Harmonising plant functional type distributions for evaluating Earth system models

Abstract. Dynamic vegetation models simulate global vegetation in terms of fractional coverage of a few plant functional types (PFTs). Although these models often share the same concept, they differ with respect to the number and kind of PFTs, complicating the comparability of simulated vegetation distributions. Pollen-based vegetation reconstructions are initially only available in the form of time series of individual taxa that are not distinguished in the models. Thus, to evaluate simulated vegetation distributions, the modelling results and pollen-based vegetation reconstructions have to be converted into a comparable format. The classical approach is the method of biomisation, but hitherto PFT-based biomisation methods were only available for individual models. We introduce and evaluate a simple, universally applicable technique to harmonise PFT distributions by assigning them into nine mega-biomes, using only assumptions on the minimum PFT cover fractions and few bioclimatic constraints (based on the 2 m temperature). These constraints mainly follow the limitation rules used in the classical biome models (here BIOME4). We test the method for six state-of-the-art dynamic vegetation models that are included in Earth system models based on pre-industrial, mid-Holocene and Last Glacial Maximum simulations. The method works well, independent of the spatial resolution or the complexity of the models. Large biome belts (such as tropical forest) are generally better represented than regionally confined biomes (warm–temperate forest, savanna). The comparison with biome distributions inferred via the classical biomisation approach of forcing biome models (here BIOME1) with the simulated climate states shows that the PFT-based biomisation is even able to keep up with the classical method. However, as the new method considers the PFT distributions actually calculated by the Earth system models, it allows for a direct comparison and evaluation of simulated vegetation distributions which the classical method cannot do. Thereby, the new method provides a powerful tool for the evaluation of Earth system models in general.