Ceramic Petrography Laboratory
Detailed Methods
Development of the Hand Sample Identification Model
The development of thin section based petrofacies models is ideal for the application of rigorous quantitative techniques to the problem of temper provenance. However, it is not easily applied to the bulk of the archaeological ceramics recovered during excavation. Our early efforts at developing a hand sample identification protocol were cumbersome: the large numbers of identified grain types made descriptions difficult to apply to individual samples. After several attempts, we have developed a means of using the petrographic thin section data and the discriminant analysis to build a descriptive key to the sands as they are seen in hand sample. This key forms the bridge between the high- and low-power microscopy used to analyze thin sections and the bulk of the sherds, respectively. Without it, the detailed sand composition data would not be worth the time and effort expended to collect them.
Choosing Sand Samples to Describe
The first step in developing reliable descriptions of the sands in hand sample is to choose how many (and which) sands to describe. It was obvious soon after the process of writing sand descriptions commenced that describing each sample would generate much more data than was necessary to complete the study. Trying to account for too much variability in the first approximation was what doomed the original descriptive key.
The "predicted group probability scores" generated by the discriminant analysis were selected as the means of culling the "quintessential" sands for each petrofacies from the full set of sands assigned to a petrofacies. The discriminant analysis assigns a posterior probability score to each sample; this score represents the likelihood of the sample's predicted group assignment being correct. For any given sample, the probability of a random assignment is termed the "prior probability;" it is dependent on the number of groups and samples in each discriminant analysis. For most petrofacies, only those samples with posterior probabilities of greater than 75 percent are used to derive initial hand sample descriptions for the petrofacies.
Description of Sand In Hand Sample
For each petrofacies, the samples chosen for description are examined under a binocular stereomicroscope at magnifications of 6x to 30x. The grain types found in each sample are identified and described, and the quantity of each grain type is identified as one of five ordinal categories. The ordinal abundance categories were chosen because they can be assigned relatively easily by estimate. The categories indicate the abundance of each grain type with more precision than a presence-absence scheme would allow, without taxing the ability of the petrologist or the ceramicist to truly characterize abundance. A more complex scheme would be less believable and accurate, and a less complex scheme would not allow for the rigorous characterizations needed to separate sands of many similar petrofacies.
The use of the Gazzi-Dickinson method to point count thin sections precludes the use of point count data to describe hand sample abundance estimates directly. As stated previously, the Gazzi-Dickinson technique requires that all monomineralic grains sand size and larger be counted as their mineral type, rather than as a rock fragment. Because of this, a sand made up of a coarse grained rock such as granite will have high point count values for quartz and feldspars, even though it looks like granite in hand sample. Because our goal is to facilitate identification of sand tempers as they appear in hand sample to the nongeologist, point count data are used only as a starting point for the hand sample analysis. All hand sample abundance estimates are derived from examination of the hand sample, not from the point count data. In this case, a "traditional" point counting technique, in which all rock fragments are counted as such regardless of their grain size, would allow the use of point count data as a direct measure of abundance in hand sample. However, sand and/or sand temper samples of differing grain sizes would not be directly comparable using a traditional point count technique. We feel that it is far more important to maintain the integrity of the model by minimizing potential differences between sands and sand temper than to increase the precision of the abundance estimate in the sand hand samples. Furthermore, abundance estimates in the bulk of the sherds are based strictly on hand sample observation so there would have been no real value in increasing the precision of the sand hand sample data set.
Once the "quintessential" samples in a petrofacies are described, the results are tabulated, and the most commonly occurring ordinal category is chosen as representative of each grain type in the petrofacies. For instance, if a petrofacies had five described samples, and quartz is found to be present in one, common in three, and abundant in one, then quartz is described as common for the petrofacies. Using the modal occurrence of the abundance category for each grain type in the sands of each petrofacies allows the description of a "modal sand" for the petrofacies. The "modal sand" is not exactly like any particular sand in the petrofacies, but represents the central tendencies of the petrofacies. Sand, like all complex natural systems, can be highly variable on the localized scales at which it is collected by either petrologists or potters. The initial sand descriptions by petrofacies identify how the sands in each petrofacies usually appear.
The abundance values for the "modal sand," and the grain descriptions collected during the hand sample description process are assembled into a master description by petrofacies for all defined petrofacies. A grain box for each petrofacies is prepared. The grain box is a small matchbox divided into 20 spaces, or "slots." Each slot contains several examples of grains representative of one of the grain types listed in the master description. To learn the characteristics of any given petrofacies, a petrologist or ceramicist can examine the grains and the descriptions at the same time, thus preparing for the identification of the sand as a temper. By using this approach, the ceramicist becomes familiar with all of the characteristics of the sand available from an area rather than relying solely on one or two grains. The clear advantage of this technique over a key grain approach is that provenance assignments are made on the basis of a simultaneous match between a number of variables in sand and sand temper, thus increasing the likelihood of a correct assignment.
