives the average difference between Global Positioning System (GPS) and reported location as 35 m. The comparison locations at sites located and marked in 1986 were mapped from temporary datums established throughout the monument. UTM coordinates were reported to the nearest meter. Interestingly, the NPS assessment found local directional trends, with GPS shots in one portion of the canyon trending north of earlier reported locations and, in other portions, trending southwest, south, and so forth.
Our assessment of locational quality involved relocating previously reported archaeological sites and then fixing the site location using GPS technology, which references the point location on the earth's surface with respect to the known positions satellites. The difference between reported and GPS site location is therefore the basis for our evaluation. Most Nebraska sites are unmarked, and therefore it was critical for us to be sure that we were indeed at the site we were attempting to relocate. Using the site paper record on file with the NSHS and comparing it with field observations, the field crew ranked their confidence in site relocation. Reasons for less than complete confidence in relocating specific sites included incomplete data on site forms, landmark changes since the site was first described, and the lack of detailed site maps.
The ability to accurately establish site locations using GPS was also critical. We used widely available, reasonably priced (lease $1,500/month), and reliable equipment for this task. A Magellan ProMark V recorded the location of each site relative to satellite position. Simultaneous observations were made with an unattended Magellan MBS-1 base station, with the associated antenna situated at first-order monument (discussed below). A Dell laptop computer with an 80-mb capacity logged the observations. Power for GPS unit and computer was supplied by a deep cycle battery. In Nebraska, one can leave $10,000 worth of equipment in the middle of nowhere without risk. In other parts of the country, base station security will be an important consideration.
Other GPS tips to help plan your resource needs include the fact that the rover, base station, and antenna each consume a lot of power. Over a four-day period of operation, approximately 28 AA batteries were used. Also, just getting all units to talk to each other, rounding up the right cables and power converters, and so forth consumed several days. Some field workers log the GPS base station data with small-capacity computers. They may make do with a short field day, or may require communication with a coworker who turns the base station on and off as needed. In urban areas, it is possible to subscribe to a base station service, something that should be available in rural areas in the near future.
Selective Availability refers to the deliberate degradation of satellite positional data by the U.S. military to make the determination of accurate, real-time GPS positions difficult. Selective Availability can be compensated for by using differential post-processing to correlate the satellite observations made by both the rover unit and the base station unit situated at a known point. At the end of the day, distance readings made to the same satellites by the roving and stationary unit were compared. The difference between the two ("differential") with respect to the coordinates of the known point allows for the determination of very accurate rover locations.
A digression on known points, i.e., monuments, and their accuracy is important here. We may speak of the ability of GPS hardware to assess locational accuracy to the sub-meter level, but, the question remains, with respect to what? The "what" is the location of the base station, but where is the base station? The base station may be located on an arbitrary site or regional datum, but usually we want to measure location in a geographical coordinate system such as UTM. If this is the case, the base station must be located at a known point in the coordinate system. Where does one find such points and how do we determine their locational error?
As most archaeologists know, a series of monuments distributed across the United States serve as either horizontal datums or vertical benches within the UTM coordinate system. Most vertical benches have their horizontal position interpolated from USGS maps and should not be used for GPS work. Information on the location and accuracy of National Geodetic Survey (NGS) monuments is available on CD-ROM and is updated annually. Note that the coordinates for the monuments are in the process of revision as part of the High Accuracy Reference Network (HARN) surveys, which will see a shift in monument position by .3-.8 m.
For the Nebraska Cultural Resources GIS project, we were interested in examining the accuracy of site location information from across the state and according to site type and other recorded data. We were interested in identifying site location on an existing geographic coordinate system rather than on an arbitrary site grid of a regional project system. We therefore needed to establish our base station at monuments throughout Nebraska. For Nebraska, however, HARN monuments and corrections had not yet been completed. In lieu of this, only monuments with both horizontal and vertical first-order ratings were used. For most rural counties, this restricted our field crew to about three usable monuments at which to locate the base station. Given that some monuments could not be located, had been destroyed, or were in use by other GPS surveyors, our crew was sometimes left with only a single monument per county for the GPS base station. In all cases, a monument was used that was within 150 km of the sites to be revisited. For maximum accuracy, most GPS manufacturers suggest that the rover-base station distance be under 600 km.
Having acquired satellite readings with the rover and base station GPS units, pseudorange differential post-processing was used to compute the UTM coordinates at a site. With our equipment, a single acceptable-quality GPS location required 10 minutes for acquiring adequate satellite data (1 observation/second) and another 10 minutes for processing that information at the end of the field day. By taking readings with the GPS units on two first-order monuments, we independently determined the accuracy of this method to be 1.5 m.
From August to October 1994, as fields were cleared and the surface became visible, 287 successfully processed GPS locations were calculated for 105 archaeological sites. If a site was larger than 25 m2, an attempt was made to define the site boundaries, and multiple GPS readings were taken along this boundary. The measurements were not all of equal quality. For some, the number of satellites fluctuated during the measurement period and only a less accurate two-dimensional, rather than three-dimensional, location was possible. Positional Dilution of Precision (PDOP) is a unitless measure that describes the geometry of the satellite configuration relative to the rover GPS position; clustered satellites (high PDOP value) provide less accurate locational information while GPS readings made with dispersed satellites (low PDOP value) provide more accurate information. The PDOP for each observation sequence varied, sometimes outside of acceptable limits. In other cases, the number of satellite observations simultaneously captured by the rover and the base station were low, resulting in inferior locational information. Finally, sometimes the post-processing of the GPS data failed, in which case the site was revisited and GPS readings logged again. Our field crew was initially inexperienced with GPS equipment, and we found a learning curve of several weeks necessary for its proper use. With more experience, we learned to take care to disable poorly performing satellites and to immediately inspect the PDOP values. With these precautions, good GPS locations were acquired during the first visit to the site 95 percent of the time.
Table 1 summarizes our results according to both strict- and acceptable-quality thresholds. The strict-quality thresholds are those specified by the GPS manufacturers. We have reviewed requests for proposals by agencies for site survey and location that specify the lesser-quality thresholds. As seen in Table 1, we found little difference in the quality of the location information by quality threshold.
Table 1--Summary of Locational Accuracy Data Using GPS
Strict Quality Acceptable Quality
PDOP 6 PDOP 8
N 400 N 180
GPS observations 238 257
Sites 96 101
GPS-Reported Location Differences for High Confidence Sites
No. sites 24 24
Mean difference 167 m 166 m
75% quantile 211 m 211 m
90% quantile 376 m 353 m
Reported Legal Description
No. sites 36 37
Mean difference 635 m 627 m
75% quantile 486 m 455 m
90% quantil 1,002 m 1,000 m
Of the 105 sites visited, only 65 could be verified as sites reported to the NSHS and ones relocated. Of these, acceptable-quality GPS information was collected for 61 sites. Mean UTM values were calculated for those sites with multiple GPS readings. Focusing only on those sites that were relocated with high confidence, Table 1 presents statistics for the distances calculated between reported and GPS site locations, which are all converted to the UTM coordinate system using NAD83 and GRS80. A further distinction is made here between those sites for which only legal descriptions had originally been reported to the NSHS and those for which UTM information was available. Where the Mesa Verde study found localized directional trends, we see nothing of this sort, probably because of the larger regional scale at which we worked.
Considering the acceptable-quality GPS shots, Figure 1 gives the frequency distributions for distances between GPS site locations and reported (UTMs reported to NSHS) and estimated (legal description reported to NSHS) site locations. Note that the outliers, which probably represent miscodes, are extreme. For this reason, means and standard deviations are less useful descriptors. To characterize the accuracy of the locational data, a value between the 75 percent quantile (recommended by statisticians in cases like this) and the 90 percent quantile (the egregious miscodes probably lie above this) is suggested.
Figure 2 plots site locations in western Nebraska buffered with the conservative 90 percent accuracy value. This map should be read to mean that there is a 90 percent chance that the actual site locations fall inside the site buffer. Sites for which only legal descriptions are available have larger (1,000 m) buffers; smaller (353 m) buffers are plotted for sites for which UTM coordinates were reported. Road improvement along State 88, for example, should thus take into account not only the reported site location, but also the accuracy of that information. Rights-of-way that intersect site buffers would require additional consideration.
Several other observations and recommendations follow from these experiences. First, cultural resource database design should consider several modifications. For one, a datum variable is necessary. Also useful is a variable describing the source legal description, UTMs estimated from USGS map, differential GPS, and GPS of the locational information. To identify egregious site coordinate miscodes, it may be useful to include the distance and general direction from an archaeological site to a major land feature or town. Address checking (an advanced GIS feature) could be used to identify the 10-20 percent miscodes expected in the site databases.
Archaeological sites containing both high- and low-density archaeological deposits are arbitrarily defined and highly variable entities. They come and go as surface conditions change, and their boundaries fluctuate according to fieldworker experience and surface conditions. Our experiences in attempting to relocate unmarked sites suggest that it is valuable to require the use of GPS equipment in locating site boundaries and survey unit boundaries. Indeed, for some projects, the U.S. Army currently requires survey transects to be mapped using GPS equipment.
In Nebraska we have some idea of the accuracy of the cultural resources layer. The accuracies of other GIS layers, such as soil polygons, hydrology, and roads, are often unknown, but likely even less accurate since their sources are usually maps at a scale of 1:250,000 or 1:100,000. This means that to answer questions of interest to us, archaeologists will have to create data layers to their specification. High-resolution SPOT or Soviet satellite imagery may be useful here. To manage the cultural landscape, these and many other critical issues will need to be addressed. We trust that our experiences will be useful to others as they make their archaeological databases GIS compatible.
LuAnn Wandsnider is assistant professor of anthropology at the University of Nebraska-Lincoln. Christopher D. Dore is a principal in the firm Archaeological Mapping Specialists and also is an adjunct assistant professor of anthropology at the University of Nebraska-Lincoln.
Part I was published in SAA Bulletin 13(4):25-26