Previous analyses of the regional distribution of forest plants have revealed that some species have a biased distribution towards either ancient or recent forest patches. It has been therefore proposed to classify forest plant species according to their affinity for ancient or recent patches. In this contribution, we aimed at providing a dynamical perspective on this static concept of ancient/recent forest species. We developed an inference method to estimate the metapopulation dynamics of forest plants based on landscape history and static (and incomplete) biodiversity data accounting for false absence records. We assessed the power of this novel inference method using simulated metapopulation and landscape dynamics. We finally applied this method to occupancy data of 174 and 121 forest species in two networks of 9208 and 9201 forest patches respectively. These patch networks are located in two French regions, with corresponding landscape historical data based on two forest maps conceived in 1840 and 2000. Our analyses revealed that i) the inference method enables to satisfactorily infer three parameters of the metapopulation model: the rates of colonization, extinction and decay of colonization with distance; ii) this three parameter metapopulation model is sufficient to reproduce the occurrence spatial structure of 72% and 61% of the investigated species in the two regions; iii) forest plant species are distributed along a spectrum of turn-over speed, in which small (large) colonization capacity are associated with small (large) propensity to extinction; iv) species position along this spectrum can be partially predicted from plant functional traits; and v) species metapopulation dynamics are driven by the interaction between landscape and species intrinsic properties, suggesting that ancient/recent forest species lists are unlikely to be easily extrapolated to novel landscapes.