Identification of starch granules using a two-step identification method

Abstract

Starch analysis has proven to be a powerful method applicable to recover microbotanical remains of starchy foods in archaeological contexts, and morphometric analysis is the most commonly used methodological approach for identifying starch granules. However, it is sometimes not easy to achieve a high level of accuracy in identification, if several coexisting taxa in an assemblage exhibit similar starch morphology. The current study attempts to use both traditional morphometric observation and also a computer-based discriminant analysis to create a multivariate model, in order to separate Job's tears (Coix lacryma-jobi) from foxtail millet (Setaria italica ssp. italica) and broomcorn millet (Panicum miliaceum) that show considerable overlapping in starch morphology and size. The two-step identification method generated in this study shows a greater power of discrimination for identifying these three taxa with high success rates. The model was then used for identification in ancient assemblages with satisfactory efficiency and accuracy. This method will be most useful for application to ancient starch assemblages recovered from sites where dry-land farming was a significant part of the subsistence strategy, such as in East Asian Neolithic and Bronze Ages.