Deposition times in the northeastern United States during the Holocene: establishing valid priors for Bayesian age models

Age depth relationships in sedimentary archives such as lakes, wetlands and bogs are non-linear with irregular probability distributions associated with calibrated radiocarbon dates. Bayesian approaches are thus well-suited to understanding relationships between age and depth for use in paleoecological studies. Bayesian models for the accumulation of sediment and organic matter within basins combine dated material from one or more records with prior information about the behavior of deposition times (yr/cm) based on expert knowledge. Well-informed priors are essential to good modeling of the age depth relationship, but are particularly important in cases where data may be sparse (e.g., few radiocarbon dates), or unclear (e.g., age-reversals, coincident dates, age offsets, outliers and dates within a radiocarbon plateau). Here we assessed Holocene deposition times using 204 age depth models obtained from the Neotoma Paleoecology Database (www.neotomadb.org) for both lacustrine and palustrine environments across the northeastern United States. These age depth models were augmented using biostratigraphic events identifiable within pollen records from the northeastern United States during the Holocene and late-Pleistocene. Deposition times are significantly related to depositional environment (palustrine and lacustrine), sediment age, and sediment depth. Spatial variables had non-significant relationships with deposition time when site effects were considered. The best-fit model was a generalized additive mixed model that relates deposition time to age, stratified by depositional environment with site as a random factor. The best-fit model accounts for 63.3% of the total deviance in deposition times. The strongly increasing accumulation rates of the last 500-1000 years indicate that gamma distributions describing lacustrine deposition times (alpha = 1.08, beta = 18.28) and palustrine deposition times (alpha = 1.23, beta = 22.32) for the entire Holocene may be insufficient for Bayesian approaches since there is strong variation in the gamma parameters both in the most recent sediments and throughout the Holocene. Time-averaged gamma distributions for lacustrine (alpha = 1.35, beta = 19.64) and palustrine samples (alpha = 1.40,beta = 20.72) show lower overall deposition times, but variability remains. The variation in gamma parameters through time may require the use of multiple gamma distributions during the Holocene to generate accurate age depth models. We present estimates of gamma parameters for deposition times at 1000 yr intervals. The parameters generated in this study can be used directly within Bacon to act as Bayesian priors for sedimentary age models. (C) 2012 Elsevier Ltd. All rights reserved.