Modelling Late Miocene vegetation in Europe: Results of the CARAIB model and comparison with palaeovegetation data

The CARAIB (CARbon Assimilation In the Biosphere) model is used to study the vegetation distribution during the Late Miocene (Tortonian). In this version, the plant classification is specifically adapted to best represent Miocene European vegetation. Compared to other plant classifications used in global models, this adapted classification is more refined, since it is specifically developed for European vegetation and it includes various thermophylous tree types, which were present in Europe during the Miocene. The corresponding climatic tolerance parameters are based on the study of Laurent et al. (Journal of Vegetation Science, 15, 739-746, 2004) for the tree types currently present in Europe and on the distribution of analogue species in southeastern Asia and North/Central America for the thermophylous (sub-tropical) trees. The same classification is used to characterize the palaeoflora at the available Late Miocene localities, allowing a model-data comparison at the plant functional type level, rather than at the biome level. The climatic inputs to CARAIB are obtained from the COSMOS atmosphere-ocean general circulation model. The climatic anomalies (Tortonian minus Present) derived from COSMOS are interpolated to a higher spatial resolution before being used in the vegetation model. These anomalies are combined with a modern climatology to produce climatic fields with high spatial resolution (10' x 10'). This procedure has the advantage of making apparent relief features smaller than the grid cells of the climate model and, hence, makes easier the comparison with local vegetation data, although it does not really improve the quality of the Tortonian climate reconstruction. The new version of CARAIB was run over Europe at this higher spatial resolution. It calculates the potential distribution of 13 different classes of trees (including cold/cool/warm-temperate, subtropical and tropical types), together with their cover fractions, net primary productivities and biomasses. The resulting model vegetation distribution reconstructed for the Tortonian is compared to available palaeovegetation and pollen data. Before performing this comparison, the tree taxa present at the various data sites are assigned to one or several model classes, depending on the identification level of the taxa. If several classes are possible for a taxon, only those that can co-exist with the other tree classes identified at the site are retained. This methodology is similar to the co-existence approach used in palaeoclimatic reconstructions based on vegetation data. It narrows the range of tree types present at the various sites, by suppressing in the data the extreme types, such as the cold boreal/temperate and tropical trees. The method allows a comparison with the model simulation on a presence/absence basis. This comparison provides an overall agreement of 53% between the model and the data, when all sites and tree types are considered. The agreement is high (>85%) for needle-leaved summergreen boreal/temperate cold trees (Larix sp.) and for tropical trees, intermediate (>40%) for other boreal/temperate cold trees and for needle-leaved evergreen temperate cool trees, broadleaved summergreen temperate cool trees and broadleaved evergreen warm-temperate trees, and poor (<40%) for most temperate perhumid warm trees. In many cases, the model is shown to be better at predicting the absence than the presence, as observed for tropical trees. The modelled distributions of cold boreal/temperate trees tend to extend too much towards the south compared to the data. B contrast, model sub-tropical trees (temperate perhumid warm and needle-leaf summergreen temperate warm trees) appear to be restricted to some limited areas in southern Europe, while they are present in the data from central Europe up to at least 50 degrees N. Consequently, modelled Late Miocene climate appears to remain too cold to produce assemblages of trees consistent with the data. The predicted modelled trends from the past to the present are in the right direction, but the amplitude remains too small. For the simulations to be in a better agreement with the data, higher CO2 levels may be necessary in the climate simulations, or possibly other oceanic boundary conditions may be required, such as different bathymetry in the Panama seaway. (C) 2011 Elsevier B.V. All rights reserved.