THE EFFECT OF SAMPLE SIZE AND DECISION RULES ON PALEOENVIRONMENTAL RECONSTRUCTIONS USING THE MODERN ANALOG TECHNIQUE

Use of the modern analog technique (MAT) to reconstruct paleoenvironmental conditions has become widespread. Two of the basic issues that affect the reconstruction of vegetation when the MAT is implemented are: 1) how count size affects the quality of vegetation reconstructions; and 2) how analogs identified by the MAT are converted into vegetation reconstructions. We examine these issues in a Monte Carlo framework using simulated pollen assemblages derived from four fossil pollen assemblages from northern Michigan, U.S.A. For each fossil stratum, 1000 simulated assemblages were generated at each of eight count sizes between 75 and 1000 grains. Simulated and fossil assemblages were then compared (using the squared chord distance) to a coupled pollen-vegetation data set used to reconstruct forest communities in northern Michigan. We contrast the ability of simulated samples to accurately reconstruct forest community composition (as defined by the fossil pollen assemblages) using two quantitative reconstruction schemes. In the single-tier scheme, a single critical value is used to define analogs, and each analog is weighted equally. In the two-tier scheme, two critical values are used; the first tier defines unambiguous analogy of vegetation, while the second tier defines a somewhat less-close relationship. Second tier analogs receive one-half the weight of first tier analogs. The sensitivity of reconstruction accuracy to these schemes and to different levels of the minimum number of analogs required (one analog versus two or more analogs) is examined with count size in a four-way factorial experimental design. The general result of this research is that the combination of single-tier cutoffs with a requirement of two or more analogs to reconstruct represents a worst-case set of selection criteria, incapable of passing a 95% test for accuracy at any count size <1000 grains. In the best-case combination, two-tier cutoffs with a one analog minimum, reduction of accuracy at lower count sizes is small for three of the four fossil samples' simulations. At count sizes °Ý150 grains, the reconstruction accuracy for these three fossil layers exceeds 90%; for simulated samples derived from a homogeneous fossil assemblage, this level of accuracy is achieved at 75-grain counts.