crestr An R package to perform probabilistic climate reconstructions using fossil proxies

Statistical climate reconstruction techniques are fundamental tools to study past climate variability from fossil proxy data. In particular, the methods based on probability density functions (or PDFs) have the potential to be used in various environments and with different climate proxies because they rely on elementary calibration data (i.e. modern geolocalised presence data). However, the access and curation of these calibration data, as well as the complexity of interpreting probabilistic results, often limit their use in palaeoclimatological studies. I introduce a new R package (crestr) to apply the CREST method (Climate REconstruction SofTware) on diverse palaeoecological datasets and address these problems. crestr includes a globally curated calibration dataset for six common climate proxies (i.e. plants, beetles, chironomids, rodents, foraminifera, and dinoflagellate cysts) associated with an extensive range of climate variables that enables its use in most terrestrial and marine environments. Private data collections can also be used instead of, or in combination with, the provided calibration dataset. The package includes a suite of graphical diagnostic tools to represent the data at each step of the reconstruction process and provide insights into the effect of the different modelling assumptions and external factors that underlie a reconstruction. With this R package, the CREST method can now be used in a scriptable environment, thus simplifying its use and integration in existing workflows. It is hoped that crestr will contribute to producing the much-needed quantified records from the many regions where climate reconstructions are currently lacking, despite the availability of suitable fossil records. The use of the package will be illustrated with a recent application to produce a 790,000 year long mean annual temperature reconstruction based on a pollen record from southeastern Africa.