Chapter Six Table of Contents

VI. DISCUSSION

The GBRI cultural resources planning model provides a statistically useful indicator for predicting the likelihood of cultural resources on a landscape level. The statistical technique of Weights of Evidence provides an adequate means to evaluate the distribution of sites within the chosen evidential themes, but some caution is required due to biases resulting from site to unit area ratios. That bias holds through calculations of predictive responses. Intersecting themes provided the most reliable means of identifying probability. The probability of encountering cultural resources is highest in areas where multiple predictive evidential themes intersect, while the likelihood of encountering cultural resources lessens as fewer predictive themes are encountered.

Table 6.1. Summary of weighting factors for prehistoric site presence in analytical units

Analytical Unit	Positive Factors	Negative Factors*
Pilot/Thousand Springs	Piedmont, desert shrub, within 1000m of potential wetland	(no strong negatives)
Ruby/Long	<1000m from wetland, piedmont	>3000m from wetland, >2000m from water course
Spring/Steptoe	valley floors, flats, sagebrush	piedmont, <500m from water course, >5000m from wetland (rare)
Salt Lake	wetland and potential wetland proximity (1000m), piedmont slopes, proximity to piedmont and wetland and piedmont/montane margin (rockshelters?)	areas away from piedmont and wetland/potential wetland combinations
Snake	0-200m from water courses	>1000m from water courses (uplands and interfluves); secondary correlation to vegetation pattern of juniper due to upland

*High slope values are always a negative factor

Table 6.1 identifies predictive classes for prehistoric cultural resources within each evidential theme for each analytic unit within the study area. Each predictive class identifies a landscape element as a potentially predictive surface, and combinations of predictive surfaces increase the probability of encountering a cultural resource within that area. For example, within the Pilot/Thousand Springs Analytic Unit, activities within the desert shrub vegetation zone, lying upon piedmont landforms, and within 1000 meters of potential wetland areas would have the highest probability of encountering sites, while sites less likely to occur in areas where those conditions are not met. With the possible exception of potential wetlands, the evidential classes are readily identifiable at the field level and provide a basis for evaluating probability of encountering cultural resources.

Historic resources are much easier to predict – they are nearly always within 1000 meters of perennial water sources and (not surprisingly) within 500 meters of roads.

An analysis of composite probabilities for the study area as a whole quantifies the effectiveness of the model. For prehistoric sites, high probability zones cover 17% of the entire study and contain 33% of the all site areas, while low probability zones extend over 60% of the study area, but contain only 25% of the sites. (Table 6.2) (Figure 6.1) The remaining 41% of the site area falls within 46% of the study area. Historic sites and probability zones exhibit a similar pattern. (Table 6.2) (Figure 6.2) Twenty-five percent of the study area contains 50% of the historic site area while 40% of the study area and 8% of the historic site area falls within the low probability zone.

From its inception, the GBRI cultural resources probability model functions was conceived as a pattern recognition tool rather than as an explanatory model relating to human adaptive response. Buffers within evidential themes were chosen since they represent potential foraging radii, or in the case of slope, habitable ground, but the results are never synthesized to suggest a causal relationship. Evidential themes provide only a recognizable landscape layer that can be contrasted against site density patterns.

Since the model is based upon pattern recognition, subsequent inventories and new site data may provide subtle, or in some cases dramatic changes to the distributional patterns. Certain classes within an evidential theme may have been inadequately sampled during previous investigations, or sites poorly reported. Newly acquired data may effectively increase both inventoried strata and drive results towards more or less predictable distributions. Recently acquired data supports that proposition.

The Spring/Steptoe Valley analytic unit revealed that in addition to several other themes, flats, within landform, were moderately predictive. A recent field investigation, not included as data in generating the current model, identified 163 sites within approximately 20,550 acres (83.16 km²) along the eastern slope of the Egan Range south and west of Ely, Nevada. Approximately 4300 acres (17 km²) were previously inventoried, resulting in a net increase of 16,250 acres (65.77 km²) of inventory. (Table 6.3) Most of the inventory was conducted within the piedmont landform, effectively increasing the investigated area within that evidential class from 2.7% to 5.7% of the analytic unit Additional sites reported within the piedmont, increase site distributions within that zone from 20% of all sites (using the model data) to 27% of all sites within the analytic unit. Slightly more than 20% of the analytic unit comprises the piedmont zone.

With the original data, weights tables show that flats are strongly predictive while other landforms reveal a negative contrast. (Table 6.4) When the new sites are included in the analysis, 27% of the sites fall within 20% of the analytic unit characterized as piedmont. Resulting contrasts within flats and piedmont are similar – indicating that the piedmont is now revealed to be “more sensitive” than before. Flats remain slightly more predictive than piedmont.

The important lesson of the example above is that we can anticipate the model to change as more information becomes available. A second survey again helps to illustrate how this can occur, especially when initial inventories are small. In Spring/Steptoe, the model data contains 65 sites within 77 square kilometers of inventoried piedmont (16% of all of the sites in the analytical unit, 20% of the inventoried area within the analytical unit as a whole). However, a single new inventory that covered 35 square kilometers of piedmont, revealed 102 new sites. With the inclusion of this new data, 29% of all sites fall within the piedmont zone, and the piedmont zone comprises 25% of all inventoried ground. The statistical significance of the ratio differences (percent of sites: percent of inventory) may be questionable, but it illustrates how new inventory will change our picture of specific analytical units.

Regardless of the effect of the new data, the model derived for Spring and Steptoe Valleys did adequately predict probability of encountering resources within the new inventory. While the spatial extent of sites that were identified during the new reconnaissance was not available, each site was buffered to a 2 acre extent so that site area per probability zone could be calculated. (Table 6.5) Over 50% of the inventory area falls within the medium probability zone and less than 12% falls within areas of high probability. Utilizing the derived site extent, 76% of the site areas lie within high to moderate probability zones with the remainder falling within the low probability area that accounts for 38% of the inventoried extent.

In addition to providing probability layers useful for long range planning, the model also brings together site and inventory information useful for short and long term planning. Summary tables and related shapefiles identify the percentage of inventory within each analytic unit, and assess the relative densities of site area to cumulative inventory blocks and within each analytic unit. They also identify proportional survey coverage within specific environmental settings, allowing the cultural resources manager to better assess the range of coverage within a resource area.

Field experience and expert knowledge of the regions within the project area provide the best means to verify model results. If regional expertise has intuitively predicted that most sites are found within 200 meters of water sources, and the evidential theme reflects a similar pattern, then that theme is most likely valid. Likewise, the model may direct confirmatory evaluation. If a composite theme is identified as predictive in the model, but has never been explored or evaluated by regional experts, subsequent projects can be tailored to validate the model's findings.

The IMACS assemblage data provides a useful tool for deciphering cultural patterns and compilation of overview information. Unfortunately, the quality of data and its completeness are variable, often dependent upon age of the record. The IMACS encoding form itself also lacks a level of information that could answer more specific research questions. Lithic assembalge characteristics, such as frequency by material type, are not preserved in the encoding format and the assessments of lithic stages are inconsistent. Older site records pose additional constraints to completeness of the assemblage database since IMACS classifications must be derived from narrative descriptions.

Several problems were also encountered in the creation of a comprehensive assemblage database that was compiled from electronic data maintained by three separate entities. Administrative data is consistent across the three database used in the analysis, but assemblage data varies from complete IMACS encoding to descriptive summaries of the cultural assemblage. Consistency in reporting National Register eligibility also varies between agency and archive. In some cases, current status is maintained in IMACS format within the site database, in others a separate database contains that information.

Shortcomings of the databases can be overcome in future projects by scaling database contents to fit the project goals. A broad based predictive model can be constructed from existing assemblage data with minimal effort if research questions are limited in scope. Where do we expect prehistoric sites? Where do we expect historic sites? Where do National Register sites occur? More detailed synthesis requires mining data from any combination of existing electronic data and paper records for completeness and missing information. What types of materials comprise the lithic debitage? Are lithic tools manufactured from materials available locally?

Scale of the analytic area should also be adjusted to fit the research questions. Another problem encountered with the anthropological analysis was the validity of generalizing results to fit such a broad research area. Variations in survey quality, site reporting, and archival data over an area in excess of 78,000 square kilometers can only elucidate very general patterns. As research questions become more pointed, the research area needs to be scaled down, evidential themes refined to be more specific, and site information scrutinized to assure validity of the observations. Patterns unique to the Upper Snake hydrologic unit may not be valid for the Southern Great Salt Lake Desert.

Planning Models As Cultural Resource Forecasts

The GBRI cultural resources model study is phrased as two map layers (distributed as Environmental Systems Research Institute (ESRI) GRID format files (Ancillary CD 1). One map layer is for prehistoric resources, the other for historic resources. The user of these map layers must be well-aware of several important aspects of them.

The map layers are summaries of models, not “known” data. Just as fire managers do not really know the accuracy of their fuel regime models until fire actually consumes a spot, we cannot know how accurate the models are currently. The examples above suggest that survey bias, differences in reporting styles (sites vs. isolates), and imprecision in the baseline data will all contribute to inaccuracy. Below, we discuss long-term strategies for coping with these problems; here, we wish to call attention to the nature of the models and maps.

Models and maps are planning tools, not compliance tools. One cannot use the GIS data to say an area will be devoid of cultural resources just because it has a LOW value associated with it. For these reasons, we prefer to call the GBRI model a planning model rather than a predictive model. The maps (paper or electronic), which summarize are current planning-level knowledge, are thus forecasts. The simile to meteorology is not accidental, for we do not fully understand the system that generated the cultural resources we are attempting to forecast. Yet, just as a forecaster can state that a particular weather pattern is highly likely to yield snow in the Sierra – without necessarily understanding why the pattern occurs – the GBRI model can forecast areas of highest and lowest likelihood of cultural resources. If one thinks of the models and map summaries as forecasts, rather than facts, appropriately cautious planning will likely ensue.

One must also bear in mind data quality limitations that went in to the creation of the planning models. Digital terrain data is fairly good – 30 meter intervals between fairly accurate elevations – but vegetation data is rather poor. Vegetation data was derived in part from 500 meter grid cells of predicted natural vegetation. Thus, the worst common spatial denominator in the model is 500 meters. This has a major effect on the boundary between very different vegetation regimes, such as the piedmont to montane margin.

The solution to many of these limitations lies in utilizing the model frequently. Actively noting inconsistencies (and consistencies) with forecast values will point out areas of poor baseline data, insufficient archaeological knowledge, or both. Both deficiencies can be remedied. Baseline data can be fixed on a local level, and more inventory in poorly-represented settings can be a management goal. From a land use perspective, confirming LOW forecast areas may be the highest priority.

Maintaining the model is critical to its utility. Field protocols for gathering model data are straightforward. A simple tally sheet for each inventory can be created that summarizes the areal coverage in each model zone, and the revealed size density within each zone. Each inventory and resource should be held in GIS, verified, and flagged as not having contributed to the current generation of the model. Periodically, the model maintainers need to review new information and decide what effort should be put in to model revisions. This could be as simple as just changing the forecast maps without statistical re-analysis or as comprehensive as running entirely new tallies and contrasts.

Resource distributions, overall, are relatively sensible. It is not difficult to understand the distribution of historic resources within the sensitivity model. They tend to lie near to water and near to transportation routes. This generalization shows clearly within each of the analytical unit studies. Nevertheless, as a forecast of where significant or interesting historic resources will be found, the map layers should be used cautiously. For example, a recent inventory of 10,000 acres within the Ely Field Office management area revealed few large historic sites, but dozens of dispersed small scatters of cans and some glassware. These were likely sheep camps. Careful analysis and locating of these seemingly insignificant, uninteresting, sites revealed a good deal about the settlement pattern of early twentieth century sheep-rearing. Each individual site would not have been considered significant; together they are a potential National Register landscape (Clay, personal communication 2002).

Prehistoric resources are more difficult to understand in simple factorial ways. The variation from one study unit to the next is somewhat unexpected. In some units, there are “sensible” reasons. For example, marshes and dunes (which lie along flat-piedmont interfaces) are important areas for food resources in the Bonneville Basin. Sites tend to be more frequent in these places. Other results are less “sensible”. Why should sagebrush flats be more likely to contain archaeology than piedmont in several of the Nevada study units? Why is water sometimes a negative factor? As promised, we offer no answers for these questions. They do make clear the importance of continuing to develop explanatory (causal, deductive) models alongside of correlation forecasts. The kind of study presented here, (an example of the latter activity) will be improved by creating forecasts from a better understanding of the rationale behind prehistoric behavior.

Planning Models As Tools

The utility and limitations for the planning model have been discussed above, but it is worth reiterating these again. The appropriate use of these tools is:

· Long range planning

· High probability relates to greatest likely overall expense

· Low probability equates with fewer resources, lower overall expense.

· If fewer sites are encountered, then testing, mitigation costs are reduced.

· Low probability does not mean no sites and does not obviate the need for fieldwork. But fieldwork should be faster and cheaper, on average.

· As model is verified further, cultural resource managers may want to examine different level of investigation within low probability areas.

· Models and forecasts articulate current state of knowledge. Thus they need maintenance.

Closing Perspectives

The GBRI model study was, we think, successful. Success is always a relative term. For this study, one goal was to evaluate the feasibility of building extensive landscape-level models. A feasibility evaluation was achieved: yes, this is feasible. Another goal was then to build such a model. This goal was also met in a series of analytical unit studies. A third goal was to examine how the model could best be used. This goal was partially met. Above, we proposed ways in which the model and forecasts can be used and improved. This should be a continual work in progress as managers and scientists work together to improve ways in which decisions are made.

Future work and elaboration needs to focus on improving survey methods, continually evaluating the model and forecast reliability, “fixing” the forecasts, and revising the model with better baseline environmental data. All of these future action recommendations pre-suppose agency use of the forecasts in the first place, and an audience for them. The agency staff, external researchers, and managers with whom we worked on this study are all sincerely interested in the success of landscape-level analyses. We think this is the best forecast, in itself, of continued use for this research.

VI. DISCUSSION

Planning Models As Tools The utility and limitations for the planning model have been discussed above, but it is worth reiterating these again. The appropriate use of these tools is:

Return to Top

Planning Models As Tools

The utility and limitations for the planning model have been discussed above, but it is worth reiterating these again. The appropriate use of these tools is: