Pollen-derived biomes in the Eastern Mediterranean-Black Sea-Caspian-Corridor

AIM: To evaluate the biomization technique for reconstructing past vegetation in the Eastern Mediterranean–Black Sea–Caspian‐Corridor using an extensive modern pollen data set and comparing reconstructions to potential vegetation and observed land cover data. LOCATION: The region between 28–48°N and 22–62°E. METHODS: We apply the biomization technique to 1,387 modern pollen samples, representing 1,107 entities, to reconstruct the distribution of 13 broad vegetation categories (biomes). We assess the results using estimates of potential natural vegetation from the European Vegetation Map and the Physico‐Geographic Atlas of the World. We test whether anthropogenic disturbance affects reconstruction quality using land use information from the Global Land Cover data set. RESULTS: The biomization scheme successfully predicts the broadscale patterns of vegetation across the region, including changes with elevation. The technique discriminates deserts from shrublands, the prevalence of woodlands in moister lowland sites, and the presence of temperate and mixed forests at higher elevations. Quantitative assessment of the reconstructions is less satisfactory: the biome is predicted correctly at 44% of the sites in Europe and 33% of the sites overall. The low success rate is not a reflection of anthropogenic impacts: only 33% of the samples are correctly assigned after the removal of sites in anthropogenically altered environments. Open vegetation is less successfully predicted (33%) than forest types (73%), reflecting the under‐representation of herbaceous taxa in pollen assemblages and the impact of long‐distance pollen transport into open environments. Samples from small basins (<1 km²) are more likely to be reconstructed accurately, with 58% of the sites in Europe and 66% of all sites correctly predicted, probably because they sample an appropriate pollen source area to reflect regional vegetation patterns in relatively heterogeneous landscapes. While methodological biase