Recognising and measuring birch pollen faster using AI
Publication of the working group of Prof. Dr. Brigitte Urban of the Institute of Ecology in "Ecology and Evolution"
2024-07-01 Prof. Dr. Brigitte Urban's research group at the Institute of Ecology is working intensively on the automatic recognition of microscopic objects (pollen, spores, etc.) using AI-based methods. A publication on this has now been accepted in "Ecology and Evolution" and was published online on 14th June 2024. This study is about the automatic measurement of birch pollen. These measurements make it possible to determine the proportion of very similar tree and dwarf birch pollen in a sample. The necessary size measurements were previously only possible manually, with a great deal of effort. The working group has now been able to develop an automatic method for the first time.
The analysis of pollen from lake sediments and peat is a central tool for the reconstruction of past vegetation. Birch pollen is often represented in high proportions. Today, this pollen comes from our native tree birches (Betula pubescens, Betula pendula), but older deposits may also contain pollen from the dwarf birch (Betula nana). Dwarf birch is currently native to the Arctic and alpine highlands, but was also widespread in Central Europe in earlier glacial periods. In order to reconstruct the former distribution of boreal forests with tree birches and open steppe tundra with dwarf birches in such glacial phases, it is necessary to distinguish between the very similar pollen of tree and dwarf birches.
From a purely morphological point of view, the pollen can only be distinguished with great uncertainty. Alternatively, the proportion of tree and dwarf birch pollen in a sample can be estimated using size statistics. Until now, pollen had to be measured manually using light microscopy, which is very time-consuming. In a new study, Prof. Brigitte Urban's working group at Leuphana University Lüneburg, in cooperation with the University of Greifswald, has developed an automated, AI-based method for measuring birch pollen that can automatically measure many thousands of pollen grains per hour. In a recent publication, the method is presented with two example applications from the Neanderthal site in Lichtenberg, Lüchow-Dannenberg district in Lower Saxony, and the Kieshofer Moor near Greifswald. Both examples show clear shifts in the proportion of tree and dwarf birch as a result of climatic changes over the past ~130,000 years. The approach is therefore an important tool, e.g. for reconstructing the Neanderthal palaeoenvironment in the ongoing joint project ‘Investigating climate and landscape changes and human adaptation strategies since the last interglacial period in northern Germany’ (Climate Change and Early Humans in the North, CCEHN).
The approach is also suitable for investigating whether and for how long the climate- and location-dependent dwarf birch, which also grows in shrub form, has survived in lowland peatlands in Lower Saxony since the end of the last ice age and which environmental factors have displaced it.
In the next step, the AI approach will be extended to other pollen types (e.g. grasses) and other microscopic objects of high environmental or medical relevance (e.g. algae, fungal spores or microscopic charcoal particles), where size parameters provide valuable additional information for taxonomic detection.
The publication is available at the following link.