Modelling landscape dynamics in a glacial refugium – or the spatial and temporal fluctuations of tree line altitudes
Correspondence: Silvio Marta, Department of Biology, University of Rome ‘Tor Vergata’, Via Cracovia 1, 00133 Rome, Italy.
To produce a dynamic model of tree line position and habitat suitability for temperate and warm temperate forests, with high spatial and temporal resolution from the Last Glacial Maximum to the present, using an approach based on the quasi-constant altitudinal difference between the tree line altitude (TLA) and the equilibrium line altitude (ELA) of glaciers.
Data sets of current tree line position and ELA were integrated and five different scenarios of tree line shifts were simulated at a millennial scale. The model was parameterized using a dense palynological data set (964 time points, representing 121 pollen sampling sites). The simulated tree line fluctuations were compared with those of the boreal forest reconstructed using field data from the southern Alps.
The reconstructed evolution of TLA yielded good results for the interval of tree line formation (correct assignment rate: lower limit = 98.29%; upper limit = 94.29%) and the best-fitting scenario within each millennium tree line (combined scenarios: AUC ± 2 SD = 0.877 ± 0.047; Kappa ± 2 SD = 0.651 ± 0.100). There was also strong agreement between the simulated and the reconstructed tree line fluctuations for both the timing and magnitude of tree line shift.
Although all the analyses support the hypothesis of a quasi-constant difference between TLA and ELA, we found a major relative upward shift of the tree line position within the interval of tree line formation, probably due to the increase in both precipitation and atmospheric CO2 concentration since the onset of the Bølling–Allerød Interstadial. Palaeodistribution maps may be useful for drawing inferences about the biogeography of single temperate and warm temperate species or for recolonization simulations; however, model-based inferences will need to account for several variables, including local climate variability, fire and herbivore disturbance, and lack of complete spatial association between modelled forests and species of interest.