5.5. Mobile GIS for archeological survey

Mobile GIS can be incorporated into archaeological survey methods with varying degrees of change to traditional archaeological survey techniques. This section describes the methods implemented in 2003 where a predominantly digital recording method was used.

5.5.1. Standard survey practice

The standard survey practice in the 2003 season consisted of a single team of four to six surveyors spaced 15m apart. The team swept across hillslopes following contours, and at the end of each survey transect the team would sweep around and return towards the opposite direction in the adjacent transect following a boustrophedon configuration. The survey team would assemble to investigate sites when they were encountered, although the team would not necessarily assemble for isolated finds.

/misc/image031.gif

Figure5-3. An example of a pedestrian survey line following a river terrace at a 15 meter interval. In this survey, only one mobile GIS unit is used. GPS units carried by the surveyors at either end of the survey line mapped the extent of all surveyed areas.

The portable equipment carried on the survey by each team member included basic day hiking equipment such as personal gear, lunch, copies of maps and a compass, and everyone had a small FM walkie-talkie. In order to map the extent of daily survey coverage, and to quantify the distances between surveyors, those hiking on either end of the survey line carried a Trimble GeoExplorer GPS logging a polyline attributed with the side of the survey that they were walking (right or left) as well as the number of surveyors hiking that day. GPS datalogger tracking devices are more widespread today (in 2006) and pair of small USB based dataloggers such as the Sony GPS-CS1 for geotagging photos are also suitable for continuous data stream mapping on either end of a survey transect. A separate mobile GIS system consisting of Dell Axim PDA with a Trimble Pathfinder Pocket GPS receiver was carried for mapping archaeological sites into Arcpad as will be described in more detail below.

5.5.2. The contribution of mobile GIS

Current mobile GIS technology contributes to traditional archaeological survey methods in several ways. First, mobile GIS aids surveyors with navigation because the anticipated survey transects, and some other relevant guidance information, can be clearly indicated in conjunction with the current GPS location. Second, mobile GIS allows researchers to record new vector data along with attribute forms that are more flexible than those provided by GPS or by data dictionary approaches in the past. Finally, mobile GIS allows researchers to transport digital datasets into the field so that they can do error checking immediately, review the work of other research teams, and perform queries on large existing volumes of data in digital form.

When a surveyor encountered an archaeological feature the surveyor would first determine if the feature exceeded the specification for isolates and then, should the feature be a site, the surveyor would call a halt to the survey line. Site boundaries were established for two reasons in the 2003 fieldwork. First, in GIS it is generally required that a geographical feature be delimited and that a database record is created before it can be attributed. Thus, one cannot describe a site that has not yet mapped, unless some kind of more complicated work-around is employed such as the creation of a temporary attribute record. A second beneficial effect of delimiting sites as an initial step, however, is that the team is forced to travel over the site completely and assess the extent and variability before an attempt was made to describe it.

5.5.3. Hardware configuration

Mobile GIS systems permit surveyors to attribute spatial locations with a variety of data types. Currently the GPS unit is the primary digital input into the mobile GIS and this permits the mapping of point, lines, and polygons delimiting archaeological features. In the GIS the spatial data is attributed and once post-processing is complete the new data joins the larger GIS database.

/misc/image032.jpg

Figure5-4. Mobile GIS implementation with ESRI Arcpad 6. New data sources from external instruments are shown in the top row. Where post-processing is needed new data is not integrated with other data until later. New and existing data can be summarized and displayed together.

Existing inputs to the mobile GIS system in 2003 are relatively limited. Important attributes that currently must be entered manually include the digital photo numbers associated with each archaeological feature, and relative measurements collected at the site. Additional instruments in the future might include wireless tapes that can transmit precise lengths back to the mobile GIS of features like the dimensions of a doorway. Alternately, for in-field lab analysis on non-collection survey, wireless calipers or scales could be used to transmit the size of an artifact to the mobile GIS linked to the spatial provenience.

5.5.4. Defining loci and sites

When an archaeological site is encountered during survey in this Arcpad system the site must be delimited first, using the Site-A polygon, and then loci located within the site are delimited and described as related in two accounts (Tripcevich 2004;Tripcevich 2004). The locations of individual artifacts of interest are mapped and bagged separately using a Lithic_P or Ceramic_P geometry type. These include diagnostic artifacts or other materials of specific interest.

Locus / Site

Min. Density Artifacts

High Density

10+ artifacts per m2

Medium Density

5-10 artifacts per m2

Low Density

1-3 artifacts per 2m2

Site

2 artifacts per 10 m2

Table 5-6. Locus and Site artifact density definitions.

Pin flags were used to delimit these features of interest, and generally in the case of most medium and high density loci, the result is a "fried-egg" model of artifact density polygons. In recording these polygon features, one generally went from the geographically largest to smallest entity because, as is also true in desktop GIS, features that are created later appear "on top" of features created earlier and thus larger, later features would visually obscure earlier features. This condition has to be corrected back in the laboratory and thus it was simply easiest to map largest to smallest. When a feature is mapped in Arcpad with the GPS then, subsequently, an attribute form appears that allows for explicit description of the feature.

5.5.5. Attribute Forms

Aside from the site datum points and site boundary polygons, three dominant feature types characterized the archaeological data set in the mobile GIS. Each archaeological data type had an attribute form associated with it that recorded information appropriate for a given feature. Page One of the digital forms comprised a unique ID number generated from a script and a range of numbers for digital photos (JPEG files) documenting a feature.

(a)

/misc/image033.jpg

(b)

/Figs_Ch5/arcpad_shot1.jpg

Figure 5-5. (a) Arcpad screen showing a large site with loci and points. (b) Example of page two of a lithic locus form in Arcpad showing Category 1 and Category 2 columns; in the background, two sites and contour lines are displayed on top of a 15m resolution ASTER satellite image.

Page Two of the attribute forms (Figure 5-5b) contained specific information about the feature type, such as Site, Locus, or Point information. The third page contained eight pull-down menus with environmental attributes for geology, exposure, and other local variables. These values were usually the same within a given site so that the values were "sticky"; they were stored in temporary memory between recording events, and the editable form was repopulated automatically unless a new site feature was being recorded. The final page contained a "Comments" field that accepted up to 255 characters and included a button that would open Pocket Word application with a text file named for the unique ID #, allowing the entry of additional notes if necessary. A link to a separate application that permitted MP3 compression of voice-based comments was available as well, but because the processor demands of sound encoding overly hampered the functionality of the Pocket PC for the GIS application, the feature went unused.

5.5.6. Variability within a Locus

A basic complexity of archaeological survey is that artifact concentrations frequently contain a variety of artifact types, perhaps dating to completely different occupations. This variability presents a particular challenge for a fast, mobile GIS based recording system because in lieu of sampling, all that the archaeologist has time to do is to document his or her rapid assessment of the artifacts that are found within individual loci geometry mapped into the GIS. Additionally, despite of the variability present within the locus, the archaeologist must attempt to generate data over the course of the field season that are consistent and comparable. During the Upper Colca Survey this difficulty was addressed by estimating the characteristics of a primary and secondary attribute category, dubbed Category 1 and Category 2 (see Figure 5-5b), that best characterizes the locus using the custom interface developed for the project.

The problem: How does one evaluate and map a scatter of, say, 5,000 stone flakes in less than one hour, as well as estimate the percentage of obsidian to another material type, such as chert?

In order to achieve statistical rigor and reliability, a sampling strategy was needed. Sampling and collecting artifacts is time consuming, and sampling at every concentration of lithics near a quarry is also unrealistic because there are so many lithic artifacts in such areas. Sampling was therefore carried out at "High Density Loci" with artifact concentrations deemed most worthwhile given the research goals, while a less rigorous approach was applied for artifact distributions of lesser importance. A solution was devised that is geared for conducting cursory inventory, not an in-depth assessment. This solution captured variability by estimating the proportions of the two dominant attribute categories within a given polygon.

/Figs_Ch5/Site_maps.jpg

(a) (b)

Figure5-6. Maps for two different hypothetical sites recorded in less than one hour. (a) A conventional, low precision sketch map showing only major site features and perhaps subdivided into site sectors (b) Mobile GIS site map with 1-2m dGPS error. Internal distributions, such as the fried-egg density gradient model shown here, can be assessed and rapidly mapped.

/misc/image034.jpg

(a) (b)

Figure 5-7. (a) Structure of the archaeological Shapefiles with names and descriptions. Each of the Shapefiles had a form associated with it that prompted the user with fields appropriate to that data type. (b) An example of a part of the ID # system that prioritizes spatial provenience in the field.

A hypothetical site description makes the site recording strategy more clear. This example takes place at a site with a large, low-density lithic locus (Figure 5-6b), where the concentration of stone artifacts was mostly obsidian material but also included artifacts made from chert, chalcedony, and quartzite. The mobile GIS user walks around the locus with the GPS running, and the area was recorded into the "Lithics-A" ShapeFile (Figure 5-7a). Lithic concentrations of medium and high density are found inside the locus, creating a 'fried-egg' density map. Subsequent to delineating the locus with a GPS, the custom form (Figure 5-5b) appears. Several steps are followed in filling out the form.

(1) The primary "axis" of variability is determined. In this case, it is stone material type.

(2) Using this variable, the largest group is characterized. This attribute category 1 (C1) was described as "Material: Obsidian," and other attributes of interest to lithic analysis such as amount of cortex, size of debitage, and artifact density in the attribute category, were rapidly estimated. In our case, the density was "Low."

(3) The second most represented group, attribute category 2 (C2), is characterized and its attributes are evaluated, again as quickly as possible. Any subsequent groups were disregarded for expediency and because of the error in estimation and low reliability of the method.

(4) The proportion of stone artifacts in the polygon estimated to meet the description of C1 is entered in the field labeled "C1% of Locus," and an estimate is also generated for category 2.

The method works for a rapid inventory, and it provides a general estimation of materials along with the characteristics and densities within loci. Using this system, archaeologists are encouraged to describe the variability between category 1 and 2 in terms of only one variable at a time. For example, if there were notable differences in both Material Type and Debitage Size in a particular locus, then a second polygon was created. Alternatively, the first polygon was copied, and the different "axes" of variability were distinguished independently. Instant types (i.e., attribute categories) were generated for each polygon by emphasizing the greatest variability within the locus, and this was considerably more spatially explicit than rapid archaeological survey had been in the past despite a relatively small investment in time. Time efficiency was a major objective of the Colca Survey with recording all but the very highest density lithic concentrations, and this approach allowed for rapid feature mapping. A variety of new possibilities for custom field applications are becoming available now that modern digital equipment, such as the mobile GIS used in this archaeological survey, can be modified and streamlined by the archaeologists to suit the needs of research without recourse to professional programmers.

5.5.7. Sampling High-density Loci

For the purposes of the Upper Colca project High density lociwere defined as areas where the density of the artifact scatter appeared to exceed 10 artifacts per m2. As with all loci, these concentrations were mapped using the mobile GIS interface, but then High-density loci were further characterized by collecting all artifacts within two or more 1x1 m sample squares for later analysis back in the lab. The Arcpad SampleDesignscript was used to pseudo-randomly place, using an unaligned-grid method, a sufficient a number of square sample units to cover at least 0.01 of the Shape Area (m2)of the locus as reported in Arcpad. This works out to a 1 m2collection area for every 100 m2of polygon area. The GPS indicator was used to navigate to the randomly generated point locations. When documenting each sample an overhead photo was taken of the 1x1m area from near-nadir for later georeferencing, and then artifacts were completely collected. One or more units were randomly placed somewhere within the polygon, and one unit was always placed right on the location of estimated highest density. During the 2003 season, such collections resulted in an average sampling fraction of 0.014 among the twenty-two samples that were collected during the course of the field season in this process of sampling high density loci.

5.5.8. Collection during survey

Traditionally, it has been impractical for archaeologists to retain precise spatial provenance for surface artifacts that are not particularly interesting or rare. Collected artifacts are aggregated by site, sector, or by locus. However, artifact collection is increasingly seen as a destructive practice. The collection strategy used in the Upper Colca Survey consisted of assigning a unique ID number (ArchID) from a single number series to all spatial proveniences, point locations, loci, or entire sites-very much like postal zip codes for street addresses. After four months of fieldwork, 1100 spatial provenance numbers had been assigned from the series. As described previously, individual artifacts collected from a given provenience were assigned key ID#s after a decimal point. An interesting alternative to handwriting the unique ID# on labels for sample bags collected in the field is to bring a sheet of pre-printed barcode stickers. As the sticker is placed on the sample container, a serial barcode scanning wand can scan the barcode value directly into the GIS record. The barcode scanner approach is somewhat restrictive, however, because the mobile GIS unit must to be available to scan every collection bag.

5.5.9. Other Data Types

As a systematic pedestrian survey of extensive areas, the Upper Colca Project survey presented an opportunity to collect other field data as well. During survey work a separate set of GIS data was collected that consisted of non-archaeological data. These included geological sources of stone material such as chert outcrops and natural obsidian flows. Similarly, fresh-water springs and other resources of use to past peoples were mapped in. Mountain summits, trails that may follow Prehispanic trade routes, and other such environmental features were also mapped. Thousands of digital photos were taken, including a number of stitched panorama photos. The location of these photos was mapped with the mobile GIS using a form to enter the JPEG file numbers, as well as the cardinal direction and an estimate of distance for photographs of distant objects. The variety of data types that were determined to be "worth recording" during this survey project underscores the need for individual flexibility in recording methods.

5.5.10. Processing steps with mobile GIS

The time investment in implementing the mobile GIS approach is still considerable, as it involved both pre-fieldwork and post-fieldwork processing steps. Pre-fieldwork tasks, discussed above, include acquiring and preparing regional datasets, and designing digital forms that are appropriate for the project. Post-fieldwork processing involved standard issues such as downloading all data to a laptop from various devices, tagging folders of digital photos with the associated ArchID, GPS post-processing, as well as analytical processing steps such as deriving meaningful indices from the digital data. Post-fieldwork tasks also involved some unanticipated and time consuming labor, such as cleaning inconsistent datasets to prepare them for general analysis, and other management issues. These inconsistent data include the records gathered during two periods when the system was not functioning smoothly as described in Table 5-7.

Issue

Problem

Resolution in future projects

Pre-fieldwork preparedness

During the first two weeks of fieldwork software and hardware debugging were still underway.

Allow sufficient time for designing and debugging fieldwork forms and equipment prior to work in remote locations.

Equipment failure

During the last 10 days of the field season the cable connecting the mobile GIS PocketPC to the GPS unit failed and all sites were recorded with older Trimble Geoexplorer GPS units.

Purchasing two similar mobile GIS units, with one acting as a backup device so that no break in data recording would have occurred in case of equipment failure.

Table 5-7. Sources of inconsistent data during the 2003 project, and means of avoiding these problems in future projects.

Some of the processing steps are the result of employing GPS based mapping, and are therefore largely inevitable. Despite post-processing, the polygons and polylines gathered using GPS in Streaming mode contained a lot of redundancy and required geometry validation. These redundant positions were especially abundant where the person mapping a feature had to slow down or stop in the process of delimiting the feature. In these cases a number of vertices would be gathered as a small cloud (within the positional error of the GPS) and this resulted in a line intersection and short segments. Processing the data to resolve these GPS derived problems was accomplished in ArcGIS 9 using the following processing steps

(1) The "Repair Geometry" function was used to resolve intersections in a single polygon or polyline.

(2) Polygons were converted to polylines and the ArcToolbox > Data Management > Features > Simplify Line operation with a 10 cm tolerance was applied the data. The 10cm range for simplifying the lines reflects the data quality. In this case, the post-processed accuracy of the spatial data was 1-2m as stated by Trimble Pathfinder Office.

More recent versions of Arcpad (v7) and other mobile GIS software such as Terrasync provide a movement filter for datalogging that allows the user to specify a minimum distance setting, so that data are not logged unless one continues moving. For example, during streaming mode in 2003 the GPS was logging a vertex from the average of every 2 positions for a smoother line (at a 0.5 second rate this resulted in a vertex every second). A feature of the newer Arcpad 7 allows the user to specify that a vertex should be logged only after 3m of movement.

This option would have resulted in cleaner data during the 2003 season, although with low real-time accuracy the movement filter is relatively coarse and it is probably accurate to only within 10-15m. In North America and Europe, where SBAS (WAAS or EGNOS) correction signals are available, real-time positions are approximately 5-10m and these movement filters for datalogging function adequately on low-cost receivers. However in South America SBAS correction will not be available in the foreseeable future and movement filters are probably most relevant for faster GPS mapping tasks such as mapping from a moving vehicle. It is conceivable that GPS movement filters will actually result in noisier data because exceptionally large position measurements are logged, but the majority of positions are filtered out because the distance of movement is not sufficient.

5.5.11. Implications ofMobile GIS for Fieldwork

For the Upper Colca Project the mobile GIS recording system produced both cartographic data for site report mapping, and GIS vectors for analysis. The results of these efforts are presented in chapter 6, where the cartographic output appears in local and regional maps. Automated cartography methods, such as one-to-many labeling through a VB script (discussed below), permitted the automated labeling of lab results based on field collections from polygons or points.

In the bigger picture, the incorporation of mobile GIS for scientific field research seems inevitable although the applicability of mobile GIS to specific applications depends largely on the extent to which mobile GIS meets research needs. Minor benefits of mobile GIS, such as the time and date stamp associated with every measurement, improve the data that are being gathered in unobtrusive ways. A more elaborate system might gather extensive metadata concerning research methods and data structure into an automatically generated digital log file. Additional tools, such as statistical summaries and visualization applications would have proved useful during the Upper Colca Survey but these are not yet available in a mobile GIS platform. The ability to estimate spatial variation measured on archaeological variables would have been useful for a more informed selection of sampling strategies, and perhaps for guiding the placement of test excavation units (Hodder and Orton 1976;Redman 1987). When researchers are able to investigate new spatial data in conjunction with existing datasets using the exploratory data analysis approach (Tukey 1977) while in the field it will open up original research strategies by combining information from new and existing digital datasets. Statistical indices, such as the degree of spatial autocorrelation among particular classes of data, would be useful to know in the field. Geostatistical methods such as kriging, familiar to archaeologists in lab analysis (Lloyd and Atkinson 2004), will have application in fieldwork contexts as well when these tools become available in future mobile GIS systems.