Reliability and resolution of the coexistence approach — A revalidation using modern-day data

The coexistence approach (CA) is widely used to reconstruct palaeoclimates for the Cenozoic. Most published CA analyses relied on climate data for nearest living relatives (NLRs) stored in the Palaeoflora database (PFDB). Here, we used more than two-hundred modern relevés (taxon lists of forest stands) from North American, Caucasian and East Asian forest regions in order to test the ability of CA/PFDB to estimate palaeoclimate. Since only data for mean annual temperature (MAT) are publicly available from the PFDB, we concentrated on this climate parameter. Two criteria were tested: ‘resolution’ and ‘reliability’ of CA/PFDB analyses. The CA assumes that for a given climate parameter (e.g. MAT; mean annual precipitation; coldest month mean temperature etc.) the interval shared by all or nearly all NLRs for a fossil assemblage is best describing the past climatic conditions. Narrow, i.e. well-resolved, intervals are desirable, since they describe most precisely the climate. Our results show that CA/PFDB is unable to reliably reconstruct the actual climates of most of the relevés analysed. CA/PFDB performed best for lowland and mid-altitude stands with MAT of ca. 13–16 °C, while producing remarkably incorrect results for warmer lowland stands and cooler stands at higher elevations. This is mainly due to generally incorrect entries of MAT ranges of NLRs in the PFDB. Using corrected MAT tolerances, the reconstructed, low-resolved intervals (3 °C in exceptional cases, typically 5–10 °C) fall within the actual climates. Hence, only dramatic climate changes are likely to be captured in a CA analysis. This renders the coexistence approach useless for the quantitative reconstruction of palaeoclimate and calls for alternative approaches of investigating past climates by means of fossil plants.

► Coexistence approach (CA) does not work.
► MAT (mean annual temperature) data in Palaeoflora database (PFDB) are erroneous.
► CA with realistic MAT ranges has little climatic resolution.
► Other climate parameters in PFDB need to be validated.
► Straightforward descriptive statistics of minimum climatic tolerances may work.