Comparing and improving methods for reconstructing peatland water-table depth from testate amoebae

TitleComparing and improving methods for reconstructing peatland water-table depth from testate amoebae
Publication TypeJournal Article
Year of Publication2019
AuthorsNolan, C, Tipton, J, Booth, RK, Hooten, MB, Jackson, ST
JournalThe Holocene
Pagination1350 - 1361
Date PublishedJan-08-2020
KeywordsBayesian statistics, compositional data, Holocene, hydrology, North America, Northeast United States, Paleoclimate, peatlands, statistical methods, testate amoebae

Proxies that use changes in the composition of ecological communities to reconstruct temporal changes in an environmental covariate are commonly used in paleoclimatology and paleolimnology. Existing methods, such as weighted averaging and modern analog technique, relate compositional data to the covariate in very simple ways, and different methods are seldom compared systematically. We present a new Bayesian model that better represents the underlying data and the complexity in the relationships between species’ abundances and a paleoenvironmental covariate. Using testate amoeba–based reconstructions of water-table depth as a test case, we systematically compare new and existing models in a cross-validation experiment on a large training dataset from North America. We then apply the different models to a new 7500-year record of testate amoeba assemblages from Caribou Bog in Maine and compare the resulting water-table depth reconstructions. We find that Bayesian models represent an improvement over existing methods in three key ways: more complete use of the underlying compositional data, full and meaningful treatment of uncertainty, and clear paths toward methodological improvements. Furthermore, we highlight how developing and systematically comparing methods lead to an improved understanding of the proxy system. This paper focuses on testate amoebae and water-table depth, but the framework and ideas are widely applicable to other proxies based on compositional data.