DISQOVER workshop 2020

Because of the complications caused by the COVID-19 virus, the workshop will be postponed. New dates are not yet fixed.

March 31. – April 3. 2020 in Greifswald/Germany

Idea and content
Quantitative methods are of increasing importance in palynology. In the
workshop we introduce the principle of some main methods and explain their
application in the R environment for statistical computing, using tools from
the disqover package. We practice application using real data. The work-
shop gives room to discuss current problems and future developments

Preliminary program
Day 1: Regional scale vegetation reconstruction with REVEALS, ROPES and EDA
Day 2: Local scale vegetation reconstruction with Marco Polo and the LRA
Day 3: Calibrating PPEs with both surface and core data
Day 4: Practicing + Discussion of future plans

DISQOVER package
The R-package DISQOVER is a collection of quantitative methods in palynology. We started its development for three main reasons:

  • to provide implementations that are easy to use and that can be automatised to handle large data sets
  • to provide full flexibility in the underlying parameters, including pollen dispersal
  • to stimulate critical discussion on the methods and parameters as a basis of future development.

During the course we introduce the principles of each approach, introduce and train application in R and discuss strength and weaknesses of the approaches.

The number of participants is limited to 15.
Fees: There is no workshop fee, we just ask for ~20 € for snacks and coffee.
Organizers: Almut Mrotzek, John Couwenberg, Martin Theuerkauf
Contact: martin.theuerkauf@uni-greifswald.de

The simplest ROPES example

For an introduction of ROPES at the EPPC2018 in Dublin, I used this spreadsheat table (ROPES and how it works) with a very simple example of a ROPES application. The example has just two taxa, A and B, and two samples. Cover is first translated into pollen data (PARs and counts) and then again into cover with REVEALS (using some simplications). Playing with the PPEs (that of B could be used as the reference) shows the principle of ROPES – changes in the reconstructed cover should be similar to changes in the PARs.

DISQOVER workshop 2018

DISQOVER Workshop September 2018


27-28. of September 2018 in Greifswald

Aims and scope

The R-package DISQOVER is a collection of quantitative methods in palynology. We started its development for three main reasons:

  1. to provide implementations that are easy to use and that can be automatised to handle large data sets
  2. to provide full flexibility in the underlying parameters, including pollen dispersal
  3. to stimulate critical discussion on the methods and parameters as a basis of future development.

The 2-day workshop will introduce four methods from the package: REVEALS, MARCO POLO,
EDA and ROPES. We introduce the principle of each approach, introduce and train application in R and discuss strength and weaknesses of the approaches.

Aims and scope

The workshop is free, but limited to 15 participants. To register, send an email to: martin.theuerkauf@uni-greifswald.de

ROPES – Quantitative reconstructions without pollen productivity estimates

While playing with a high resolution pollen record from lake Tiefer See, John and I developed the idea of a new method to reconstruct past vegetation composition. It is now published as “ROPES” in Frontiers of Earth Science. ROPES stands for ‘Reveals withOut PpES’. So, like Shinya Sugitas REVEALS method, ROPES aims to translate pollen data from large lakes (or peatlands) into regional vegetation composition. Other than REVEALS, ROPES does not require pollen productivity estimates (such as PPE or RPPs).

Traditional use of ropes in Greifswald – Treideln of the “Weiße Düne” (that’s the boat) from the city harbour towards the Bay of Greifswald. (© Melanie Asse)


Why is it useful?
All pollen records from lakes and peatlands are biased by differences between single plant taxa in the pollen production and dispersal (to name the main problem, there are more). As already the founders of palynology recognised, pollen assemblages can therefore not simply be translated into past vegetation composition. Today, two general approaches exist to reduce the bias: the modern analogue technique (MAT) and correction factor approaches. MAT first determines modern relationships between pollen assemblages and vegetation cover and then applies these relationships to palaeo-assemblages to infer past vegetation cover. The main limitation of MAT is the lack of modern analogues. In Europe, for example, the forests that covered the continent during the mid-Holocene have largely disappeared. So, surface samples that represent these forests of the past are virtually not existing as modern analogues. Hence, reconstructions of these forests are problematic, as e.g. the paper of Zanon et al. (2018) clearly shows.

More than 30% of mid-Holocene pollen samples from European pollen records have no modern analgue in 2526 surface samples from across continent. These numbers are based on a threshold value T of 0.3. With the more strict value of 0.2 that Davis et al. 2015 proposed for a similar data set, the proportion of missing modern analogs would be even higher.

Correction factor approaches like REVEALS or STEPPS (Dawson et al. 2016) account for the production bias with taxon specific pollen productivity estimates – with the advantage that success does not depend on vegetation composition. (So, Zanon et al. 2018 actually use REVEALS to re-calibrate their MAT results). However, calibrating pollen productivity requires comprehensive sets of surface pollen samples and vegetation data. The calculations are far from simple and require a good understanding of pollen dispersal. Therefore, reliable pollen productivity estimates are thus far available from a few selected areas only. Furthermore, pollen productivity of plants is not constant but may change substantially in response to variable factors, e.g. climate, soils, land management and stand structure. So, correction factor approaches are limited through the availability and the stability of the correction factors.

ROPES translates pollen data from large lakes (or peatlands) into regional vegetation composition without any modern reference. The method is not affected by the just mentioned limitations of MAT and correction factor approaches. It therefore has potential to enable reconstructions for areas and periods, for which the existing approaches are not applicable.

How does it work?
To arrive at reconstructions from the pollen data alone, ROPES combines pollen counts and pollen accumulation rate data (PAR). Modern PARs appear to vary too strongly between sites to interpret them in a quantitative way. Still, changes in PARs through time in a single record should be meaningful: if for example the PAR of pine pollen doubles at some point, that probably means that the cover of pine has doubled (we discuss this assumption below). So while PAR values are difficult to interpret in terms of past plant abundances, they can well be interpreted in terms of changes in plant abundances. To arrive at actual cover values, ROPES combines the analysis of PAR values with REVEALS. The idea is that a REVEALS reconstruction should show the same changes as the PAR values, like the doubling in pine. This information about the past vegetation allows for applying REVEALS without pre-defined pollen productivity estimates. We instead search for those estimates that give the best solution, i.e. the solution in which the resulting changes in the cover of each taxon best match the changes in PAR values. In other words, we search for the pollen productivity estimates, with which for each taxon, the ratio between the REVEALS reconstructed cover and PAR values is (close to) constant all along the record.

Assumptions of ROPES
The fundamental assumption of ROPES is that changes in PAR values are linearly related to changes in plant abundances (the doubling of pine example…). This assumption is probably robust in stable sedimentary environments, i.e. stable lakes. Thomas Giesecke and Sonja Fontana have shown very stable pollen accumulation in three Swedish lakes over the Holocene. Changes in the sedimentary conditions due to changes in e.g. shape, size or depth of a lake may change pollen deposition at a lakes bottom and thus violate the assumption. Lakes that experience(d) such changes are not suited for ROPES.
Beyond that, ROPES also relies on all the underlying assumption of REVEALS. Sites should receive only atmospheric pollen deposition, and the surrounding vegetation has to be largely homogeneous.

New version of the disqover package

Version 0.9 of the disqover package is now available. The new version includes two new functions (MarcoPolo and edar), plotting tools for REVEALSinR and MarcoPolo, a much improved help and a number of smaller corrections.

The MarcoPolo function is an implementation of the MARCO POLO approach (Mrotzek et al. 2017). The approach aims for stand scale vegetation reconstructions from pollen data. MARCO POLO combines pollen data from very small sites (with much (extra)local pollen deposition) with pollen data from large sites (with regional pollen deposition). The approach is based on iterative  manipulations of the pollen sum. In the package, the MarcoPolo function comes with example data from the publication for illustration and testing. The package also provides the plotMP() function for plotting the results.

The edar function is an implementation of the Extended Downscaling Approach (EDA, Theuerkauf and Couwenberg 2017). The approach aims to detect past vegetation patterns and plant communities within landscapes. To that end, EDA applies iterative forward modelling to fit vegetation composition to robust landscape patterns by comparing simulated with observed pollen deposition. Again, the R function comes with example data for illustration and testing.

The new disqover version includes some changes for the REVEALSinR function. Most importantly, the first column of the data should now include ages or depth. See help for the details. Plotting is now available with the plotREVEALS() function. Two main options are available. First, a diagram with seperate columns (type = “single”):

Example plot of REVEALSinR with the Lake Tiefer See data (single column version)

The second version is a stacked diagram (type = “stacked”):

Example plot of REVEALSinR with the Lake Tiefer See data (stacked version)

Please contact us about any problems or comments!

first disqover workshop is over

The first disqover workshop is over. 11 participants from all around the globe came to Greifswald to discussed application of quantitative methods in palynology with R. They brought many questions and ideas for the future package development (e.g. plotting and GIS functions). Some of them will be implemented in the upcoming version 0.9. When the rain stopped after 2.5 days, we even had some time for sightseeing. Thanks to all the participants for making this workshop a success!

EDA paper out

The EDA (extended downscaling approach) aims to find the pattern in past vegetation. We here illustrate what EDA does – and how. In the paper, we have tested performance of the approach, now implemented in R, in five synthetic scenarios. In all scenarios, landcover is correctly reconstructed for over 90% of the model landscapes.

REVEALS and the homogeneity bias in pollen data

When Lennart von Post presented the first pollen diagrams and thus laid the foundations of palynology in 1916, the limitations of the method were immediately recognised. Plants produce pollen in very different amounts, so there is a production bias in pollen data. Pollen types differ in fall speed, so there is a dispersal bias in pollen data. Thirdly, plants growing near a site contribute more pollen than plants growing at greater distance. This problem, first raised by Hesselman, may be called the homogeneity bias because it is only a problem if vegetation composition near the site is different from vegetation composition further away. The problem reflects the difference between the ‘world as a pollen record sees it’ and the ‘world as humans see it’. Pollen samples provide a distance weighted assessment of the relative cover of various taxa, whereas humans see vegetation patches at various distances. These two views only match when the vegetation cover is homogeneous.

The three biases have hampered the obviously necessary translation of pollen data into past plant abundances for many decades. Only in 2007, Shinya Sugita presented a solution: the REVEALS model. REVEALS reduces (ideally eliminates) the production bias in pollen percentage data with pollen productivity estimates (PPEs) and the dispersal bias with the K factor. K factors are calculated with a dispersal model. Since then, REVEALS has become an important, regularly applied instrument in palynology. And we hope that our R implementation with a state-of-the art pollen dispersal modelling helps to further broaden its application.

It is often ignored, however, what REVEALS cannot achieve: it does not account for the homogeneity bias. REVEALS output can only be interpreted in terms of mean regional vegetation composition, but this interpretation is only solid if vegetation composition is homogeneous across the pollen source area. A simple example illustrates the problem. Assume a pollen sample from a large lake with 50% pollen of taxon A (in green) and 50% pollen of taxon B (in grey). For simplicity, both have similar pollen productivity and fall speed of pollen (3 cm s-1). The lake has a radius of 500 m, is nicely circular and has no inflow. Pollen dispersal follows the LS model, which predicts with the given settings that 50% of total pollen deposition arrives from within 25 km distance and 95% from within 100 km distance. So to arrive at 50% of total pollen deposition, taxon A may cover all the area within 25 km distance (case 1), or half of the full source area (case 2), or all the area between 25 km and somewhat more than 100 km distance (case 3).

Absolute cover of taxon A would be ~2000 km² in the first case, 16,000 m² in the second case or 30,000 km² in the third case: very different cover but still the same pollen deposition.

This example is certainly extreme and will rarely apply in reality. It still illustrates that the inherent postulation of homogeneity may not be ignored in REVEALS modelling. Whether regional vegetation was indeed homogeneous can be validated by comparing the pollen records of a number of sites. Similar pollen records will prove homogeneity. Different pollen records indicate that vegetation composition was not homogeneous across the region – and that REVEALS output can be misleading in terms of representing actual vegetation composition. In such situations, ever so common in a Central European context, the EDA will help explore the vegetation pattern.