The AMPD constitutes the most comprehensive, up-to-date, taxonomically harmonised modern pollen calibration dataset from Amazonia, with particular focus on the lowlands, yet compiled. This provides empirical basis for pollen-based reconstructions in the region. The AMPD was then used to investigate diversity gradients and community structure across the Amazon Basin, to identify climatically-related vegetation distribution patterns and to evaluate the ecological representativeness and calibration potential of the dataset.
The new river pollen trap design and related methodological approach, which was tested in rivers in the NW Amazon in Peru and Ecuador, represents the first device specifically made for time-integrated, instream pollen sampling method under continuous and natural water flow, enabling the investigation of fluvial pollen transport in Amazonia and potentially elsewhere as well. Our investigation (submitted to Journal of Ecology) revealed that the fluvial pollen signal reflects the composition of vegetation occurring along the river’s course, although the representation of vegetation communities depends on the size, sediment load, and discharge of the river. This device may also provide a new tool for calibrating the waterborne pollen component in pollen records. In addition, we also used the deployment of river traps in Peru to study the taphonomic processes that might influence the pollen signal by combining different pollen collectors (river traps, aerial traps, lake traps, surface samples) and relate them to modern vegetation data.
First application of quantitative pollen-based climatic reconstruction techniques to Amazonian fossil records, proving that this method is applicable in this region if enough calibration data is available, though further tests are required to ensure the accuracy of the models. So far, comparison of results with regional climate records allowed the identification of broad coeval trends between reconstructed climatic variables and other records of temperature, precipitation and ENSO (El Niño Southern Oscillation) variability. These comparisons also proved useful for identifying discrepancies between reconstruction outputs and regional records.
The application of Latent Dirichlet Allocation (LDA) methods to tropical pollen is innovative, and helped to identify shifts in ecological assemblages through time. This is a valuable method for the tropics, where the high diversity, distribution overlap (same species occurring in different environments) and large variation in taxonomic identification (several families identified to family or genus and few to species level) mean that few taxa can carry a meaningful climatic or ecological signal. The implementation of this method has opened future research avenues to implement LDA-derived topics for transfer functions rather than pollen data, which opens a promising new direction for reducing noise in complex tropical fossil pollen records.
Our Python package provides a fully open-source, accessible and interactive interface for proxy-based palaeoclimatic reconstruction (not just pollen), offering a valuable and transparent tool that can enhance the accuracy and validity of results. The application of this tool on our datasets helped to improve accuracy of the reconstructions, validate results, and reveal climatic gradients across the Amazon Basin.
