Occupancy Models Reveal Regional Differences In Detectability And Improve Relative Abundance Estimations In Fossil Pollen Assemblages

The late Quaternary fossil record provides crucial data that demonstrate how organisms respond to climate change. These records have been used to great effect, demonstrating that no-analog communities frequently occur during periods of no-analog climate, and that taxa demonstrate individualistic responses to change. However, our efforts to reconstruct biotic responses to environmental change are frequently hampered by inconsistent sampling and differential preservation of fossil taxa. We leveraged occupancy modeling methods and the fossil pollen record across eastern North America to identify circumstances under which occupancy modeling improves our ability to estimate relative abundance in four pollen taxa (Corpus, Fagus, Picea, and Pinus) through time (15 kya to present) and to identify localities where data are unreliable reflections of the local community. We found that integrating observed pollen abundance and detectability improves model performance. Low genus richness and large basin area were consistently important determinants of low detection. Our occupancy models were most informative for taxa with high enough variation in observed pollen abundance for models to be adequately calibrated. We combined occupancy model estimates of pollen abundance and detectability with a Getis-Ord statistical approach to identify spatial clusters of high or low detectability, identifying regions where a taxon's pollen is more (or less) reliable. This work will advance the integration of ecological and paleontological sciences by allowing us to better identify whether a pollen taxon is truly absent from a fossil site or if it has simply gone undetected, allowing us to produce more robust paleoecological models. This approach will bolster our ability to identify the responses of plant communities to past climatic and anthropogenic change so that we can improve our predictions of future responses. (C) 2020 Elsevier Ltd. All rights reserved.