Predicting long-term development of abandoned subalpine conifer forests in the Swiss National Park

In the past 35 years, various kinds of dynamic models have been used to study vegetation development during primary or secondary succession. Typically, one specific model or models with the same conceptual background were employed. It remains largely unknown to what extent such model-based findings, e.g., on the speed of succession, depend on the specific model approach. To address this issue, we estimated the time elapsing during secondary succession in subalpine conifer forests of the Swiss National Park using three models of different conceptual background: (i) a forest gap model, (ii) a Markov chain model, and (iii) a minimum spanning tree model. Starting from a 95- to 125-year-old mountain pine (Pinus montana Miller) forest, all three models predicted a similar successional development. Even though the forest gap model and the Markov chain model are based on totally different approaches and were calibrated using different data sets, they both forecasted that it would take 500-550 years to reach a late-successional forest stage. The minimum spanning tree model, which only reveals a certain number of time steps yielding a minimum time estimate, showed a development of tree density (stems/ha) that was similar to the results of the forest gap model, but a strict quantitative comparison is not feasible. Our study shows that modeling forest development using three different approaches is quite powerful to obtain a robust estimate of the speed of forest succession. In our case, this estimate is higher than what has been suggested in previous studies that investigated secondary forest succession. The use of several approaches allows for a more comprehensive analysis in terms of variables covered (e.g., relative forest cover in the Markov approach vs. stand-scale species composition in the forest gap model). We recommend that in studies focusing on the speed of succession, several models should be employed simultaneously to identify inconsistencies in our knowledge and to increase confidence in the results. (C) 2009 Elsevier B.V. All rights reserved.