Quantifying the Drivers of Palaeoecological Change

Past environmental conditions can be reconstructed from microfossil assemblages in lake or ocean sediments with a transfer function that uses the modern relationship between species distributions and the environment in a modern calibration set. The palaeoDrivers project is working to develop methods to assess how good these reconstructions are. PalaeoDrivers has worked with palaeoecologists who use chironomids to reconstruct air temperature and testate amoeba to reconstruct water table depths in bogs to identify which reconstructions are statistically significant and to try to determine which factors make a significant reconstruction more likely. We have found although reconstruction that span the deglaciation are usually statistically significant, surprising few Holocene reconstuctions are. The reasons for this are still under debate, but probably reflect the lower amplitude of the climatic signal. However, we believe that even if the climate signal in a single reconstruction is weak, the climate signal in a regional compilation of reconstruction is probably robust. Spatial autocorrelation in modern calibration sets tends to make the performance of transfer functions appear better than it is, but has been largely ignored by palaeoecologists, in part because of absence of methods to account for it. A paper from palaeoDrivers which is currently in review at the open access journal Climate of the Past, develops methods to correct the performance estimates and uses simulated species assemblages to test the methods. The use of simulated species assemblages is a relatively underused tool; the great advantage of simulated over real data for testing methods is that we know what the truth is and can check that our methods are accurate. Code for simulating species assemblages has been included a new version of the palaeoSig R package which is currently available on CRAN. In collaboration with researchers from AWI-Potsdam, palaeoDrivers has used the output of a coupled vegetation-climate simulation that spans most the last 6000 years to test the ability of transfer functions to reconstruct climate of different seasons and to test whether standard reconstruction diagnostics and the reconstruction tests developed by palaeoDrivers are useful. We find that reconstructions of climatic variables that are not ecological important can have gross errors in their trend. This is not surprising, but should inform the common practice of reconstructing many environmental variables without considering how important they are likely to be. The reconstruction significance test proved to be a skillful tool for identifying useful reconstructions. I have been using a blog [quantpalaeo.wordpress.com] to popularise the results of palaeoDrivers. This blog receives about 30000 page views per year.

Past environmental conditions can be reconstructed from fossil pollen and other microfossil assemblages using the modern relationship between the species and the environment. Such palaeoenvironmental reconstructions have been important evidence for severa l policy-relevant scientific debates. For example, diatom-based reconstructions of lake-pH contributed to the acid rain debate by showing that Norwegian lakes had acidified recently and simultaneously with increased atmospheric pollution, and reconstructi ons of past climate, using pollen preserved in lake sediments to infer air temperatures, or foraminifera in ocean sediments to infer sea surface temperatures, have helped understand natural climatic variability. Despite the widespread use of palaeoenviro nmental reconstructions, it has been difficult to assess how robust they are, potentially allowing weak or erroneous reconstructions to be published. Recent work by Telford and Birks (2011) developed a method to determine the statistical significance of p alaeoenvironmental reconstructions by testing if the reconstruction explains more of the variance in the fossil data than most reconstructions of random environmental variables. This project will extend these significance tests. The key challenge will be to develop appropriate null models. With our novel methods, we will be able to test which environmental variable, or combination of variables, best explains biotic changes. For example, using pollen data, we will determine whether precipitation or temper ature changes best explain vegetation dynamics on the Tibetan Plateau or in North America during the Holocene. We will also generate guidelines for the design of modern species-environment calibration sets to maximise the likelihood of obtaining statistic ally significant reconstructions rather than the current practice of attempting to maximise cross-validation performance. This project will help make palaeoecology a more quantitative, rigorous and robust scienc