Comparison of simulated and proxy-based climate reconstructions for Mid-Holocene Europe reveals high uncertainty

Gridded and time-continuous paleoclimate model outputs are increasingly used to inform high-resolution models in biogeography and other disciplines. However, few studies quantitatively evaluate such outputs for bias, uncertainty, and consistency with other climate reconstructions of similar scale and scope of interest. Here we evaluate downscaled and bias-corrected output from two paleoclimate models against proxy-based reconstructions from European sites in the Mid-Holocene (9000–5000 years before present) by comparing absolute values of annual precipitation sum, and mean January and July air temperature. We paired proxy-based climate values with simulated ones from the same time and place. For pairs within each site we checked for correlation. Then we pooled pair-wise differences of all sites together to check for systematic over- or underestimation. Correlation analysis revealed that climate models reproduce a winter warming trend in the northern half of Europe. Pair-wise differences indicate that all gridded datasets show both over- and underestimation bias compared to proxies in different climate variables, but without one dataset performing consistently better. Distinguishing performance between these datasets is complicated by high uncertainties in proxy-based reconstructions. Our results indicate that, at least on the whole-European scale, time-continuous model outputs resolve relevant Mid-Holocene climatic changes only to a very limited degree. In order to account for current uncertainties in reconstructing paleoclimate we advise users not to rely on only one gridded dataset and to evaluate reconstructions from models against proxy data from the area of interest.