Mon 19 Jun 2006
The Vermont Archaeological Sensitivity Model (VTASM)
Posted by Matt under Archaeology , CRM , Prehistoric , ESRI
As I mentioned last week, the current Spring 2006 issue of ESRI ArcNews, has a series of articles on the use of GIS in Archaeology and more specifically Cultural Resource Management. This post is a short synopsis of one particular article on the use of GIS for the creation of a state wide archaeological sessitiity model for Vermont.
Overview of VTASM (link to article)
The Vermont Archaeological Sensitivity Model (VTASM) is a joint project between the Vermont Division of Historic Preservation (DHP) and the Vermont Agency of Transportation (VTrans), created with the guidance of the University of Maine at Farmington Archaeology Reaserch Center (UMFARC), the University of Vermont Consulting Archaeology Program (UVMCAP), and ESRI business partner, Earth Analytic Inc.
In general, this model is an inductive environmentally based model using ESRI
ArcGIS ModelBuilder, Spatial Analyst, and 3D analyst to produce a state wide, 10 m resultion coverage, demonstrating archaeological sensitivity. The creation of a state wide sensitvitiy model, such as this, follows a trend set by other states, paticulary, Minnisota and North Carolina.
Used during a projects planning phase, a model such as the VTASM, allows archaeologist and transprtation planners the ability to query the archaeological impact of a project or variations of a particualr design. By gauging the possible extent of impact on archaeoligcal resources, the planning agency can create a much more efficient budget or stear clear of potential large archaeological investigations. While this is the underlying theory behind wide scale models, as the stakeholders of the VTASM has recognized, sensitivity models are only abostractions of recorded archaeoligcal knowledge mated with modern environmnetal conditions to create a “best guess” of where sites may be located. Even though the VTASM produces high scores for predicability, it is a guide and not a subsititute for archaeological field survey.
As mentioned, the VTASM is created through the weighted sum of correlative distance buffers of a selection of environmental features. Through years of survey and knowledge, archaeoligsts build up mental models of archaeological site location. These models often include, distance from water, slope angle, proximity to wetland. The VTASM, and similar models, quantify site location knowledge, through a rule base or correlation statistics, into raster layers which when overlain are summed to achieve the overall sensitivity for each cell.
In the VTASM, the environemtnal atributes are computed into 11 environmental compnent models (ECMs). Six of the ECMs are computed for water related features such as streams, confulences, and wetlands. The remaining five ECMs are realted to lakes, floodplains, soils, slope, and glacial features. Each ECM is a sesitivity raster which assigns a weighted value based on the proximity to one or more environemtal attributes. For example, through correlation or survey findings, the project team knows that archaeoligical sites are more likely within a range of 100m to 200m of a stream confluence. Using this, the ECM is weigthed higher in the 100m to 200m buffer distance.
The final VTASM model, which is computed on the fly for the area under investigation, is the weighted sum total of all overlapping ECMs. The weights assigned to each ECM are adjustable and can be fine tuned based on the environmental character of the region under investigation.
A Bit about Sesitivtiy Modeling
An archaeological sensitivity model, often referred to as predictive models, is simply an expression of a single or multiple attributes that demonstrates the sensitivity (probablitity or possibility) that a specifiic location on earth has been utilized by people in the past. “In the past” may refer to 100 years ago or 100,000 years ago. In Vermont and the Eastern US, models such as this generally related to ~12,000 to ~500 years ago.
Sesitivty models, s described above, are often cell based raster grids created through map algebra of one form or another. The unit of analysis for raster based sensitivity models can be of any size, but generally, in accordance with the use of USGS Digital Elevation Models (DEM), a 10 meter or 30 meter resolution is used. In the Vermont example, the overal model
is at a 10m resolution, which uses a combination of 30m and 8m Lidar DEMs.
Although the general technological underpinnings of sesntivity models are often the same, the theoretical guidlines can be substantially different. The two main catagories that archaeoligical sensitivty models are put into consist of Inductive or Deductive. The claimed difference between the two is “explanitory power”. Abstractly, an Inductive model (aka emprical model, correlative model) correlates known site location to environmental features to create a “fingerprint” for where sites most often are found. This type of model does not attempt to exmplain why sites are located where they are. On the other hand, Deductive sensitivity models attempt to use what we know of past human behaviors to create a explantion of why sites may be found in certain places. Explination of archaeoligcal phenomina is the primary goal of the deductive model.
Here is a great PDF [1.25 megs] that explains various types of sesitivity models.
The inductive/deductive divsion, in my mind, is akin to a major politcal debate drawn down party lines where each side is arguing the same darn thing but spun in thier favor. For your sake, my thoughts on the difference between the two will be save for a seperate post, but what is important to mention is that within Cultural Resource Managment (CRM) in America, the inductive model is most frequently used. This model has a solid history of test and application. The technical methodology is documented, repeatable, and testable. And most inportantly, it is the most cost effective way to consult the client on where it will cost them lots of money to put a road.
I’m sure some of these reasons are why the Vermont DHR and VTrans decided to build the VTASM the way they did. The model they created is flexible and testable. As archaeoligsts learn more and continue to survey the landscape, the model can be added too and tweak to ablige. The use of VTASM as a common model helps foster communication and trust between the DHR and the Transportation Athourity leading to the likleyhood of better archaeoligcal resource protection. As with Minnesota, North Carolina, and now Vermont, more states are bound to follow this example.
One Response to “The Vermont Archaeological Sensitivity Model (VTASM)”
Leave a Reply
You must be logged in to post a comment.
July 13th, 2006 at 11:35 pm
Just a comment on your coverage of the VT modelling effort: The ESRI article makes no mention of the theoretical background behind the factors used to ‘predict’ site locations in this model.
What is not said is that the model uses the same assumptions about site locations that have been made for almost 100 years. Further, no effort has ever been made to test the model’s validity in predicting locations. When put to use, the model accurately predicts the locations of (but not the reasons for those locations) about 50% of the state’s known archaeological (Native American) sites. Even more striking is that the environmental variables used to score landforms are based completely on surveys within one physiographic region that comprises less than 1/3 of the state’s total area (coincidentally this is also the area that has historically held the largest population density). No attempt has been made to examine the remaining 2/3s of the state and to develop predictive models for mountainous areas or the piedmont of northeast Vermont.
Under a purely research-oriented system, this would be a great application - the model would/should/could be continually tested and refined to move beyond pure induction. However, as it is being promoted by the developers, the model will be used to highlight areas where archaeological testing is necessary (high sensitivity areas) and those where no testing is necessary (everything else).
This creates and perpetuates a self-validating prophecy. Despite the air of high-tech accuracy and the glitzy graphics, the ideas that are used to develop the model are essentially unchanged from those developed from “collector knowledge” of the 1920s and the first formalized models developed in the late 1970s. Have we really learned nothing in the past century that would refine our ideas about where and why sites are?
This isn’t intended to slam all use of GIS for modelling, rather, to point out the dangers of repackaging untested and unvalidated models into a GIS system to be expressly used for urban planning. Contrary to the Vermont model, the Minnesota model was initially inductively developed, and then rigorously tested and refined. Finally, a ’survey bias’ was also introduced into the system allowing users to equally weight those areas that are systematically under- and over-represented. The VT model does none of these, and it makes no attempt to correct for these deficiencies. Instead, it restates some of the most obvious assumptions about where archaeological sites will be located.
The real power of GIS lies in being able to reexamine such assumptions and to develop models that more accurately reflect reality. Further, though such inductive models are sometimes necessary starting points, there most always be some amount of systematic testing of the results. Without any such testing, the model only tells you where people have looked for and found the highest numbers of archaeological sites in the past, and if they’ve been looking in the same places since 1860, chances are your model is going to have some serious holes.