GEOSTATISTICAL ANALYSIS: COMBINING CATEGORICAL GEOLOGIC AND CULTURAL DATA TO CREATE A PREDICTABILITY MODEL FOR ARCHAEOLOGICAL SENSITIVITY

Being able to create areas on which to focus a cultural resources inventory is essential in the Bureau of Land Management (BLM) to better expedite the archaeological process. Cultural inventories must be conducted before any BLM project can start or be renewed whether it is related to a new oil and gas project, range grazing permits, or any other surface disturbing event. The BLM office in Pinedale, WY is in a rich cultural area due to its location at the heart of historic westward expansion on the Oregon-California Trail System and within a landscape heavily used by prehistoric Native Americans.

In creating the probability model categorical data was categorized by, low likelihood features were determined by the archaeologist to be slopes greater than 15 degrees, clay flat areas, extremely rocky or shallow soils, and any pre-existing disturbance. Moderate likelihood features were areas within 100 feet of any intermittent/dry drainage or within 100 feet of a basin or playa environment. High likelihood features were sandy soil environments, areas within a quarter mile of perennial water, 100 feet around ridges/outcrops, and 100 feet around a previously-identified cultural site. Taking this process one-step further by quantifying the likelihoods and running geostatistical models within ArcMap, what was once a probability model is now a predictive model; an advanced practice not commonly used.

The benefit of having a predictive model versus a probability model is the increased spatial refinement and quantified likelihood of a site. Probability models are a generalized likelihood of an event, but do not take into account the statistical hierarchy of how likely a site will be.

When quantifying these features, a 41,586-point grid (one point per 100m²) was laid over the study area and a point given to each feature the point intersected. Then, the totals were multiplied by a factor of 3 and 6 for moderate and high -likelihood features. From here, a geostatistical analysis was conducted using the Empirical Bayesian Kriging method with a predictive output with an empirical transformation and further examined with a K-Bessel semivariogram. Further analysis was conducted with Geographic Weighted Regression and Hot Spot Analysis.